From 7dc4965f34e37b37f4fab69859fcce6476f87811 Mon Sep 17 00:00:00 2001 From: Reynold Xin Date: Thu, 5 Feb 2015 23:02:40 -0800 Subject: [PATCH 01/53] [SPARK-5639][SQL] Support DataFrame.renameColumn. Author: Reynold Xin Closes #4410 from rxin/df-renameCol and squashes the following commits: a6a796e [Reynold Xin] [SPARK-5639][SQL] Support DataFrame.renameColumn. --- .../org/apache/spark/sql/DataFrame.scala | 9 +++++++- .../org/apache/spark/sql/DataFrameImpl.scala | 8 +++++++ .../apache/spark/sql/IncomputableColumn.scala | 2 ++ .../org/apache/spark/sql/DataFrameSuite.scala | 21 +++++++++++++++++++ 4 files changed, 39 insertions(+), 1 deletion(-) diff --git a/sql/core/src/main/scala/org/apache/spark/sql/DataFrame.scala b/sql/core/src/main/scala/org/apache/spark/sql/DataFrame.scala index 92e04ce17c2e7..8ad6526f872e5 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/DataFrame.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/DataFrame.scala @@ -36,7 +36,8 @@ private[sql] object DataFrame { /** - * A collection of rows that have the same columns. + * :: Experimental :: + * A distributed collection of data organized into named columns. * * A [[DataFrame]] is equivalent to a relational table in Spark SQL, and can be created using * various functions in [[SQLContext]]. @@ -72,6 +73,7 @@ private[sql] object DataFrame { * }}} */ // TODO: Improve documentation. +@Experimental trait DataFrame extends RDDApi[Row] { val sqlContext: SQLContext @@ -425,6 +427,11 @@ trait DataFrame extends RDDApi[Row] { */ def addColumn(colName: String, col: Column): DataFrame + /** + * Returns a new [[DataFrame]] with a column renamed. + */ + def renameColumn(existingName: String, newName: String): DataFrame + /** * Returns the first `n` rows. */ diff --git a/sql/core/src/main/scala/org/apache/spark/sql/DataFrameImpl.scala b/sql/core/src/main/scala/org/apache/spark/sql/DataFrameImpl.scala index 4911443dd6dde..789bcf6184b3e 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/DataFrameImpl.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/DataFrameImpl.scala @@ -238,6 +238,14 @@ private[sql] class DataFrameImpl protected[sql]( select(Column("*"), col.as(colName)) } + override def renameColumn(existingName: String, newName: String): DataFrame = { + val colNames = schema.map { field => + val name = field.name + if (name == existingName) Column(name).as(newName) else Column(name) + } + select(colNames :_*) + } + override def head(n: Int): Array[Row] = limit(n).collect() override def head(): Row = head(1).head diff --git a/sql/core/src/main/scala/org/apache/spark/sql/IncomputableColumn.scala b/sql/core/src/main/scala/org/apache/spark/sql/IncomputableColumn.scala index fedd7f06ef50a..6043fb4dee01d 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/IncomputableColumn.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/IncomputableColumn.scala @@ -108,6 +108,8 @@ private[sql] class IncomputableColumn(protected[sql] val expr: Expression) exten override def addColumn(colName: String, col: Column): DataFrame = err() + override def renameColumn(existingName: String, newName: String): DataFrame = err() + override def head(n: Int): Array[Row] = err() override def head(): Row = err() diff --git a/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala index 77fd3165f151f..5aa3db720c886 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala @@ -313,6 +313,27 @@ class DataFrameSuite extends QueryTest { ) } + test("addColumn") { + val df = testData.toDataFrame.addColumn("newCol", col("key") + 1) + checkAnswer( + df, + testData.collect().map { case Row(key: Int, value: String) => + Row(key, value, key + 1) + }.toSeq) + assert(df.schema.map(_.name).toSeq === Seq("key", "value", "newCol")) + } + + test("renameColumn") { + val df = testData.toDataFrame.addColumn("newCol", col("key") + 1) + .renameColumn("value", "valueRenamed") + checkAnswer( + df, + testData.collect().map { case Row(key: Int, value: String) => + Row(key, value, key + 1) + }.toSeq) + assert(df.schema.map(_.name).toSeq === Seq("key", "valueRenamed", "newCol")) + } + test("apply on query results (SPARK-5462)") { val df = testData.sqlContext.sql("select key from testData") checkAnswer(df("key"), testData.select('key).collect().toSeq) From 6b88825a25a0a072c13bbcc57bbfdb102a3f133d Mon Sep 17 00:00:00 2001 From: Xiangrui Meng Date: Thu, 5 Feb 2015 23:32:09 -0800 Subject: [PATCH 02/53] [SPARK-5604][MLLIB] remove checkpointDir from trees This is the second part of SPARK-5604, which removes checkpointDir from tree strategies. Note that this is a break change. I will mention it in the migration guide. Author: Xiangrui Meng Closes #4407 from mengxr/SPARK-5604-1 and squashes the following commits: 13a276d [Xiangrui Meng] remove checkpointDir from trees --- .../spark/examples/mllib/DecisionTreeRunner.scala | 3 ++- .../org/apache/spark/mllib/tree/RandomForest.scala | 1 - .../spark/mllib/tree/configuration/Strategy.scala | 10 ++++------ .../apache/spark/mllib/tree/impl/NodeIdCache.scala | 12 ------------ 4 files changed, 6 insertions(+), 20 deletions(-) diff --git a/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala b/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala index 205d80dd02682..262fd2c9611d0 100644 --- a/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala +++ b/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala @@ -272,6 +272,8 @@ object DecisionTreeRunner { case Variance => impurity.Variance } + params.checkpointDir.foreach(sc.setCheckpointDir) + val strategy = new Strategy( algo = params.algo, @@ -282,7 +284,6 @@ object DecisionTreeRunner { minInstancesPerNode = params.minInstancesPerNode, minInfoGain = params.minInfoGain, useNodeIdCache = params.useNodeIdCache, - checkpointDir = params.checkpointDir, checkpointInterval = params.checkpointInterval) if (params.numTrees == 1) { val startTime = System.nanoTime() diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala index 45b0154c5e4cb..db01f2e229e5a 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala @@ -204,7 +204,6 @@ private class RandomForest ( Some(NodeIdCache.init( data = baggedInput, numTrees = numTrees, - checkpointDir = strategy.checkpointDir, checkpointInterval = strategy.checkpointInterval, initVal = 1)) } else { diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala index 3308adb6752ff..8d5c36da32bdb 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala @@ -62,11 +62,10 @@ import org.apache.spark.mllib.tree.configuration.QuantileStrategy._ * @param subsamplingRate Fraction of the training data used for learning decision tree. * @param useNodeIdCache If this is true, instead of passing trees to executors, the algorithm will * maintain a separate RDD of node Id cache for each row. - * @param checkpointDir If the node Id cache is used, it will help to checkpoint - * the node Id cache periodically. This is the checkpoint directory - * to be used for the node Id cache. * @param checkpointInterval How often to checkpoint when the node Id cache gets updated. - * E.g. 10 means that the cache will get checkpointed every 10 updates. + * E.g. 10 means that the cache will get checkpointed every 10 updates. If + * the checkpoint directory is not set in + * [[org.apache.spark.SparkContext]], this setting is ignored. */ @Experimental class Strategy ( @@ -82,7 +81,6 @@ class Strategy ( @BeanProperty var maxMemoryInMB: Int = 256, @BeanProperty var subsamplingRate: Double = 1, @BeanProperty var useNodeIdCache: Boolean = false, - @BeanProperty var checkpointDir: Option[String] = None, @BeanProperty var checkpointInterval: Int = 10) extends Serializable { def isMulticlassClassification = @@ -165,7 +163,7 @@ class Strategy ( def copy: Strategy = { new Strategy(algo, impurity, maxDepth, numClasses, maxBins, quantileCalculationStrategy, categoricalFeaturesInfo, minInstancesPerNode, minInfoGain, - maxMemoryInMB, subsamplingRate, useNodeIdCache, checkpointDir, checkpointInterval) + maxMemoryInMB, subsamplingRate, useNodeIdCache, checkpointInterval) } } diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala index 83011b48b7d9b..bdd0f576b048d 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala @@ -71,15 +71,12 @@ private[tree] case class NodeIndexUpdater( * The nodeIdsForInstances RDD needs to be updated at each iteration. * @param nodeIdsForInstances The initial values in the cache * (should be an Array of all 1's (meaning the root nodes)). - * @param checkpointDir The checkpoint directory where - * the checkpointed files will be stored. * @param checkpointInterval The checkpointing interval * (how often should the cache be checkpointed.). */ @DeveloperApi private[tree] class NodeIdCache( var nodeIdsForInstances: RDD[Array[Int]], - val checkpointDir: Option[String], val checkpointInterval: Int) { // Keep a reference to a previous node Ids for instances. @@ -91,12 +88,6 @@ private[tree] class NodeIdCache( private val checkpointQueue = mutable.Queue[RDD[Array[Int]]]() private var rddUpdateCount = 0 - // If a checkpoint directory is given, and there's no prior checkpoint directory, - // then set the checkpoint directory with the given one. - if (checkpointDir.nonEmpty && nodeIdsForInstances.sparkContext.getCheckpointDir.isEmpty) { - nodeIdsForInstances.sparkContext.setCheckpointDir(checkpointDir.get) - } - /** * Update the node index values in the cache. * This updates the RDD and its lineage. @@ -184,7 +175,6 @@ private[tree] object NodeIdCache { * Initialize the node Id cache with initial node Id values. * @param data The RDD of training rows. * @param numTrees The number of trees that we want to create cache for. - * @param checkpointDir The checkpoint directory where the checkpointed files will be stored. * @param checkpointInterval The checkpointing interval * (how often should the cache be checkpointed.). * @param initVal The initial values in the cache. @@ -193,12 +183,10 @@ private[tree] object NodeIdCache { def init( data: RDD[BaggedPoint[TreePoint]], numTrees: Int, - checkpointDir: Option[String], checkpointInterval: Int, initVal: Int = 1): NodeIdCache = { new NodeIdCache( data.map(_ => Array.fill[Int](numTrees)(initVal)), - checkpointDir, checkpointInterval) } } From dc0c4490a12ecedd8ca5a1bb256c7ccbdf0be04f Mon Sep 17 00:00:00 2001 From: "Joseph K. Bradley" Date: Thu, 5 Feb 2015 23:43:47 -0800 Subject: [PATCH 03/53] [SPARK-4789] [SPARK-4942] [SPARK-5031] [mllib] Standardize ML Prediction APIs MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This is part (1a) of the updates from the design doc in [https://docs.google.com/document/d/1BH9el33kBX8JiDdgUJXdLW14CA2qhTCWIG46eXZVoJs] **UPDATE**: Most of the APIs are being kept private[spark] to allow further discussion. Here is a list of changes which are public: * new output columns: rawPrediction, probabilities * The “score” column is now called “rawPrediction” * Classifiers now provide numClasses * Params.get and .set are now protected instead of private[ml]. * ParamMap now has a size method. * new classes: LinearRegression, LinearRegressionModel * LogisticRegression now has an intercept. ### Sketch of APIs (most of which are private[spark] for now) Abstract classes for learning algorithms (+ corresponding Model abstractions): * Classifier (+ ClassificationModel) * ProbabilisticClassifier (+ ProbabilisticClassificationModel) * Regressor (+ RegressionModel) * Predictor (+ PredictionModel) * *For all of these*: * There is no strongly typed training-time API. * There is a strongly typed test-time (prediction) API which helps developers implement new algorithms. Concrete classes: learning algorithms * LinearRegression * LogisticRegression (updated to use new abstract classes) * Also, removed "score" in favor of "probability" output column. Changed BinaryClassificationEvaluator to match. (SPARK-5031) Other updates: * params.scala: Changed Params.set/get to be protected instead of private[ml] * This was needed for the example of defining a class from outside of the MLlib namespace. * VectorUDT: Will later change from private[spark] to public. * This is needed for outside users to write their own validateAndTransformSchema() methods using vectors. * Also, added equals() method.f * SPARK-4942 : ML Transformers should allow output cols to be turned on,off * Update validateAndTransformSchema * Update transform * (Updated examples, test suites according to other changes) New examples: * DeveloperApiExample.scala (example of defining algorithm from outside of the MLlib namespace) * Added Java version too Test Suites: * LinearRegressionSuite * LogisticRegressionSuite * + Java versions of above suites CC: mengxr etrain shivaram Author: Joseph K. Bradley Closes #3637 from jkbradley/ml-api-part1 and squashes the following commits: 405bfb8 [Joseph K. Bradley] Last edits based on code review. Small cleanups fec348a [Joseph K. Bradley] Added JavaDeveloperApiExample.java and fixed other issues: Made developer API private[spark] for now. Added constructors Java can understand to specialized Param types. 8316d5e [Joseph K. Bradley] fixes after rebasing on master fc62406 [Joseph K. Bradley] fixed test suites after last commit bcb9549 [Joseph K. Bradley] Fixed issues after rebasing from master (after move from SchemaRDD to DataFrame) 9872424 [Joseph K. Bradley] fixed JavaLinearRegressionSuite.java Java sql api f542997 [Joseph K. Bradley] Added MIMA excludes for VectorUDT (now public), and added DeveloperApi annotation to it 216d199 [Joseph K. Bradley] fixed after sql datatypes PR got merged f549e34 [Joseph K. Bradley] Updates based on code review. Major ones are: * Created weakly typed Predictor.train() method which is called by fit() so that developers do not have to call schema validation or copy parameters. * Made Predictor.featuresDataType have a default value of VectorUDT. * NOTE: This could be dangerous since the FeaturesType type parameter cannot have a default value. 343e7bd [Joseph K. Bradley] added blanket mima exclude for ml package 82f340b [Joseph K. Bradley] Fixed bug in LogisticRegression (introduced in this PR). Fixed Java suites 0a16da9 [Joseph K. Bradley] Fixed Linear/Logistic RegressionSuites c3c8da5 [Joseph K. Bradley] small cleanup 934f97b [Joseph K. Bradley] Fixed bugs from previous commit. 1c61723 [Joseph K. Bradley] * Made ProbabilisticClassificationModel into a subclass of ClassificationModel. Also introduced ProbabilisticClassifier. * This was to support output column “probabilityCol” in transform(). 4e2f711 [Joseph K. Bradley] rat fix bc654e1 [Joseph K. Bradley] Added spark.ml LinearRegressionSuite 8d13233 [Joseph K. Bradley] Added methods: * Classifier: batch predictRaw() * Predictor: train() without paramMap ProbabilisticClassificationModel.predictProbabilities() * Java versions of all above batch methods + others 1680905 [Joseph K. Bradley] Added JavaLabeledPointSuite.java for spark.ml, and added constructor to LabeledPoint which defaults weight to 1.0 adbe50a [Joseph K. Bradley] * fixed LinearRegression train() to use embedded paramMap * added Predictor.predict(RDD[Vector]) method * updated Linear/LogisticRegressionSuites 58802e3 [Joseph K. Bradley] added train() to Predictor subclasses which does not take a ParamMap. 57d54ab [Joseph K. Bradley] * Changed semantics of Predictor.train() to merge the given paramMap with the embedded paramMap. * remove threshold_internal from logreg * Added Predictor.copy() * Extended LogisticRegressionSuite e433872 [Joseph K. Bradley] Updated docs. Added LabeledPointSuite to spark.ml 54b7b31 [Joseph K. Bradley] Fixed issue with logreg threshold being set correctly 0617d61 [Joseph K. Bradley] Fixed bug from last commit (sorting paramMap by parameter names in toString). Fixed bug in persisting logreg data. Added threshold_internal to logreg for faster test-time prediction (avoiding map lookup). 601e792 [Joseph K. Bradley] Modified ParamMap to sort parameters in toString. Cleaned up classes in class hierarchy, before implementing tests and examples. d705e87 [Joseph K. Bradley] Added LinearRegression and Regressor back from ml-api branch 52f4fde [Joseph K. Bradley] removing everything except for simple class hierarchy for classification d35bb5d [Joseph K. Bradley] fixed compilation issues, but have not added tests yet bfade12 [Joseph K. Bradley] Added lots of classes for new ML API: --- .../ml/JavaCrossValidatorExample.java | 6 +- .../examples/ml/JavaDeveloperApiExample.java | 217 ++++++++++++++++ .../examples/ml/JavaSimpleParamsExample.java | 10 +- .../JavaSimpleTextClassificationPipeline.java | 4 +- .../examples/ml/CrossValidatorExample.scala | 7 +- .../examples/ml/DeveloperApiExample.scala | 184 ++++++++++++++ .../examples/ml/SimpleParamsExample.scala | 16 +- .../ml/SimpleTextClassificationPipeline.scala | 7 +- .../scala/org/apache/spark/ml/Estimator.scala | 9 +- .../spark/ml/classification/Classifier.scala | 206 +++++++++++++++ .../classification/LogisticRegression.scala | 212 ++++++++++------ .../ProbabilisticClassifier.scala | 147 +++++++++++ .../BinaryClassificationEvaluator.scala | 24 +- .../apache/spark/ml/feature/Tokenizer.scala | 4 +- .../spark/ml/impl/estimator/Predictor.scala | 234 ++++++++++++++++++ .../org/apache/spark/ml/param/params.scala | 68 ++++- .../apache/spark/ml/param/sharedParams.scala | 28 ++- .../ml/regression/LinearRegression.scala | 96 +++++++ .../spark/ml/regression/Regressor.scala | 78 ++++++ .../apache/spark/mllib/linalg/Vectors.scala | 13 + .../apache/spark/ml/JavaPipelineSuite.java | 2 +- .../JavaLogisticRegressionSuite.java | 91 ++++++- .../regression/JavaLinearRegressionSuite.java | 89 +++++++ .../LogisticRegressionSuite.scala | 86 ++++++- .../ml/regression/LinearRegressionSuite.scala | 65 +++++ project/MimaExcludes.scala | 6 + 26 files changed, 1753 insertions(+), 156 deletions(-) create mode 100644 examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java create mode 100644 examples/src/main/scala/org/apache/spark/examples/ml/DeveloperApiExample.scala create mode 100644 mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala create mode 100644 mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala create mode 100644 mllib/src/main/scala/org/apache/spark/ml/impl/estimator/Predictor.scala create mode 100644 mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala create mode 100644 mllib/src/main/scala/org/apache/spark/ml/regression/Regressor.scala create mode 100644 mllib/src/test/java/org/apache/spark/ml/regression/JavaLinearRegressionSuite.java create mode 100644 mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaCrossValidatorExample.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaCrossValidatorExample.java index 0fbee6e433608..5041e0b6d34b0 100644 --- a/examples/src/main/java/org/apache/spark/examples/ml/JavaCrossValidatorExample.java +++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaCrossValidatorExample.java @@ -116,10 +116,12 @@ public static void main(String[] args) { // Make predictions on test documents. cvModel uses the best model found (lrModel). cvModel.transform(test).registerTempTable("prediction"); - DataFrame predictions = jsql.sql("SELECT id, text, score, prediction FROM prediction"); + DataFrame predictions = jsql.sql("SELECT id, text, probability, prediction FROM prediction"); for (Row r: predictions.collect()) { - System.out.println("(" + r.get(0) + ", " + r.get(1) + ") --> score=" + r.get(2) + System.out.println("(" + r.get(0) + ", " + r.get(1) + ") --> prob=" + r.get(2) + ", prediction=" + r.get(3)); } + + jsc.stop(); } } diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java new file mode 100644 index 0000000000000..42d4d7d0bef26 --- /dev/null +++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java @@ -0,0 +1,217 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.examples.ml; + +import java.util.List; + +import com.google.common.collect.Lists; + +import org.apache.spark.SparkConf; +import org.apache.spark.api.java.JavaRDD; +import org.apache.spark.api.java.JavaSparkContext; +import org.apache.spark.ml.classification.Classifier; +import org.apache.spark.ml.classification.ClassificationModel; +import org.apache.spark.ml.param.IntParam; +import org.apache.spark.ml.param.ParamMap; +import org.apache.spark.ml.param.Params; +import org.apache.spark.ml.param.Params$; +import org.apache.spark.mllib.linalg.BLAS; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.mllib.regression.LabeledPoint; +import org.apache.spark.sql.DataFrame; +import org.apache.spark.sql.Row; +import org.apache.spark.sql.SQLContext; + + +/** + * A simple example demonstrating how to write your own learning algorithm using Estimator, + * Transformer, and other abstractions. + * This mimics {@link org.apache.spark.ml.classification.LogisticRegression}. + * + * Run with + * 
+ * bin/run-example ml.JavaDeveloperApiExample
+ *
+ */ +public class JavaDeveloperApiExample { + + public static void main(String[] args) throws Exception { + SparkConf conf = new SparkConf().setAppName("JavaDeveloperApiExample"); + JavaSparkContext jsc = new JavaSparkContext(conf); + SQLContext jsql = new SQLContext(jsc); + + // Prepare training data. + List localTraining = Lists.newArrayList( + new LabeledPoint(1.0, Vectors.dense(0.0, 1.1, 0.1)), + new LabeledPoint(0.0, Vectors.dense(2.0, 1.0, -1.0)), + new LabeledPoint(0.0, Vectors.dense(2.0, 1.3, 1.0)), + new LabeledPoint(1.0, Vectors.dense(0.0, 1.2, -0.5))); + DataFrame training = jsql.applySchema(jsc.parallelize(localTraining), LabeledPoint.class); + + // Create a LogisticRegression instance. This instance is an Estimator. + MyJavaLogisticRegression lr = new MyJavaLogisticRegression(); + // Print out the parameters, documentation, and any default values. + System.out.println("MyJavaLogisticRegression parameters:\n" + lr.explainParams() + "\n"); + + // We may set parameters using setter methods. + lr.setMaxIter(10); + + // Learn a LogisticRegression model. This uses the parameters stored in lr. + MyJavaLogisticRegressionModel model = lr.fit(training); + + // Prepare test data. + List localTest = Lists.newArrayList( + new LabeledPoint(1.0, Vectors.dense(-1.0, 1.5, 1.3)), + new LabeledPoint(0.0, Vectors.dense(3.0, 2.0, -0.1)), + new LabeledPoint(1.0, Vectors.dense(0.0, 2.2, -1.5))); + DataFrame test = jsql.applySchema(jsc.parallelize(localTest), LabeledPoint.class); + + // Make predictions on test documents. cvModel uses the best model found (lrModel). + DataFrame results = model.transform(test); + double sumPredictions = 0; + for (Row r : results.select("features", "label", "prediction").collect()) { + sumPredictions += r.getDouble(2); + } + if (sumPredictions != 0.0) { + throw new Exception("MyJavaLogisticRegression predicted something other than 0," + + " even though all weights are 0!"); + } + + jsc.stop(); + } +} + +/** + * Example of defining a type of {@link Classifier}. + * + * NOTE: This is private since it is an example. In practice, you may not want it to be private. + */ +class MyJavaLogisticRegression + extends Classifier + implements Params { + + /** + * Param for max number of iterations + *

+ * NOTE: The usual way to add a parameter to a model or algorithm is to include: + * - val myParamName: ParamType + * - def getMyParamName + * - def setMyParamName + */ + IntParam maxIter = new IntParam(this, "maxIter", "max number of iterations"); + + int getMaxIter() { return (int)get(maxIter); } + + public MyJavaLogisticRegression() { + setMaxIter(100); + } + + // The parameter setter is in this class since it should return type MyJavaLogisticRegression. + MyJavaLogisticRegression setMaxIter(int value) { + return (MyJavaLogisticRegression)set(maxIter, value); + } + + // This method is used by fit(). + // In Java, we have to make it public since Java does not understand Scala's protected modifier. + public MyJavaLogisticRegressionModel train(DataFrame dataset, ParamMap paramMap) { + // Extract columns from data using helper method. + JavaRDD oldDataset = extractLabeledPoints(dataset, paramMap).toJavaRDD(); + + // Do learning to estimate the weight vector. + int numFeatures = oldDataset.take(1).get(0).features().size(); + Vector weights = Vectors.zeros(numFeatures); // Learning would happen here. + + // Create a model, and return it. + return new MyJavaLogisticRegressionModel(this, paramMap, weights); + } +} + +/** + * Example of defining a type of {@link ClassificationModel}. + * + * NOTE: This is private since it is an example. In practice, you may not want it to be private. + */ +class MyJavaLogisticRegressionModel + extends ClassificationModel implements Params { + + private MyJavaLogisticRegression parent_; + public MyJavaLogisticRegression parent() { return parent_; } + + private ParamMap fittingParamMap_; + public ParamMap fittingParamMap() { return fittingParamMap_; } + + private Vector weights_; + public Vector weights() { return weights_; } + + public MyJavaLogisticRegressionModel( + MyJavaLogisticRegression parent_, + ParamMap fittingParamMap_, + Vector weights_) { + this.parent_ = parent_; + this.fittingParamMap_ = fittingParamMap_; + this.weights_ = weights_; + } + + // This uses the default implementation of transform(), which reads column "features" and outputs + // columns "prediction" and "rawPrediction." + + // This uses the default implementation of predict(), which chooses the label corresponding to + // the maximum value returned by [[predictRaw()]]. + + /** + * Raw prediction for each possible label. + * The meaning of a "raw" prediction may vary between algorithms, but it intuitively gives + * a measure of confidence in each possible label (where larger = more confident). + * This internal method is used to implement [[transform()]] and output [[rawPredictionCol]]. + * + * @return vector where element i is the raw prediction for label i. + * This raw prediction may be any real number, where a larger value indicates greater + * confidence for that label. + * + * In Java, we have to make this method public since Java does not understand Scala's protected + * modifier. + */ + public Vector predictRaw(Vector features) { + double margin = BLAS.dot(features, weights_); + // There are 2 classes (binary classification), so we return a length-2 vector, + // where index i corresponds to class i (i = 0, 1). + return Vectors.dense(-margin, margin); + } + + /** + * Number of classes the label can take. 2 indicates binary classification. + */ + public int numClasses() { return 2; } + + /** + * Create a copy of the model. + * The copy is shallow, except for the embedded paramMap, which gets a deep copy. + *

+ * This is used for the defaul implementation of [[transform()]]. + * + * In Java, we have to make this method public since Java does not understand Scala's protected + * modifier. + */ + public MyJavaLogisticRegressionModel copy() { + MyJavaLogisticRegressionModel m = + new MyJavaLogisticRegressionModel(parent_, fittingParamMap_, weights_); + Params$.MODULE$.inheritValues(this.paramMap(), this, m); + return m; + } +} diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleParamsExample.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleParamsExample.java index eaaa344be49c8..cc69e6315fdda 100644 --- a/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleParamsExample.java +++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleParamsExample.java @@ -81,7 +81,7 @@ public static void main(String[] args) { // One can also combine ParamMaps. ParamMap paramMap2 = new ParamMap(); - paramMap2.put(lr.scoreCol().w("probability")); // Change output column name + paramMap2.put(lr.probabilityCol().w("myProbability")); // Change output column name ParamMap paramMapCombined = paramMap.$plus$plus(paramMap2); // Now learn a new model using the paramMapCombined parameters. @@ -98,14 +98,16 @@ public static void main(String[] args) { // Make predictions on test documents using the Transformer.transform() method. // LogisticRegression.transform will only use the 'features' column. - // Note that model2.transform() outputs a 'probability' column instead of the usual 'score' - // column since we renamed the lr.scoreCol parameter previously. + // Note that model2.transform() outputs a 'myProbability' column instead of the usual + // 'probability' column since we renamed the lr.probabilityCol parameter previously. model2.transform(test).registerTempTable("results"); DataFrame results = - jsql.sql("SELECT features, label, probability, prediction FROM results"); + jsql.sql("SELECT features, label, myProbability, prediction FROM results"); for (Row r: results.collect()) { System.out.println("(" + r.get(0) + ", " + r.get(1) + ") -> prob=" + r.get(2) + ", prediction=" + r.get(3)); } + + jsc.stop(); } } diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleTextClassificationPipeline.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleTextClassificationPipeline.java index 82d665a3e1386..d929f1ad2014a 100644 --- a/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleTextClassificationPipeline.java +++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleTextClassificationPipeline.java @@ -85,8 +85,10 @@ public static void main(String[] args) { model.transform(test).registerTempTable("prediction"); DataFrame predictions = jsql.sql("SELECT id, text, score, prediction FROM prediction"); for (Row r: predictions.collect()) { - System.out.println("(" + r.get(0) + ", " + r.get(1) + ") --> score=" + r.get(2) + System.out.println("(" + r.get(0) + ", " + r.get(1) + ") --> prob=" + r.get(2) + ", prediction=" + r.get(3)); } + + jsc.stop(); } } diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/CrossValidatorExample.scala b/examples/src/main/scala/org/apache/spark/examples/ml/CrossValidatorExample.scala index b6c30a007d88f..a2893f78e0fec 100644 --- a/examples/src/main/scala/org/apache/spark/examples/ml/CrossValidatorExample.scala +++ b/examples/src/main/scala/org/apache/spark/examples/ml/CrossValidatorExample.scala @@ -23,6 +23,7 @@ import org.apache.spark.ml.classification.LogisticRegression import org.apache.spark.ml.evaluation.BinaryClassificationEvaluator import org.apache.spark.ml.feature.{HashingTF, Tokenizer} import org.apache.spark.ml.tuning.{ParamGridBuilder, CrossValidator} +import org.apache.spark.mllib.linalg.Vector import org.apache.spark.sql.{Row, SQLContext} /** @@ -100,10 +101,10 @@ object CrossValidatorExample { // Make predictions on test documents. cvModel uses the best model found (lrModel). cvModel.transform(test) - .select("id", "text", "score", "prediction") + .select("id", "text", "probability", "prediction") .collect() - .foreach { case Row(id: Long, text: String, score: Double, prediction: Double) => - println("(" + id + ", " + text + ") --> score=" + score + ", prediction=" + prediction) + .foreach { case Row(id: Long, text: String, prob: Vector, prediction: Double) => + println(s"($id, $text) --> prob=$prob, prediction=$prediction") } sc.stop() diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/DeveloperApiExample.scala b/examples/src/main/scala/org/apache/spark/examples/ml/DeveloperApiExample.scala new file mode 100644 index 0000000000000..aed44238939c7 --- /dev/null +++ b/examples/src/main/scala/org/apache/spark/examples/ml/DeveloperApiExample.scala @@ -0,0 +1,184 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.examples.ml + +import org.apache.spark.{SparkConf, SparkContext} +import org.apache.spark.ml.classification.{Classifier, ClassifierParams, ClassificationModel} +import org.apache.spark.ml.param.{Params, IntParam, ParamMap} +import org.apache.spark.mllib.linalg.{BLAS, Vector, Vectors} +import org.apache.spark.mllib.regression.LabeledPoint +import org.apache.spark.sql.{DataFrame, Row, SQLContext} + + +/** + * A simple example demonstrating how to write your own learning algorithm using Estimator, + * Transformer, and other abstractions. + * This mimics [[org.apache.spark.ml.classification.LogisticRegression]]. + * Run with + * {{{ + * bin/run-example ml.DeveloperApiExample + * }}} + */ +object DeveloperApiExample { + + def main(args: Array[String]) { + val conf = new SparkConf().setAppName("DeveloperApiExample") + val sc = new SparkContext(conf) + val sqlContext = new SQLContext(sc) + import sqlContext.implicits._ + + // Prepare training data. + val training = sc.parallelize(Seq( + LabeledPoint(1.0, Vectors.dense(0.0, 1.1, 0.1)), + LabeledPoint(0.0, Vectors.dense(2.0, 1.0, -1.0)), + LabeledPoint(0.0, Vectors.dense(2.0, 1.3, 1.0)), + LabeledPoint(1.0, Vectors.dense(0.0, 1.2, -0.5)))) + + // Create a LogisticRegression instance. This instance is an Estimator. + val lr = new MyLogisticRegression() + // Print out the parameters, documentation, and any default values. + println("MyLogisticRegression parameters:\n" + lr.explainParams() + "\n") + + // We may set parameters using setter methods. + lr.setMaxIter(10) + + // Learn a LogisticRegression model. This uses the parameters stored in lr. + val model = lr.fit(training) + + // Prepare test data. + val test = sc.parallelize(Seq( + LabeledPoint(1.0, Vectors.dense(-1.0, 1.5, 1.3)), + LabeledPoint(0.0, Vectors.dense(3.0, 2.0, -0.1)), + LabeledPoint(1.0, Vectors.dense(0.0, 2.2, -1.5)))) + + // Make predictions on test data. + val sumPredictions: Double = model.transform(test) + .select("features", "label", "prediction") + .collect() + .map { case Row(features: Vector, label: Double, prediction: Double) => + prediction + }.sum + assert(sumPredictions == 0.0, + "MyLogisticRegression predicted something other than 0, even though all weights are 0!") + + sc.stop() + } +} + +/** + * Example of defining a parameter trait for a user-defined type of [[Classifier]]. + * + * NOTE: This is private since it is an example. In practice, you may not want it to be private. + */ +private trait MyLogisticRegressionParams extends ClassifierParams { + + /** + * Param for max number of iterations + * + * NOTE: The usual way to add a parameter to a model or algorithm is to include: + * - val myParamName: ParamType + * - def getMyParamName + * - def setMyParamName + * Here, we have a trait to be mixed in with the Estimator and Model (MyLogisticRegression + * and MyLogisticRegressionModel). We place the setter (setMaxIter) method in the Estimator + * class since the maxIter parameter is only used during training (not in the Model). + */ + val maxIter: IntParam = new IntParam(this, "maxIter", "max number of iterations") + def getMaxIter: Int = get(maxIter) +} + +/** + * Example of defining a type of [[Classifier]]. + * + * NOTE: This is private since it is an example. In practice, you may not want it to be private. + */ +private class MyLogisticRegression + extends Classifier[Vector, MyLogisticRegression, MyLogisticRegressionModel] + with MyLogisticRegressionParams { + + setMaxIter(100) // Initialize + + // The parameter setter is in this class since it should return type MyLogisticRegression. + def setMaxIter(value: Int): this.type = set(maxIter, value) + + // This method is used by fit() + override protected def train( + dataset: DataFrame, + paramMap: ParamMap): MyLogisticRegressionModel = { + // Extract columns from data using helper method. + val oldDataset = extractLabeledPoints(dataset, paramMap) + + // Do learning to estimate the weight vector. + val numFeatures = oldDataset.take(1)(0).features.size + val weights = Vectors.zeros(numFeatures) // Learning would happen here. + + // Create a model, and return it. + new MyLogisticRegressionModel(this, paramMap, weights) + } +} + +/** + * Example of defining a type of [[ClassificationModel]]. + * + * NOTE: This is private since it is an example. In practice, you may not want it to be private. + */ +private class MyLogisticRegressionModel( + override val parent: MyLogisticRegression, + override val fittingParamMap: ParamMap, + val weights: Vector) + extends ClassificationModel[Vector, MyLogisticRegressionModel] + with MyLogisticRegressionParams { + + // This uses the default implementation of transform(), which reads column "features" and outputs + // columns "prediction" and "rawPrediction." + + // This uses the default implementation of predict(), which chooses the label corresponding to + // the maximum value returned by [[predictRaw()]]. + + /** + * Raw prediction for each possible label. + * The meaning of a "raw" prediction may vary between algorithms, but it intuitively gives + * a measure of confidence in each possible label (where larger = more confident). + * This internal method is used to implement [[transform()]] and output [[rawPredictionCol]]. + * + * @return vector where element i is the raw prediction for label i. + * This raw prediction may be any real number, where a larger value indicates greater + * confidence for that label. + */ + override protected def predictRaw(features: Vector): Vector = { + val margin = BLAS.dot(features, weights) + // There are 2 classes (binary classification), so we return a length-2 vector, + // where index i corresponds to class i (i = 0, 1). + Vectors.dense(-margin, margin) + } + + /** Number of classes the label can take. 2 indicates binary classification. */ + override val numClasses: Int = 2 + + /** + * Create a copy of the model. + * The copy is shallow, except for the embedded paramMap, which gets a deep copy. + * + * This is used for the defaul implementation of [[transform()]]. + */ + override protected def copy(): MyLogisticRegressionModel = { + val m = new MyLogisticRegressionModel(parent, fittingParamMap, weights) + Params.inheritValues(this.paramMap, this, m) + m + } +} diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/SimpleParamsExample.scala b/examples/src/main/scala/org/apache/spark/examples/ml/SimpleParamsExample.scala index 4d1530cd1349f..80c9f5ff5781e 100644 --- a/examples/src/main/scala/org/apache/spark/examples/ml/SimpleParamsExample.scala +++ b/examples/src/main/scala/org/apache/spark/examples/ml/SimpleParamsExample.scala @@ -72,7 +72,7 @@ object SimpleParamsExample { paramMap.put(lr.regParam -> 0.1, lr.threshold -> 0.55) // Specify multiple Params. // One can also combine ParamMaps. - val paramMap2 = ParamMap(lr.scoreCol -> "probability") // Change output column name + val paramMap2 = ParamMap(lr.probabilityCol -> "myProbability") // Change output column name val paramMapCombined = paramMap ++ paramMap2 // Now learn a new model using the paramMapCombined parameters. @@ -80,21 +80,21 @@ object SimpleParamsExample { val model2 = lr.fit(training, paramMapCombined) println("Model 2 was fit using parameters: " + model2.fittingParamMap) - // Prepare test documents. + // Prepare test data. val test = sc.parallelize(Seq( LabeledPoint(1.0, Vectors.dense(-1.0, 1.5, 1.3)), LabeledPoint(0.0, Vectors.dense(3.0, 2.0, -0.1)), LabeledPoint(1.0, Vectors.dense(0.0, 2.2, -1.5)))) - // Make predictions on test documents using the Transformer.transform() method. + // Make predictions on test data using the Transformer.transform() method. // LogisticRegression.transform will only use the 'features' column. - // Note that model2.transform() outputs a 'probability' column instead of the usual 'score' - // column since we renamed the lr.scoreCol parameter previously. + // Note that model2.transform() outputs a 'myProbability' column instead of the usual + // 'probability' column since we renamed the lr.probabilityCol parameter previously. model2.transform(test) - .select("features", "label", "probability", "prediction") + .select("features", "label", "myProbability", "prediction") .collect() - .foreach { case Row(features: Vector, label: Double, prob: Double, prediction: Double) => - println("(" + features + ", " + label + ") -> prob=" + prob + ", prediction=" + prediction) + .foreach { case Row(features: Vector, label: Double, prob: Vector, prediction: Double) => + println("($features, $label) -> prob=$prob, prediction=$prediction") } sc.stop() diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/SimpleTextClassificationPipeline.scala b/examples/src/main/scala/org/apache/spark/examples/ml/SimpleTextClassificationPipeline.scala index dbbe01dd5ce8e..968cb292120d8 100644 --- a/examples/src/main/scala/org/apache/spark/examples/ml/SimpleTextClassificationPipeline.scala +++ b/examples/src/main/scala/org/apache/spark/examples/ml/SimpleTextClassificationPipeline.scala @@ -23,6 +23,7 @@ import org.apache.spark.{SparkConf, SparkContext} import org.apache.spark.ml.Pipeline import org.apache.spark.ml.classification.LogisticRegression import org.apache.spark.ml.feature.{HashingTF, Tokenizer} +import org.apache.spark.mllib.linalg.Vector import org.apache.spark.sql.{Row, SQLContext} @BeanInfo @@ -79,10 +80,10 @@ object SimpleTextClassificationPipeline { // Make predictions on test documents. model.transform(test) - .select("id", "text", "score", "prediction") + .select("id", "text", "probability", "prediction") .collect() - .foreach { case Row(id: Long, text: String, score: Double, prediction: Double) => - println("(" + id + ", " + text + ") --> score=" + score + ", prediction=" + prediction) + .foreach { case Row(id: Long, text: String, prob: Vector, prediction: Double) => + println("($id, $text) --> prob=$prob, prediction=$prediction") } sc.stop() diff --git a/mllib/src/main/scala/org/apache/spark/ml/Estimator.scala b/mllib/src/main/scala/org/apache/spark/ml/Estimator.scala index bc3defe968afd..eff7ef925dfbd 100644 --- a/mllib/src/main/scala/org/apache/spark/ml/Estimator.scala +++ b/mllib/src/main/scala/org/apache/spark/ml/Estimator.scala @@ -34,7 +34,8 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage with Params { * Fits a single model to the input data with optional parameters. * * @param dataset input dataset - * @param paramPairs optional list of param pairs (overwrite embedded params) + * @param paramPairs Optional list of param pairs. + * These values override any specified in this Estimator's embedded ParamMap. * @return fitted model */ @varargs @@ -47,7 +48,8 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage with Params { * Fits a single model to the input data with provided parameter map. * * @param dataset input dataset - * @param paramMap parameter map + * @param paramMap Parameter map. + * These values override any specified in this Estimator's embedded ParamMap. * @return fitted model */ def fit(dataset: DataFrame, paramMap: ParamMap): M @@ -58,7 +60,8 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage with Params { * Subclasses could overwrite this to optimize multi-model training. * * @param dataset input dataset - * @param paramMaps an array of parameter maps + * @param paramMaps An array of parameter maps. + * These values override any specified in this Estimator's embedded ParamMap. * @return fitted models, matching the input parameter maps */ def fit(dataset: DataFrame, paramMaps: Array[ParamMap]): Seq[M] = { diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala new file mode 100644 index 0000000000000..1bf8eb4640d11 --- /dev/null +++ b/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala @@ -0,0 +1,206 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.classification + +import org.apache.spark.annotation.{DeveloperApi, AlphaComponent} +import org.apache.spark.ml.impl.estimator.{PredictionModel, Predictor, PredictorParams} +import org.apache.spark.ml.param.{Params, ParamMap, HasRawPredictionCol} +import org.apache.spark.mllib.linalg.{Vector, VectorUDT} +import org.apache.spark.sql.Dsl._ +import org.apache.spark.sql.DataFrame +import org.apache.spark.sql.types.{DataType, DoubleType, StructType} + + +/** + * :: DeveloperApi :: + * Params for classification. + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@DeveloperApi +private[spark] trait ClassifierParams extends PredictorParams + with HasRawPredictionCol { + + override protected def validateAndTransformSchema( + schema: StructType, + paramMap: ParamMap, + fitting: Boolean, + featuresDataType: DataType): StructType = { + val parentSchema = super.validateAndTransformSchema(schema, paramMap, fitting, featuresDataType) + val map = this.paramMap ++ paramMap + addOutputColumn(parentSchema, map(rawPredictionCol), new VectorUDT) + } +} + +/** + * :: AlphaComponent :: + * Single-label binary or multiclass classification. + * Classes are indexed {0, 1, ..., numClasses - 1}. + * + * @tparam FeaturesType Type of input features. E.g., [[Vector]] + * @tparam E Concrete Estimator type + * @tparam M Concrete Model type + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@AlphaComponent +private[spark] abstract class Classifier[ + FeaturesType, + E <: Classifier[FeaturesType, E, M], + M <: ClassificationModel[FeaturesType, M]] + extends Predictor[FeaturesType, E, M] + with ClassifierParams { + + def setRawPredictionCol(value: String): E = + set(rawPredictionCol, value).asInstanceOf[E] + + // TODO: defaultEvaluator (follow-up PR) +} + +/** + * :: AlphaComponent :: + * Model produced by a [[Classifier]]. + * Classes are indexed {0, 1, ..., numClasses - 1}. + * + * @tparam FeaturesType Type of input features. E.g., [[Vector]] + * @tparam M Concrete Model type + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@AlphaComponent +private[spark] +abstract class ClassificationModel[FeaturesType, M <: ClassificationModel[FeaturesType, M]] + extends PredictionModel[FeaturesType, M] with ClassifierParams { + + def setRawPredictionCol(value: String): M = set(rawPredictionCol, value).asInstanceOf[M] + + /** Number of classes (values which the label can take). */ + def numClasses: Int + + /** + * Transforms dataset by reading from [[featuresCol]], and appending new columns as specified by + * parameters: + * - predicted labels as [[predictionCol]] of type [[Double]] + * - raw predictions (confidences) as [[rawPredictionCol]] of type [[Vector]]. + * + * @param dataset input dataset + * @param paramMap additional parameters, overwrite embedded params + * @return transformed dataset + */ + override def transform(dataset: DataFrame, paramMap: ParamMap): DataFrame = { + // This default implementation should be overridden as needed. + + // Check schema + transformSchema(dataset.schema, paramMap, logging = true) + val map = this.paramMap ++ paramMap + + // Prepare model + val tmpModel = if (paramMap.size != 0) { + val tmpModel = this.copy() + Params.inheritValues(paramMap, parent, tmpModel) + tmpModel + } else { + this + } + + val (numColsOutput, outputData) = + ClassificationModel.transformColumnsImpl[FeaturesType](dataset, tmpModel, map) + if (numColsOutput == 0) { + logWarning(s"$uid: ClassificationModel.transform() was called as NOOP" + + " since no output columns were set.") + } + outputData + } + + /** + * :: DeveloperApi :: + * + * Predict label for the given features. + * This internal method is used to implement [[transform()]] and output [[predictionCol]]. + * + * This default implementation for classification predicts the index of the maximum value + * from [[predictRaw()]]. + */ + @DeveloperApi + override protected def predict(features: FeaturesType): Double = { + predictRaw(features).toArray.zipWithIndex.maxBy(_._1)._2 + } + + /** + * :: DeveloperApi :: + * + * Raw prediction for each possible label. + * The meaning of a "raw" prediction may vary between algorithms, but it intuitively gives + * a measure of confidence in each possible label (where larger = more confident). + * This internal method is used to implement [[transform()]] and output [[rawPredictionCol]]. + * + * @return vector where element i is the raw prediction for label i. + * This raw prediction may be any real number, where a larger value indicates greater + * confidence for that label. + */ + @DeveloperApi + protected def predictRaw(features: FeaturesType): Vector + +} + +private[ml] object ClassificationModel { + + /** + * Added prediction column(s). This is separated from [[ClassificationModel.transform()]] + * since it is used by [[org.apache.spark.ml.classification.ProbabilisticClassificationModel]]. + * @param dataset Input dataset + * @param map Parameter map. This will NOT be merged with the embedded paramMap; the merge + * should already be done. + * @return (number of columns added, transformed dataset) + */ + def transformColumnsImpl[FeaturesType]( + dataset: DataFrame, + model: ClassificationModel[FeaturesType, _], + map: ParamMap): (Int, DataFrame) = { + + // Output selected columns only. + // This is a bit complicated since it tries to avoid repeated computation. + var tmpData = dataset + var numColsOutput = 0 + if (map(model.rawPredictionCol) != "") { + // output raw prediction + val features2raw: FeaturesType => Vector = model.predictRaw + tmpData = tmpData.select($"*", + callUDF(features2raw, new VectorUDT, + col(map(model.featuresCol))).as(map(model.rawPredictionCol))) + numColsOutput += 1 + if (map(model.predictionCol) != "") { + val raw2pred: Vector => Double = (rawPred) => { + rawPred.toArray.zipWithIndex.maxBy(_._1)._2 + } + tmpData = tmpData.select($"*", callUDF(raw2pred, DoubleType, + col(map(model.rawPredictionCol))).as(map(model.predictionCol))) + numColsOutput += 1 + } + } else if (map(model.predictionCol) != "") { + // output prediction + val features2pred: FeaturesType => Double = model.predict + tmpData = tmpData.select($"*", + callUDF(features2pred, DoubleType, + col(map(model.featuresCol))).as(map(model.predictionCol))) + numColsOutput += 1 + } + (numColsOutput, tmpData) + } + +} diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala index b46a5cd8bdf29..c146fe244c66e 100644 --- a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala +++ b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala @@ -18,61 +18,32 @@ package org.apache.spark.ml.classification import org.apache.spark.annotation.AlphaComponent -import org.apache.spark.ml._ import org.apache.spark.ml.param._ import org.apache.spark.mllib.classification.LogisticRegressionWithLBFGS -import org.apache.spark.mllib.linalg.{BLAS, Vector, VectorUDT} -import org.apache.spark.mllib.regression.LabeledPoint -import org.apache.spark.sql._ +import org.apache.spark.mllib.linalg.{VectorUDT, BLAS, Vector, Vectors} +import org.apache.spark.sql.DataFrame import org.apache.spark.sql.Dsl._ -import org.apache.spark.sql.types.{DoubleType, StructField, StructType} +import org.apache.spark.sql.types.DoubleType import org.apache.spark.storage.StorageLevel + /** - * :: AlphaComponent :: * Params for logistic regression. */ -@AlphaComponent -private[classification] trait LogisticRegressionParams extends Params - with HasRegParam with HasMaxIter with HasLabelCol with HasThreshold with HasFeaturesCol - with HasScoreCol with HasPredictionCol { +private[classification] trait LogisticRegressionParams extends ProbabilisticClassifierParams + with HasRegParam with HasMaxIter with HasThreshold - /** - * Validates and transforms the input schema with the provided param map. - * @param schema input schema - * @param paramMap additional parameters - * @param fitting whether this is in fitting - * @return output schema - */ - protected def validateAndTransformSchema( - schema: StructType, - paramMap: ParamMap, - fitting: Boolean): StructType = { - val map = this.paramMap ++ paramMap - val featuresType = schema(map(featuresCol)).dataType - // TODO: Support casting Array[Double] and Array[Float] to Vector. - require(featuresType.isInstanceOf[VectorUDT], - s"Features column ${map(featuresCol)} must be a vector column but got $featuresType.") - if (fitting) { - val labelType = schema(map(labelCol)).dataType - require(labelType == DoubleType, - s"Cannot convert label column ${map(labelCol)} of type $labelType to a double column.") - } - val fieldNames = schema.fieldNames - require(!fieldNames.contains(map(scoreCol)), s"Score column ${map(scoreCol)} already exists.") - require(!fieldNames.contains(map(predictionCol)), - s"Prediction column ${map(predictionCol)} already exists.") - val outputFields = schema.fields ++ Seq( - StructField(map(scoreCol), DoubleType, false), - StructField(map(predictionCol), DoubleType, false)) - StructType(outputFields) - } -} /** + * :: AlphaComponent :: + * * Logistic regression. + * Currently, this class only supports binary classification. */ -class LogisticRegression extends Estimator[LogisticRegressionModel] with LogisticRegressionParams { +@AlphaComponent +class LogisticRegression + extends ProbabilisticClassifier[Vector, LogisticRegression, LogisticRegressionModel] + with LogisticRegressionParams { setRegParam(0.1) setMaxIter(100) @@ -80,68 +51,151 @@ class LogisticRegression extends Estimator[LogisticRegressionModel] with Logisti def setRegParam(value: Double): this.type = set(regParam, value) def setMaxIter(value: Int): this.type = set(maxIter, value) - def setLabelCol(value: String): this.type = set(labelCol, value) def setThreshold(value: Double): this.type = set(threshold, value) - def setFeaturesCol(value: String): this.type = set(featuresCol, value) - def setScoreCol(value: String): this.type = set(scoreCol, value) - def setPredictionCol(value: String): this.type = set(predictionCol, value) - override def fit(dataset: DataFrame, paramMap: ParamMap): LogisticRegressionModel = { - transformSchema(dataset.schema, paramMap, logging = true) - val map = this.paramMap ++ paramMap - val instances = dataset.select(map(labelCol), map(featuresCol)) - .map { case Row(label: Double, features: Vector) => - LabeledPoint(label, features) - }.persist(StorageLevel.MEMORY_AND_DISK) + override protected def train(dataset: DataFrame, paramMap: ParamMap): LogisticRegressionModel = { + // Extract columns from data. If dataset is persisted, do not persist oldDataset. + val oldDataset = extractLabeledPoints(dataset, paramMap) + val handlePersistence = dataset.rdd.getStorageLevel == StorageLevel.NONE + if (handlePersistence) { + oldDataset.persist(StorageLevel.MEMORY_AND_DISK) + } + + // Train model val lr = new LogisticRegressionWithLBFGS lr.optimizer - .setRegParam(map(regParam)) - .setNumIterations(map(maxIter)) - val lrm = new LogisticRegressionModel(this, map, lr.run(instances).weights) - instances.unpersist() - // copy model params - Params.inheritValues(map, this, lrm) - lrm - } + .setRegParam(paramMap(regParam)) + .setNumIterations(paramMap(maxIter)) + val oldModel = lr.run(oldDataset) + val lrm = new LogisticRegressionModel(this, paramMap, oldModel.weights, oldModel.intercept) - private[ml] override def transformSchema(schema: StructType, paramMap: ParamMap): StructType = { - validateAndTransformSchema(schema, paramMap, fitting = true) + if (handlePersistence) { + oldDataset.unpersist() + } + lrm } } + /** * :: AlphaComponent :: + * * Model produced by [[LogisticRegression]]. */ @AlphaComponent class LogisticRegressionModel private[ml] ( override val parent: LogisticRegression, override val fittingParamMap: ParamMap, - weights: Vector) - extends Model[LogisticRegressionModel] with LogisticRegressionParams { + val weights: Vector, + val intercept: Double) + extends ProbabilisticClassificationModel[Vector, LogisticRegressionModel] + with LogisticRegressionParams { + + setThreshold(0.5) def setThreshold(value: Double): this.type = set(threshold, value) - def setFeaturesCol(value: String): this.type = set(featuresCol, value) - def setScoreCol(value: String): this.type = set(scoreCol, value) - def setPredictionCol(value: String): this.type = set(predictionCol, value) - private[ml] override def transformSchema(schema: StructType, paramMap: ParamMap): StructType = { - validateAndTransformSchema(schema, paramMap, fitting = false) + private val margin: Vector => Double = (features) => { + BLAS.dot(features, weights) + intercept + } + + private val score: Vector => Double = (features) => { + val m = margin(features) + 1.0 / (1.0 + math.exp(-m)) } override def transform(dataset: DataFrame, paramMap: ParamMap): DataFrame = { + // This is overridden (a) to be more efficient (avoiding re-computing values when creating + // multiple output columns) and (b) to handle threshold, which the abstractions do not use. + // TODO: We should abstract away the steps defined by UDFs below so that the abstractions + // can call whichever UDFs are needed to create the output columns. + + // Check schema transformSchema(dataset.schema, paramMap, logging = true) + val map = this.paramMap ++ paramMap - val scoreFunction = udf { v: Vector => - val margin = BLAS.dot(v, weights) - 1.0 / (1.0 + math.exp(-margin)) + + // Output selected columns only. + // This is a bit complicated since it tries to avoid repeated computation. + // rawPrediction (-margin, margin) + // probability (1.0-score, score) + // prediction (max margin) + var tmpData = dataset + var numColsOutput = 0 + if (map(rawPredictionCol) != "") { + val features2raw: Vector => Vector = (features) => predictRaw(features) + tmpData = tmpData.select($"*", + callUDF(features2raw, new VectorUDT, col(map(featuresCol))).as(map(rawPredictionCol))) + numColsOutput += 1 + } + if (map(probabilityCol) != "") { + if (map(rawPredictionCol) != "") { + val raw2prob: Vector => Vector = { (rawPreds: Vector) => + val prob1 = 1.0 / (1.0 + math.exp(-rawPreds(1))) + Vectors.dense(1.0 - prob1, prob1) + } + tmpData = tmpData.select($"*", + callUDF(raw2prob, new VectorUDT, col(map(rawPredictionCol))).as(map(probabilityCol))) + } else { + val features2prob: Vector => Vector = (features: Vector) => predictProbabilities(features) + tmpData = tmpData.select($"*", + callUDF(features2prob, new VectorUDT, col(map(featuresCol))).as(map(probabilityCol))) + } + numColsOutput += 1 } - val t = map(threshold) - val predictFunction = udf { score: Double => - if (score > t) 1.0 else 0.0 + if (map(predictionCol) != "") { + val t = map(threshold) + if (map(probabilityCol) != "") { + val predict: Vector => Double = { probs: Vector => + if (probs(1) > t) 1.0 else 0.0 + } + tmpData = tmpData.select($"*", + callUDF(predict, DoubleType, col(map(probabilityCol))).as(map(predictionCol))) + } else if (map(rawPredictionCol) != "") { + val predict: Vector => Double = { rawPreds: Vector => + val prob1 = 1.0 / (1.0 + math.exp(-rawPreds(1))) + if (prob1 > t) 1.0 else 0.0 + } + tmpData = tmpData.select($"*", + callUDF(predict, DoubleType, col(map(rawPredictionCol))).as(map(predictionCol))) + } else { + val predict: Vector => Double = (features: Vector) => this.predict(features) + tmpData = tmpData.select($"*", + callUDF(predict, DoubleType, col(map(featuresCol))).as(map(predictionCol))) + } + numColsOutput += 1 } - dataset - .select($"*", scoreFunction(col(map(featuresCol))).as(map(scoreCol))) - .select($"*", predictFunction(col(map(scoreCol))).as(map(predictionCol))) + if (numColsOutput == 0) { + this.logWarning(s"$uid: LogisticRegressionModel.transform() was called as NOOP" + + " since no output columns were set.") + } + tmpData + } + + override val numClasses: Int = 2 + + /** + * Predict label for the given feature vector. + * The behavior of this can be adjusted using [[threshold]]. + */ + override protected def predict(features: Vector): Double = { + println(s"LR.predict with threshold: ${paramMap(threshold)}") + if (score(features) > paramMap(threshold)) 1 else 0 + } + + override protected def predictProbabilities(features: Vector): Vector = { + val s = score(features) + Vectors.dense(1.0 - s, s) + } + + override protected def predictRaw(features: Vector): Vector = { + val m = margin(features) + Vectors.dense(0.0, m) + } + + override protected def copy(): LogisticRegressionModel = { + val m = new LogisticRegressionModel(parent, fittingParamMap, weights, intercept) + Params.inheritValues(this.paramMap, this, m) + m } } diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala new file mode 100644 index 0000000000000..1202528ca654e --- /dev/null +++ b/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala @@ -0,0 +1,147 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.classification + +import org.apache.spark.annotation.{AlphaComponent, DeveloperApi} +import org.apache.spark.ml.param.{HasProbabilityCol, ParamMap, Params} +import org.apache.spark.mllib.linalg.{Vector, VectorUDT} +import org.apache.spark.sql.DataFrame +import org.apache.spark.sql.Dsl._ +import org.apache.spark.sql.types.{DataType, StructType} + + +/** + * Params for probabilistic classification. + */ +private[classification] trait ProbabilisticClassifierParams + extends ClassifierParams with HasProbabilityCol { + + override protected def validateAndTransformSchema( + schema: StructType, + paramMap: ParamMap, + fitting: Boolean, + featuresDataType: DataType): StructType = { + val parentSchema = super.validateAndTransformSchema(schema, paramMap, fitting, featuresDataType) + val map = this.paramMap ++ paramMap + addOutputColumn(parentSchema, map(probabilityCol), new VectorUDT) + } +} + + +/** + * :: AlphaComponent :: + * + * Single-label binary or multiclass classifier which can output class conditional probabilities. + * + * @tparam FeaturesType Type of input features. E.g., [[Vector]] + * @tparam E Concrete Estimator type + * @tparam M Concrete Model type + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@AlphaComponent +private[spark] abstract class ProbabilisticClassifier[ + FeaturesType, + E <: ProbabilisticClassifier[FeaturesType, E, M], + M <: ProbabilisticClassificationModel[FeaturesType, M]] + extends Classifier[FeaturesType, E, M] with ProbabilisticClassifierParams { + + def setProbabilityCol(value: String): E = set(probabilityCol, value).asInstanceOf[E] +} + + +/** + * :: AlphaComponent :: + * + * Model produced by a [[ProbabilisticClassifier]]. + * Classes are indexed {0, 1, ..., numClasses - 1}. + * + * @tparam FeaturesType Type of input features. E.g., [[Vector]] + * @tparam M Concrete Model type + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@AlphaComponent +private[spark] abstract class ProbabilisticClassificationModel[ + FeaturesType, + M <: ProbabilisticClassificationModel[FeaturesType, M]] + extends ClassificationModel[FeaturesType, M] with ProbabilisticClassifierParams { + + def setProbabilityCol(value: String): M = set(probabilityCol, value).asInstanceOf[M] + + /** + * Transforms dataset by reading from [[featuresCol]], and appending new columns as specified by + * parameters: + * - predicted labels as [[predictionCol]] of type [[Double]] + * - raw predictions (confidences) as [[rawPredictionCol]] of type [[Vector]] + * - probability of each class as [[probabilityCol]] of type [[Vector]]. + * + * @param dataset input dataset + * @param paramMap additional parameters, overwrite embedded params + * @return transformed dataset + */ + override def transform(dataset: DataFrame, paramMap: ParamMap): DataFrame = { + // This default implementation should be overridden as needed. + + // Check schema + transformSchema(dataset.schema, paramMap, logging = true) + val map = this.paramMap ++ paramMap + + // Prepare model + val tmpModel = if (paramMap.size != 0) { + val tmpModel = this.copy() + Params.inheritValues(paramMap, parent, tmpModel) + tmpModel + } else { + this + } + + val (numColsOutput, outputData) = + ClassificationModel.transformColumnsImpl[FeaturesType](dataset, tmpModel, map) + + // Output selected columns only. + if (map(probabilityCol) != "") { + // output probabilities + val features2probs: FeaturesType => Vector = (features) => { + tmpModel.predictProbabilities(features) + } + outputData.select($"*", + callUDF(features2probs, new VectorUDT, col(map(featuresCol))).as(map(probabilityCol))) + } else { + if (numColsOutput == 0) { + this.logWarning(s"$uid: ProbabilisticClassificationModel.transform() was called as NOOP" + + " since no output columns were set.") + } + outputData + } + } + + /** + * :: DeveloperApi :: + * + * Predict the probability of each class given the features. + * These predictions are also called class conditional probabilities. + * + * WARNING: Not all models output well-calibrated probability estimates! These probabilities + * should be treated as confidences, not precise probabilities. + * + * This internal method is used to implement [[transform()]] and output [[probabilityCol]]. + */ + @DeveloperApi + protected def predictProbabilities(features: FeaturesType): Vector +} diff --git a/mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala b/mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala index 1979ab9eb6516..f21a30627e540 100644 --- a/mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala +++ b/mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala @@ -18,19 +18,22 @@ package org.apache.spark.ml.evaluation import org.apache.spark.annotation.AlphaComponent -import org.apache.spark.ml._ +import org.apache.spark.ml.Evaluator import org.apache.spark.ml.param._ import org.apache.spark.mllib.evaluation.BinaryClassificationMetrics +import org.apache.spark.mllib.linalg.{Vector, VectorUDT} import org.apache.spark.sql.{DataFrame, Row} import org.apache.spark.sql.types.DoubleType + /** * :: AlphaComponent :: + * * Evaluator for binary classification, which expects two input columns: score and label. */ @AlphaComponent class BinaryClassificationEvaluator extends Evaluator with Params - with HasScoreCol with HasLabelCol { + with HasRawPredictionCol with HasLabelCol { /** param for metric name in evaluation */ val metricName: Param[String] = new Param(this, "metricName", @@ -38,23 +41,20 @@ class BinaryClassificationEvaluator extends Evaluator with Params def getMetricName: String = get(metricName) def setMetricName(value: String): this.type = set(metricName, value) - def setScoreCol(value: String): this.type = set(scoreCol, value) + def setScoreCol(value: String): this.type = set(rawPredictionCol, value) def setLabelCol(value: String): this.type = set(labelCol, value) override def evaluate(dataset: DataFrame, paramMap: ParamMap): Double = { val map = this.paramMap ++ paramMap val schema = dataset.schema - val scoreType = schema(map(scoreCol)).dataType - require(scoreType == DoubleType, - s"Score column ${map(scoreCol)} must be double type but found $scoreType") - val labelType = schema(map(labelCol)).dataType - require(labelType == DoubleType, - s"Label column ${map(labelCol)} must be double type but found $labelType") + checkInputColumn(schema, map(rawPredictionCol), new VectorUDT) + checkInputColumn(schema, map(labelCol), DoubleType) - val scoreAndLabels = dataset.select(map(scoreCol), map(labelCol)) - .map { case Row(score: Double, label: Double) => - (score, label) + // TODO: When dataset metadata has been implemented, check rawPredictionCol vector length = 2. + val scoreAndLabels = dataset.select(map(rawPredictionCol), map(labelCol)) + .map { case Row(rawPrediction: Vector, label: Double) => + (rawPrediction(1), label) } val metrics = new BinaryClassificationMetrics(scoreAndLabels) val metric = map(metricName) match { diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/Tokenizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/Tokenizer.scala index e622a5cf9e6f3..0b1f90daa7d8e 100644 --- a/mllib/src/main/scala/org/apache/spark/ml/feature/Tokenizer.scala +++ b/mllib/src/main/scala/org/apache/spark/ml/feature/Tokenizer.scala @@ -29,11 +29,11 @@ import org.apache.spark.sql.types.{DataType, StringType, ArrayType} @AlphaComponent class Tokenizer extends UnaryTransformer[String, Seq[String], Tokenizer] { - protected override def createTransformFunc(paramMap: ParamMap): String => Seq[String] = { + override protected def createTransformFunc(paramMap: ParamMap): String => Seq[String] = { _.toLowerCase.split("\\s") } - protected override def validateInputType(inputType: DataType): Unit = { + override protected def validateInputType(inputType: DataType): Unit = { require(inputType == StringType, s"Input type must be string type but got $inputType.") } diff --git a/mllib/src/main/scala/org/apache/spark/ml/impl/estimator/Predictor.scala b/mllib/src/main/scala/org/apache/spark/ml/impl/estimator/Predictor.scala new file mode 100644 index 0000000000000..89b53f3890ea3 --- /dev/null +++ b/mllib/src/main/scala/org/apache/spark/ml/impl/estimator/Predictor.scala @@ -0,0 +1,234 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.impl.estimator + +import org.apache.spark.annotation.{AlphaComponent, DeveloperApi} +import org.apache.spark.ml.{Estimator, Model} +import org.apache.spark.ml.param._ +import org.apache.spark.mllib.linalg.{VectorUDT, Vector} +import org.apache.spark.mllib.regression.LabeledPoint +import org.apache.spark.rdd.RDD +import org.apache.spark.sql.{DataFrame, Row} +import org.apache.spark.sql.Dsl._ +import org.apache.spark.sql.types.{DataType, DoubleType, StructType} + + +/** + * :: DeveloperApi :: + * + * Trait for parameters for prediction (regression and classification). + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@DeveloperApi +private[spark] trait PredictorParams extends Params + with HasLabelCol with HasFeaturesCol with HasPredictionCol { + + /** + * Validates and transforms the input schema with the provided param map. + * @param schema input schema + * @param paramMap additional parameters + * @param fitting whether this is in fitting + * @param featuresDataType SQL DataType for FeaturesType. + * E.g., [[org.apache.spark.mllib.linalg.VectorUDT]] for vector features. + * @return output schema + */ + protected def validateAndTransformSchema( + schema: StructType, + paramMap: ParamMap, + fitting: Boolean, + featuresDataType: DataType): StructType = { + val map = this.paramMap ++ paramMap + // TODO: Support casting Array[Double] and Array[Float] to Vector when FeaturesType = Vector + checkInputColumn(schema, map(featuresCol), featuresDataType) + if (fitting) { + // TODO: Allow other numeric types + checkInputColumn(schema, map(labelCol), DoubleType) + } + addOutputColumn(schema, map(predictionCol), DoubleType) + } +} + +/** + * :: AlphaComponent :: + * + * Abstraction for prediction problems (regression and classification). + * + * @tparam FeaturesType Type of features. + * E.g., [[org.apache.spark.mllib.linalg.VectorUDT]] for vector features. + * @tparam Learner Specialization of this class. If you subclass this type, use this type + * parameter to specify the concrete type. + * @tparam M Specialization of [[PredictionModel]]. If you subclass this type, use this type + * parameter to specify the concrete type for the corresponding model. + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@AlphaComponent +private[spark] abstract class Predictor[ + FeaturesType, + Learner <: Predictor[FeaturesType, Learner, M], + M <: PredictionModel[FeaturesType, M]] + extends Estimator[M] with PredictorParams { + + def setLabelCol(value: String): Learner = set(labelCol, value).asInstanceOf[Learner] + def setFeaturesCol(value: String): Learner = set(featuresCol, value).asInstanceOf[Learner] + def setPredictionCol(value: String): Learner = set(predictionCol, value).asInstanceOf[Learner] + + override def fit(dataset: DataFrame, paramMap: ParamMap): M = { + // This handles a few items such as schema validation. + // Developers only need to implement train(). + transformSchema(dataset.schema, paramMap, logging = true) + val map = this.paramMap ++ paramMap + val model = train(dataset, map) + Params.inheritValues(map, this, model) // copy params to model + model + } + + /** + * :: DeveloperApi :: + * + * Train a model using the given dataset and parameters. + * Developers can implement this instead of [[fit()]] to avoid dealing with schema validation + * and copying parameters into the model. + * + * @param dataset Training dataset + * @param paramMap Parameter map. Unlike [[fit()]]'s paramMap, this paramMap has already + * been combined with the embedded ParamMap. + * @return Fitted model + */ + @DeveloperApi + protected def train(dataset: DataFrame, paramMap: ParamMap): M + + /** + * :: DeveloperApi :: + * + * Returns the SQL DataType corresponding to the FeaturesType type parameter. + * + * This is used by [[validateAndTransformSchema()]]. + * This workaround is needed since SQL has different APIs for Scala and Java. + * + * The default value is VectorUDT, but it may be overridden if FeaturesType is not Vector. + */ + @DeveloperApi + protected def featuresDataType: DataType = new VectorUDT + + private[ml] override def transformSchema(schema: StructType, paramMap: ParamMap): StructType = { + validateAndTransformSchema(schema, paramMap, fitting = true, featuresDataType) + } + + /** + * Extract [[labelCol]] and [[featuresCol]] from the given dataset, + * and put it in an RDD with strong types. + */ + protected def extractLabeledPoints(dataset: DataFrame, paramMap: ParamMap): RDD[LabeledPoint] = { + val map = this.paramMap ++ paramMap + dataset.select(map(labelCol), map(featuresCol)) + .map { case Row(label: Double, features: Vector) => + LabeledPoint(label, features) + } + } +} + +/** + * :: AlphaComponent :: + * + * Abstraction for a model for prediction tasks (regression and classification). + * + * @tparam FeaturesType Type of features. + * E.g., [[org.apache.spark.mllib.linalg.VectorUDT]] for vector features. + * @tparam M Specialization of [[PredictionModel]]. If you subclass this type, use this type + * parameter to specify the concrete type for the corresponding model. + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@AlphaComponent +private[spark] abstract class PredictionModel[FeaturesType, M <: PredictionModel[FeaturesType, M]] + extends Model[M] with PredictorParams { + + def setFeaturesCol(value: String): M = set(featuresCol, value).asInstanceOf[M] + + def setPredictionCol(value: String): M = set(predictionCol, value).asInstanceOf[M] + + /** + * :: DeveloperApi :: + * + * Returns the SQL DataType corresponding to the FeaturesType type parameter. + * + * This is used by [[validateAndTransformSchema()]]. + * This workaround is needed since SQL has different APIs for Scala and Java. + * + * The default value is VectorUDT, but it may be overridden if FeaturesType is not Vector. + */ + @DeveloperApi + protected def featuresDataType: DataType = new VectorUDT + + private[ml] override def transformSchema(schema: StructType, paramMap: ParamMap): StructType = { + validateAndTransformSchema(schema, paramMap, fitting = false, featuresDataType) + } + + /** + * Transforms dataset by reading from [[featuresCol]], calling [[predict()]], and storing + * the predictions as a new column [[predictionCol]]. + * + * @param dataset input dataset + * @param paramMap additional parameters, overwrite embedded params + * @return transformed dataset with [[predictionCol]] of type [[Double]] + */ + override def transform(dataset: DataFrame, paramMap: ParamMap): DataFrame = { + // This default implementation should be overridden as needed. + + // Check schema + transformSchema(dataset.schema, paramMap, logging = true) + val map = this.paramMap ++ paramMap + + // Prepare model + val tmpModel = if (paramMap.size != 0) { + val tmpModel = this.copy() + Params.inheritValues(paramMap, parent, tmpModel) + tmpModel + } else { + this + } + + if (map(predictionCol) != "") { + val pred: FeaturesType => Double = (features) => { + tmpModel.predict(features) + } + dataset.select($"*", callUDF(pred, DoubleType, col(map(featuresCol))).as(map(predictionCol))) + } else { + this.logWarning(s"$uid: Predictor.transform() was called as NOOP" + + " since no output columns were set.") + dataset + } + } + + /** + * :: DeveloperApi :: + * + * Predict label for the given features. + * This internal method is used to implement [[transform()]] and output [[predictionCol]]. + */ + @DeveloperApi + protected def predict(features: FeaturesType): Double + + /** + * Create a copy of the model. + * The copy is shallow, except for the embedded paramMap, which gets a deep copy. + */ + protected def copy(): M +} diff --git a/mllib/src/main/scala/org/apache/spark/ml/param/params.scala b/mllib/src/main/scala/org/apache/spark/ml/param/params.scala index 5fb4379e23c2f..17ece897a6c55 100644 --- a/mllib/src/main/scala/org/apache/spark/ml/param/params.scala +++ b/mllib/src/main/scala/org/apache/spark/ml/param/params.scala @@ -22,8 +22,10 @@ import scala.collection.mutable import java.lang.reflect.Modifier -import org.apache.spark.annotation.AlphaComponent +import org.apache.spark.annotation.{AlphaComponent, DeveloperApi} import org.apache.spark.ml.Identifiable +import org.apache.spark.sql.types.{DataType, StructField, StructType} + /** * :: AlphaComponent :: @@ -65,37 +67,47 @@ class Param[T] ( // specialize primitive-typed params because Java doesn't recognize scala.Double, scala.Int, ... /** Specialized version of [[Param[Double]]] for Java. */ -class DoubleParam(parent: Params, name: String, doc: String, defaultValue: Option[Double] = None) +class DoubleParam(parent: Params, name: String, doc: String, defaultValue: Option[Double]) extends Param[Double](parent, name, doc, defaultValue) { + def this(parent: Params, name: String, doc: String) = this(parent, name, doc, None) + override def w(value: Double): ParamPair[Double] = super.w(value) } /** Specialized version of [[Param[Int]]] for Java. */ -class IntParam(parent: Params, name: String, doc: String, defaultValue: Option[Int] = None) +class IntParam(parent: Params, name: String, doc: String, defaultValue: Option[Int]) extends Param[Int](parent, name, doc, defaultValue) { + def this(parent: Params, name: String, doc: String) = this(parent, name, doc, None) + override def w(value: Int): ParamPair[Int] = super.w(value) } /** Specialized version of [[Param[Float]]] for Java. */ -class FloatParam(parent: Params, name: String, doc: String, defaultValue: Option[Float] = None) +class FloatParam(parent: Params, name: String, doc: String, defaultValue: Option[Float]) extends Param[Float](parent, name, doc, defaultValue) { + def this(parent: Params, name: String, doc: String) = this(parent, name, doc, None) + override def w(value: Float): ParamPair[Float] = super.w(value) } /** Specialized version of [[Param[Long]]] for Java. */ -class LongParam(parent: Params, name: String, doc: String, defaultValue: Option[Long] = None) +class LongParam(parent: Params, name: String, doc: String, defaultValue: Option[Long]) extends Param[Long](parent, name, doc, defaultValue) { + def this(parent: Params, name: String, doc: String) = this(parent, name, doc, None) + override def w(value: Long): ParamPair[Long] = super.w(value) } /** Specialized version of [[Param[Boolean]]] for Java. */ -class BooleanParam(parent: Params, name: String, doc: String, defaultValue: Option[Boolean] = None) +class BooleanParam(parent: Params, name: String, doc: String, defaultValue: Option[Boolean]) extends Param[Boolean](parent, name, doc, defaultValue) { + def this(parent: Params, name: String, doc: String) = this(parent, name, doc, None) + override def w(value: Boolean): ParamPair[Boolean] = super.w(value) } @@ -158,7 +170,7 @@ trait Params extends Identifiable with Serializable { /** * Sets a parameter in the embedded param map. */ - private[ml] def set[T](param: Param[T], value: T): this.type = { + protected def set[T](param: Param[T], value: T): this.type = { require(param.parent.eq(this)) paramMap.put(param.asInstanceOf[Param[Any]], value) this @@ -174,7 +186,7 @@ trait Params extends Identifiable with Serializable { /** * Gets the value of a parameter in the embedded param map. */ - private[ml] def get[T](param: Param[T]): T = { + protected def get[T](param: Param[T]): T = { require(param.parent.eq(this)) paramMap(param) } @@ -183,9 +195,40 @@ trait Params extends Identifiable with Serializable { * Internal param map. */ protected val paramMap: ParamMap = ParamMap.empty + + /** + * Check whether the given schema contains an input column. + * @param colName Parameter name for the input column. + * @param dataType SQL DataType of the input column. + */ + protected def checkInputColumn(schema: StructType, colName: String, dataType: DataType): Unit = { + val actualDataType = schema(colName).dataType + require(actualDataType.equals(dataType), + s"Input column $colName must be of type $dataType" + + s" but was actually $actualDataType. Column param description: ${getParam(colName)}") + } + + protected def addOutputColumn( + schema: StructType, + colName: String, + dataType: DataType): StructType = { + if (colName.length == 0) return schema + val fieldNames = schema.fieldNames + require(!fieldNames.contains(colName), s"Prediction column $colName already exists.") + val outputFields = schema.fields ++ Seq(StructField(colName, dataType, nullable = false)) + StructType(outputFields) + } } -private[ml] object Params { +/** + * :: DeveloperApi :: + * + * Helper functionality for developers. + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@DeveloperApi +private[spark] object Params { /** * Copies parameter values from the parent estimator to the child model it produced. @@ -279,7 +322,7 @@ class ParamMap private[ml] (private val map: mutable.Map[Param[Any], Any]) exten def copy: ParamMap = new ParamMap(map.clone()) override def toString: String = { - map.map { case (param, value) => + map.toSeq.sortBy(_._1.name).map { case (param, value) => s"\t${param.parent.uid}-${param.name}: $value" }.mkString("{\n", ",\n", "\n}") } @@ -310,6 +353,11 @@ class ParamMap private[ml] (private val map: mutable.Map[Param[Any], Any]) exten ParamPair(param, value) } } + + /** + * Number of param pairs in this set. + */ + def size: Int = map.size } object ParamMap { diff --git a/mllib/src/main/scala/org/apache/spark/ml/param/sharedParams.scala b/mllib/src/main/scala/org/apache/spark/ml/param/sharedParams.scala index ef141d3eb2b06..32fc74462ef4a 100644 --- a/mllib/src/main/scala/org/apache/spark/ml/param/sharedParams.scala +++ b/mllib/src/main/scala/org/apache/spark/ml/param/sharedParams.scala @@ -17,6 +17,12 @@ package org.apache.spark.ml.param +/* NOTE TO DEVELOPERS: + * If you mix these parameter traits into your algorithm, please add a setter method as well + * so that users may use a builder pattern: + * val myLearner = new MyLearner().setParam1(x).setParam2(y)... + */ + private[ml] trait HasRegParam extends Params { /** param for regularization parameter */ val regParam: DoubleParam = new DoubleParam(this, "regParam", "regularization parameter") @@ -42,12 +48,6 @@ private[ml] trait HasLabelCol extends Params { def getLabelCol: String = get(labelCol) } -private[ml] trait HasScoreCol extends Params { - /** param for score column name */ - val scoreCol: Param[String] = new Param(this, "scoreCol", "score column name", Some("score")) - def getScoreCol: String = get(scoreCol) -} - private[ml] trait HasPredictionCol extends Params { /** param for prediction column name */ val predictionCol: Param[String] = @@ -55,6 +55,22 @@ private[ml] trait HasPredictionCol extends Params { def getPredictionCol: String = get(predictionCol) } +private[ml] trait HasRawPredictionCol extends Params { + /** param for raw prediction column name */ + val rawPredictionCol: Param[String] = + new Param(this, "rawPredictionCol", "raw prediction (a.k.a. confidence) column name", + Some("rawPrediction")) + def getRawPredictionCol: String = get(rawPredictionCol) +} + +private[ml] trait HasProbabilityCol extends Params { + /** param for predicted class conditional probabilities column name */ + val probabilityCol: Param[String] = + new Param(this, "probabilityCol", "column name for predicted class conditional probabilities", + Some("probability")) + def getProbabilityCol: String = get(probabilityCol) +} + private[ml] trait HasThreshold extends Params { /** param for threshold in (binary) prediction */ val threshold: DoubleParam = new DoubleParam(this, "threshold", "threshold in prediction") diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala new file mode 100644 index 0000000000000..d5a7bdafcb623 --- /dev/null +++ b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala @@ -0,0 +1,96 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.regression + +import org.apache.spark.annotation.AlphaComponent +import org.apache.spark.ml.param.{Params, ParamMap, HasMaxIter, HasRegParam} +import org.apache.spark.mllib.linalg.{BLAS, Vector} +import org.apache.spark.mllib.regression.LinearRegressionWithSGD +import org.apache.spark.sql.DataFrame +import org.apache.spark.storage.StorageLevel + + +/** + * Params for linear regression. + */ +private[regression] trait LinearRegressionParams extends RegressorParams + with HasRegParam with HasMaxIter + + +/** + * :: AlphaComponent :: + * + * Linear regression. + */ +@AlphaComponent +class LinearRegression extends Regressor[Vector, LinearRegression, LinearRegressionModel] + with LinearRegressionParams { + + setRegParam(0.1) + setMaxIter(100) + + def setRegParam(value: Double): this.type = set(regParam, value) + def setMaxIter(value: Int): this.type = set(maxIter, value) + + override protected def train(dataset: DataFrame, paramMap: ParamMap): LinearRegressionModel = { + // Extract columns from data. If dataset is persisted, do not persist oldDataset. + val oldDataset = extractLabeledPoints(dataset, paramMap) + val handlePersistence = dataset.rdd.getStorageLevel == StorageLevel.NONE + if (handlePersistence) { + oldDataset.persist(StorageLevel.MEMORY_AND_DISK) + } + + // Train model + val lr = new LinearRegressionWithSGD() + lr.optimizer + .setRegParam(paramMap(regParam)) + .setNumIterations(paramMap(maxIter)) + val model = lr.run(oldDataset) + val lrm = new LinearRegressionModel(this, paramMap, model.weights, model.intercept) + + if (handlePersistence) { + oldDataset.unpersist() + } + lrm + } +} + +/** + * :: AlphaComponent :: + * + * Model produced by [[LinearRegression]]. + */ +@AlphaComponent +class LinearRegressionModel private[ml] ( + override val parent: LinearRegression, + override val fittingParamMap: ParamMap, + val weights: Vector, + val intercept: Double) + extends RegressionModel[Vector, LinearRegressionModel] + with LinearRegressionParams { + + override protected def predict(features: Vector): Double = { + BLAS.dot(features, weights) + intercept + } + + override protected def copy(): LinearRegressionModel = { + val m = new LinearRegressionModel(parent, fittingParamMap, weights, intercept) + Params.inheritValues(this.paramMap, this, m) + m + } +} diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/Regressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/Regressor.scala new file mode 100644 index 0000000000000..d679085eeafe1 --- /dev/null +++ b/mllib/src/main/scala/org/apache/spark/ml/regression/Regressor.scala @@ -0,0 +1,78 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.regression + +import org.apache.spark.annotation.{DeveloperApi, AlphaComponent} +import org.apache.spark.ml.impl.estimator.{PredictionModel, Predictor, PredictorParams} + +/** + * :: DeveloperApi :: + * Params for regression. + * Currently empty, but may add functionality later. + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@DeveloperApi +private[spark] trait RegressorParams extends PredictorParams + +/** + * :: AlphaComponent :: + * + * Single-label regression + * + * @tparam FeaturesType Type of input features. E.g., [[org.apache.spark.mllib.linalg.Vector]] + * @tparam Learner Concrete Estimator type + * @tparam M Concrete Model type + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@AlphaComponent +private[spark] abstract class Regressor[ + FeaturesType, + Learner <: Regressor[FeaturesType, Learner, M], + M <: RegressionModel[FeaturesType, M]] + extends Predictor[FeaturesType, Learner, M] + with RegressorParams { + + // TODO: defaultEvaluator (follow-up PR) +} + +/** + * :: AlphaComponent :: + * + * Model produced by a [[Regressor]]. + * + * @tparam FeaturesType Type of input features. E.g., [[org.apache.spark.mllib.linalg.Vector]] + * @tparam M Concrete Model type. + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. + */ +@AlphaComponent +private[spark] abstract class RegressionModel[FeaturesType, M <: RegressionModel[FeaturesType, M]] + extends PredictionModel[FeaturesType, M] with RegressorParams { + + /** + * :: DeveloperApi :: + * + * Predict real-valued label for the given features. + * This internal method is used to implement [[transform()]] and output [[predictionCol]]. + */ + @DeveloperApi + protected def predict(features: FeaturesType): Double + +} diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala index 77785bdbd03d9..480bbfb5fe94a 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala @@ -26,6 +26,7 @@ import scala.collection.JavaConverters._ import breeze.linalg.{DenseVector => BDV, SparseVector => BSV, Vector => BV} import org.apache.spark.SparkException +import org.apache.spark.annotation.DeveloperApi import org.apache.spark.mllib.util.NumericParser import org.apache.spark.sql.Row import org.apache.spark.sql.catalyst.expressions.GenericMutableRow @@ -110,9 +111,14 @@ sealed trait Vector extends Serializable { } /** + * :: DeveloperApi :: + * * User-defined type for [[Vector]] which allows easy interaction with SQL * via [[org.apache.spark.sql.DataFrame]]. + * + * NOTE: This is currently private[spark] but will be made public later once it is stabilized. */ +@DeveloperApi private[spark] class VectorUDT extends UserDefinedType[Vector] { override def sqlType: StructType = { @@ -169,6 +175,13 @@ private[spark] class VectorUDT extends UserDefinedType[Vector] { override def pyUDT: String = "pyspark.mllib.linalg.VectorUDT" override def userClass: Class[Vector] = classOf[Vector] + + override def equals(o: Any): Boolean = { + o match { + case v: VectorUDT => true + case _ => false + } + } } /** diff --git a/mllib/src/test/java/org/apache/spark/ml/JavaPipelineSuite.java b/mllib/src/test/java/org/apache/spark/ml/JavaPipelineSuite.java index 56a9dbdd58b64..50995ffef9ad5 100644 --- a/mllib/src/test/java/org/apache/spark/ml/JavaPipelineSuite.java +++ b/mllib/src/test/java/org/apache/spark/ml/JavaPipelineSuite.java @@ -65,7 +65,7 @@ public void pipeline() { .setStages(new PipelineStage[] {scaler, lr}); PipelineModel model = pipeline.fit(dataset); model.transform(dataset).registerTempTable("prediction"); - DataFrame predictions = jsql.sql("SELECT label, score, prediction FROM prediction"); + DataFrame predictions = jsql.sql("SELECT label, probability, prediction FROM prediction"); predictions.collectAsList(); } } diff --git a/mllib/src/test/java/org/apache/spark/ml/classification/JavaLogisticRegressionSuite.java b/mllib/src/test/java/org/apache/spark/ml/classification/JavaLogisticRegressionSuite.java index f4ba23c44563e..26284023b0f69 100644 --- a/mllib/src/test/java/org/apache/spark/ml/classification/JavaLogisticRegressionSuite.java +++ b/mllib/src/test/java/org/apache/spark/ml/classification/JavaLogisticRegressionSuite.java @@ -18,17 +18,22 @@ package org.apache.spark.ml.classification; import java.io.Serializable; +import java.lang.Math; import java.util.List; import org.junit.After; import org.junit.Before; import org.junit.Test; +import org.apache.spark.api.java.JavaRDD; import org.apache.spark.api.java.JavaSparkContext; +import static org.apache.spark.mllib.classification.LogisticRegressionSuite.generateLogisticInputAsList; +import org.apache.spark.mllib.linalg.Vector; import org.apache.spark.mllib.regression.LabeledPoint; import org.apache.spark.sql.DataFrame; import org.apache.spark.sql.SQLContext; -import static org.apache.spark.mllib.classification.LogisticRegressionSuite.generateLogisticInputAsList; +import org.apache.spark.sql.Row; + public class JavaLogisticRegressionSuite implements Serializable { @@ -36,12 +41,17 @@ public class JavaLogisticRegressionSuite implements Serializable { private transient SQLContext jsql; private transient DataFrame dataset; + private transient JavaRDD datasetRDD; + private double eps = 1e-5; + @Before public void setUp() { jsc = new JavaSparkContext("local", "JavaLogisticRegressionSuite"); jsql = new SQLContext(jsc); List points = generateLogisticInputAsList(1.0, 1.0, 100, 42); - dataset = jsql.applySchema(jsc.parallelize(points, 2), LabeledPoint.class); + datasetRDD = jsc.parallelize(points, 2); + dataset = jsql.applySchema(datasetRDD, LabeledPoint.class); + dataset.registerTempTable("dataset"); } @After @@ -51,29 +61,88 @@ public void tearDown() { } @Test - public void logisticRegression() { + public void logisticRegressionDefaultParams() { LogisticRegression lr = new LogisticRegression(); + assert(lr.getLabelCol().equals("label")); LogisticRegressionModel model = lr.fit(dataset); model.transform(dataset).registerTempTable("prediction"); - DataFrame predictions = jsql.sql("SELECT label, score, prediction FROM prediction"); + DataFrame predictions = jsql.sql("SELECT label, probability, prediction FROM prediction"); predictions.collectAsList(); + // Check defaults + assert(model.getThreshold() == 0.5); + assert(model.getFeaturesCol().equals("features")); + assert(model.getPredictionCol().equals("prediction")); + assert(model.getProbabilityCol().equals("probability")); } @Test public void logisticRegressionWithSetters() { + // Set params, train, and check as many params as we can. LogisticRegression lr = new LogisticRegression() .setMaxIter(10) - .setRegParam(1.0); + .setRegParam(1.0) + .setThreshold(0.6) + .setProbabilityCol("myProbability"); LogisticRegressionModel model = lr.fit(dataset); - model.transform(dataset, model.threshold().w(0.8)) // overwrite threshold - .registerTempTable("prediction"); - DataFrame predictions = jsql.sql("SELECT label, score, prediction FROM prediction"); - predictions.collectAsList(); + assert(model.fittingParamMap().apply(lr.maxIter()) == 10); + assert(model.fittingParamMap().apply(lr.regParam()).equals(1.0)); + assert(model.fittingParamMap().apply(lr.threshold()).equals(0.6)); + assert(model.getThreshold() == 0.6); + + // Modify model params, and check that the params worked. + model.setThreshold(1.0); + model.transform(dataset).registerTempTable("predAllZero"); + DataFrame predAllZero = jsql.sql("SELECT prediction, myProbability FROM predAllZero"); + for (Row r: predAllZero.collectAsList()) { + assert(r.getDouble(0) == 0.0); + } + // Call transform with params, and check that the params worked. + model.transform(dataset, model.threshold().w(0.0), model.probabilityCol().w("myProb")) + .registerTempTable("predNotAllZero"); + DataFrame predNotAllZero = jsql.sql("SELECT prediction, myProb FROM predNotAllZero"); + boolean foundNonZero = false; + for (Row r: predNotAllZero.collectAsList()) { + if (r.getDouble(0) != 0.0) foundNonZero = true; + } + assert(foundNonZero); + + // Call fit() with new params, and check as many params as we can. + LogisticRegressionModel model2 = lr.fit(dataset, lr.maxIter().w(5), lr.regParam().w(0.1), + lr.threshold().w(0.4), lr.probabilityCol().w("theProb")); + assert(model2.fittingParamMap().apply(lr.maxIter()) == 5); + assert(model2.fittingParamMap().apply(lr.regParam()).equals(0.1)); + assert(model2.fittingParamMap().apply(lr.threshold()).equals(0.4)); + assert(model2.getThreshold() == 0.4); + assert(model2.getProbabilityCol().equals("theProb")); } + @SuppressWarnings("unchecked") @Test - public void logisticRegressionFitWithVarargs() { + public void logisticRegressionPredictorClassifierMethods() { LogisticRegression lr = new LogisticRegression(); - lr.fit(dataset, lr.maxIter().w(10), lr.regParam().w(1.0)); + LogisticRegressionModel model = lr.fit(dataset); + assert(model.numClasses() == 2); + + model.transform(dataset).registerTempTable("transformed"); + DataFrame trans1 = jsql.sql("SELECT rawPrediction, probability FROM transformed"); + for (Row row: trans1.collect()) { + Vector raw = (Vector)row.get(0); + Vector prob = (Vector)row.get(1); + assert(raw.size() == 2); + assert(prob.size() == 2); + double probFromRaw1 = 1.0 / (1.0 + Math.exp(-raw.apply(1))); + assert(Math.abs(prob.apply(1) - probFromRaw1) < eps); + assert(Math.abs(prob.apply(0) - (1.0 - probFromRaw1)) < eps); + } + + DataFrame trans2 = jsql.sql("SELECT prediction, probability FROM transformed"); + for (Row row: trans2.collect()) { + double pred = row.getDouble(0); + Vector prob = (Vector)row.get(1); + double probOfPred = prob.apply((int)pred); + for (int i = 0; i < prob.size(); ++i) { + assert(probOfPred >= prob.apply(i)); + } + } } } diff --git a/mllib/src/test/java/org/apache/spark/ml/regression/JavaLinearRegressionSuite.java b/mllib/src/test/java/org/apache/spark/ml/regression/JavaLinearRegressionSuite.java new file mode 100644 index 0000000000000..5bd616e74d86c --- /dev/null +++ b/mllib/src/test/java/org/apache/spark/ml/regression/JavaLinearRegressionSuite.java @@ -0,0 +1,89 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.regression; + +import java.io.Serializable; +import java.util.List; + +import org.junit.After; +import org.junit.Before; +import org.junit.Test; + +import org.apache.spark.api.java.JavaRDD; +import org.apache.spark.api.java.JavaSparkContext; +import static org.apache.spark.mllib.classification.LogisticRegressionSuite + .generateLogisticInputAsList; +import org.apache.spark.mllib.regression.LabeledPoint; +import org.apache.spark.sql.DataFrame; +import org.apache.spark.sql.SQLContext; + + +public class JavaLinearRegressionSuite implements Serializable { + + private transient JavaSparkContext jsc; + private transient SQLContext jsql; + private transient DataFrame dataset; + private transient JavaRDD datasetRDD; + + @Before + public void setUp() { + jsc = new JavaSparkContext("local", "JavaLinearRegressionSuite"); + jsql = new SQLContext(jsc); + List points = generateLogisticInputAsList(1.0, 1.0, 100, 42); + datasetRDD = jsc.parallelize(points, 2); + dataset = jsql.applySchema(datasetRDD, LabeledPoint.class); + dataset.registerTempTable("dataset"); + } + + @After + public void tearDown() { + jsc.stop(); + jsc = null; + } + + @Test + public void linearRegressionDefaultParams() { + LinearRegression lr = new LinearRegression(); + assert(lr.getLabelCol().equals("label")); + LinearRegressionModel model = lr.fit(dataset); + model.transform(dataset).registerTempTable("prediction"); + DataFrame predictions = jsql.sql("SELECT label, prediction FROM prediction"); + predictions.collect(); + // Check defaults + assert(model.getFeaturesCol().equals("features")); + assert(model.getPredictionCol().equals("prediction")); + } + + @Test + public void linearRegressionWithSetters() { + // Set params, train, and check as many params as we can. + LinearRegression lr = new LinearRegression() + .setMaxIter(10) + .setRegParam(1.0); + LinearRegressionModel model = lr.fit(dataset); + assert(model.fittingParamMap().apply(lr.maxIter()) == 10); + assert(model.fittingParamMap().apply(lr.regParam()).equals(1.0)); + + // Call fit() with new params, and check as many params as we can. + LinearRegressionModel model2 = + lr.fit(dataset, lr.maxIter().w(5), lr.regParam().w(0.1), lr.predictionCol().w("thePred")); + assert(model2.fittingParamMap().apply(lr.maxIter()) == 5); + assert(model2.fittingParamMap().apply(lr.regParam()).equals(0.1)); + assert(model2.getPredictionCol().equals("thePred")); + } +} diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala index 33e40dc7410cc..b3d1bfcfbee0f 100644 --- a/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala +++ b/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala @@ -20,44 +20,108 @@ package org.apache.spark.ml.classification import org.scalatest.FunSuite import org.apache.spark.mllib.classification.LogisticRegressionSuite.generateLogisticInput +import org.apache.spark.mllib.linalg.Vector import org.apache.spark.mllib.util.MLlibTestSparkContext -import org.apache.spark.sql.{SQLContext, DataFrame} +import org.apache.spark.mllib.util.TestingUtils._ +import org.apache.spark.sql.{DataFrame, Row, SQLContext} + class LogisticRegressionSuite extends FunSuite with MLlibTestSparkContext { @transient var sqlContext: SQLContext = _ @transient var dataset: DataFrame = _ + private val eps: Double = 1e-5 override def beforeAll(): Unit = { super.beforeAll() sqlContext = new SQLContext(sc) dataset = sqlContext.createDataFrame( - sc.parallelize(generateLogisticInput(1.0, 1.0, 100, 42), 2)) + sc.parallelize(generateLogisticInput(1.0, 1.0, nPoints = 100, seed = 42), 2)) } - test("logistic regression") { + test("logistic regression: default params") { val lr = new LogisticRegression + assert(lr.getLabelCol == "label") + assert(lr.getFeaturesCol == "features") + assert(lr.getPredictionCol == "prediction") + assert(lr.getRawPredictionCol == "rawPrediction") + assert(lr.getProbabilityCol == "probability") val model = lr.fit(dataset) model.transform(dataset) - .select("label", "prediction") + .select("label", "probability", "prediction", "rawPrediction") .collect() + assert(model.getThreshold === 0.5) + assert(model.getFeaturesCol == "features") + assert(model.getPredictionCol == "prediction") + assert(model.getRawPredictionCol == "rawPrediction") + assert(model.getProbabilityCol == "probability") } test("logistic regression with setters") { + // Set params, train, and check as many params as we can. val lr = new LogisticRegression() .setMaxIter(10) .setRegParam(1.0) + .setThreshold(0.6) + .setProbabilityCol("myProbability") val model = lr.fit(dataset) - model.transform(dataset, model.threshold -> 0.8) // overwrite threshold - .select("label", "score", "prediction") + assert(model.fittingParamMap.get(lr.maxIter) === Some(10)) + assert(model.fittingParamMap.get(lr.regParam) === Some(1.0)) + assert(model.fittingParamMap.get(lr.threshold) === Some(0.6)) + assert(model.getThreshold === 0.6) + + // Modify model params, and check that the params worked. + model.setThreshold(1.0) + val predAllZero = model.transform(dataset) + .select("prediction", "myProbability") .collect() + .map { case Row(pred: Double, prob: Vector) => pred } + assert(predAllZero.forall(_ === 0), + s"With threshold=1.0, expected predictions to be all 0, but only" + + s" ${predAllZero.count(_ === 0)} of ${dataset.count()} were 0.") + // Call transform with params, and check that the params worked. + val predNotAllZero = + model.transform(dataset, model.threshold -> 0.0, model.probabilityCol -> "myProb") + .select("prediction", "myProb") + .collect() + .map { case Row(pred: Double, prob: Vector) => pred } + assert(predNotAllZero.exists(_ !== 0.0)) + + // Call fit() with new params, and check as many params as we can. + val model2 = lr.fit(dataset, lr.maxIter -> 5, lr.regParam -> 0.1, lr.threshold -> 0.4, + lr.probabilityCol -> "theProb") + assert(model2.fittingParamMap.get(lr.maxIter).get === 5) + assert(model2.fittingParamMap.get(lr.regParam).get === 0.1) + assert(model2.fittingParamMap.get(lr.threshold).get === 0.4) + assert(model2.getThreshold === 0.4) + assert(model2.getProbabilityCol == "theProb") } - test("logistic regression fit and transform with varargs") { + test("logistic regression: Predictor, Classifier methods") { + val sqlContext = this.sqlContext val lr = new LogisticRegression - val model = lr.fit(dataset, lr.maxIter -> 10, lr.regParam -> 1.0) - model.transform(dataset, model.threshold -> 0.8, model.scoreCol -> "probability") - .select("label", "probability", "prediction") - .collect() + + val model = lr.fit(dataset) + assert(model.numClasses === 2) + + val threshold = model.getThreshold + val results = model.transform(dataset) + + // Compare rawPrediction with probability + results.select("rawPrediction", "probability").collect().map { + case Row(raw: Vector, prob: Vector) => + assert(raw.size === 2) + assert(prob.size === 2) + val probFromRaw1 = 1.0 / (1.0 + math.exp(-raw(1))) + assert(prob(1) ~== probFromRaw1 relTol eps) + assert(prob(0) ~== 1.0 - probFromRaw1 relTol eps) + } + + // Compare prediction with probability + results.select("prediction", "probability").collect().map { + case Row(pred: Double, prob: Vector) => + val predFromProb = prob.toArray.zipWithIndex.maxBy(_._1)._2 + assert(pred == predFromProb) + } } } diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala new file mode 100644 index 0000000000000..bbb44c3e2dfc2 --- /dev/null +++ b/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala @@ -0,0 +1,65 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.regression + +import org.scalatest.FunSuite + +import org.apache.spark.mllib.classification.LogisticRegressionSuite.generateLogisticInput +import org.apache.spark.mllib.util.MLlibTestSparkContext +import org.apache.spark.sql.{DataFrame, SQLContext} + +class LinearRegressionSuite extends FunSuite with MLlibTestSparkContext { + + @transient var sqlContext: SQLContext = _ + @transient var dataset: DataFrame = _ + + override def beforeAll(): Unit = { + super.beforeAll() + sqlContext = new SQLContext(sc) + dataset = sqlContext.createDataFrame( + sc.parallelize(generateLogisticInput(1.0, 1.0, nPoints = 100, seed = 42), 2)) + } + + test("linear regression: default params") { + val lr = new LinearRegression + assert(lr.getLabelCol == "label") + val model = lr.fit(dataset) + model.transform(dataset) + .select("label", "prediction") + .collect() + // Check defaults + assert(model.getFeaturesCol == "features") + assert(model.getPredictionCol == "prediction") + } + + test("linear regression with setters") { + // Set params, train, and check as many as we can. + val lr = new LinearRegression() + .setMaxIter(10) + .setRegParam(1.0) + val model = lr.fit(dataset) + assert(model.fittingParamMap.get(lr.maxIter).get === 10) + assert(model.fittingParamMap.get(lr.regParam).get === 1.0) + + // Call fit() with new params, and check as many as we can. + val model2 = lr.fit(dataset, lr.maxIter -> 5, lr.regParam -> 0.1, lr.predictionCol -> "thePred") + assert(model2.fittingParamMap.get(lr.maxIter).get === 5) + assert(model2.fittingParamMap.get(lr.regParam).get === 0.1) + assert(model2.getPredictionCol == "thePred") + } +} diff --git a/project/MimaExcludes.scala b/project/MimaExcludes.scala index b17532c1d814c..4065a562a1a18 100644 --- a/project/MimaExcludes.scala +++ b/project/MimaExcludes.scala @@ -36,6 +36,7 @@ object MimaExcludes { case v if v.startsWith("1.3") => Seq( MimaBuild.excludeSparkPackage("deploy"), + MimaBuild.excludeSparkPackage("ml"), // These are needed if checking against the sbt build, since they are part of // the maven-generated artifacts in the 1.2 build. MimaBuild.excludeSparkPackage("unused"), @@ -142,6 +143,11 @@ object MimaExcludes { "org.apache.spark.graphx.Graph.getCheckpointFiles"), ProblemFilters.exclude[MissingMethodProblem]( "org.apache.spark.graphx.Graph.isCheckpointed") + ) ++ Seq( + // SPARK-4789 Standardize ML Prediction APIs + ProblemFilters.exclude[MissingTypesProblem]("org.apache.spark.mllib.linalg.VectorUDT"), + ProblemFilters.exclude[IncompatibleResultTypeProblem]("org.apache.spark.mllib.linalg.VectorUDT.serialize"), + ProblemFilters.exclude[IncompatibleResultTypeProblem]("org.apache.spark.mllib.linalg.VectorUDT.sqlType") ) case v if v.startsWith("1.2") => From 575d2df350e32964fffb02536cb0587a92ef20e9 Mon Sep 17 00:00:00 2001 From: Leolh Date: Fri, 6 Feb 2015 09:01:53 +0000 Subject: [PATCH 04/53] [SPARK-5380][GraphX] Solve an ArrayIndexOutOfBoundsException when build graph with a file format error When I build a graph with a file format error, there will be an ArrayIndexOutOfBoundsException Author: Leolh Closes #4176 from Leolh/patch-1 and squashes the following commits: 94f6d22 [Leolh] Update GraphLoader.scala 23767f1 [Leolh] [SPARK-3650][GraphX] There will be an ArrayIndexOutOfBoundsException if the format of the source file is wrong --- graphx/src/main/scala/org/apache/spark/graphx/GraphLoader.scala | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/graphx/src/main/scala/org/apache/spark/graphx/GraphLoader.scala b/graphx/src/main/scala/org/apache/spark/graphx/GraphLoader.scala index 4933aecba1286..21187be7678a6 100644 --- a/graphx/src/main/scala/org/apache/spark/graphx/GraphLoader.scala +++ b/graphx/src/main/scala/org/apache/spark/graphx/GraphLoader.scala @@ -77,7 +77,7 @@ object GraphLoader extends Logging { if (!line.isEmpty && line(0) != '#') { val lineArray = line.split("\\s+") if (lineArray.length < 2) { - logWarning("Invalid line: " + line) + throw new IllegalArgumentException("Invalid line: " + line) } val srcId = lineArray(0).toLong val dstId = lineArray(1).toLong From f6ba813af21f8aa277155b789a7330e356ce15da Mon Sep 17 00:00:00 2001 From: Kousuke Saruta Date: Fri, 6 Feb 2015 09:33:36 +0000 Subject: [PATCH 05/53] [Minor] Remove permission for execution from spark-shell.cmd .cmd files in bin is not set permission for execution except for spark-shell.cmd. Let's unify that. Author: Kousuke Saruta Closes #3983 from sarutak/fix-mode-of-cmd and squashes the following commits: 9d6eedc [Kousuke Saruta] Removed permission for execution from spark-shell.cmd --- bin/spark-shell.cmd | 0 1 file changed, 0 insertions(+), 0 deletions(-) mode change 100755 => 100644 bin/spark-shell.cmd diff --git a/bin/spark-shell.cmd b/bin/spark-shell.cmd old mode 100755 new mode 100644 From 24dbc50b934bfcd9fc3fd91cfd9966b75b250ad7 Mon Sep 17 00:00:00 2001 From: Kousuke Saruta Date: Fri, 6 Feb 2015 09:39:12 +0000 Subject: [PATCH 06/53] [SPARK-5157][YARN] Configure more JVM options properly when we use ConcMarkSweepGC for AM. When we set `SPARK_USE_CONC_INCR_GC`, ConcurrentMarkSweepGC works on the AM. Actually, if ConcurrentMarkSweepGC is set for the JVM, following JVM options are set automatically and implicitly. * MaxTenuringThreshold=0 * SurvivorRatio=1024 Those can not be proper value for most cases. See also http://www.oracle.com/technetwork/java/tuning-139912.html Author: Kousuke Saruta Closes #3956 from sarutak/SPARK-5157 and squashes the following commits: c15da4e [Kousuke Saruta] Set more JVM options for AM when enabling CMS --- yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala | 2 ++ 1 file changed, 2 insertions(+) diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala index 91e8574e94e2f..e7005094b5f3c 100644 --- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala +++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala @@ -418,6 +418,8 @@ private[spark] class Client( // In our expts, using (default) throughput collector has severe perf ramifications in // multi-tenant machines javaOpts += "-XX:+UseConcMarkSweepGC" + javaOpts += "-XX:MaxTenuringThreshold=31" + javaOpts += "-XX:SurvivorRatio=8" javaOpts += "-XX:+CMSIncrementalMode" javaOpts += "-XX:+CMSIncrementalPacing" javaOpts += "-XX:CMSIncrementalDutyCycleMin=0" From 856928979f7d00fbb518fc1102a2e7d80cea0f7c Mon Sep 17 00:00:00 2001 From: Marcelo Vanzin Date: Fri, 6 Feb 2015 10:07:20 +0000 Subject: [PATCH 07/53] [SPARK-5582] [history] Ignore empty log directories. Empty log directories are not useful at the moment, but if one ends up showing in the log root, it breaks the code that checks for log directories. Author: Marcelo Vanzin Closes #4352 from vanzin/SPARK-5582 and squashes the following commits: 1a6a3d4 [Marcelo Vanzin] [SPARK-5582] Fix exception when looking at empty directories. --- .../deploy/history/FsHistoryProvider.scala | 22 ++++++++++++------- .../history/FsHistoryProviderSuite.scala | 18 +++++++++++++++ 2 files changed, 32 insertions(+), 8 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala b/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala index 0ae45f4ad9130..92125f2df7d10 100644 --- a/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala +++ b/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala @@ -173,9 +173,10 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis val logInfos = statusList .filter { entry => try { - val modTime = getModificationTime(entry) - newLastModifiedTime = math.max(newLastModifiedTime, modTime) - modTime >= lastModifiedTime + getModificationTime(entry).map { time => + newLastModifiedTime = math.max(newLastModifiedTime, time) + time >= lastModifiedTime + }.getOrElse(false) } catch { case e: AccessControlException => // Do not use "logInfo" since these messages can get pretty noisy if printed on @@ -251,7 +252,7 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis appListener.appName.getOrElse(NOT_STARTED), appListener.startTime.getOrElse(-1L), appListener.endTime.getOrElse(-1L), - getModificationTime(eventLog), + getModificationTime(eventLog).get, appListener.sparkUser.getOrElse(NOT_STARTED), isApplicationCompleted(eventLog)) } finally { @@ -310,11 +311,16 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis */ private def isLegacyLogDirectory(entry: FileStatus): Boolean = entry.isDir() - private def getModificationTime(fsEntry: FileStatus): Long = { - if (fsEntry.isDir) { - fs.listStatus(fsEntry.getPath).map(_.getModificationTime()).max + /** + * Returns the modification time of the given event log. If the status points at an empty + * directory, `None` is returned, indicating that there isn't an event log at that location. + */ + private def getModificationTime(fsEntry: FileStatus): Option[Long] = { + if (isLegacyLogDirectory(fsEntry)) { + val statusList = fs.listStatus(fsEntry.getPath) + if (!statusList.isEmpty) Some(statusList.map(_.getModificationTime()).max) else None } else { - fsEntry.getModificationTime() + Some(fsEntry.getModificationTime()) } } diff --git a/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala b/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala index 3fbc1a21d10ed..1d95432258111 100644 --- a/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala +++ b/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala @@ -190,6 +190,24 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers appListAfterRename.head.logPath should not endWith(EventLoggingListener.IN_PROGRESS) } + test("SPARK-5582: empty log directory") { + val conf = new SparkConf() + .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) + val provider = new FsHistoryProvider(conf) + + val logFile1 = new File(testDir, "app1" + EventLoggingListener.IN_PROGRESS) + writeFile(logFile1, true, None, + SparkListenerApplicationStart("app1", Some("app1"), 1L, "test"), + SparkListenerApplicationEnd(2L)) + + val oldLog = new File(testDir, "old1") + oldLog.mkdir() + + provider.checkForLogs() + val appListAfterRename = provider.getListing() + appListAfterRename.size should be (1) + } + private def writeFile(file: File, isNewFormat: Boolean, codec: Option[CompressionCodec], events: SparkListenerEvent*) = { val out = From ed3aac791232560d2cfe25e3304e48873d48b338 Mon Sep 17 00:00:00 2001 From: lianhuiwang Date: Fri, 6 Feb 2015 11:00:35 +0000 Subject: [PATCH 08/53] [SPARK-5470][Core]use defaultClassLoader to load classes in KryoSerializer Now KryoSerializer load classes of classesToRegister at the time of its initialization. when we set spark.kryo.classesToRegister=class1, it will throw SparkException("Failed to load class to register with Kryo". because in KryoSerializer's initialization, classLoader cannot include class of user's jars. we need to use defaultClassLoader of Serializer in newKryo(), because executor will reset defaultClassLoader of Serializer after Serializer's initialization. thank zzcclp for reporting it to me. Author: lianhuiwang Closes #4258 from lianhuiwang/SPARK-5470 and squashes the following commits: 73b719f [lianhuiwang] do the splitting and filtering during initialization 64cf306 [lianhuiwang] use defaultClassLoader to load classes of classesToRegister in KryoSerializer --- .../org/apache/spark/serializer/KryoSerializer.scala | 11 ++--------- 1 file changed, 2 insertions(+), 9 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/serializer/KryoSerializer.scala b/core/src/main/scala/org/apache/spark/serializer/KryoSerializer.scala index d56e23ce4478a..02158aa0f866e 100644 --- a/core/src/main/scala/org/apache/spark/serializer/KryoSerializer.scala +++ b/core/src/main/scala/org/apache/spark/serializer/KryoSerializer.scala @@ -58,14 +58,6 @@ class KryoSerializer(conf: SparkConf) private val classesToRegister = conf.get("spark.kryo.classesToRegister", "") .split(',') .filter(!_.isEmpty) - .map { className => - try { - Class.forName(className) - } catch { - case e: Exception => - throw new SparkException("Failed to load class to register with Kryo", e) - } - } def newKryoOutput() = new KryoOutput(bufferSize, math.max(bufferSize, maxBufferSize)) @@ -97,7 +89,8 @@ class KryoSerializer(conf: SparkConf) // Use the default classloader when calling the user registrator. Thread.currentThread.setContextClassLoader(classLoader) // Register classes given through spark.kryo.classesToRegister. - classesToRegister.foreach { clazz => kryo.register(clazz) } + classesToRegister + .foreach { className => kryo.register(Class.forName(className, true, classLoader)) } // Allow the user to register their own classes by setting spark.kryo.registrator. userRegistrator .map(Class.forName(_, true, classLoader).newInstance().asInstanceOf[KryoRegistrator]) From f827ef4d7ead85aab65837ea625141e05e917c2e Mon Sep 17 00:00:00 2001 From: Miguel Peralvo Date: Fri, 6 Feb 2015 11:04:48 +0000 Subject: [PATCH 09/53] Update ec2-scripts.md Change spark-version from 1.1.0 to 1.2.0 in the example for spark-ec2/Launch Cluster. Author: Miguel Peralvo Closes #4300 from MiguelPeralvo/patch-1 and squashes the following commits: 38adf0b [Miguel Peralvo] Update ec2-scripts.md 1850869 [Miguel Peralvo] Update ec2-scripts.md --- docs/ec2-scripts.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/ec2-scripts.md b/docs/ec2-scripts.md index d50f445d7ecc7..8c9a1e1262d8f 100644 --- a/docs/ec2-scripts.md +++ b/docs/ec2-scripts.md @@ -52,7 +52,7 @@ identify machines belonging to each cluster in the Amazon EC2 Console. ```bash export AWS_SECRET_ACCESS_KEY=AaBbCcDdEeFGgHhIiJjKkLlMmNnOoPpQqRrSsTtU export AWS_ACCESS_KEY_ID=ABCDEFG1234567890123 -./spark-ec2 --key-pair=awskey --identity-file=awskey.pem --region=us-west-1 --zone=us-west-1a --spark-version=1.1.0 launch my-spark-cluster +./spark-ec2 --key-pair=awskey --identity-file=awskey.pem --region=us-west-1 --zone=us-west-1a launch my-spark-cluster ``` - After everything launches, check that the cluster scheduler is up and sees From cf6778e8d8e8d7e0b4e7b17637ad812624a937dd Mon Sep 17 00:00:00 2001 From: Nicholas Chammas Date: Fri, 6 Feb 2015 11:38:39 +0000 Subject: [PATCH 10/53] [Build] Set all Debian package permissions to 755 755 means the owner can read, write, and execute, and everyone else can just read and execute. I think that's what we want here since without execute permissions others cannot open directories. Inspired by [this comment on a separate PR](https://github.com/apache/spark/pull/3297#issuecomment-63286730). Author: Nicholas Chammas Closes #4277 from nchammas/patch-1 and squashes the following commits: da77fb0 [Nicholas Chammas] [Build] Set all Debian package permissions to 755 --- assembly/pom.xml | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/assembly/pom.xml b/assembly/pom.xml index 1bb5a671f5390..301ff69c2ae3b 100644 --- a/assembly/pom.xml +++ b/assembly/pom.xml @@ -39,7 +39,7 @@ spark /usr/share/spark root - 744 + 755 @@ -280,7 +280,7 @@ ${deb.user} ${deb.user} ${deb.install.path}/conf - 744 + ${deb.bin.filemode} @@ -302,7 +302,7 @@ ${deb.user} ${deb.user} ${deb.install.path}/sbin - 744 + ${deb.bin.filemode} @@ -313,7 +313,7 @@ ${deb.user} ${deb.user} ${deb.install.path}/python - 744 + ${deb.bin.filemode} From 37d35ab53b82916e4bf0eeb788044bfc33f329df Mon Sep 17 00:00:00 2001 From: Ryan Williams Date: Fri, 6 Feb 2015 12:22:25 +0000 Subject: [PATCH 11/53] [SPARK-5416] init Executor.threadPool before ExecutorSource Some ExecutorSource metrics can NPE by attempting to reference the threadpool otherwise. Author: Ryan Williams Closes #4212 from ryan-williams/threadpool and squashes the following commits: 236f2ad [Ryan Williams] init Executor.threadPool before ExecutorSource --- .../src/main/scala/org/apache/spark/executor/Executor.scala | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/executor/Executor.scala b/core/src/main/scala/org/apache/spark/executor/Executor.scala index 312bb3a1daaa3..5141483d1e745 100644 --- a/core/src/main/scala/org/apache/spark/executor/Executor.scala +++ b/core/src/main/scala/org/apache/spark/executor/Executor.scala @@ -75,6 +75,9 @@ private[spark] class Executor( Thread.setDefaultUncaughtExceptionHandler(SparkUncaughtExceptionHandler) } + // Start worker thread pool + val threadPool = Utils.newDaemonCachedThreadPool("Executor task launch worker") + val executorSource = new ExecutorSource(this, executorId) if (!isLocal) { @@ -101,9 +104,6 @@ private[spark] class Executor( // Limit of bytes for total size of results (default is 1GB) private val maxResultSize = Utils.getMaxResultSize(conf) - // Start worker thread pool - val threadPool = Utils.newDaemonCachedThreadPool("Executor task launch worker") - // Maintains the list of running tasks. private val runningTasks = new ConcurrentHashMap[Long, TaskRunner] From 9ad56ad2a2a51df449040c4f4b7c66b104883312 Mon Sep 17 00:00:00 2001 From: Travis Galoppo Date: Fri, 6 Feb 2015 10:26:51 -0800 Subject: [PATCH 12/53] [SPARK-5013] [MLlib] Added documentation and sample data file for GaussianMixture Simple description and code samples (and sample data) for GaussianMixture Author: Travis Galoppo Closes #4401 from tgaloppo/spark-5013 and squashes the following commits: c9ff9a5 [Travis Galoppo] Fixed link in mllib-clustering.md Added Gaussian mixture and power iteration as available clustering techniques in mllib-guide 2368690 [Travis Galoppo] Minor fixes 3eb41fa [Travis Galoppo] [SPARK-5013] Added documentation and sample data file for GaussianMixture --- data/mllib/gmm_data.txt | 2000 ++++++++++++++++++++++++++++++++++++++ docs/mllib-clustering.md | 131 ++- docs/mllib-guide.md | 2 + 3 files changed, 2130 insertions(+), 3 deletions(-) create mode 100644 data/mllib/gmm_data.txt diff --git a/data/mllib/gmm_data.txt b/data/mllib/gmm_data.txt new file mode 100644 index 0000000000000..934ee4a83a2df --- /dev/null +++ b/data/mllib/gmm_data.txt @@ -0,0 +1,2000 @@ + 2.59470454e+00 2.12298217e+00 + 1.15807024e+00 -1.46498723e-01 + 2.46206638e+00 6.19556894e-01 + -5.54845070e-01 -7.24700066e-01 + -3.23111426e+00 -1.42579084e+00 + 3.02978115e+00 7.87121753e-01 + 1.97365907e+00 1.15914704e+00 + -6.44852101e+00 -3.18154314e+00 + 1.30963349e+00 1.62866434e-01 + 4.26482541e+00 2.15547996e+00 + 3.79927257e+00 1.50572445e+00 + 4.17452609e-01 -6.74032760e-01 + 4.21117627e-01 4.45590255e-01 + -2.80425571e+00 -7.77150554e-01 + 2.55928797e+00 7.03954218e-01 + 1.32554059e+00 -9.46663152e-01 + -3.39691439e+00 -1.49005743e+00 + -2.26542270e-01 3.60052515e-02 + 1.04994198e+00 5.29825685e-01 + -1.51566882e+00 -1.86264432e-01 + -3.27928172e-01 -7.60859110e-01 + -3.18054866e-01 3.97719805e-01 + 1.65579418e-01 -3.47232033e-01 + 6.47162333e-01 4.96059961e-02 + -2.80776647e-01 4.79418757e-01 + 7.45069752e-01 1.20790281e-01 + 2.13604102e-01 1.59542555e-01 + -3.08860224e+00 -1.43259870e+00 + 8.97066497e-01 1.10206801e+00 + -2.23918874e-01 -1.07267267e+00 + 2.51525708e+00 2.84761973e-01 + 9.98052532e-01 1.08333783e+00 + 1.76705588e+00 8.18866778e-01 + 5.31555163e-02 -1.90111151e-01 + -2.17405059e+00 7.21854582e-02 + -2.13772505e+00 -3.62010387e-01 + 2.95974057e+00 1.31602381e+00 + 2.74053561e+00 1.61781757e+00 + 6.68135448e-01 2.86586009e-01 + 2.82323739e+00 1.74437257e+00 + 8.11540288e-01 5.50744478e-01 + 4.10050897e-01 5.10668402e-03 + 9.58626136e-01 -3.49633680e-01 + 4.66599798e+00 1.49964894e+00 + 4.94507794e-01 2.58928077e-01 + -2.36029742e+00 -1.61042909e+00 + -4.99306804e-01 -8.04984769e-01 + 1.07448510e+00 9.39605828e-01 + -1.80448949e+00 -1.05983264e+00 + -3.22353821e-01 1.73612093e-01 + 1.85418702e+00 1.15640643e+00 + 6.93794163e-01 6.59993560e-01 + 1.99399102e+00 1.44547123e+00 + 3.38866124e+00 1.23379290e+00 + -4.24067720e+00 -1.22264282e+00 + 6.03230201e-02 2.95232729e-01 + -3.59341813e+00 -7.17453726e-01 + 4.87447372e-01 -2.00733911e-01 + 1.20149195e+00 4.07880197e-01 + -2.13331464e+00 -4.58518077e-01 + -3.84091083e+00 -1.71553950e+00 + -5.37279250e-01 2.64822629e-02 + -2.10155227e+00 -1.32558103e+00 + -1.71318897e+00 -7.12098563e-01 + -1.46280695e+00 -1.84868337e-01 + -3.59785325e+00 -1.54832434e+00 + -5.77528081e-01 -5.78580857e-01 + 3.14734283e-01 5.80184639e-01 + -2.71164714e+00 -1.19379432e+00 + 1.09634489e+00 7.20143887e-01 + -3.05527722e+00 -1.47774064e+00 + 6.71753586e-01 7.61350020e-01 + 3.98294144e+00 1.54166484e+00 + -3.37220384e+00 -2.21332064e+00 + 1.81222914e+00 7.41212752e-01 + 2.71458282e-01 1.36329078e-01 + -3.97815359e-01 1.16766886e-01 + -1.70192814e+00 -9.75851571e-01 + -3.46803804e+00 -1.09965988e+00 + -1.69649627e+00 -5.76045801e-01 + -1.02485636e-01 -8.81841246e-01 + -3.24194667e-02 2.55429276e-01 + -2.75343168e+00 -1.51366320e+00 + -2.78676702e+00 -5.22360489e-01 + 1.70483164e+00 1.19769805e+00 + 4.92022579e-01 3.24944706e-01 + 2.48768464e+00 1.00055363e+00 + 4.48786400e-01 7.63902870e-01 + 2.93862696e+00 1.73809968e+00 + -3.55019305e+00 -1.97875558e+00 + 1.74270784e+00 6.90229224e-01 + 5.13391994e-01 4.58374016e-01 + 1.78379499e+00 9.08026381e-01 + 1.75814147e+00 7.41449784e-01 + -2.30687792e-01 3.91009729e-01 + 3.92271353e+00 1.44006290e+00 + 2.93361679e-01 -4.99886375e-03 + 2.47902690e-01 -7.49542503e-01 + -3.97675355e-01 1.36824887e-01 + 3.56535953e+00 1.15181329e+00 + 3.22425301e+00 1.28702383e+00 + -2.94192478e-01 -2.42382557e-01 + 8.02068864e-01 -1.51671475e-01 + 8.54133530e-01 -4.89514885e-02 + -1.64316316e-01 -5.34642346e-01 + -6.08485405e-01 -2.10332352e-01 + -2.18940059e+00 -1.07024952e+00 + -1.71586960e+00 -2.83333492e-02 + 1.70200448e-01 -3.28031178e-01 + -1.97210346e+00 -5.39948532e-01 + 2.19500160e+00 1.05697170e+00 + -1.76239935e+00 -1.09377438e+00 + 1.68314744e+00 6.86491164e-01 + -2.99852288e+00 -1.46619067e+00 + -2.23769560e+00 -9.15008355e-01 + 9.46887516e-01 5.58410503e-01 + 5.02153123e-01 1.63851235e-01 + -9.70297062e-01 3.14625374e-01 + -1.29405593e+00 -8.20994131e-01 + 2.72516079e+00 7.85839947e-01 + 1.45788024e+00 3.37487353e-01 + -4.36292749e-01 -5.42150480e-01 + 2.21304711e+00 1.25254042e+00 + -1.20810271e-01 4.79632898e-01 + -3.30884511e+00 -1.50607586e+00 + -6.55882455e+00 -1.94231256e+00 + -3.17033630e+00 -9.94678930e-01 + 1.42043617e+00 7.28808957e-01 + -1.57546099e+00 -1.10320497e+00 + -3.22748754e+00 -1.64174579e+00 + 2.96776017e-03 -3.16191512e-02 + -2.25986054e+00 -6.13123197e-01 + 2.49434243e+00 7.73069183e-01 + 9.08494049e-01 -1.53926853e-01 + -2.80559090e+00 -1.37474221e+00 + 4.75224286e-01 2.53153674e-01 + 4.37644006e+00 8.49116998e-01 + 2.27282959e+00 6.16568202e-01 + 1.16006880e+00 1.65832798e-01 + -1.67163193e+00 -1.22555386e+00 + -1.38231118e+00 -7.29575504e-01 + -3.49922750e+00 -2.26446675e+00 + -3.73780110e-01 -1.90657869e-01 + 1.68627679e+00 1.05662987e+00 + -3.28891792e+00 -1.11080334e+00 + -2.59815798e+00 -1.51410198e+00 + -2.61203309e+00 -6.00143552e-01 + 6.58964943e-01 4.47216094e-01 + -2.26711381e+00 -7.26512923e-01 + -5.31429009e-02 -1.97925341e-02 + 3.19749807e+00 9.20425476e-01 + -1.37595787e+00 -6.58062732e-01 + 8.09900278e-01 -3.84286160e-01 + -5.07741280e+00 -1.97683808e+00 + -2.99764250e+00 -1.50753777e+00 + -9.87671815e-01 -4.63255889e-01 + 1.65390765e+00 6.73806615e-02 + 5.51252659e+00 2.69842267e+00 + -2.23724309e+00 -4.77624004e-01 + 4.99726228e+00 1.74690949e+00 + 1.75859162e-01 -1.49350995e-01 + 4.13382789e+00 1.31735161e+00 + 2.69058117e+00 4.87656923e-01 + 1.07180318e+00 1.01426954e+00 + 3.37216869e+00 1.05955377e+00 + -2.95006781e+00 -1.57048303e+00 + -2.46401648e+00 -8.37056374e-01 + 1.19012962e-01 7.54702770e-01 + 3.34142539e+00 4.81938295e-01 + 2.92643913e+00 1.04301050e+00 + 2.89697751e+00 1.37551442e+00 + -1.03094242e+00 2.20903962e-01 + -5.13914589e+00 -2.23355387e+00 + -8.81680780e-01 1.83590000e-01 + 2.82334775e+00 1.26650464e+00 + -2.81042540e-01 -3.26370240e-01 + 2.97995487e+00 8.34569452e-01 + -1.39857135e+00 -1.15798385e+00 + 4.27186506e+00 9.04253702e-01 + 6.98684517e-01 7.91167305e-01 + 3.52233095e+00 1.29976473e+00 + 2.21448029e+00 2.73213379e-01 + -3.13505683e-01 -1.20593774e-01 + 3.70571571e+00 1.06220876e+00 + 9.83881041e-01 5.67713803e-01 + -2.17897705e+00 2.52925205e-01 + 1.38734039e+00 4.61287066e-01 + -1.41181602e+00 -1.67248955e-02 + -1.69974639e+00 -7.17812071e-01 + -2.01005793e-01 -7.49662056e-01 + 1.69016336e+00 3.24687979e-01 + -2.03250179e+00 -2.76108460e-01 + 3.68776848e-01 4.12536941e-01 + 7.66238259e-01 -1.84750637e-01 + -2.73989147e-01 -1.72817250e-01 + -2.18623745e+00 -2.10906798e-01 + -1.39795625e-01 3.26066094e-02 + -2.73826912e-01 -6.67586097e-02 + -1.57880654e+00 -4.99395900e-01 + 4.55950908e+00 2.29410489e+00 + -7.36479631e-01 -1.57861857e-01 + 1.92082888e+00 1.05843391e+00 + 4.29192810e+00 1.38127810e+00 + 1.61852879e+00 1.95871986e-01 + -1.95027403e+00 -5.22448168e-01 + -1.67446281e+00 -9.41497162e-01 + 6.07097859e-01 3.44178029e-01 + -3.44004683e+00 -1.49258461e+00 + 2.72114752e+00 6.00728991e-01 + 8.80685522e-01 -2.53243336e-01 + 1.39254928e+00 3.42988512e-01 + 1.14194836e-01 -8.57945694e-02 + -1.49387332e+00 -7.60860481e-01 + -1.98053285e+00 -4.86039865e-01 + 3.56008568e+00 1.08438692e+00 + 2.27833961e-01 1.09441881e+00 + -1.16716710e+00 -6.54778242e-01 + 2.02156613e+00 5.42075758e-01 + 1.08429178e+00 -7.67420693e-01 + 6.63058455e-01 4.61680991e-01 + -1.06201537e+00 1.38862846e-01 + 3.08701875e+00 8.32580273e-01 + -4.96558108e-01 -2.47031257e-01 + 7.95109987e-01 7.59314147e-02 + -3.39903524e-01 8.71565566e-03 + 8.68351357e-01 4.78358641e-01 + 1.48750819e+00 7.63257420e-01 + -4.51224101e-01 -4.44056898e-01 + -3.02734750e-01 -2.98487961e-01 + 5.46846609e-01 7.02377629e-01 + 1.65129778e+00 3.74008231e-01 + -7.43336512e-01 3.95723531e-01 + -5.88446605e-01 -6.47520211e-01 + 3.58613167e+00 1.95024937e+00 + 3.11718883e+00 8.37984715e-01 + 1.80919244e+00 9.62644986e-01 + 5.43856371e-02 -5.86297543e-01 + -1.95186766e+00 -1.02624212e-01 + 8.95628057e-01 5.91812281e-01 + 4.97691627e-02 5.31137156e-01 + -1.07633113e+00 -2.47392788e-01 + -1.17257986e+00 -8.68528265e-01 + -8.19227665e-02 5.80579434e-03 + -2.86409787e-01 1.95812924e-01 + 1.10582671e+00 7.42853240e-01 + 4.06429774e+00 1.06557476e+00 + -3.42521792e+00 -7.74327139e-01 + 1.28468671e+00 6.20431661e-01 + 6.01201008e-01 -1.16799728e-01 + -1.85058727e-01 -3.76235293e-01 + 5.44083324e+00 2.98490868e+00 + 2.69273070e+00 7.83901153e-01 + 1.88938036e-01 -4.83222152e-01 + 1.05667256e+00 -2.57003165e-01 + 2.99711662e-01 -4.33131912e-01 + 7.73689216e-02 -1.78738364e-01 + 9.58326279e-01 6.38325706e-01 + -3.97727049e-01 2.27314759e-01 + 3.36098175e+00 1.12165237e+00 + 1.77804871e+00 6.46961933e-01 + -2.86945546e+00 -1.00395518e+00 + 3.03494815e+00 7.51814612e-01 + -1.43658194e+00 -3.55432244e-01 + -3.08455105e+00 -1.51535106e+00 + -1.55841975e+00 3.93454820e-02 + 7.96073412e-01 -3.11036969e-01 + -9.84125401e-01 -1.02064649e+00 + -7.75688143e+00 -3.65219926e+00 + 1.53816429e+00 7.65926670e-01 + -4.92712738e-01 2.32244240e-02 + -1.93166919e+00 -1.07701304e+00 + 2.03029875e-02 -7.54055699e-01 + 2.52177489e+00 1.01544979e+00 + 3.65109048e-01 -9.48328494e-01 + -1.28849143e-01 2.51947174e-01 + -1.02428075e+00 -9.37767116e-01 + -3.04179748e+00 -9.97926994e-01 + -2.51986980e+00 -1.69117413e+00 + -1.24900838e+00 -4.16179917e-01 + 2.77943992e+00 1.22842327e+00 + -4.37434557e+00 -1.70182693e+00 + -1.60019319e+00 -4.18345639e-01 + -1.67613646e+00 -9.44087262e-01 + -9.00843245e-01 8.26378089e-02 + 3.29770621e-01 -9.07870444e-01 + -2.84650535e+00 -9.00155396e-01 + 1.57111705e+00 7.07432268e-01 + 1.24948552e+00 1.04812849e-01 + 1.81440558e+00 9.53545082e-01 + -1.74915794e+00 -1.04606288e+00 + 1.20593269e+00 -1.12607147e-02 + 1.36004919e-01 -1.09828044e+00 + 2.57480693e-01 3.34941541e-01 + 7.78775385e-01 -5.32494732e-01 + -1.79155126e+00 -6.29994129e-01 + -1.75706839e+00 -8.35100126e-01 + 4.29512012e-01 7.81426910e-02 + 3.08349370e-01 -1.27359861e-01 + 1.05560329e+00 4.55150640e-01 + 1.95662574e+00 1.17593217e+00 + 8.77376632e-01 6.57866662e-01 + 7.71311255e-01 9.15134334e-02 + -6.36978275e+00 -2.55874241e+00 + -2.98335339e+00 -1.59567024e+00 + -3.67104587e-01 1.85315291e-01 + 1.95347407e+00 -7.15503113e-02 + 8.45556363e-01 6.51256415e-02 + 9.42868521e-01 3.56647624e-01 + 2.99321875e+00 1.07505254e+00 + -2.91030538e-01 -3.77637183e-01 + 1.62870918e+00 3.37563671e-01 + 2.05773173e-01 3.43337416e-01 + -8.40879199e-01 -1.35600767e-01 + 1.38101624e+00 5.99253495e-01 + -6.93715607e+00 -2.63580662e+00 + -1.04423404e+00 -8.32865050e-01 + 1.33448476e+00 1.04863475e+00 + 6.01675207e-01 1.98585194e-01 + 2.31233993e+00 7.98628331e-01 + 1.85201313e-01 -1.76070247e+00 + 1.92006354e+00 8.45737582e-01 + 1.06320415e+00 2.93426068e-01 + -1.20360141e+00 -1.00301288e+00 + 1.95926629e+00 6.26643532e-01 + 6.04483978e-02 5.72643059e-01 + -1.04568563e+00 -5.91021496e-01 + 2.62300678e+00 9.50997831e-01 + -4.04610275e-01 3.73150879e-01 + 2.26371902e+00 8.73627529e-01 + 2.12545313e+00 7.90640352e-01 + 7.72181917e-03 1.65718952e-02 + 1.00422340e-01 -2.05562936e-01 + -1.22989802e+00 -1.01841681e-01 + 3.09064082e+00 1.04288010e+00 + 5.18274167e+00 1.34749259e+00 + -8.32075153e-01 -1.97592029e-01 + 3.84126764e-02 5.58171345e-01 + 4.99560727e-01 -4.26154438e-02 + 4.79071151e+00 2.19728942e+00 + -2.78437968e+00 -1.17812590e+00 + -2.22804226e+00 -4.31174255e-01 + 8.50762292e-01 -1.06445261e-01 + 1.10812830e+00 -2.59118812e-01 + -2.91450155e-01 6.42802679e-01 + -1.38631532e-01 -5.88585623e-01 + -5.04120983e-01 -2.17094915e-01 + 3.41410820e+00 1.67897767e+00 + -2.23697326e+00 -6.62735244e-01 + -3.55961064e-01 -1.27647226e-01 + -3.55568274e+00 -2.49011369e+00 + -8.77586408e-01 -9.38268065e-03 + 1.52382384e-01 -5.62155760e-01 + 1.55885574e-01 1.07617069e-01 + -8.37129973e-01 -5.22259081e-01 + -2.92741750e+00 -1.35049428e+00 + -3.54670781e-01 5.69205952e-02 + 2.21030255e+00 1.34689986e+00 + 1.60787722e+00 5.75984706e-01 + 1.32294221e+00 5.31577509e-01 + 7.05672928e-01 3.34241244e-01 + 1.41406179e+00 1.15783408e+00 + -6.92172228e-01 -2.84817896e-01 + 3.28358655e-01 -2.66910083e-01 + 1.68013644e-01 -4.28016549e-02 + 2.07365974e+00 7.76496211e-01 + -3.92974907e-01 2.46796730e-01 + -5.76078636e-01 3.25676963e-01 + -1.82547204e-01 -5.06410543e-01 + 3.04754906e+00 1.16174496e+00 + -3.01090632e+00 -1.09195183e+00 + -1.44659696e+00 -6.87838682e-01 + 2.11395861e+00 9.10495785e-01 + 1.40962871e+00 1.13568678e+00 + -1.66653234e-01 -2.10012503e-01 + 3.17456029e+00 9.74502922e-01 + 2.15944820e+00 8.62807189e-01 + -3.45418719e+00 -1.33647548e+00 + -3.41357732e+00 -8.47048920e-01 + -3.06702448e-01 -6.64280634e-01 + -2.86930714e-01 -1.35268264e-01 + -3.15835557e+00 -5.43439253e-01 + 2.49541440e-01 -4.71733570e-01 + 2.71933912e+00 4.13308399e-01 + -2.43787038e+00 -1.08050547e+00 + -4.90234490e-01 -6.64069865e-01 + 8.99524451e-02 5.76180541e-01 + 5.00500404e+00 2.12125521e+00 + -1.73107940e-01 -2.28506575e-02 + 5.44938858e-01 -1.29523352e-01 + 5.13526842e+00 1.68785993e+00 + 1.70228304e+00 1.02601138e+00 + 3.58957507e+00 1.54396196e+00 + 1.85615738e+00 4.92916197e-01 + 2.55772147e+00 7.88438908e-01 + -1.57008279e+00 -4.17377300e-01 + -1.42548604e+00 -3.63684860e-01 + -8.52026118e-01 2.72052686e-01 + -5.10563077e+00 -2.35665994e+00 + -2.95517031e+00 -1.84945297e+00 + -2.91947959e+00 -1.66016784e+00 + -4.21462387e+00 -1.41131535e+00 + 6.59901121e-01 4.87156314e-01 + -9.75352532e-01 -4.50231285e-01 + -5.94084444e-01 -1.16922670e+00 + 7.50554615e-01 -9.83692552e-01 + 1.07054926e+00 2.77143030e-01 + -3.88079578e-01 -4.17737309e-02 + -9.59373733e-01 -8.85454886e-01 + -7.53560665e-02 -5.16223870e-02 + 9.84108158e-01 -5.89290700e-02 + 1.87272961e-01 -4.34238391e-01 + 6.86509981e-01 -3.15116460e-01 + -1.07762538e+00 6.58984161e-02 + 6.09266592e-01 6.91808473e-02 + -8.30529954e-01 -7.00454791e-01 + -9.13179464e-01 -6.31712891e-01 + 7.68744851e-01 1.09840676e+00 + -1.07606690e+00 -8.78390282e-01 + -1.71038184e+00 -5.73606033e-01 + 8.75982765e-01 3.66343143e-01 + -7.04919009e-01 -8.49182590e-01 + -1.00274668e+00 -7.99573611e-01 + -1.05562848e+00 -5.84060076e-01 + 4.03490015e+00 1.28679206e+00 + -3.53484804e+00 -1.71381255e+00 + 2.31527363e-01 1.04179397e-01 + -3.58592392e-02 3.74895739e-01 + 3.92253428e+00 1.81852726e+00 + -7.27384249e-01 -6.45605128e-01 + 4.65678097e+00 2.41379899e+00 + 1.16750534e+00 7.60718205e-01 + 1.15677059e+00 7.96225550e-01 + -1.42920261e+00 -4.66946295e-01 + 3.71148192e+00 1.88060191e+00 + 2.44052407e+00 3.84472199e-01 + -1.64535035e+00 -8.94530036e-01 + -3.69608753e+00 -1.36402754e+00 + 2.24419208e+00 9.69744889e-01 + 2.54822427e+00 1.22613039e+00 + 3.77484909e-01 -5.98521878e-01 + -3.61521175e+00 -1.11123912e+00 + 3.28113127e+00 1.52551775e+00 + -3.51030902e+00 -1.53913980e+00 + -2.44874505e+00 -6.30246005e-01 + -3.42516153e-01 -5.07352665e-01 + 1.09110502e+00 6.36821628e-01 + -2.49434967e+00 -8.02827146e-01 + 1.41763139e+00 -3.46591820e-01 + 1.61108619e+00 5.93871102e-01 + 3.97371717e+00 1.35552499e+00 + -1.33437177e+00 -2.83908670e-01 + -1.41606483e+00 -1.76402601e-01 + 2.23945322e-01 -1.77157065e-01 + 2.60271569e+00 2.40778251e-01 + -2.82213895e-02 1.98255474e-01 + 4.20727940e+00 1.31490863e+00 + 3.36944889e+00 1.57566635e+00 + 3.53049396e+00 1.73579350e+00 + -1.29170202e+00 -1.64196290e+00 + 9.27295604e-01 9.98808036e-01 + 1.75321843e-01 -2.83267817e-01 + -2.19069578e+00 -1.12814358e+00 + 1.66606031e+00 7.68006933e-01 + -7.13826035e-01 5.20881684e-02 + -3.43821888e+00 -2.36137021e+00 + -5.93210310e-01 1.21843813e-01 + -4.09800822e+00 -1.39893953e+00 + 2.74110954e+00 1.52728606e+00 + 1.72652512e+00 -1.25435113e-01 + 1.97722357e+00 6.40667481e-01 + 4.18635780e-01 3.57018509e-01 + -1.78303569e+00 -2.11864764e-01 + -3.52809366e+00 -2.58794450e-01 + -4.72407090e+00 -1.63870734e+00 + 1.73917807e+00 8.73251829e-01 + 4.37979356e-01 8.49210569e-01 + 3.93791881e+00 1.76269490e+00 + 2.79065411e+00 1.04019042e+00 + -8.47426142e-01 -3.40136892e-01 + -4.24389181e+00 -1.80253120e+00 + -1.86675870e+00 -7.64558265e-01 + 9.46212675e-01 -7.77681445e-02 + -2.82448462e+00 -1.33592449e+00 + -2.57938567e+00 -1.56554690e+00 + -2.71615767e+00 -6.27667233e-01 + -1.55999166e+00 -5.81013466e-01 + -4.24696864e-01 -7.44673250e-01 + 1.67592970e+00 7.68164292e-01 + 8.48455216e-01 -6.05681126e-01 + 6.12575454e+00 1.65607584e+00 + 1.38207327e+00 2.39261863e-01 + 3.13364450e+00 1.17154698e+00 + 1.71694858e+00 1.26744905e+00 + -1.61746367e+00 -8.80098073e-01 + -8.52196756e-01 -9.27299728e-01 + -1.51562462e-01 -8.36552490e-02 + -7.04792753e-01 -1.24726713e-02 + -3.35265757e+00 -1.82176312e+00 + 3.32173170e-01 -1.33405580e-01 + 4.95841013e-01 4.58292712e-01 + 1.57713955e+00 7.79272991e-01 + 2.09743109e+00 9.23542557e-01 + 3.90450311e-03 -8.42873164e-01 + 2.59519038e+00 7.56479591e-01 + -5.77643976e-01 -2.36401904e-01 + -5.22310654e-01 1.34187830e-01 + -2.22096086e+00 -7.75507719e-01 + 1.35907831e+00 7.80197510e-01 + 3.80355868e+00 1.16983476e+00 + 3.82746596e+00 1.31417718e+00 + 3.30451183e+00 1.55398159e+00 + -3.42917814e-01 -8.62281222e-02 + -2.59093020e+00 -9.29883526e-01 + 1.40928562e+00 1.08398346e+00 + 1.54400137e-01 3.35881092e-01 + 1.59171586e+00 1.18855802e+00 + -5.25164002e-01 -1.03104220e-01 + 2.20067959e+00 1.37074713e+00 + 6.97860830e-01 6.27718548e-01 + -4.59743507e-01 1.36061163e-01 + -1.04691963e-01 -2.16271727e-01 + -1.08905573e+00 -5.95510769e-01 + -1.00826983e+00 -5.38509162e-02 + -3.16402719e+00 -1.33414216e+00 + 1.47870874e-01 1.75234619e-01 + -2.57078234e-01 7.03316889e-02 + 1.81073945e+00 4.26901462e-01 + 2.65476530e+00 6.74217273e-01 + 1.27539811e+00 6.22914081e-01 + -3.76750499e-01 -1.20629449e+00 + 1.00177595e+00 -1.40660091e-01 + -2.98919265e+00 -1.65145013e+00 + -2.21557682e+00 -8.11123452e-01 + -3.22635378e+00 -1.65639056e+00 + -2.72868553e+00 -1.02812087e+00 + 1.26042797e+00 8.49005248e-01 + -9.38318534e-01 -9.87588651e-01 + 3.38013194e-01 -1.00237461e-01 + 1.91175691e+00 8.48716369e-01 + 4.30244344e-01 6.05539915e-02 + 2.21783435e+00 3.03268204e-01 + 1.78019576e+00 1.27377108e+00 + 1.59733274e+00 4.40674687e-02 + 3.97428484e+00 2.20881566e+00 + -2.41108677e+00 -6.01410418e-01 + -2.50796499e+00 -5.71169866e-01 + -3.71957427e+00 -1.38195726e+00 + -1.57992670e+00 1.32068593e-01 + -1.35278851e+00 -6.39349270e-01 + 1.23075932e+00 2.40445409e-01 + 1.35606530e+00 4.33180078e-01 + 9.60968518e-02 2.26734255e-01 + 6.22975063e-01 5.03431915e-02 + -1.47624851e+00 -3.60568238e-01 + -2.49337808e+00 -1.15083052e+00 + 2.15717792e+00 1.03071559e+00 + -3.07814376e-02 1.38700314e-02 + 4.52049499e-02 -4.86409775e-01 + 2.58231061e+00 1.14327809e-01 + 1.10999138e+00 -5.18568405e-01 + -2.19426443e-01 -5.37505538e-01 + -4.44740298e-01 6.78099955e-01 + 4.03379080e+00 1.49825720e+00 + -5.13182408e-01 -4.90201950e-01 + -6.90139716e-01 1.63875126e-01 + -8.17281461e-01 2.32155064e-01 + -2.92357619e-01 -8.02573544e-01 + -1.80769841e+00 -7.58907326e-01 + 2.16981590e+00 1.06728873e+00 + 1.98995203e-01 -6.84176682e-02 + -2.39546753e+00 -2.92873789e-01 + -4.24251021e+00 -1.46255564e+00 + -5.01411291e-01 -5.95712813e-03 + 2.68085809e+00 1.42883780e+00 + -4.13289873e+00 -1.62729388e+00 + 1.87957843e+00 3.63341638e-01 + -1.15270744e+00 -3.03563774e-01 + -4.43994248e+00 -2.97323905e+00 + -7.17067733e-01 -7.08349542e-01 + -3.28870393e+00 -1.19263863e+00 + -7.55325944e-01 -5.12703329e-01 + -2.07291938e+00 -2.65025085e-01 + -7.50073814e-01 -1.70771041e-01 + -8.77381404e-01 -5.47417325e-01 + -5.33725862e-01 5.15837119e-01 + 8.45056431e-01 2.82125560e-01 + -1.59598637e+00 -1.38743235e+00 + 1.41362902e+00 1.06407789e+00 + 1.02584504e+00 -3.68219466e-01 + -1.04644488e+00 -1.48769392e-01 + 2.66990191e+00 8.57633492e-01 + -1.84251857e+00 -9.82430175e-01 + 9.71404204e-01 -2.81934209e-01 + -2.50177989e+00 -9.21260335e-01 + -1.31060074e+00 -5.84488113e-01 + -2.12129400e-01 -3.06244708e-02 + -5.28933882e+00 -2.50663129e+00 + 1.90220541e+00 1.08662918e+00 + -3.99366086e-02 -6.87178973e-01 + -4.93417342e-01 4.37354182e-01 + 2.13494486e+00 1.37679569e+00 + 2.18396765e+00 5.81023868e-01 + -3.07866587e+00 -1.45384974e+00 + 6.10894119e-01 -4.17050124e-01 + -1.88766952e+00 -8.86160058e-01 + 3.34527253e+00 1.78571260e+00 + 6.87769059e-01 -5.01157336e-01 + 2.60470837e+00 1.45853560e+00 + -6.49315691e-01 -9.16112805e-01 + -1.29817687e+00 -2.15924339e-01 + -1.20100409e-03 -4.03137422e-01 + -1.36471594e+00 -6.93266356e-01 + 1.38682062e+00 7.15131598e-01 + 2.47830103e+00 1.24862305e+00 + -2.78288147e+00 -1.03329235e+00 + -7.33443403e-01 -6.11041652e-01 + -4.12745671e-01 -5.96133390e-02 + -2.58632336e+00 -4.51557058e-01 + -1.16570367e+00 -1.27065510e+00 + 2.76187104e+00 2.21895451e-01 + -3.80443767e+00 -1.66319902e+00 + 9.84658633e-01 6.81475569e-01 + 9.33814584e-01 -4.89335563e-02 + -4.63427997e-01 1.72989539e-01 + 1.82401546e+00 3.60164021e-01 + -5.36521077e-01 -8.08691351e-01 + -1.37367030e+00 -1.02126160e+00 + -3.70310682e+00 -1.19840844e+00 + -1.51894242e+00 -3.89510223e-01 + -3.67347940e-01 -3.25540516e-02 + -1.00988595e+00 1.82802194e-01 + 2.01622795e+00 7.86367901e-01 + 1.02440231e+00 8.79780360e-01 + -3.05971480e+00 -8.40901527e-01 + 2.73909457e+00 1.20558628e+00 + 2.39559056e+00 1.10786694e+00 + 1.65471544e+00 7.33824651e-01 + 2.18546787e+00 6.41168955e-01 + 1.47152266e+00 3.91839132e-01 + 1.45811155e+00 5.21820495e-01 + -4.27531469e-02 -3.52343068e-03 + -9.54948010e-01 -1.52313876e-01 + 7.57151215e-01 -5.68728854e-03 + -8.46205751e-01 -7.54580229e-01 + 4.14493548e+00 1.45532780e+00 + 4.58688968e-01 -4.54012803e-02 + -1.49295381e+00 -4.57471758e-01 + 1.80020351e+00 8.13724973e-01 + -5.82727738e+00 -2.18269581e+00 + -2.09017809e+00 -1.18305177e+00 + -2.31628303e+00 -7.21600235e-01 + -8.09679091e-01 -1.49101752e-01 + 8.88005605e-01 8.57940857e-01 + -1.44148219e+00 -3.10926299e-01 + 3.68828186e-01 -3.08848059e-01 + -6.63267389e-01 -8.58950139e-02 + -1.14702569e+00 -6.32147854e-01 + -1.51741715e+00 -8.53330564e-01 + -1.33903718e+00 -1.45875547e-01 + 4.12485387e+00 1.85620435e+00 + -2.42353639e+00 -2.92669850e-01 + 1.88708583e+00 9.35984730e-01 + 2.15585179e+00 6.30469051e-01 + -1.13627973e-01 -1.62554045e-01 + 2.04540494e+00 1.36599834e+00 + 2.81591381e+00 1.60897941e+00 + 3.02736260e-02 3.83255815e-03 + 7.97634013e-02 -2.82035099e-01 + -3.24607473e-01 -5.30065956e-01 + -3.91862894e+00 -1.94083334e+00 + 1.56360901e+00 7.93882743e-01 + -1.03905772e+00 6.25590229e-01 + 2.54746492e+00 1.64233560e+00 + -4.80774423e-01 -8.92298032e-02 + 9.06979990e-02 1.05020427e+00 + -2.47521290e+00 -1.78275982e-01 + -3.91871729e-01 3.80285423e-01 + 1.00658382e+00 4.58947483e-01 + 4.68102941e-01 1.02992741e+00 + 4.44242568e-01 2.89870239e-01 + 3.29684452e+00 1.44677474e+00 + -2.24983007e+00 -9.65574499e-01 + -3.54453926e-01 -3.99020325e-01 + -3.87429665e+00 -1.90079739e+00 + 2.02656674e+00 1.12444894e+00 + 3.77011621e+00 1.43200852e+00 + 1.61259275e+00 4.65417399e-01 + 2.28725434e+00 6.79181395e-01 + 2.75421009e+00 2.27327345e+00 + -2.40894409e+00 -1.03926359e+00 + 1.52996651e-01 -2.73373046e-02 + -2.63218977e+00 -7.22802821e-01 + 2.77688169e+00 1.15310186e+00 + 1.18832341e+00 4.73457165e-01 + -2.35536326e+00 -1.08034554e+00 + -5.84221627e-01 1.03505984e-02 + 2.96730300e+00 1.33478306e+00 + -8.61947692e-01 6.09137051e-02 + 8.22343921e-01 -8.14155286e-02 + 1.75809015e+00 1.07921470e+00 + 1.19501279e+00 1.05309972e+00 + -1.75901792e+00 9.75320161e-02 + 1.64398635e+00 9.54384323e-01 + -2.21878052e-01 -3.64847144e-01 + -2.03128968e+00 -8.57866419e-01 + 1.86750633e+00 7.08524487e-01 + 8.03972976e-01 3.47404314e-01 + 3.41203749e+00 1.39810900e+00 + 4.22397681e-01 -6.41440488e-01 + -4.88493360e+00 -1.58967816e+00 + -1.67649284e-01 -1.08485915e-01 + 2.11489023e+00 1.50506158e+00 + -1.81639929e+00 -3.85542192e-01 + 2.24044819e-01 -1.45100577e-01 + -3.39262411e+00 -1.44394324e+00 + 1.68706599e+00 2.29199618e-01 + -1.94093257e+00 -1.65975814e-01 + 8.28143367e-01 5.92109281e-01 + -8.29587998e-01 -9.57130831e-01 + -1.50011401e+00 -8.36802092e-01 + 2.40770449e+00 9.32820177e-01 + 7.41391309e-02 3.12878473e-01 + 1.87745264e-01 6.19231425e-01 + 9.57622692e-01 -2.20640033e-01 + 3.18479243e+00 1.02986233e+00 + 2.43133846e+00 8.41302677e-01 + -7.09963834e-01 1.99718943e-01 + -2.88253498e-01 -3.62772094e-01 + 5.14052574e+00 1.79304595e+00 + -3.27930993e+00 -1.29177973e+00 + -1.16723536e+00 1.29519656e-01 + 1.04801056e+00 3.41508300e-01 + -3.99256195e+00 -2.51176471e+00 + -7.62824318e-01 -6.84242153e-01 + 2.71524986e-02 5.35157164e-02 + 3.26430102e+00 1.34887262e+00 + -1.72357766e+00 -4.94524388e-01 + -3.81149536e+00 -1.28121944e+00 + 3.36919354e+00 1.10672075e+00 + -3.14841757e+00 -7.10713767e-01 + -3.16463676e+00 -7.58558435e-01 + -2.44745969e+00 -1.08816514e+00 + 2.79173264e-01 -2.19652051e-02 + 4.15309883e-01 6.07502790e-01 + -9.51007417e-01 -5.83976336e-01 + -1.47929839e+00 -8.39850409e-01 + 2.38335703e+00 6.16055149e-01 + -7.47749031e-01 -5.56164928e-01 + -3.65643622e-01 -5.06684411e-01 + -1.76634163e+00 -7.86382097e-01 + 6.76372222e-01 -3.06592181e-01 + -1.33505058e+00 -1.18301441e-01 + 3.59660179e+00 2.00424178e+00 + -7.88912762e-02 8.71956146e-02 + 1.22656397e+00 1.18149583e+00 + 4.24919729e+00 1.20082355e+00 + 2.94607456e+00 1.00676505e+00 + 7.46061275e-02 4.41761753e-02 + -2.47738025e-02 1.92737701e-01 + -2.20509316e-01 -3.79163193e-01 + -3.50222190e-01 3.58727299e-01 + -3.64788014e+00 -1.36107312e+00 + 3.56062799e+00 9.27032742e-01 + 1.04317289e+00 6.08035970e-01 + 4.06718718e-01 3.00628051e-01 + 4.33158086e+00 2.25860714e+00 + 2.13917145e-01 -1.72757967e-01 + -1.40637998e+00 -1.14119465e+00 + 3.61554872e+00 1.87797348e+00 + 1.01726871e+00 5.70255097e-01 + -7.04902551e-01 2.16444147e-01 + -2.51492186e+00 -8.52997369e-01 + 1.85097530e+00 1.15124496e+00 + -8.67569714e-01 -3.05682432e-01 + 8.07550858e-01 5.88901608e-01 + 1.85186755e-01 -1.94589367e-01 + -1.23378238e+00 -7.84128347e-01 + -1.22713161e+00 -4.21218235e-01 + 2.97751165e-01 2.81055275e-01 + 4.77703554e+00 1.66265524e+00 + 2.51549669e+00 7.49980674e-01 + 2.76510822e-01 1.40456909e-01 + 1.98740905e+00 -1.79608212e-01 + 9.35429145e-01 8.44344180e-01 + -1.20854492e+00 -5.00598453e-01 + 2.29936219e+00 8.10236668e-01 + 6.92555544e-01 -2.65891331e-01 + -1.58050994e+00 2.31237821e-01 + -1.50864880e+00 -9.49661690e-01 + -1.27689206e+00 -7.18260016e-01 + -3.12517127e+00 -1.75587113e+00 + 8.16062912e-02 -6.56551804e-01 + -5.02479939e-01 -4.67162543e-01 + -5.47435788e+00 -2.47799576e+00 + 1.95872901e-02 5.80874076e-01 + -1.59064958e+00 -6.34554756e-01 + -3.77521478e+00 -1.74301790e+00 + 5.89628224e-01 8.55736553e-01 + -1.81903543e+00 -7.50011008e-01 + 1.38557775e+00 3.71490991e-01 + 9.70032652e-01 -7.11356016e-01 + 2.63539625e-01 -4.20994771e-01 + 2.12154222e+00 8.19081400e-01 + -6.56977937e-01 -1.37810098e-01 + 8.91309581e-01 2.77864361e-01 + -7.43693195e-01 -1.46293770e-01 + 2.24447769e+00 4.00911438e-01 + -2.25169262e-01 2.04148801e-02 + 1.68744684e+00 9.47573007e-01 + 2.73086373e-01 3.30877195e-01 + 5.54294414e+00 2.14198009e+00 + -8.49238733e-01 3.65603298e-02 + 2.39685712e+00 1.17951039e+00 + -2.58230528e+00 -5.52116673e-01 + 2.79785277e+00 2.88833717e-01 + -1.96576188e-01 1.11652123e+00 + -4.69383301e-01 1.96496282e-01 + -1.95011845e+00 -6.15235169e-01 + 1.03379890e-02 2.33701239e-01 + 4.18933607e-01 2.77939814e-01 + -1.18473337e+00 -4.10051126e-01 + -7.61499744e-01 -1.43658094e+00 + -1.65586092e+00 -3.41615303e-01 + -5.58523700e-02 -5.21837080e-01 + -2.40331088e+00 -2.64521583e-01 + 2.24925206e+00 6.79843335e-02 + 1.46360479e+00 1.04271443e+00 + -3.09255443e+00 -1.82548953e+00 + 2.11325841e+00 1.14996627e+00 + -8.70657797e-01 1.02461839e-01 + -5.71056521e-01 9.71232588e-02 + -3.37870752e+00 -1.54091877e+00 + 1.03907189e+00 -1.35661392e-01 + 8.40057486e-01 6.12172413e-02 + -1.30998234e+00 -1.34077226e+00 + 7.53744974e-01 1.49447350e-01 + 9.13995056e-01 -1.81227962e-01 + 2.28386229e-01 3.74498520e-01 + 2.54829151e-01 -2.88802704e-01 + 1.61709009e+00 2.09319193e-01 + -1.12579380e+00 -5.95955338e-01 + -2.69610726e+00 -2.76222736e-01 + -2.63773329e+00 -7.84491970e-01 + -2.62167427e+00 -1.54792874e+00 + -4.80639856e-01 -1.30582102e-01 + -1.26130891e+00 -8.86841840e-01 + -1.24951950e+00 -1.18182622e+00 + -1.40107574e+00 -9.13695575e-01 + 4.99872179e-01 4.69014702e-01 + -2.03550193e-02 -1.48859738e-01 + -1.50189069e+00 -2.97714278e-02 + -2.07846113e+00 -7.29937809e-01 + -5.50576792e-01 -7.03151525e-01 + -3.88069238e+00 -1.63215295e+00 + 2.97032988e+00 6.43571144e-01 + -1.85999273e-01 1.18107620e+00 + 1.79249709e+00 6.65356160e-01 + 2.68842472e+00 1.35703255e+00 + 1.07675417e+00 1.39845588e-01 + 8.01226349e-01 2.11392275e-01 + 9.64329379e-01 3.96146195e-01 + -8.22529511e-01 1.96080831e-01 + 1.92481841e+00 4.62985744e-01 + 3.69756927e-01 3.77135799e-01 + 1.19807835e+00 8.87715050e-01 + -1.01363587e+00 -2.48151636e-01 + 8.53071010e-01 4.96887868e-01 + -3.41120553e+00 -1.35401843e+00 + -2.64787381e+00 -1.08690563e+00 + -1.11416759e+00 -4.43848915e-01 + 1.46242648e+00 6.17106076e-02 + -7.52968881e-01 -9.20972209e-01 + -1.22492228e+00 -5.40327617e-01 + 1.08001827e+00 5.29593785e-01 + -2.58706464e-01 1.13022085e-01 + -4.27394011e-01 1.17864354e-02 + -3.20728413e+00 -1.71224737e-01 + 1.71398530e+00 8.68885893e-01 + 2.12067866e+00 1.45092772e+00 + 4.32782616e-01 -3.34117769e-01 + 7.80084374e-01 -1.35100217e-01 + -2.05547729e+00 -4.70217750e-01 + 2.38379736e+00 1.09186058e+00 + -2.80825477e+00 -1.03320187e+00 + 2.63434576e+00 1.15671733e+00 + -1.60936214e+00 1.91843035e-01 + -5.02298769e+00 -2.32820708e+00 + 1.90349195e+00 1.45215416e+00 + 3.00232888e-01 3.24412586e-01 + -2.46503943e+00 -1.19550010e+00 + 1.06304233e+00 2.20136246e-01 + -2.99101388e+00 -1.58299318e+00 + 2.30071719e+00 1.12881362e+00 + -2.37587247e+00 -8.08298336e-01 + 7.27006308e-01 3.80828984e-01 + 2.61199061e+00 1.56473491e+00 + 8.33936357e-01 -1.42189425e-01 + 3.13291605e+00 1.77771210e+00 + 2.21917371e+00 5.68427075e-01 + 2.38867649e+00 9.06637262e-01 + -6.92959466e+00 -3.57682881e+00 + 2.57904824e+00 5.93959108e-01 + 2.71452670e+00 1.34436199e+00 + 4.39988761e+00 2.13124672e+00 + 5.71783077e-01 5.08346173e-01 + -3.65399429e+00 -1.18192861e+00 + 4.46176453e-01 3.75685594e-02 + -2.97501495e+00 -1.69459236e+00 + 1.60855728e+00 9.20930014e-01 + -1.44270290e+00 -1.93922306e-01 + 1.67624229e+00 1.66233866e+00 + -1.42579598e+00 -1.44990145e-01 + 1.19923176e+00 4.58490278e-01 + -9.00068460e-01 5.09701825e-02 + -1.69391694e+00 -7.60070300e-01 + -1.36576440e+00 -5.24244256e-01 + -1.03016748e+00 -3.44625878e-01 + 2.40519313e+00 1.09947587e+00 + 1.50365433e+00 1.06464802e+00 + -1.07609727e+00 -3.68897187e-01 + 2.44969069e+00 1.28486192e+00 + -1.25610307e+00 -1.14644789e+00 + 2.05962899e+00 4.31162369e-01 + -7.15886908e-01 -6.11587804e-02 + -6.92354119e-01 -7.85019920e-01 + -1.63016508e+00 -5.96944975e-01 + 1.90352536e+00 1.28197457e+00 + -4.01535243e+00 -1.81934488e+00 + -1.07534435e+00 -2.10544784e-01 + 3.25500866e-01 7.69603661e-01 + 2.18443365e+00 6.59773335e-01 + 8.80856790e-01 6.39505913e-01 + -2.23956372e-01 -4.65940132e-01 + -1.06766519e+00 -5.38388505e-03 + 7.25556863e-01 -2.91123488e-01 + -4.69451411e-01 7.89182650e-02 + 2.58146587e+00 1.29653243e+00 + 1.53747468e-01 7.69239075e-01 + -4.61152262e-01 -4.04151413e-01 + 1.48183517e+00 8.10079506e-01 + -1.83402614e+00 -1.36939322e+00 + 1.49315501e+00 7.95225425e-01 + 1.41922346e+00 1.05582774e-01 + 1.57473493e-01 9.70795657e-01 + -2.67603254e+00 -7.48562280e-01 + -8.49156216e-01 -6.05762529e-03 + 1.12944274e+00 3.67741591e-01 + 1.94228071e-01 5.28188141e-01 + -3.65610158e-01 4.05851838e-01 + -1.98839111e+00 -1.38452764e+00 + 2.73765752e+00 8.24150530e-01 + 7.63728641e-01 3.51617707e-01 + 5.78307267e+00 1.68103612e+00 + 2.27547227e+00 3.60876164e-01 + -3.50681697e+00 -1.74429984e+00 + 4.01241184e+00 1.26227829e+00 + 2.44946343e+00 9.06119057e-01 + -2.96638941e+00 -9.01532322e-01 + 1.11267643e+00 -3.43333381e-01 + -6.61868994e-01 -3.44666391e-01 + -8.34917179e-01 5.69478372e-01 + -1.91888454e+00 -3.03791075e-01 + 1.50397636e+00 8.31961240e-01 + 6.12260198e+00 2.16851807e+00 + 1.34093127e+00 8.86649385e-01 + 1.48748519e+00 8.26273697e-01 + 7.62243068e-01 2.64841396e-01 + -2.17604986e+00 -3.54219958e-01 + 2.64708640e-01 -4.38136718e-02 + 1.44725372e+00 1.18499914e-01 + -6.71259446e-01 -1.19526851e-01 + 2.40134595e-01 -8.90042323e-02 + -3.57238199e+00 -1.23166201e+00 + -3.77626645e+00 -1.19533443e+00 + -3.81101035e-01 -4.94160532e-01 + -3.02758757e+00 -1.18436066e+00 + 2.59116298e-01 1.38023047e+00 + 4.17900116e+00 1.12065959e+00 + 1.54598848e+00 2.89806755e-01 + 1.00656475e+00 1.76974511e-01 + -4.15730234e-01 -6.22681694e-01 + -6.00903565e-01 -1.43256959e-01 + -6.03652508e-01 -5.09936379e-01 + -1.94096658e+00 -9.48789544e-01 + -1.74464105e+00 -8.50491590e-01 + 1.17652544e+00 1.88118317e+00 + 2.35507776e+00 1.44000205e+00 + 2.63067924e+00 1.06692988e+00 + 2.88805386e+00 1.23924715e+00 + 8.27595008e-01 5.75364692e-01 + 3.91384216e-01 9.72781920e-02 + -1.03866816e+00 -1.37567768e+00 + -1.34777969e+00 -8.40266025e-02 + -4.12904508e+00 -1.67618340e+00 + 1.27918111e+00 3.52085961e-01 + 4.15361174e-01 6.28896189e-01 + -7.00539496e-01 4.80447955e-02 + -1.62332639e+00 -5.98236485e-01 + 1.45957300e+00 1.00305154e+00 + -3.06875603e+00 -1.25897545e+00 + -1.94708176e+00 4.85143006e-01 + 3.55744156e+00 -1.07468822e+00 + 1.21602223e+00 1.28768827e-01 + 1.89093098e+00 -4.70835659e-01 + -6.55759125e+00 2.70114082e+00 + 8.96843535e-01 -3.98115252e-01 + 4.13450429e+00 -2.32069236e+00 + 2.37764218e+00 -1.09098890e+00 + -1.11388901e+00 6.27083097e-01 + -6.34116929e-01 4.62816387e-01 + 2.90203079e+00 -1.33589143e+00 + 3.17457598e+00 -5.13575945e-01 + -1.76362299e+00 5.71820693e-01 + 1.66103362e+00 -8.99466249e-01 + -2.53947433e+00 8.40084780e-01 + 4.36631397e-01 7.24234261e-02 + -1.87589394e+00 5.08529113e-01 + 4.49563965e+00 -9.43365992e-01 + 1.78876299e+00 -1.27076149e+00 + -1.16269107e-01 -4.55078316e-01 + 1.92966079e+00 -8.05371385e-01 + 2.20632583e+00 -9.00919345e-01 + 1.52387824e+00 -4.82391996e-01 + 8.04004564e-01 -2.73650595e-01 + -7.75326067e-01 1.07469566e+00 + 1.83226282e+00 -4.52173344e-01 + 1.25079758e-01 -3.52895417e-02 + -9.90957437e-01 8.55993130e-01 + 1.71623322e+00 -7.08691667e-01 + -2.86175924e+00 6.75160955e-01 + -8.40817853e-01 -1.00361809e-01 + 1.33393000e+00 -4.65788123e-01 + 5.29394114e-01 -5.44881619e-02 + -8.07435599e-01 8.27353370e-01 + -4.33165824e+00 1.97299638e+00 + 1.26452422e+00 -8.34070486e-01 + 1.45996394e-02 2.97736043e-01 + -1.64489287e+00 6.72839598e-01 + -5.74234578e+00 3.20975117e+00 + 2.13841341e-02 3.64514015e-01 + 6.68084924e+00 -2.27464254e+00 + -3.22881590e+00 8.01879324e-01 + 3.02534313e-01 -4.56222796e-01 + -5.84520734e+00 1.95678162e+00 + 2.81515232e+00 -1.72101318e+00 + -2.39620908e-01 2.69145522e-01 + -7.41669691e-01 -2.30283281e-01 + -2.15682714e+00 3.45313021e-01 + 1.23475788e+00 -7.32276553e-01 + -1.71816113e-01 1.20419560e-02 + 1.89174235e+00 2.27435901e-01 + -3.64511114e-01 1.72260361e-02 + -3.24143860e+00 6.50125817e-01 + -2.25707409e+00 5.66970751e-01 + 1.03901456e+00 -1.00588433e+00 + -5.09159710e+00 1.58736109e+00 + 1.45534075e+00 -5.83787452e-01 + 4.28879587e+00 -1.58006866e+00 + 8.52384427e-01 -1.11042299e+00 + 4.51431615e+00 -2.63844265e+00 + -4.33042648e+00 1.86497078e+00 + -2.13568046e+00 5.82559743e-01 + -4.42568887e+00 1.26131214e+00 + 3.15821315e+00 -1.61515905e+00 + -3.14125204e+00 8.49604386e-01 + 6.54152300e-01 -2.04624711e-01 + -3.73374317e-01 9.94187820e-02 + -3.96177282e+00 1.27245623e+00 + 9.59825199e-01 -1.15547861e+00 + 3.56902055e+00 -1.46591091e+00 + 1.55433633e-02 6.93544345e-01 + 1.15684646e+00 -4.99836352e-01 + 3.11824573e+00 -4.75900506e-01 + -8.61706369e-01 -3.50774059e-01 + 9.89057391e-01 -7.16878802e-01 + -4.94787870e+00 2.09137481e+00 + 1.37777347e+00 -1.34946349e+00 + -1.13161577e+00 8.05114754e-01 + 8.12020675e-01 -1.04849421e+00 + 4.73783881e+00 -2.26718812e+00 + 8.99579366e-01 -8.89764451e-02 + 4.78524868e+00 -2.25795843e+00 + 1.75164590e+00 -1.73822209e-01 + 1.30204590e+00 -7.26724717e-01 + -7.26526403e-01 -5.23925361e-02 + 2.01255351e+00 -1.69965366e+00 + 9.87852740e-01 -4.63577220e-01 + 2.45957762e+00 -1.29278962e+00 + -3.13817948e+00 1.64433038e+00 + -1.76302159e+00 9.62784302e-01 + -1.91106331e+00 5.81460008e-01 + -3.30883001e+00 1.30378978e+00 + 5.54376450e-01 3.78814272e-01 + 1.09982111e+00 -1.47969612e+00 + -2.61300705e-02 -1.42573464e-01 + -2.22096157e+00 7.75684440e-01 + 1.70319323e+00 -2.89738444e-01 + -1.43223842e+00 6.39284281e-01 + 2.34360959e-01 -1.64379268e-01 + -2.67147991e+00 9.46548086e-01 + 1.51131425e+00 -4.91594395e-01 + -2.48446856e+00 1.01286123e+00 + 1.50534658e-01 -2.94620246e-01 + -1.66966792e+00 1.67755508e+00 + -1.50094241e+00 3.30163095e-01 + 2.27681194e+00 -1.08064317e+00 + 2.05122965e+00 -1.15165939e+00 + -4.23509309e-01 -6.56906167e-02 + 1.80084023e+00 -1.07228556e+00 + -2.65769521e+00 1.18023206e+00 + 2.02852676e+00 -8.06793574e-02 + -4.49544185e+00 2.68200163e+00 + -7.50043216e-01 1.17079331e+00 + 6.80060893e-02 3.99055351e-01 + -3.83634635e+00 1.38406887e+00 + 3.24858545e-01 -9.25273218e-02 + -2.19895100e+00 1.47819500e+00 + -3.61569522e-01 -1.03188739e-01 + 1.12180375e-01 -9.52696354e-02 + -1.31477803e+00 1.79900570e-01 + 2.39573628e+00 -6.09739269e-01 + -1.00135700e+00 6.02837296e-01 + -4.11994589e+00 2.49599192e+00 + -1.54196236e-01 -4.84921951e-01 + 5.92569908e-01 -1.87310359e-01 + 3.85407741e+00 -1.50979925e+00 + 5.17802528e+00 -2.26032607e+00 + -1.37018916e+00 1.87111822e-01 + 8.46682996e-01 -3.56676331e-01 + -1.17559949e+00 5.29057734e-02 + -5.56475671e-02 6.79049243e-02 + 1.07851745e+00 -5.14535101e-01 + -2.71622446e+00 1.00151846e+00 + -1.08477208e+00 8.81391054e-01 + 5.50755824e-01 -5.20577727e-02 + 4.70885495e+00 -2.04220397e+00 + -1.87375336e-01 -6.16962830e-02 + 3.52097100e-01 2.21163550e-01 + 7.07929984e-01 -1.75827590e-01 + -1.22149219e+00 1.83084346e-01 + 2.58247412e+00 -6.15914898e-01 + -6.01206182e-01 -2.29832987e-01 + 9.83360449e-01 -3.75870060e-01 + -3.20027685e+00 1.35467480e+00 + 1.79178978e+00 -1.38531981e+00 + -3.30376867e-01 -1.16250192e-01 + -1.89053055e+00 5.68463567e-01 + -4.20604849e+00 1.65429681e+00 + -1.01185529e+00 1.92801240e-01 + -6.18819882e-01 5.42206996e-01 + -5.08091672e+00 2.61598591e+00 + -2.62570344e+00 2.51590658e+00 + 3.05577906e+00 -1.49090609e+00 + 2.77609677e+00 -1.37681378e+00 + -7.93515301e-02 4.28072744e-01 + -2.08359471e+00 8.94334295e-01 + 2.20163801e+00 4.01127167e-02 + -1.18145785e-01 -2.06822464e-01 + -2.74788298e-01 2.96250607e-01 + 1.59613555e+00 -3.87246203e-01 + -3.82971472e-01 -3.39716093e-02 + -4.20311307e-02 3.88529510e-01 + 1.52128574e+00 -9.33138876e-01 + -9.06584458e-01 -2.75016094e-02 + 3.56216834e+00 -9.99384622e-01 + 2.11964220e+00 -9.98749118e-02 + 4.01203480e+00 -2.03032745e+00 + -1.24171557e+00 1.97596725e-01 + -1.57230455e+00 4.14126609e-01 + -1.85484741e+00 5.40041563e-01 + 1.76329831e+00 -6.95967734e-01 + -2.29439232e-01 5.08669245e-01 + -5.45124276e+00 2.26907549e+00 + -5.71364288e-02 5.04476476e-01 + 3.12468018e+00 -1.46358879e+00 + 8.20017359e-01 6.51949028e-01 + -1.33977500e+00 2.83634232e-04 + -1.83311685e+00 1.23947117e+00 + 6.31205922e-01 1.19792164e-02 + -2.21967834e+00 6.94056232e-01 + -1.41693842e+00 9.93526233e-01 + -7.58885703e-01 6.78547347e-01 + 3.60239086e+00 -1.08644935e+00 + 6.72217073e-02 3.00036011e-02 + -3.42680958e-01 -3.48049352e-01 + 1.87546079e+00 -4.78018246e-01 + 7.00485821e-01 -3.52905383e-01 + -8.54580948e-01 8.17330861e-01 + 8.19123706e-01 -5.73927281e-01 + 2.70855639e-01 -3.08940052e-01 + -1.05059952e+00 3.27873168e-01 + 1.08282999e+00 4.84559349e-02 + -7.89899220e-01 1.22291138e+00 + -2.87939816e+00 7.17403497e-01 + -2.08429452e+00 8.87409226e-01 + 1.58409232e+00 -4.74123532e-01 + 1.26882735e+00 1.59162510e-01 + -2.53782993e+00 6.18253491e-01 + -8.92757445e-01 3.35979011e-01 + 1.31867900e+00 -1.17355054e+00 + 1.14918879e-01 -5.35184038e-01 + -1.70288738e-01 5.35868087e-02 + 4.21355121e-01 5.41848690e-02 + 2.07926943e+00 -5.72538144e-01 + 4.08788970e-01 3.77655777e-01 + -3.39631381e+00 9.84216764e-01 + 2.94170163e+00 -1.83120916e+00 + -7.94798752e-01 7.39889052e-01 + 1.46555463e+00 -4.62275563e-01 + 2.57255955e+00 -1.04671434e+00 + 8.45042540e-01 -1.96952892e-01 + -3.23526646e+00 1.60049846e+00 + 3.21948565e+00 -8.88376674e-01 + 1.43005104e+00 -9.21561086e-01 + 8.82360506e-01 2.98403872e-01 + -8.91168097e-01 1.01319072e+00 + -5.13215241e-01 -2.47182649e-01 + -1.35759444e+00 7.07450608e-02 + -4.04550983e+00 2.23534867e+00 + 1.39348883e+00 3.81637747e-01 + -2.85676418e+00 1.53240862e+00 + -1.37183120e+00 6.37977425e-02 + -3.88195859e+00 1.73887145e+00 + 1.19509776e+00 -6.25013512e-01 + -2.80062734e+00 1.79840585e+00 + 1.96558429e+00 -4.70997234e-01 + 1.93111352e+00 -9.70318441e-01 + 3.57991190e+00 -1.65065116e+00 + 2.12831714e+00 -1.11531708e+00 + -3.95661018e-01 -8.54339904e-02 + -2.41630441e+00 1.65166304e+00 + 7.55412624e-01 -1.53453579e-01 + -1.77043450e+00 1.39928715e+00 + -9.32631260e-01 8.73649199e-01 + 1.53342205e+00 -8.39569765e-01 + -6.29846924e-02 1.25023084e-01 + 3.31509049e+00 -1.10733235e+00 + -2.18957109e+00 3.07376993e-01 + -2.35740747e+00 6.47437564e-01 + -2.22142438e+00 8.47318938e-01 + -6.51401147e-01 3.48398562e-01 + 2.75763095e+00 -1.21390708e+00 + 1.12550484e+00 -5.61412847e-01 + -5.65053161e-01 6.74365205e-02 + 1.68952456e+00 -6.57566096e-01 + 8.95598401e-01 3.96738993e-01 + -1.86537066e+00 9.44129208e-01 + -2.59933294e+00 2.57423247e-01 + -6.59598267e-01 1.91828851e-02 + -2.64506676e+00 8.41783205e-01 + -1.25911802e+00 5.52425066e-01 + -1.39754507e+00 3.73689222e-01 + 5.49550729e-02 1.35071215e+00 + 3.31874811e+00 -1.05682424e+00 + 3.63159604e+00 -1.42864695e+00 + -4.45944617e+00 1.42889446e+00 + 5.87314342e-01 -4.88892988e-01 + -7.26130820e-01 1.51936106e-01 + -1.79246441e+00 6.05888105e-01 + -5.50948207e-01 6.21443081e-01 + -3.17246063e-01 1.77213880e-01 + -2.00098937e+00 1.23799074e+00 + 4.33790961e+00 -1.08490465e+00 + -2.03114114e+00 1.31613237e+00 + -6.29216542e+00 1.92406317e+00 + -1.60265624e+00 8.87947500e-01 + 8.64465062e-01 -8.37416270e-01 + -2.14273937e+00 8.05485900e-01 + -2.36844256e+00 6.17915124e-01 + -1.40429636e+00 6.78296866e-01 + 9.99019988e-01 -5.84297572e-01 + 7.38824546e-01 1.68838678e-01 + 1.45681238e+00 3.04641461e-01 + 2.15914949e+00 -3.43089227e-01 + -1.23895930e+00 1.05339864e-01 + -1.23162264e+00 6.46629863e-01 + 2.28183862e+00 -9.24157063e-01 + -4.29615882e-01 5.69130863e-01 + -1.37449121e+00 -9.12032183e-01 + -7.33890904e-01 -3.91865471e-02 + 8.41400661e-01 -4.76002200e-01 + -1.73349274e-01 -6.84143467e-02 + 3.16042891e+00 -1.32651856e+00 + -3.78244609e+00 2.38619718e+00 + -3.69634380e+00 2.22368561e+00 + 1.83766344e+00 -1.65675953e+00 + -1.63206002e+00 1.19484469e+00 + 3.68480064e-01 -5.70764494e-01 + 3.61982479e-01 1.04274409e-01 + 2.48863048e+00 -1.13285542e+00 + -2.81896488e+00 9.47958768e-01 + 5.74952901e-01 -2.75959392e-01 + 3.72783275e-01 -3.48937848e-01 + 1.95935716e+00 -1.06750415e+00 + 5.19357531e+00 -2.32070803e+00 + 4.09246149e+00 -1.89976700e+00 + -3.36666087e-01 8.17645057e-02 + 1.85453493e-01 3.76913151e-01 + -3.06458262e+00 1.34106402e+00 + -3.13796566e+00 7.00485099e-01 + 1.42964058e+00 -1.35536932e-01 + -1.23440423e-01 4.60094177e-02 + -2.86753037e+00 -5.21724160e-02 + 2.67113726e+00 -1.83746924e+00 + -1.35335062e+00 1.28238073e+00 + -2.43569899e+00 1.25998539e+00 + 1.26036740e-01 -2.35416844e-01 + -1.35725745e+00 7.37788491e-01 + -3.80897538e-01 3.30757889e-01 + 6.58694434e-01 -1.07566603e+00 + 2.11273640e+00 -9.02260632e-01 + 4.00755057e-01 -2.49229150e-02 + -1.80095812e+00 9.73099742e-01 + -2.68408372e+00 1.63737364e+00 + -2.66079826e+00 7.47289412e-01 + -9.92321439e-02 -1.49331396e-01 + 4.45678251e+00 -1.80352394e+00 + 1.35962915e+00 -1.31554389e+00 + -7.76601417e-01 -9.66173523e-02 + 1.68096348e+00 -6.27235133e-01 + 1.53081227e-01 -3.54216830e-01 + -1.54913095e+00 3.43689269e-01 + 5.29187357e-02 -6.73916964e-01 + -2.06606084e+00 8.34784242e-01 + 1.73701179e+00 -6.06467340e-01 + 1.55856757e+00 -2.58642780e-01 + 1.04349101e+00 -4.43027348e-01 + -1.02397719e+00 1.01308824e+00 + -2.13860204e-01 -4.73347361e-01 + -2.59004955e+00 1.43367853e+00 + 7.98457679e-01 2.18621627e-02 + -1.32974762e+00 4.61802208e-01 + 3.21419359e-01 2.30723316e-02 + 2.87201888e-02 6.24566672e-02 + -1.22261418e+00 6.02340363e-01 + 1.28750335e+00 -3.34839548e-02 + -9.67952623e-01 4.34470505e-01 + 2.02850324e+00 -9.05160255e-01 + -4.13946010e+00 2.33779091e+00 + -4.47508806e-01 3.06440495e-01 + -3.91543394e+00 1.68251022e+00 + -6.45193001e-01 5.29781162e-01 + -2.15518916e-02 5.07278355e-01 + -2.83356868e+00 1.00670227e+00 + 1.82989749e+00 -1.37329222e+00 + -1.09330213e+00 1.08560688e+00 + 1.90533722e+00 -1.28905879e+00 + 2.33986084e+00 2.30642626e-02 + 8.01940220e-01 -1.63986962e+00 + -4.23415165e+00 2.07530423e+00 + 9.33382522e-01 -7.62917211e-01 + -1.84033954e+00 1.07469401e+00 + -2.81938669e+00 1.07342024e+00 + -7.05169988e-01 2.13124943e-01 + 5.09598137e-01 1.32725493e-01 + -2.34558226e+00 8.62383168e-01 + -1.70322072e+00 2.70893796e-01 + 1.23652660e+00 -7.53216034e-02 + 2.84660646e+00 -3.48178304e-02 + 2.50250128e+00 -1.27770855e+00 + -1.00279469e+00 8.77194218e-01 + -4.34674121e-02 -2.12091350e-01 + -5.84151289e-01 1.50382340e-01 + -1.79024013e+00 4.24972808e-01 + -1.23434666e+00 -8.85546570e-02 + 1.36575412e+00 -6.42639880e-01 + -1.98429947e+00 2.27650336e-01 + 2.36253589e+00 -1.51340773e+00 + 8.79157643e-01 6.84142159e-01 + -2.18577755e+00 2.76526200e-01 + -3.55473434e-01 8.29976561e-01 + 1.16442595e+00 -5.97699411e-01 + -7.35528097e-01 2.40318183e-01 + -1.73702631e-01 7.33788663e-02 + -1.40451745e+00 3.24899628e-01 + -2.05434385e+00 5.68123738e-01 + 8.47876642e-01 -5.74224294e-01 + -6.91955602e-01 1.26009087e+00 + 2.56574498e+00 -1.15602581e+00 + 3.93306545e+00 -1.38398209e+00 + -2.73230251e+00 4.89062581e-01 + -1.04315474e+00 6.06335547e-01 + 1.23231431e+00 -4.46675065e-01 + -3.93035285e+00 1.43287651e+00 + -1.02132111e+00 9.58919791e-01 + -1.49425352e+00 1.06456165e+00 + -6.26485337e-01 1.03791402e+00 + -6.61772998e-01 2.63275425e-01 + -1.80940386e+00 5.70767403e-01 + 9.83720450e-01 -1.39449756e-01 + -2.24619662e+00 9.01044870e-01 + 8.94343014e-01 5.31038678e-02 + 1.95518199e-01 -2.81343295e-01 + -2.30533019e-01 -1.74478106e-01 + -2.01550361e+00 5.55958010e-01 + -4.36281469e+00 1.94374226e+00 + -5.18530457e+00 2.89278357e+00 + 2.67289101e+00 -2.98511449e-01 + -1.53566179e+00 -1.00588944e-01 + -6.09943217e-02 -1.56986047e-01 + -5.22146452e+00 1.66209208e+00 + -3.69777478e+00 2.26154873e+00 + 2.24607181e-01 -4.86934960e-01 + 2.49909450e+00 -1.03033370e+00 + -1.07841120e+00 8.22388054e-01 + -3.20697089e+00 1.09536143e+00 + 3.43524232e+00 -1.47289362e+00 + -5.65784134e-01 4.60365175e-01 + -1.76714734e+00 1.57752346e-01 + -7.77620365e-01 5.60153443e-01 + 6.34399352e-01 -5.22339836e-01 + 2.91011875e+00 -9.72623380e-01 + -1.19286824e+00 6.32370253e-01 + -2.18327609e-01 8.23953181e-01 + 3.42430842e-01 1.37098055e-01 + 1.28658034e+00 -9.11357320e-01 + 2.06914465e+00 -6.67556382e-01 + -6.69451020e-01 -6.38605102e-01 + -2.09312398e+00 1.16743634e+00 + -3.63778357e+00 1.91919157e+00 + 8.74685911e-01 -1.09931208e+00 + -3.91496791e+00 1.00808357e+00 + 1.29621330e+00 -8.32239802e-01 + 9.00222045e-01 -1.31159793e+00 + -1.12242062e+00 1.98517079e-01 + -3.71932852e-01 1.31667093e-01 + -2.23829610e+00 1.26328346e+00 + -2.08365062e+00 9.93385336e-01 + -1.91082720e+00 7.45866855e-01 + 4.38024917e+00 -2.05901118e+00 + -2.28872886e+00 6.85279335e-01 + 1.01274497e-01 -3.26227153e-01 + -5.04447572e-01 -3.18619513e-01 + 1.28537006e+00 -1.04573551e+00 + -7.83175212e-01 1.54791645e-01 + -3.89239175e+00 1.60017929e+00 + -8.87877111e-01 -1.04968005e-01 + 9.32215179e-01 -5.58691113e-01 + -6.44977127e-01 -2.23018375e-01 + 1.10141900e+00 -1.00666432e+00 + 2.92755687e-01 -1.45480350e-01 + 7.73580681e-01 -2.21150567e-01 + -1.40873709e+00 7.61548044e-01 + -8.89031805e-01 -3.48542923e-01 + 4.16844267e-01 -2.39914494e-01 + -4.64265832e-01 7.29581138e-01 + 1.99835179e+00 -7.70542813e-01 + 4.20523191e-02 -2.18783563e-01 + -6.32611758e-01 -3.09926115e-01 + 6.82912198e-02 -8.48327050e-01 + 1.92425229e+00 -1.37876951e+00 + 3.49461782e+00 -1.88354255e+00 + -3.25209026e+00 1.49809395e+00 + 6.59273182e-01 -2.37435654e-01 + -1.15517300e+00 8.46134387e-01 + 1.26756151e+00 -4.58988026e-01 + -3.99178418e+00 2.04153008e+00 + 7.05687841e-01 -6.83433306e-01 + -1.61997342e+00 8.16577004e-01 + -3.89750399e-01 4.29753250e-01 + -2.53026432e-01 4.92861432e-01 + -3.16788324e+00 4.44285524e-01 + -7.86248901e-01 1.12753716e+00 + -3.02351433e+00 1.28419015e+00 + -1.30131355e+00 1.71226678e+00 + -4.08843475e+00 1.62063214e+00 + -3.09209403e+00 1.19958520e+00 + 1.49102271e+00 -1.11834864e+00 + -3.18059348e+00 5.74587042e-01 + 2.06054867e+00 3.25797860e-03 + -3.50999200e+00 2.02412428e+00 + -8.26610023e-01 3.46528211e-01 + 2.00546034e+00 -4.07333110e-01 + -9.69941653e-01 4.80953753e-01 + 4.47925660e+00 -2.33127314e+00 + 2.03845790e+00 -9.90439915e-01 + -1.11349191e+00 4.31183918e-01 + -4.03628396e+00 1.68509679e+00 + -1.48177601e+00 7.74322088e-01 + 3.07369385e+00 -9.57465886e-01 + 2.39011286e+00 -6.44506921e-01 + 2.91561991e+00 -8.78627328e-01 + 1.10212733e+00 -4.21637388e-01 + 5.31985231e-01 -6.17445696e-01 + -6.82340929e-01 -2.93529716e-01 + 1.94290679e+00 -4.64268634e-01 + 1.92262116e+00 -7.93142835e-01 + 4.73762800e+00 -1.63654174e+00 + -3.17848641e+00 8.05791391e-01 + 4.08739432e+00 -1.80816807e+00 + -7.60648826e-01 1.24216138e-01 + -2.24716400e+00 7.90020937e-01 + 1.64284052e+00 -7.18784070e-01 + 1.04410012e-01 -7.11195880e-02 + 2.18268225e+00 -7.01767831e-01 + 2.06218013e+00 -8.70251746e-01 + -1.35266581e+00 7.08456358e-01 + -1.38157779e+00 5.14401086e-01 + -3.28326008e+00 1.20988399e+00 + 8.85358917e-01 -8.12213495e-01 + -2.34067500e+00 3.67657353e-01 + 3.96878127e+00 -1.66841450e+00 + 1.36518053e+00 -8.33436812e-01 + 5.25771988e-01 -5.06121987e-01 + -2.25948361e+00 1.30663765e+00 + -2.57662070e+00 6.32114628e-01 + -3.43134685e+00 2.38106008e+00 + 2.31571924e+00 -1.56566818e+00 + -2.95397202e+00 1.05661888e+00 + -1.35331242e+00 6.76383411e-01 + 1.40977132e+00 -1.17775938e+00 + 1.52561996e+00 -9.83147176e-01 + 2.26550832e+00 -2.10464123e-02 + 6.23371684e-01 -5.30768122e-01 + -4.42356624e-01 9.72226986e-01 + 2.31517901e+00 -1.08468105e+00 + 1.97236640e+00 -1.42016619e+00 + 3.18618687e+00 -1.45056343e+00 + -2.75880360e+00 5.40254980e-01 + -1.92916581e+00 1.45029864e-01 + 1.90022524e+00 -6.03805754e-01 + -1.05446211e+00 5.74361752e-01 + 1.45990390e+00 -9.28233993e-01 + 5.14960557e+00 -2.07564096e+00 + -7.53104842e-01 1.55876958e-01 + 8.09490983e-02 -8.58886384e-02 + -1.56894969e+00 4.53497227e-01 + 1.36944658e-01 5.60670875e-01 + -5.32635329e-01 4.40309945e-01 + 1.32507853e+00 -5.83670099e-01 + 1.20676031e+00 -8.02296831e-01 + -3.65023422e+00 1.17211368e+00 + 1.53393850e+00 -6.17771312e-01 + -3.99977129e+00 1.71415137e+00 + 5.70705058e-01 -4.60771539e-01 + -2.20608002e+00 1.07866596e+00 + -1.09040244e+00 6.77441076e-01 + -5.09886482e-01 -1.97282128e-01 + -1.58062785e+00 6.18333697e-01 + -1.53295020e+00 4.02168701e-01 + -5.18580598e-01 2.25767177e-01 + 1.59514316e+00 -2.54983617e-01 + -5.91938655e+00 2.68223782e+00 + 2.84200509e+00 -1.04685313e+00 + 1.31298664e+00 -1.16672614e+00 + -2.36660033e+00 1.81359460e+00 + 6.94163290e-02 3.76658816e-01 + 2.33973934e+00 -8.33173023e-01 + -8.24640389e-01 7.83717285e-01 + -1.02888281e+00 1.04680766e+00 + 1.34750745e+00 -5.89568160e-01 + -2.48761231e+00 7.44199284e-01 + -1.04501559e+00 4.72326911e-01 + -3.14610089e+00 1.89843692e+00 + 2.13003416e-01 5.76633620e-01 + -1.69239608e+00 5.66070021e-01 + 1.80491280e+00 -9.31701080e-01 + -6.94362572e-02 6.96026587e-01 + 1.36502578e+00 -6.85599000e-02 + -7.76764337e-01 3.64328661e-01 + -2.67322167e+00 6.80150021e-01 + 1.84338485e+00 -1.18487494e+00 + 2.88009231e+00 -1.25700411e+00 + 1.17114433e+00 -7.69727080e-01 + 2.11576167e+00 2.81502116e-01 + -1.51470088e+00 2.61553540e-01 + 1.18923669e-01 -1.17890202e-01 + 4.48359786e+00 -1.81427466e+00 + -1.27055948e+00 9.92388998e-01 + -8.00276606e-01 9.11326621e-02 + 7.51764024e-01 -1.03676498e-01 + 1.35769348e-01 -2.11470084e-01 + 2.50731332e+00 -1.12418270e+00 + -2.49752781e-01 7.81224033e-02 + -6.23037902e-01 3.16599691e-01 + -3.93772902e+00 1.37195391e+00 + 1.74256361e+00 -1.12363582e+00 + -1.49737281e+00 5.98828310e-01 + 7.75592115e-01 -4.64733802e-01 + -2.26027693e+00 1.36991118e+00 + -1.62849836e+00 7.36899107e-01 + 2.36850751e+00 -9.32126872e-01 + 5.86169745e+00 -2.49342512e+00 + -5.37092226e-01 1.23821274e+00 + 2.80535867e+00 -1.93363302e+00 + -1.77638106e+00 9.10050276e-01 + 3.02692018e+00 -1.60774676e+00 + 1.97833084e+00 -1.50636531e+00 + 9.09168906e-01 -8.83799359e-01 + 2.39769655e+00 -7.56977869e-01 + 1.47283981e+00 -1.06749890e+00 + 2.92060943e-01 -6.07040605e-01 + -2.09278201e+00 7.71858590e-01 + 7.10015905e-01 -5.42768432e-01 + -2.16826169e-01 1.56897896e-01 + 4.56288247e+00 -2.08912680e+00 + -6.63374020e-01 6.67325183e-01 + 1.80564442e+00 -9.76366134e-01 + 3.28720168e+00 -4.66575145e-01 + -1.60463695e-01 -2.58428153e-01 + 1.78590750e+00 -3.96427146e-01 + 2.75950306e+00 -1.82102856e+00 + -1.18234310e+00 6.28073320e-01 + 4.11415835e+00 -2.33551216e+00 + 1.38721004e+00 -2.77450622e-01 + -2.94903545e+00 1.74813352e+00 + 8.67290400e-01 -6.51667894e-01 + 2.70022274e+00 -8.11832480e-01 + -2.06766146e+00 8.24047249e-01 + 3.90717142e+00 -1.20155758e+00 + -2.95102809e+00 1.36667968e+00 + 6.08815147e+00 -2.60737974e+00 + 2.78576476e+00 -7.86628755e-01 + -3.26258407e+00 1.09302450e+00 + 1.59849422e+00 -1.09705202e+00 + -2.50600710e-01 1.63243175e-01 + -4.90477087e-01 -4.57729572e-01 + -1.24837181e+00 3.22157840e-01 + -2.46341049e+00 1.06517849e+00 + 9.62880751e-01 4.56962496e-01 + 3.99964487e-01 2.07472802e-01 + 6.36657705e-01 -3.46400942e-02 + 4.91231407e-02 -1.40289235e-02 + -4.66683524e-02 -3.72326100e-01 + -5.22049702e-01 -1.70440260e-01 + 5.27062938e-01 -2.32628395e-01 + -2.69440318e+00 1.18914874e+00 + 3.65087539e+00 -1.53427267e+00 + -1.16546364e-01 4.93245392e-02 + 7.55931384e-01 -3.02980139e-01 + 2.06338745e+00 -6.24841225e-01 + 1.31177908e-01 7.29338183e-01 + 1.48021784e+00 -6.39509896e-01 + -5.98656707e-01 2.84525503e-01 + -2.18611080e+00 1.79549812e+00 + -2.91673624e+00 2.15772237e-01 + -8.95591350e-01 7.68250538e-01 + 1.36139762e+00 -1.93845144e-01 + 5.45730414e+00 -2.28114404e+00 + 3.22747247e-01 9.33582332e-01 + -1.46384504e+00 1.12801186e-01 + 4.26728166e-01 -2.33481242e-01 + -1.41327270e+00 8.16103740e-01 + -2.53998067e-01 1.44906646e-01 + -1.32436467e+00 1.87556361e-01 + -3.77313086e+00 1.32896038e+00 + 3.77651731e+00 -1.76548043e+00 + -2.45297093e+00 1.32571926e+00 + -6.55900588e-01 3.56921462e-01 + 9.25558722e-01 -4.51988954e-01 + 1.20732231e+00 -3.02821614e-01 + 3.72660154e-01 -1.89365208e-01 + -1.77090939e+00 9.18087975e-01 + 3.01127567e-01 2.67965829e-01 + -1.76708900e+00 4.62069259e-01 + -2.71812099e+00 1.57233508e+00 + -5.35297633e-01 4.99231535e-01 + 1.50507631e+00 -9.85763646e-01 + 3.00424787e+00 -1.29837562e+00 + -4.99311105e-01 3.91086482e-01 + 1.30125207e+00 -1.26247924e-01 + 4.01699483e-01 -4.46909391e-01 + -1.33635257e+00 5.12068703e-01 + 1.39229757e+00 -9.10974858e-01 + -1.74229508e+00 1.49475978e+00 + -1.21489414e+00 4.04193753e-01 + -3.36537605e-01 -6.74335427e-01 + -2.79186828e-01 8.48314720e-01 + -2.03080140e+00 1.66599815e+00 + -3.53064281e-01 -7.68582906e-04 + -5.30305657e+00 2.91091546e+00 + -1.20049972e+00 8.26578358e-01 + 2.95906989e-01 2.40215920e-01 + -1.42955534e+00 4.63480310e-01 + -1.87856619e+00 8.21459385e-01 + -2.71124720e+00 1.80246843e+00 + -3.06933780e+00 1.22235760e+00 + 5.21935582e-01 -1.27298218e+00 + -1.34175797e+00 7.69018937e-01 + -1.81962785e+00 1.15528991e+00 + -3.99227550e-01 2.93821598e-01 + 1.22533179e+00 -4.73846323e-01 + -2.08068359e-01 -1.75039817e-01 + -2.03068526e+00 1.50370503e+00 + -3.27606113e+00 1.74906330e+00 + -4.37802587e-01 -2.26956048e-01 + -7.69774213e-02 -3.54922468e-01 + 6.47160749e-02 -2.07334721e-01 + -1.37791524e+00 4.43766709e-01 + 3.29846803e+00 -1.04060799e+00 + -3.63704046e+00 1.05800226e+00 + -1.26716116e+00 1.13077353e+00 + 1.98549075e+00 -1.31864807e+00 + 1.85159500e+00 -5.78629560e-01 + -1.55295206e+00 1.23655857e+00 + 6.76026255e-01 9.18824125e-02 + 1.23418960e+00 -4.68162027e-01 + 2.43186642e+00 -9.22422440e-01 + -3.18729701e+00 1.77582673e+00 + -4.02945613e+00 1.14303496e+00 + -1.92694576e-01 1.03301431e-01 + 1.89554730e+00 -4.60128096e-01 + -2.55626581e+00 1.16057084e+00 + 6.89144365e-01 -9.94982900e-01 + -4.44680606e+00 2.19751983e+00 + -3.15196193e+00 1.18762993e+00 + -1.17434977e+00 1.04534656e+00 + 8.58386984e-02 -1.03947487e+00 + 3.33354973e-01 5.54813610e-01 + -9.37631808e-01 3.33450150e-01 + -2.50232471e+00 5.39720635e-01 + 1.03611949e+00 -7.16304095e-02 + -2.05556816e-02 -3.28992265e-01 + -2.24176201e+00 1.13077506e+00 + 4.53583688e+00 -1.10710212e+00 + 4.77389762e-01 -8.99445512e-01 + -2.69075551e+00 6.83176866e-01 + -2.21779724e+00 1.16916849e+00 + -1.09669056e+00 2.10044765e-01 + -8.45367920e-01 -8.45951423e-02 + 4.37558941e-01 -6.95904256e-01 + 1.84884195e+00 -1.71205136e-01 + -8.36371957e-01 5.62862478e-01 + 1.27786531e+00 -1.33362147e+00 + 2.90684492e+00 -7.49892184e-01 + -3.38652716e+00 1.51180670e+00 + -1.30945978e+00 7.09261928e-01 + -7.50471924e-01 -5.24637889e-01 + 1.18580718e+00 -9.97943971e-04 + -7.55395645e+00 3.19273590e+00 + 1.72822535e+00 -1.20996962e+00 + 5.67374320e-01 6.19573416e-01 + -2.99163781e+00 1.79721534e+00 + 1.49862187e+00 -6.05631846e-02 + 1.79503506e+00 -4.90419706e-01 + 3.85626054e+00 -1.95396324e+00 + -9.39188410e-01 7.96498057e-01 + 2.91986664e+00 -1.29392724e+00 + -1.54265750e+00 6.40727933e-01 + 1.14919794e+00 1.20834257e-01 + 2.00936817e+00 -1.53728359e+00 + 3.72468420e+00 -1.38704612e+00 + -1.27794802e+00 3.48543179e-01 + 3.63294077e-01 5.70623314e-01 + 1.49381016e+00 -6.04500534e-01 + 2.98912256e+00 -1.72295726e+00 + -1.80833817e+00 2.94907625e-01 + -3.19669622e+00 1.31888700e+00 + 1.45889401e+00 -8.88448639e-01 + -2.80045388e+00 1.01207060e+00 + -4.78379567e+00 1.48646520e+00 + 2.25510003e+00 -7.13372461e-01 + -9.74441433e-02 -2.17766373e-01 + 2.64468496e-01 -3.60842698e-01 + -5.98821713e+00 3.20197892e+00 + 2.67030213e-01 -5.36386416e-01 + 2.24546960e+00 -8.13464649e-01 + -4.89171414e-01 3.86255031e-01 + -7.45713706e-01 6.29800380e-01 + -3.30460503e-01 3.85127284e-01 + -4.19588147e+00 1.52793198e+00 + 5.42078582e-01 -2.61642741e-02 + 4.24938513e-01 -5.72936751e-01 + 2.82717288e+00 -6.75355024e-01 + -1.44741788e+00 5.03578028e-01 + -1.65547573e+00 7.76444277e-01 + 2.20361170e+00 -1.40835680e+00 + -3.69540235e+00 2.32953767e+00 + -1.41909357e-01 2.28989778e-01 + 1.92838879e+00 -8.72525737e-01 + 1.40708100e+00 -6.81849638e-02 + 1.24988112e+00 -1.39470590e-01 + -2.39435855e+00 7.26587655e-01 + 7.03985028e-01 4.85403277e-02 + 4.05214529e+00 -9.16928318e-01 + 3.74198837e-01 -5.04192358e-01 + -8.43374127e-01 2.36064018e-01 + -3.32253349e-01 7.47840055e-01 + -6.03725210e+00 1.95173337e+00 + 4.60829865e+00 -1.51191309e+00 + -1.46247098e+00 1.11140916e+00 + -9.60111157e-01 -1.23189114e-01 + -7.49613187e-01 4.53614129e-01 + -5.77838219e-01 2.07366469e-02 + 8.07652950e-01 -5.16272662e-01 + -6.02556049e-01 5.05318649e-01 + -1.28712445e-01 2.57836512e-01 + -5.27662820e+00 2.11790737e+00 + 5.40819308e+00 -2.15366022e+00 + 9.37742513e-02 -1.60221751e-01 + 4.55902865e+00 -1.24646307e+00 + -9.06582589e-01 1.92928110e-01 + 2.99928996e+00 -8.04301218e-01 + -3.24317381e+00 1.80076061e+00 + 3.20421743e-01 8.76524679e-01 + -5.29606705e-01 -3.16717696e-01 + -1.77264560e+00 7.52686776e-01 + -1.51706824e+00 8.43755103e-01 + 1.52759111e+00 -7.86814243e-01 + 4.74845617e-01 4.21319700e-01 + 6.97829149e-01 -8.15664881e-01 + 3.09564973e+00 -1.06202469e+00 + 2.95320379e+00 -1.98963943e+00 + -4.23033224e+00 1.41013338e+00 + 1.48576206e+00 8.02908511e-02 + 4.52041627e+00 -2.04620399e+00 + 6.58403922e-01 -7.60781799e-01 + 2.10667543e-01 1.15241731e-01 + 1.77702583e+00 -8.10271859e-01 + 2.41277385e+00 -1.46972042e+00 + 1.50685525e+00 -1.99272545e-01 + 7.61665522e-01 -4.11276152e-01 + 1.18352312e+00 -9.59908608e-01 + -3.32031305e-01 8.07500132e-02 + 1.16813118e+00 -1.73095194e-01 + 1.18363346e+00 -5.41565052e-01 + 5.17702179e-01 -7.62442035e-01 + 4.57401006e-01 -1.45951115e-02 + 1.49377115e-01 2.99571605e-01 + 1.40399453e+00 -1.30160353e+00 + 5.26231567e-01 3.52783752e-01 + -1.91136514e+00 4.24228635e-01 + 1.74156701e+00 -9.92076776e-01 + -4.89323391e+00 2.32483507e+00 + 2.54011209e+00 -8.80366295e-01 + -5.56925706e-01 1.48842026e-01 + -2.35904668e+00 9.60474853e-01 + 1.42216971e+00 -4.67062761e-01 + -1.10809680e+00 7.68684300e-01 + 4.09674726e+00 -1.90795680e+00 + -2.23048923e+00 9.03812542e-01 + 6.57025763e-01 1.36514871e-01 + 2.10944145e+00 -9.78897838e-02 + 1.22552525e+00 -2.50303867e-01 + 2.84620103e-01 -5.30164020e-01 + -2.13562585e+00 1.03503056e+00 + 1.32414902e-01 -8.14190240e-03 + -5.82433561e-01 3.21020292e-01 + -5.06473247e-01 3.11530419e-01 + 1.57162465e+00 -1.20763919e+00 + -1.43155284e+00 -2.51203698e-02 + -1.47093713e+00 -1.39620999e-01 + -2.65765643e+00 1.06091403e+00 + 2.45992927e+00 -5.88815836e-01 + -1.28440162e+00 -1.99377398e-01 + 6.11257504e-01 -3.73577401e-01 + -3.46606103e-01 6.06081290e-01 + 3.76687505e+00 -8.80181424e-01 + -1.03725103e+00 1.45177517e+00 + 2.76659936e+00 -1.09361320e+00 + -3.61311296e+00 9.75032455e-01 + 3.22878655e+00 -9.69497365e-01 + 1.43560379e+00 -5.52524585e-01 + 2.94042153e+00 -1.79747037e+00 + 1.30739580e+00 2.47989248e-01 + -4.05056982e-01 1.22831715e+00 + -2.25827421e+00 2.30604626e-01 + 3.69262926e-01 4.32714650e-02 + -5.52064063e-01 6.07806340e-01 + 7.03325987e+00 -2.17956730e+00 + -2.37823835e-01 -8.28068639e-01 + -4.84279888e-01 5.67765194e-01 + -3.15863410e+00 1.02241617e+00 + -3.39561593e+00 1.36876374e+00 + -2.78482934e+00 6.81641104e-01 + -4.37604334e+00 2.23826340e+00 + -2.54049692e+00 8.22676745e-01 + 3.73264822e+00 -9.93498732e-01 + -3.49536064e+00 1.84771519e+00 + 9.81801604e-01 -5.21278776e-01 + 1.52996831e+00 -1.27386206e+00 + -9.23490293e-01 5.29099482e-01 + -2.76999461e+00 9.24831872e-01 + -3.30029834e-01 -2.49645555e-01 + -1.71156166e+00 5.44940854e-01 + -2.37009487e+00 5.83826982e-01 + -3.03216865e+00 1.04922722e+00 + -2.19539936e+00 1.37558730e+00 + 1.15350207e+00 -6.15318535e-01 + 4.62011792e+00 -2.46714517e+00 + 1.52627952e-02 -1.00618283e-01 + -1.10399342e+00 4.87413533e-01 + 3.55448194e+00 -9.10394190e-01 + -5.21890321e+00 2.44710745e+00 + 1.54289749e+00 -6.54269311e-01 + 2.67935674e+00 -9.92758863e-01 + 1.05801310e+00 2.60054285e-02 + 1.52509097e+00 -4.08768600e-01 + 3.27576917e+00 -1.28769406e+00 + 1.71008412e-01 -2.68739994e-01 + -9.83351344e-04 7.02495897e-02 + -7.60795056e-03 1.61968285e-01 + -1.80620472e+00 4.24934471e-01 + 2.32023297e-02 -2.57284559e-01 + 3.98219478e-01 -4.65361935e-01 + 6.63476988e-01 -3.29823196e-02 + 4.00154707e+00 -1.01792211e+00 + -1.50286870e+00 9.46875359e-01 + -2.22717585e+00 7.50636195e-01 + -3.47381508e-01 -6.51596975e-01 + 2.08076453e+00 -8.22800165e-01 + 2.05099963e+00 -4.00868250e-01 + 3.52576988e-02 -2.54418565e-01 + 1.57342042e+00 -7.62166492e-02 + -1.47019722e+00 3.40861172e-01 + -1.21156090e+00 3.21891246e-01 + 3.79729047e+00 -1.54350764e+00 + 1.26459678e-02 6.99203693e-01 + 1.53974177e-01 4.68643204e-01 + -1.73923561e-01 -1.26229768e-01 + 4.54644993e+00 -2.13951783e+00 + 1.46022547e-01 -4.57084165e-01 + 6.50048037e+00 -2.78872609e+00 + -1.51934912e+00 1.03216768e+00 + -3.06483575e+00 1.81101446e+00 + -2.38212125e+00 9.19559042e-01 + -1.81319611e+00 8.10545112e-01 + 1.70951294e+00 -6.10712680e-01 + 1.67974156e+00 -1.51241453e+00 + -5.94795113e+00 2.56893813e+00 + 3.62633110e-01 -7.46965304e-01 + -2.44042594e+00 8.52761797e-01 + 3.32412550e+00 -1.28439899e+00 + 4.74860766e+00 -1.72821964e+00 + 1.29072541e+00 -8.24872902e-01 + -1.69450702e+00 4.09600876e-01 + 1.29705411e+00 1.22300809e-01 + -2.63597613e+00 8.55612913e-01 + 9.28467301e-01 -2.63550114e-02 + 2.44670264e+00 -4.10123002e-01 + 1.06408206e+00 -5.03361942e-01 + 5.12384049e-02 -1.27116595e-02 + -1.06731272e+00 -1.76205029e-01 + -9.45454582e-01 3.74404917e-01 + 2.54343689e+00 -7.13810545e-01 + -2.54460335e+00 1.31590265e+00 + 1.89864233e+00 -3.98436339e-01 + -1.93990133e+00 6.01474630e-01 + -1.35938824e+00 4.00751788e-01 + 2.38567018e+00 -6.13904880e-01 + 2.18748050e-01 2.62631712e-01 + -2.01388788e+00 1.41474031e+00 + 2.74014581e+00 -1.27448105e+00 + -2.13828583e+00 1.13616144e+00 + 5.98730932e+00 -2.53430080e+00 + -1.72872795e+00 1.53702057e+00 + -2.53263962e+00 1.27342410e+00 + 1.34326968e+00 -1.99395088e-01 + 3.83352666e-01 -1.25683065e-01 + -2.35630657e+00 5.54116983e-01 + -1.94900838e+00 5.76270178e-01 + -1.36699108e+00 -3.40904824e-01 + -2.34727346e+00 -1.93054940e-02 + -3.82779777e+00 1.83025664e+00 + -4.31602080e+00 9.21605705e-01 + 5.54098133e-01 2.33991419e-01 + -4.53591188e+00 1.99833353e+00 + -3.92715909e+00 1.83231482e+00 + 3.91344440e-01 -1.11355111e-01 + 3.48576363e+00 -1.41379449e+00 + -1.42858690e+00 3.84532286e-01 + 1.79519859e+00 -9.23486448e-01 + 8.49691242e-01 -1.76551084e-01 + 1.53618138e+00 8.23835015e-02 + 5.91476520e-02 3.88296940e-02 + 1.44837346e+00 -7.24097604e-01 + -6.79008418e-01 4.04078097e-01 + 2.87555510e+00 -9.51825076e-01 + -1.12379101e+00 2.93457714e-01 + 1.45263980e+00 -6.01960544e-01 + -2.55741621e-01 9.26233518e-01 + 3.54570714e+00 -1.41521877e+00 + -1.61542388e+00 6.57844512e-01 + -3.22844269e-01 3.02823546e-01 + 1.03523913e+00 -6.92730711e-01 + 1.11084909e+00 -3.50823642e-01 + 3.41268693e+00 -1.90865862e+00 + 7.67062858e-01 -9.48792160e-01 + -5.49798016e+00 1.71139960e+00 + 1.14865798e+00 -6.12669150e-01 + -2.18256680e+00 7.78634462e-01 + 4.78857389e+00 -2.55555085e+00 + -1.85555569e+00 8.04311615e-01 + -4.22278799e+00 2.01162524e+00 + -1.56556149e+00 1.54353907e+00 + -3.11527864e+00 1.65973526e+00 + 2.66342611e+00 -1.20449402e+00 + 1.57635314e+00 -1.48716308e-01 + -6.35606865e-01 2.59701180e-01 + 1.02431976e+00 -6.76929904e-01 + 1.12973772e+00 1.49473892e-02 + -9.12758116e-01 2.21533933e-01 + -2.98014470e+00 1.71651189e+00 + 2.74016965e+00 -9.47893923e-01 + -3.47830591e+00 1.34941430e+00 + 1.74757562e+00 -3.72503752e-01 + 5.55820383e-01 -6.47992466e-01 + -1.19871928e+00 9.82429151e-01 + -2.53040133e+00 2.10671307e+00 + -1.94085605e+00 1.38938137e+00 diff --git a/docs/mllib-clustering.md b/docs/mllib-clustering.md index 413b824e369da..1e9ef345b7435 100644 --- a/docs/mllib-clustering.md +++ b/docs/mllib-clustering.md @@ -16,9 +16,12 @@ often used for exploratory analysis and/or as a component of a hierarchical supervised learning pipeline (in which distinct classifiers or regression models are trained for each cluster). -MLlib supports -[k-means](http://en.wikipedia.org/wiki/K-means_clustering) clustering, one of -the most commonly used clustering algorithms that clusters the data points into +MLlib supports the following models: + +### k-means + +[k-means](http://en.wikipedia.org/wiki/K-means_clustering) is one of the +most commonly used clustering algorithms that clusters the data points into a predefined number of clusters. The MLlib implementation includes a parallelized variant of the [k-means++](http://en.wikipedia.org/wiki/K-means%2B%2B) method called [kmeans||](http://theory.stanford.edu/~sergei/papers/vldb12-kmpar.pdf). @@ -34,6 +37,20 @@ a given dataset, the algorithm returns the best clustering result). * *initializationSteps* determines the number of steps in the k-means\|\| algorithm. * *epsilon* determines the distance threshold within which we consider k-means to have converged. +### Gaussian mixture + +A [Gaussian Mixture Model](http://en.wikipedia.org/wiki/Mixture_model#Multivariate_Gaussian_mixture_model) +represents a composite distribution whereby points are drawn from one of *k* Gaussian sub-distributions, +each with its own probability. The MLlib implementation uses the +[expectation-maximization](http://en.wikipedia.org/wiki/Expectation%E2%80%93maximization_algorithm) + algorithm to induce the maximum-likelihood model given a set of samples. The implementation +has the following parameters: + +* *k* is the number of desired clusters. +* *convergenceTol* is the maximum change in log-likelihood at which we consider convergence achieved. +* *maxIterations* is the maximum number of iterations to perform without reaching convergence. +* *initialModel* is an optional starting point from which to start the EM algorithm. If this parameter is omitted, a random starting point will be constructed from the data. + ### Power Iteration Clustering Power iteration clustering is a scalable and efficient algorithm for clustering points given pointwise mutual affinity values. Internally the algorithm: @@ -56,6 +73,8 @@ Example outputs for a dataset inspired by the paper - but with five clusters ins ### Examples +#### k-means +

The following code snippets can be executed in `spark-shell`. @@ -168,6 +187,112 @@ print("Within Set Sum of Squared Error = " + str(WSSSE))
+#### GaussianMixture + +
+
+In the following example after loading and parsing data, we use a +[GaussianMixture](api/scala/index.html#org.apache.spark.mllib.clustering.GaussianMixture) +object to cluster the data into two clusters. The number of desired clusters is passed +to the algorithm. We then output the parameters of the mixture model. + +{% highlight scala %} +import org.apache.spark.mllib.clustering.GaussianMixture +import org.apache.spark.mllib.linalg.Vectors + +// Load and parse the data +val data = sc.textFile("data/mllib/gmm_data.txt") +val parsedData = data.map(s => Vectors.dense(s.trim.split(' ').map(_.toDouble))).cache() + +// Cluster the data into two classes using GaussianMixture +val gmm = new GaussianMixture().setK(2).run(parsedData) + +// output parameters of max-likelihood model +for (i <- 0 until gmm.k) { + println("weight=%f\nmu=%s\nsigma=\n%s\n" format + (gmm.weights(i), gmm.gaussians(i).mu, gmm.gaussians(i).sigma)) +} + +{% endhighlight %} +
+ +
+All of MLlib's methods use Java-friendly types, so you can import and call them there the same +way you do in Scala. The only caveat is that the methods take Scala RDD objects, while the +Spark Java API uses a separate `JavaRDD` class. You can convert a Java RDD to a Scala one by +calling `.rdd()` on your `JavaRDD` object. A self-contained application example +that is equivalent to the provided example in Scala is given below: + +{% highlight java %} +import org.apache.spark.api.java.*; +import org.apache.spark.api.java.function.Function; +import org.apache.spark.mllib.clustering.GaussianMixture; +import org.apache.spark.mllib.clustering.GaussianMixtureModel; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.SparkConf; + +public class GaussianMixtureExample { + public static void main(String[] args) { + SparkConf conf = new SparkConf().setAppName("GaussianMixture Example"); + JavaSparkContext sc = new JavaSparkContext(conf); + + // Load and parse data + String path = "data/mllib/gmm_data.txt"; + JavaRDD data = sc.textFile(path); + JavaRDD parsedData = data.map( + new Function() { + public Vector call(String s) { + String[] sarray = s.trim().split(" "); + double[] values = new double[sarray.length]; + for (int i = 0; i < sarray.length; i++) + values[i] = Double.parseDouble(sarray[i]); + return Vectors.dense(values); + } + } + ); + parsedData.cache(); + + // Cluster the data into two classes using GaussianMixture + GaussianMixtureModel gmm = new GaussianMixture().setK(2).run(parsedData.rdd()); + + // Output the parameters of the mixture model + for(int j=0; j + +
+In the following example after loading and parsing data, we use a +[GaussianMixture](api/python/pyspark.mllib.html#pyspark.mllib.clustering.GaussianMixture) +object to cluster the data into two clusters. The number of desired clusters is passed +to the algorithm. We then output the parameters of the mixture model. + +{% highlight python %} +from pyspark.mllib.clustering import GaussianMixture +from numpy import array + +# Load and parse the data +data = sc.textFile("data/mllib/gmm_data.txt") +parsedData = data.map(lambda line: array([float(x) for x in line.strip().split(' ')])) + +# Build the model (cluster the data) +gmm = GaussianMixture.train(parsedData, 2) + +# output parameters of model +for i in range(2): + print ("weight = ", gmm.weights[i], "mu = ", gmm.gaussians[i].mu, + "sigma = ", gmm.gaussians[i].sigma.toArray()) + +{% endhighlight %} +
+ +
+ In order to run the above application, follow the instructions provided in the [Self-Contained Applications](quick-start.html#self-contained-applications) section of the Spark diff --git a/docs/mllib-guide.md b/docs/mllib-guide.md index 73728bb35eb96..7779fbc9c49e4 100644 --- a/docs/mllib-guide.md +++ b/docs/mllib-guide.md @@ -25,6 +25,8 @@ filtering, dimensionality reduction, as well as underlying optimization primitiv * alternating least squares (ALS) * [Clustering](mllib-clustering.html) * k-means + * Gaussian mixture + * power iteration * [Dimensionality reduction](mllib-dimensionality-reduction.html) * singular value decomposition (SVD) * principal component analysis (PCA) From cc6e53119d7a51b95b19244f50b25814088b4d11 Mon Sep 17 00:00:00 2001 From: lianhuiwang Date: Fri, 6 Feb 2015 10:48:31 -0800 Subject: [PATCH 13/53] [SPARK-5653][YARN] In ApplicationMaster rename isDriver to isClusterMode in ApplicationMaster rename isDriver to isClusterMode,because in Client it uses isClusterMode,ApplicationMaster should keep consistent with it and uses isClusterMode.Also isClusterMode is easier to understand. andrewor14 sryza Author: lianhuiwang Closes #4430 from lianhuiwang/am-isDriver-rename and squashes the following commits: f9f3ed0 [lianhuiwang] rename isDriver to isClusterMode --- .../spark/deploy/yarn/ApplicationMaster.scala | 24 +++++++++---------- 1 file changed, 12 insertions(+), 12 deletions(-) diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala index 37e98e01fddf7..4cc320c5d59b5 100644 --- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala +++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala @@ -54,7 +54,7 @@ private[spark] class ApplicationMaster( private val sparkConf = new SparkConf() private val yarnConf: YarnConfiguration = SparkHadoopUtil.get.newConfiguration(sparkConf) .asInstanceOf[YarnConfiguration] - private val isDriver = args.userClass != null + private val isClusterMode = args.userClass != null // Default to numExecutors * 2, with minimum of 3 private val maxNumExecutorFailures = sparkConf.getInt("spark.yarn.max.executor.failures", @@ -81,7 +81,7 @@ private[spark] class ApplicationMaster( try { val appAttemptId = client.getAttemptId() - if (isDriver) { + if (isClusterMode) { // Set the web ui port to be ephemeral for yarn so we don't conflict with // other spark processes running on the same box System.setProperty("spark.ui.port", "0") @@ -139,7 +139,7 @@ private[spark] class ApplicationMaster( // doAs in order for the credentials to be passed on to the executor containers. val securityMgr = new SecurityManager(sparkConf) - if (isDriver) { + if (isClusterMode) { runDriver(securityMgr) } else { runExecutorLauncher(securityMgr) @@ -162,7 +162,7 @@ private[spark] class ApplicationMaster( * from the application code. */ final def getDefaultFinalStatus() = { - if (isDriver) { + if (isClusterMode) { FinalApplicationStatus.SUCCEEDED } else { FinalApplicationStatus.UNDEFINED @@ -243,7 +243,7 @@ private[spark] class ApplicationMaster( private def runAMActor( host: String, port: String, - isDriver: Boolean): Unit = { + isClusterMode: Boolean): Unit = { val driverUrl = AkkaUtils.address( AkkaUtils.protocol(actorSystem), @@ -251,7 +251,7 @@ private[spark] class ApplicationMaster( host, port, YarnSchedulerBackend.ACTOR_NAME) - actor = actorSystem.actorOf(Props(new AMActor(driverUrl, isDriver)), name = "YarnAM") + actor = actorSystem.actorOf(Props(new AMActor(driverUrl, isClusterMode)), name = "YarnAM") } private def runDriver(securityMgr: SecurityManager): Unit = { @@ -272,7 +272,7 @@ private[spark] class ApplicationMaster( runAMActor( sc.getConf.get("spark.driver.host"), sc.getConf.get("spark.driver.port"), - isDriver = true) + isClusterMode = true) registerAM(sc.ui.map(_.appUIAddress).getOrElse(""), securityMgr) userClassThread.join() } @@ -427,7 +427,7 @@ private[spark] class ApplicationMaster( sparkConf.set("spark.driver.host", driverHost) sparkConf.set("spark.driver.port", driverPort.toString) - runAMActor(driverHost, driverPort.toString, isDriver = false) + runAMActor(driverHost, driverPort.toString, isClusterMode = false) } /** Add the Yarn IP filter that is required for properly securing the UI. */ @@ -435,7 +435,7 @@ private[spark] class ApplicationMaster( val proxyBase = System.getenv(ApplicationConstants.APPLICATION_WEB_PROXY_BASE_ENV) val amFilter = "org.apache.hadoop.yarn.server.webproxy.amfilter.AmIpFilter" val params = client.getAmIpFilterParams(yarnConf, proxyBase) - if (isDriver) { + if (isClusterMode) { System.setProperty("spark.ui.filters", amFilter) params.foreach { case (k, v) => System.setProperty(s"spark.$amFilter.param.$k", v) } } else { @@ -491,7 +491,7 @@ private[spark] class ApplicationMaster( /** * An actor that communicates with the driver's scheduler backend. */ - private class AMActor(driverUrl: String, isDriver: Boolean) extends Actor { + private class AMActor(driverUrl: String, isClusterMode: Boolean) extends Actor { var driver: ActorSelection = _ override def preStart() = { @@ -503,7 +503,7 @@ private[spark] class ApplicationMaster( driver ! RegisterClusterManager // In cluster mode, the AM can directly monitor the driver status instead // of trying to deduce it from the lifecycle of the driver's actor - if (!isDriver) { + if (!isClusterMode) { context.system.eventStream.subscribe(self, classOf[RemotingLifecycleEvent]) } } @@ -513,7 +513,7 @@ private[spark] class ApplicationMaster( logInfo(s"Driver terminated or disconnected! Shutting down. $x") // In cluster mode, do not rely on the disassociated event to exit // This avoids potentially reporting incorrect exit codes if the driver fails - if (!isDriver) { + if (!isClusterMode) { finish(FinalApplicationStatus.SUCCEEDED, ApplicationMaster.EXIT_SUCCESS) } From 1a88f20de798030a7d5713bd267f612ba5617fca Mon Sep 17 00:00:00 2001 From: Sandy Ryza Date: Fri, 6 Feb 2015 10:53:16 -0800 Subject: [PATCH 14/53] SPARK-4337. [YARN] Add ability to cancel pending requests Author: Sandy Ryza Closes #4141 from sryza/sandy-spark-4337 and squashes the following commits: a98bd20 [Sandy Ryza] Andrew's comments cdaab7f [Sandy Ryza] SPARK-4337. Add ability to cancel pending requests to YARN --- .../spark/deploy/yarn/YarnAllocator.scala | 65 ++++++++++++------- .../deploy/yarn/YarnAllocatorSuite.scala | 54 +++++++++++++-- 2 files changed, 89 insertions(+), 30 deletions(-) diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnAllocator.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnAllocator.scala index 0dbb6154b3039..12c62a659d799 100644 --- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnAllocator.scala +++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnAllocator.scala @@ -69,8 +69,7 @@ private[yarn] class YarnAllocator( } // Visible for testing. - val allocatedHostToContainersMap = - new HashMap[String, collection.mutable.Set[ContainerId]] + val allocatedHostToContainersMap = new HashMap[String, collection.mutable.Set[ContainerId]] val allocatedContainerToHostMap = new HashMap[ContainerId, String] // Containers that we no longer care about. We've either already told the RM to release them or @@ -84,7 +83,7 @@ private[yarn] class YarnAllocator( private var executorIdCounter = 0 @volatile private var numExecutorsFailed = 0 - @volatile private var maxExecutors = args.numExecutors + @volatile private var targetNumExecutors = args.numExecutors // Keep track of which container is running which executor to remove the executors later private val executorIdToContainer = new HashMap[String, Container] @@ -133,10 +132,12 @@ private[yarn] class YarnAllocator( amClient.getMatchingRequests(RM_REQUEST_PRIORITY, location, resource).map(_.size).sum /** - * Request as many executors from the ResourceManager as needed to reach the desired total. + * Request as many executors from the ResourceManager as needed to reach the desired total. If + * the requested total is smaller than the current number of running executors, no executors will + * be killed. */ def requestTotalExecutors(requestedTotal: Int): Unit = synchronized { - maxExecutors = requestedTotal + targetNumExecutors = requestedTotal } /** @@ -147,8 +148,8 @@ private[yarn] class YarnAllocator( val container = executorIdToContainer.remove(executorId).get internalReleaseContainer(container) numExecutorsRunning -= 1 - maxExecutors -= 1 - assert(maxExecutors >= 0, "Allocator killed more executors than are allocated!") + targetNumExecutors -= 1 + assert(targetNumExecutors >= 0, "Allocator killed more executors than are allocated!") } else { logWarning(s"Attempted to kill unknown executor $executorId!") } @@ -163,15 +164,8 @@ private[yarn] class YarnAllocator( * This must be synchronized because variables read in this method are mutated by other methods. */ def allocateResources(): Unit = synchronized { - val numPendingAllocate = getNumPendingAllocate - val missing = maxExecutors - numPendingAllocate - numExecutorsRunning + updateResourceRequests() - if (missing > 0) { - logInfo(s"Will request $missing executor containers, each with ${resource.getVirtualCores} " + - s"cores and ${resource.getMemory} MB memory including $memoryOverhead MB overhead") - } - - addResourceRequests(missing) val progressIndicator = 0.1f // Poll the ResourceManager. This doubles as a heartbeat if there are no pending container // requests. @@ -201,15 +195,36 @@ private[yarn] class YarnAllocator( } /** - * Request numExecutors additional containers from YARN. Visible for testing. + * Update the set of container requests that we will sync with the RM based on the number of + * executors we have currently running and our target number of executors. + * + * Visible for testing. */ - def addResourceRequests(numExecutors: Int): Unit = { - for (i <- 0 until numExecutors) { - val request = new ContainerRequest(resource, null, null, RM_REQUEST_PRIORITY) - amClient.addContainerRequest(request) - val nodes = request.getNodes - val hostStr = if (nodes == null || nodes.isEmpty) "Any" else nodes.last - logInfo("Container request (host: %s, capability: %s".format(hostStr, resource)) + def updateResourceRequests(): Unit = { + val numPendingAllocate = getNumPendingAllocate + val missing = targetNumExecutors - numPendingAllocate - numExecutorsRunning + + if (missing > 0) { + logInfo(s"Will request $missing executor containers, each with ${resource.getVirtualCores} " + + s"cores and ${resource.getMemory} MB memory including $memoryOverhead MB overhead") + + for (i <- 0 until missing) { + val request = new ContainerRequest(resource, null, null, RM_REQUEST_PRIORITY) + amClient.addContainerRequest(request) + val nodes = request.getNodes + val hostStr = if (nodes == null || nodes.isEmpty) "Any" else nodes.last + logInfo(s"Container request (host: $hostStr, capability: $resource)") + } + } else if (missing < 0) { + val numToCancel = math.min(numPendingAllocate, -missing) + logInfo(s"Canceling requests for $numToCancel executor containers") + + val matchingRequests = amClient.getMatchingRequests(RM_REQUEST_PRIORITY, ANY_HOST, resource) + if (!matchingRequests.isEmpty) { + matchingRequests.head.take(numToCancel).foreach(amClient.removeContainerRequest) + } else { + logWarning("Expected to find pending requests, but found none.") + } } } @@ -266,7 +281,7 @@ private[yarn] class YarnAllocator( * containersToUse or remaining. * * @param allocatedContainer container that was given to us by YARN - * @location resource name, either a node, rack, or * + * @param location resource name, either a node, rack, or * * @param containersToUse list of containers that will be used * @param remaining list of containers that will not be used */ @@ -294,7 +309,7 @@ private[yarn] class YarnAllocator( private def runAllocatedContainers(containersToUse: ArrayBuffer[Container]): Unit = { for (container <- containersToUse) { numExecutorsRunning += 1 - assert(numExecutorsRunning <= maxExecutors) + assert(numExecutorsRunning <= targetNumExecutors) val executorHostname = container.getNodeId.getHost val containerId = container.getId executorIdCounter += 1 diff --git a/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnAllocatorSuite.scala b/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnAllocatorSuite.scala index 024b25f9d3365..3c224f148802e 100644 --- a/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnAllocatorSuite.scala +++ b/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnAllocatorSuite.scala @@ -107,8 +107,8 @@ class YarnAllocatorSuite extends FunSuite with Matchers with BeforeAndAfterEach test("single container allocated") { // request a single container and receive it - val handler = createAllocator() - handler.addResourceRequests(1) + val handler = createAllocator(1) + handler.updateResourceRequests() handler.getNumExecutorsRunning should be (0) handler.getNumPendingAllocate should be (1) @@ -123,8 +123,8 @@ class YarnAllocatorSuite extends FunSuite with Matchers with BeforeAndAfterEach test("some containers allocated") { // request a few containers and receive some of them - val handler = createAllocator() - handler.addResourceRequests(4) + val handler = createAllocator(4) + handler.updateResourceRequests() handler.getNumExecutorsRunning should be (0) handler.getNumPendingAllocate should be (4) @@ -144,7 +144,7 @@ class YarnAllocatorSuite extends FunSuite with Matchers with BeforeAndAfterEach test("receive more containers than requested") { val handler = createAllocator(2) - handler.addResourceRequests(2) + handler.updateResourceRequests() handler.getNumExecutorsRunning should be (0) handler.getNumPendingAllocate should be (2) @@ -162,6 +162,50 @@ class YarnAllocatorSuite extends FunSuite with Matchers with BeforeAndAfterEach handler.allocatedHostToContainersMap.contains("host4") should be (false) } + test("decrease total requested executors") { + val handler = createAllocator(4) + handler.updateResourceRequests() + handler.getNumExecutorsRunning should be (0) + handler.getNumPendingAllocate should be (4) + + handler.requestTotalExecutors(3) + handler.updateResourceRequests() + handler.getNumPendingAllocate should be (3) + + val container = createContainer("host1") + handler.handleAllocatedContainers(Array(container)) + + handler.getNumExecutorsRunning should be (1) + handler.allocatedContainerToHostMap.get(container.getId).get should be ("host1") + handler.allocatedHostToContainersMap.get("host1").get should contain (container.getId) + + handler.requestTotalExecutors(2) + handler.updateResourceRequests() + handler.getNumPendingAllocate should be (1) + } + + test("decrease total requested executors to less than currently running") { + val handler = createAllocator(4) + handler.updateResourceRequests() + handler.getNumExecutorsRunning should be (0) + handler.getNumPendingAllocate should be (4) + + handler.requestTotalExecutors(3) + handler.updateResourceRequests() + handler.getNumPendingAllocate should be (3) + + val container1 = createContainer("host1") + val container2 = createContainer("host2") + handler.handleAllocatedContainers(Array(container1, container2)) + + handler.getNumExecutorsRunning should be (2) + + handler.requestTotalExecutors(1) + handler.updateResourceRequests() + handler.getNumPendingAllocate should be (0) + handler.getNumExecutorsRunning should be (2) + } + test("memory exceeded diagnostic regexes") { val diagnostics = "Container [pid=12465,containerID=container_1412887393566_0003_01_000002] is running " + From fe3740c4c859d087b714c666741a29061bba5f58 Mon Sep 17 00:00:00 2001 From: Andrew Or Date: Fri, 6 Feb 2015 10:54:23 -0800 Subject: [PATCH 15/53] [SPARK-5636] Ramp up faster in dynamic allocation A recent patch #4051 made the initial number default to 0. With this change, any Spark application using dynamic allocation's default settings will ramp up very slowly. Since we never request more executors than needed to saturate the pending tasks, it is safe to ramp up quickly. The current default of 60 may be too slow. Author: Andrew Or Closes #4409 from andrewor14/dynamic-allocation-interval and squashes the following commits: d3cc485 [Andrew Or] Lower request interval --- .../scala/org/apache/spark/ExecutorAllocationManager.scala | 6 +++--- docs/configuration.md | 2 +- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala b/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala index 5d5288bb6e60d..8b38366e0325c 100644 --- a/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala +++ b/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala @@ -76,15 +76,15 @@ private[spark] class ExecutorAllocationManager( private val maxNumExecutors = conf.getInt("spark.dynamicAllocation.maxExecutors", Integer.MAX_VALUE) - // How long there must be backlogged tasks for before an addition is triggered + // How long there must be backlogged tasks for before an addition is triggered (seconds) private val schedulerBacklogTimeout = conf.getLong( - "spark.dynamicAllocation.schedulerBacklogTimeout", 60) + "spark.dynamicAllocation.schedulerBacklogTimeout", 5) // Same as above, but used only after `schedulerBacklogTimeout` is exceeded private val sustainedSchedulerBacklogTimeout = conf.getLong( "spark.dynamicAllocation.sustainedSchedulerBacklogTimeout", schedulerBacklogTimeout) - // How long an executor must be idle for before it is removed + // How long an executor must be idle for before it is removed (seconds) private val executorIdleTimeout = conf.getLong( "spark.dynamicAllocation.executorIdleTimeout", 600) diff --git a/docs/configuration.md b/docs/configuration.md index 4c86cb7c16238..00e973c245005 100644 --- a/docs/configuration.md +++ b/docs/configuration.md @@ -1140,7 +1140,7 @@ Apart from these, the following properties are also available, and may be useful spark.dynamicAllocation.schedulerBacklogTimeout - 60 + 5 If dynamic allocation is enabled and there have been pending tasks backlogged for more than this duration (in seconds), new executors will be requested. For more detail, see this From c01b9852ea2f7d453249b07d89e62af71bd26e3d Mon Sep 17 00:00:00 2001 From: Masayoshi TSUZUKI Date: Fri, 6 Feb 2015 10:58:26 -0800 Subject: [PATCH 16/53] [SPARK-5396] Syntax error in spark scripts on windows. Modified syntax error in spark-submit2.cmd. Command prompt doesn't have "defined" operator. Author: Masayoshi TSUZUKI Closes #4428 from tsudukim/feature/SPARK-5396 and squashes the following commits: ec18465 [Masayoshi TSUZUKI] [SPARK-5396] Syntax error in spark scripts on windows. --- bin/spark-submit2.cmd | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/bin/spark-submit2.cmd b/bin/spark-submit2.cmd index 12244a9cb04fb..446cbc74b74f9 100644 --- a/bin/spark-submit2.cmd +++ b/bin/spark-submit2.cmd @@ -25,7 +25,7 @@ set ORIG_ARGS=%* rem Reset the values of all variables used set SPARK_SUBMIT_DEPLOY_MODE=client -if not defined %SPARK_CONF_DIR% ( +if [%SPARK_CONF_DIR%] == [] ( set SPARK_CONF_DIR=%SPARK_HOME%\conf ) set SPARK_SUBMIT_PROPERTIES_FILE=%SPARK_CONF_DIR%\spark-defaults.conf From 9792bec596113a6f5f4534772b7539255403b082 Mon Sep 17 00:00:00 2001 From: Stephen Haberman Date: Fri, 6 Feb 2015 11:03:56 -0800 Subject: [PATCH 17/53] [SPARK-4877] Allow user first classes to extend classes in the parent. Previously, the classloader isolation was almost too good, such that if a child class needed to load/reference a class that was only available in the parent, it could not do so. This adds tests for that case, the user-first Fake2 class extends the only-in-parent Fake3 class. It also sneaks in a fix where only the first stage seemed to work, and on subsequent stages, a LinkageError happened because classes from the user-first classpath were getting defined twice. Author: Stephen Haberman Closes #3725 from stephenh/4877_user_first_parent_inheritance and squashes the following commits: dabcd35 [Stephen Haberman] [SPARK-4877] Respect userClassPathFirst for the driver code too. 3d0fa7c [Stephen Haberman] [SPARK-4877] Allow user first classes to extend classes in the parent. --- .../scala/org/apache/spark/TestUtils.scala | 34 +++++++++++++++---- .../org/apache/spark/deploy/SparkSubmit.scala | 17 ++++++---- .../executor/ExecutorURLClassLoader.scala | 12 ++++++- .../ExecutorURLClassLoaderSuite.scala | 16 ++++++--- 4 files changed, 61 insertions(+), 18 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/TestUtils.scala b/core/src/main/scala/org/apache/spark/TestUtils.scala index 34078142f5385..be081c3825566 100644 --- a/core/src/main/scala/org/apache/spark/TestUtils.scala +++ b/core/src/main/scala/org/apache/spark/TestUtils.scala @@ -43,11 +43,20 @@ private[spark] object TestUtils { * Note: if this is used during class loader tests, class names should be unique * in order to avoid interference between tests. */ - def createJarWithClasses(classNames: Seq[String], value: String = ""): URL = { + def createJarWithClasses( + classNames: Seq[String], + toStringValue: String = "", + classNamesWithBase: Seq[(String, String)] = Seq(), + classpathUrls: Seq[URL] = Seq()): URL = { val tempDir = Utils.createTempDir() - val files = for (name <- classNames) yield createCompiledClass(name, tempDir, value) + val files1 = for (name <- classNames) yield { + createCompiledClass(name, tempDir, toStringValue, classpathUrls = classpathUrls) + } + val files2 = for ((childName, baseName) <- classNamesWithBase) yield { + createCompiledClass(childName, tempDir, toStringValue, baseName, classpathUrls) + } val jarFile = new File(tempDir, "testJar-%s.jar".format(System.currentTimeMillis())) - createJar(files, jarFile) + createJar(files1 ++ files2, jarFile) } @@ -85,15 +94,26 @@ private[spark] object TestUtils { } /** Creates a compiled class with the given name. Class file will be placed in destDir. */ - def createCompiledClass(className: String, destDir: File, value: String = ""): File = { + def createCompiledClass( + className: String, + destDir: File, + toStringValue: String = "", + baseClass: String = null, + classpathUrls: Seq[URL] = Seq()): File = { val compiler = ToolProvider.getSystemJavaCompiler + val extendsText = Option(baseClass).map { c => s" extends ${c}" }.getOrElse("") val sourceFile = new JavaSourceFromString(className, - "public class " + className + " implements java.io.Serializable {" + - " @Override public String toString() { return \"" + value + "\"; }}") + "public class " + className + extendsText + " implements java.io.Serializable {" + + " @Override public String toString() { return \"" + toStringValue + "\"; }}") // Calling this outputs a class file in pwd. It's easier to just rename the file than // build a custom FileManager that controls the output location. - compiler.getTask(null, null, null, null, null, Seq(sourceFile)).call() + val options = if (classpathUrls.nonEmpty) { + Seq("-classpath", classpathUrls.map { _.getFile }.mkString(File.pathSeparator)) + } else { + Seq() + } + compiler.getTask(null, null, null, options, null, Seq(sourceFile)).call() val fileName = className + ".class" val result = new File(fileName) diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala index 8bbfcd2997dc6..9d25e647f1703 100644 --- a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala +++ b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala @@ -20,11 +20,9 @@ package org.apache.spark.deploy import java.io.{File, PrintStream} import java.lang.reflect.{Modifier, InvocationTargetException} import java.net.URL - import scala.collection.mutable.{ArrayBuffer, HashMap, Map} import org.apache.hadoop.fs.Path - import org.apache.ivy.Ivy import org.apache.ivy.core.LogOptions import org.apache.ivy.core.module.descriptor.{DefaultExcludeRule, DefaultDependencyDescriptor, DefaultModuleDescriptor} @@ -35,9 +33,10 @@ import org.apache.ivy.core.retrieve.RetrieveOptions import org.apache.ivy.core.settings.IvySettings import org.apache.ivy.plugins.matcher.GlobPatternMatcher import org.apache.ivy.plugins.resolver.{ChainResolver, IBiblioResolver} - import org.apache.spark.executor.ExecutorURLClassLoader import org.apache.spark.util.Utils +import org.apache.spark.executor.ChildExecutorURLClassLoader +import org.apache.spark.executor.MutableURLClassLoader /** * Main gateway of launching a Spark application. @@ -389,8 +388,14 @@ object SparkSubmit { printStream.println("\n") } - val loader = new ExecutorURLClassLoader(new Array[URL](0), - Thread.currentThread.getContextClassLoader) + val loader = + if (sysProps.getOrElse("spark.files.userClassPathFirst", "false").toBoolean) { + new ChildExecutorURLClassLoader(new Array[URL](0), + Thread.currentThread.getContextClassLoader) + } else { + new ExecutorURLClassLoader(new Array[URL](0), + Thread.currentThread.getContextClassLoader) + } Thread.currentThread.setContextClassLoader(loader) for (jar <- childClasspath) { @@ -434,7 +439,7 @@ object SparkSubmit { } } - private def addJarToClasspath(localJar: String, loader: ExecutorURLClassLoader) { + private def addJarToClasspath(localJar: String, loader: MutableURLClassLoader) { val uri = Utils.resolveURI(localJar) uri.getScheme match { case "file" | "local" => diff --git a/core/src/main/scala/org/apache/spark/executor/ExecutorURLClassLoader.scala b/core/src/main/scala/org/apache/spark/executor/ExecutorURLClassLoader.scala index 218ed7b5d2d39..8011e75944aac 100644 --- a/core/src/main/scala/org/apache/spark/executor/ExecutorURLClassLoader.scala +++ b/core/src/main/scala/org/apache/spark/executor/ExecutorURLClassLoader.scala @@ -39,7 +39,17 @@ private[spark] class ChildExecutorURLClassLoader(urls: Array[URL], parent: Class super.addURL(url) } override def findClass(name: String): Class[_] = { - super.findClass(name) + val loaded = super.findLoadedClass(name) + if (loaded != null) { + return loaded + } + try { + super.findClass(name) + } catch { + case e: ClassNotFoundException => { + parentClassLoader.loadClass(name) + } + } } } diff --git a/core/src/test/scala/org/apache/spark/executor/ExecutorURLClassLoaderSuite.scala b/core/src/test/scala/org/apache/spark/executor/ExecutorURLClassLoaderSuite.scala index e2050e95a1b88..b7912c09d1410 100644 --- a/core/src/test/scala/org/apache/spark/executor/ExecutorURLClassLoaderSuite.scala +++ b/core/src/test/scala/org/apache/spark/executor/ExecutorURLClassLoaderSuite.scala @@ -26,10 +26,14 @@ import org.apache.spark.util.Utils class ExecutorURLClassLoaderSuite extends FunSuite { - val childClassNames = List("FakeClass1", "FakeClass2") - val parentClassNames = List("FakeClass1", "FakeClass2", "FakeClass3") - val urls = List(TestUtils.createJarWithClasses(childClassNames, "1")).toArray - val urls2 = List(TestUtils.createJarWithClasses(parentClassNames, "2")).toArray + val urls2 = List(TestUtils.createJarWithClasses( + classNames = Seq("FakeClass1", "FakeClass2", "FakeClass3"), + toStringValue = "2")).toArray + val urls = List(TestUtils.createJarWithClasses( + classNames = Seq("FakeClass1"), + classNamesWithBase = Seq(("FakeClass2", "FakeClass3")), // FakeClass3 is in parent + toStringValue = "1", + classpathUrls = urls2)).toArray test("child first") { val parentLoader = new URLClassLoader(urls2, null) @@ -37,6 +41,8 @@ class ExecutorURLClassLoaderSuite extends FunSuite { val fakeClass = classLoader.loadClass("FakeClass2").newInstance() val fakeClassVersion = fakeClass.toString assert(fakeClassVersion === "1") + val fakeClass2 = classLoader.loadClass("FakeClass2").newInstance() + assert(fakeClass.getClass === fakeClass2.getClass) } test("parent first") { @@ -45,6 +51,8 @@ class ExecutorURLClassLoaderSuite extends FunSuite { val fakeClass = classLoader.loadClass("FakeClass1").newInstance() val fakeClassVersion = fakeClass.toString assert(fakeClassVersion === "2") + val fakeClass2 = classLoader.loadClass("FakeClass1").newInstance() + assert(fakeClass.getClass === fakeClass2.getClass) } test("child first can fall back") { From 6072fcc14ee1a4eba793e725fcb2cb2ffebd5b60 Mon Sep 17 00:00:00 2001 From: lianhuiwang Date: Fri, 6 Feb 2015 11:09:37 -0800 Subject: [PATCH 18/53] [SPARK-5593][Core]Replace BlockManagerListener with ExecutorListener in ExecutorAllocationListener More strictly, in ExecutorAllocationListener, we need to replace onBlockManagerAdded, onBlockManagerRemoved with onExecutorAdded,onExecutorRemoved. because at some time, onExecutorAdded and onExecutorRemoved are more accurate to express these meanings. example at SPARK-5529, BlockManager has been removed,but executor is existed. andrewor14 sryza Author: lianhuiwang Closes #4369 from lianhuiwang/SPARK-5593 and squashes the following commits: 333367c [lianhuiwang] Replace BlockManagerListener with ExecutorListener in ExecutorAllocationListener --- .../spark/ExecutorAllocationManager.scala | 9 +++--- .../ExecutorAllocationManagerSuite.scala | 32 +++++++++---------- 2 files changed, 19 insertions(+), 22 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala b/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala index 8b38366e0325c..02d54bf3b53cc 100644 --- a/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala +++ b/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala @@ -486,8 +486,8 @@ private[spark] class ExecutorAllocationManager( } } - override def onBlockManagerAdded(blockManagerAdded: SparkListenerBlockManagerAdded): Unit = { - val executorId = blockManagerAdded.blockManagerId.executorId + override def onExecutorAdded(executorAdded: SparkListenerExecutorAdded): Unit = { + val executorId = executorAdded.executorId if (executorId != SparkContext.DRIVER_IDENTIFIER) { // This guards against the race condition in which the `SparkListenerTaskStart` // event is posted before the `SparkListenerBlockManagerAdded` event, which is @@ -498,9 +498,8 @@ private[spark] class ExecutorAllocationManager( } } - override def onBlockManagerRemoved( - blockManagerRemoved: SparkListenerBlockManagerRemoved): Unit = { - allocationManager.onExecutorRemoved(blockManagerRemoved.blockManagerId.executorId) + override def onExecutorRemoved(executorRemoved: SparkListenerExecutorRemoved): Unit = { + allocationManager.onExecutorRemoved(executorRemoved.executorId) } /** diff --git a/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala b/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala index 57081ddd959a5..c2869628afbb3 100644 --- a/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala +++ b/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala @@ -22,7 +22,7 @@ import scala.collection.mutable import org.scalatest.{FunSuite, PrivateMethodTester} import org.apache.spark.executor.TaskMetrics import org.apache.spark.scheduler._ -import org.apache.spark.storage.BlockManagerId +import org.apache.spark.scheduler.cluster.ExecutorInfo /** * Test add and remove behavior of ExecutorAllocationManager. @@ -144,8 +144,8 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { // Verify that running a task reduces the cap sc.listenerBus.postToAll(SparkListenerStageSubmitted(createStageInfo(1, 3))) - sc.listenerBus.postToAll(SparkListenerBlockManagerAdded( - 0L, BlockManagerId("executor-1", "host1", 1), 100L)) + sc.listenerBus.postToAll(SparkListenerExecutorAdded( + 0L, "executor-1", new ExecutorInfo("host1", 1))) sc.listenerBus.postToAll(SparkListenerTaskStart(1, 0, createTaskInfo(0, 0, "executor-1"))) assert(numExecutorsPending(manager) === 4) assert(addExecutors(manager) === 1) @@ -578,30 +578,28 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { assert(removeTimes(manager).isEmpty) // New executors have registered - sc.listenerBus.postToAll(SparkListenerBlockManagerAdded( - 0L, BlockManagerId("executor-1", "host1", 1), 100L)) + sc.listenerBus.postToAll(SparkListenerExecutorAdded( + 0L, "executor-1", new ExecutorInfo("host1", 1))) assert(executorIds(manager).size === 1) assert(executorIds(manager).contains("executor-1")) assert(removeTimes(manager).size === 1) assert(removeTimes(manager).contains("executor-1")) - sc.listenerBus.postToAll(SparkListenerBlockManagerAdded( - 0L, BlockManagerId("executor-2", "host2", 1), 100L)) + sc.listenerBus.postToAll(SparkListenerExecutorAdded( + 0L, "executor-2", new ExecutorInfo("host2", 1))) assert(executorIds(manager).size === 2) assert(executorIds(manager).contains("executor-2")) assert(removeTimes(manager).size === 2) assert(removeTimes(manager).contains("executor-2")) // Existing executors have disconnected - sc.listenerBus.postToAll(SparkListenerBlockManagerRemoved( - 0L, BlockManagerId("executor-1", "host1", 1))) + sc.listenerBus.postToAll(SparkListenerExecutorRemoved(0L, "executor-1", "")) assert(executorIds(manager).size === 1) assert(!executorIds(manager).contains("executor-1")) assert(removeTimes(manager).size === 1) assert(!removeTimes(manager).contains("executor-1")) // Unknown executor has disconnected - sc.listenerBus.postToAll(SparkListenerBlockManagerRemoved( - 0L, BlockManagerId("executor-3", "host3", 1))) + sc.listenerBus.postToAll(SparkListenerExecutorRemoved(0L, "executor-3", "")) assert(executorIds(manager).size === 1) assert(removeTimes(manager).size === 1) } @@ -613,8 +611,8 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { assert(removeTimes(manager).isEmpty) sc.listenerBus.postToAll(SparkListenerTaskStart(0, 0, createTaskInfo(0, 0, "executor-1"))) - sc.listenerBus.postToAll(SparkListenerBlockManagerAdded( - 0L, BlockManagerId("executor-1", "host1", 1), 100L)) + sc.listenerBus.postToAll(SparkListenerExecutorAdded( + 0L, "executor-1", new ExecutorInfo("host1", 1))) assert(executorIds(manager).size === 1) assert(executorIds(manager).contains("executor-1")) assert(removeTimes(manager).size === 0) @@ -625,16 +623,16 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { val manager = sc.executorAllocationManager.get assert(executorIds(manager).isEmpty) assert(removeTimes(manager).isEmpty) - sc.listenerBus.postToAll(SparkListenerBlockManagerAdded( - 0L, BlockManagerId("executor-1", "host1", 1), 100L)) + sc.listenerBus.postToAll(SparkListenerExecutorAdded( + 0L, "executor-1", new ExecutorInfo("host1", 1))) sc.listenerBus.postToAll(SparkListenerTaskStart(0, 0, createTaskInfo(0, 0, "executor-1"))) assert(executorIds(manager).size === 1) assert(executorIds(manager).contains("executor-1")) assert(removeTimes(manager).size === 0) - sc.listenerBus.postToAll(SparkListenerBlockManagerAdded( - 0L, BlockManagerId("executor-2", "host1", 1), 100L)) + sc.listenerBus.postToAll(SparkListenerExecutorAdded( + 0L, "executor-2", new ExecutorInfo("host1", 1))) assert(executorIds(manager).size === 2) assert(executorIds(manager).contains("executor-2")) assert(removeTimes(manager).size === 1) From 4cdb26c174e479a144950d12e1ad180f361af1fd Mon Sep 17 00:00:00 2001 From: Makoto Fukuhara Date: Fri, 6 Feb 2015 11:11:38 -0800 Subject: [PATCH 19/53] [SPARK-5618][Spark Core][Minor] Optimise utility code. Author: Makoto Fukuhara Closes #4396 from fukuo33/fix-unnecessary-regex and squashes the following commits: cd07fd6 [Makoto Fukuhara] fix unnecessary regex. --- core/src/main/scala/org/apache/spark/util/Utils.scala | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/util/Utils.scala b/core/src/main/scala/org/apache/spark/util/Utils.scala index 72d15e65bcde6..61d287ca9c3ac 100644 --- a/core/src/main/scala/org/apache/spark/util/Utils.scala +++ b/core/src/main/scala/org/apache/spark/util/Utils.scala @@ -1162,9 +1162,9 @@ private[spark] object Utils extends Logging { // finding the call site of a method. val SPARK_CORE_CLASS_REGEX = """^org\.apache\.spark(\.api\.java)?(\.util)?(\.rdd)?(\.broadcast)?\.[A-Z]""".r - val SCALA_CLASS_REGEX = """^scala""".r + val SCALA_CORE_CLASS_PREFIX = "scala" val isSparkCoreClass = SPARK_CORE_CLASS_REGEX.findFirstIn(className).isDefined - val isScalaClass = SCALA_CLASS_REGEX.findFirstIn(className).isDefined + val isScalaClass = className.startsWith(SCALA_CORE_CLASS_PREFIX) // If the class is a Spark internal class or a Scala class, then exclude. isSparkCoreClass || isScalaClass } From 32e964c410e7083b43264c46291e93cd206a8038 Mon Sep 17 00:00:00 2001 From: Kostas Sakellis Date: Fri, 6 Feb 2015 11:13:00 -0800 Subject: [PATCH 20/53] SPARK-2450 Adds executor log links to Web UI Adds links to stderr/stdout in the executor tab of the webUI for: 1) Standalone 2) Yarn client 3) Yarn cluster This tries to add the log url support in a general way so as to make it easy to add support for all the cluster managers. This is done by using environment variables to pass to the executor the log urls. The SPARK_LOG_URL_ prefix is used and so additional logs besides stderr/stdout can also be added. To propagate this information to the UI we use the onExecutorAdded spark listener event. Although this commit doesn't add log urls when running on a mesos cluster, it should be possible to add using the same mechanism. Author: Kostas Sakellis Author: Josh Rosen Closes #3486 from ksakellis/kostas-spark-2450 and squashes the following commits: d190936 [Josh Rosen] Fix a few minor style / formatting nits. Reset listener after each test Don't null listener out at end of main(). 8673fe1 [Kostas Sakellis] CR feedback. Hide the log column if there are no logs available 5bf6952 [Kostas Sakellis] [SPARK-2450] [CORE] Adds exeuctor log links to Web UI --- .../spark/deploy/worker/ExecutorRunner.scala | 7 +++ .../apache/spark/deploy/worker/Worker.scala | 1 + .../CoarseGrainedExecutorBackend.scala | 8 ++- .../cluster/CoarseGrainedClusterMessage.scala | 6 +- .../CoarseGrainedSchedulerBackend.scala | 4 +- .../scheduler/cluster/ExecutorData.scala | 5 +- .../scheduler/cluster/ExecutorInfo.scala | 9 +-- .../cluster/mesos/MesosSchedulerBackend.scala | 3 +- .../apache/spark/ui/exec/ExecutorsPage.scala | 31 ++++++++-- .../apache/spark/ui/exec/ExecutorsTab.scala | 6 ++ .../org/apache/spark/util/JsonProtocol.scala | 6 +- .../spark/deploy/JsonProtocolSuite.scala | 2 +- .../spark/deploy/LogUrlsStandaloneSuite.scala | 59 +++++++++++++++++++ .../deploy/worker/ExecutorRunnerTest.scala | 2 +- .../mesos/MesosSchedulerBackendSuite.scala | 4 +- .../apache/spark/util/JsonProtocolSuite.scala | 12 +++- .../spark/deploy/yarn/ExecutorRunnable.scala | 12 +++- .../spark/deploy/yarn/YarnClusterSuite.scala | 31 +++++++++- 18 files changed, 178 insertions(+), 30 deletions(-) create mode 100644 core/src/test/scala/org/apache/spark/deploy/LogUrlsStandaloneSuite.scala diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala b/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala index bc9f78b9e5c77..0add3064da452 100644 --- a/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala +++ b/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala @@ -43,6 +43,7 @@ private[spark] class ExecutorRunner( val worker: ActorRef, val workerId: String, val host: String, + val webUiPort: Int, val sparkHome: File, val executorDir: File, val workerUrl: String, @@ -134,6 +135,12 @@ private[spark] class ExecutorRunner( // In case we are running this from within the Spark Shell, avoid creating a "scala" // parent process for the executor command builder.environment.put("SPARK_LAUNCH_WITH_SCALA", "0") + + // Add webUI log urls + val baseUrl = s"http://$host:$webUiPort/logPage/?appId=$appId&executorId=$execId&logType=" + builder.environment.put("SPARK_LOG_URL_STDERR", s"${baseUrl}stderr") + builder.environment.put("SPARK_LOG_URL_STDOUT", s"${baseUrl}stdout") + process = builder.start() val header = "Spark Executor Command: %s\n%s\n\n".format( command.mkString("\"", "\" \"", "\""), "=" * 40) diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/Worker.scala b/core/src/main/scala/org/apache/spark/deploy/worker/Worker.scala index b20f5c0c82895..10929eb516041 100755 --- a/core/src/main/scala/org/apache/spark/deploy/worker/Worker.scala +++ b/core/src/main/scala/org/apache/spark/deploy/worker/Worker.scala @@ -362,6 +362,7 @@ private[spark] class Worker( self, workerId, host, + webUiPort, sparkHome, executorDir, akkaUrl, diff --git a/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala b/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala index bc72c8970319c..3a42f8b157977 100644 --- a/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala +++ b/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala @@ -49,10 +49,16 @@ private[spark] class CoarseGrainedExecutorBackend( override def preStart() { logInfo("Connecting to driver: " + driverUrl) driver = context.actorSelection(driverUrl) - driver ! RegisterExecutor(executorId, hostPort, cores) + driver ! RegisterExecutor(executorId, hostPort, cores, extractLogUrls) context.system.eventStream.subscribe(self, classOf[RemotingLifecycleEvent]) } + def extractLogUrls: Map[String, String] = { + val prefix = "SPARK_LOG_URL_" + sys.env.filterKeys(_.startsWith(prefix)) + .map(e => (e._1.substring(prefix.length).toLowerCase, e._2)) + } + override def receiveWithLogging = { case RegisteredExecutor => logInfo("Successfully registered with driver") diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedClusterMessage.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedClusterMessage.scala index 1da6fe976da5b..9bf74f4be198d 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedClusterMessage.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedClusterMessage.scala @@ -39,7 +39,11 @@ private[spark] object CoarseGrainedClusterMessages { case class RegisterExecutorFailed(message: String) extends CoarseGrainedClusterMessage // Executors to driver - case class RegisterExecutor(executorId: String, hostPort: String, cores: Int) + case class RegisterExecutor( + executorId: String, + hostPort: String, + cores: Int, + logUrls: Map[String, String]) extends CoarseGrainedClusterMessage { Utils.checkHostPort(hostPort, "Expected host port") } diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala index 103a5c053c289..9d2fb4f3b4729 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala @@ -86,7 +86,7 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val actorSyste } def receiveWithLogging = { - case RegisterExecutor(executorId, hostPort, cores) => + case RegisterExecutor(executorId, hostPort, cores, logUrls) => Utils.checkHostPort(hostPort, "Host port expected " + hostPort) if (executorDataMap.contains(executorId)) { sender ! RegisterExecutorFailed("Duplicate executor ID: " + executorId) @@ -98,7 +98,7 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val actorSyste totalCoreCount.addAndGet(cores) totalRegisteredExecutors.addAndGet(1) val (host, _) = Utils.parseHostPort(hostPort) - val data = new ExecutorData(sender, sender.path.address, host, cores, cores) + val data = new ExecutorData(sender, sender.path.address, host, cores, cores, logUrls) // This must be synchronized because variables mutated // in this block are read when requesting executors CoarseGrainedSchedulerBackend.this.synchronized { diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/ExecutorData.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/ExecutorData.scala index eb52ddfb1eab1..5e571efe76720 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/cluster/ExecutorData.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/ExecutorData.scala @@ -33,5 +33,6 @@ private[cluster] class ExecutorData( val executorAddress: Address, override val executorHost: String, var freeCores: Int, - override val totalCores: Int -) extends ExecutorInfo(executorHost, totalCores) + override val totalCores: Int, + override val logUrlMap: Map[String, String] +) extends ExecutorInfo(executorHost, totalCores, logUrlMap) diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/ExecutorInfo.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/ExecutorInfo.scala index b4738e64c9391..7f218566146a1 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/cluster/ExecutorInfo.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/ExecutorInfo.scala @@ -25,8 +25,8 @@ import org.apache.spark.annotation.DeveloperApi @DeveloperApi class ExecutorInfo( val executorHost: String, - val totalCores: Int -) { + val totalCores: Int, + val logUrlMap: Map[String, String]) { def canEqual(other: Any): Boolean = other.isInstanceOf[ExecutorInfo] @@ -34,12 +34,13 @@ class ExecutorInfo( case that: ExecutorInfo => (that canEqual this) && executorHost == that.executorHost && - totalCores == that.totalCores + totalCores == that.totalCores && + logUrlMap == that.logUrlMap case _ => false } override def hashCode(): Int = { - val state = Seq(executorHost, totalCores) + val state = Seq(executorHost, totalCores, logUrlMap) state.map(_.hashCode()).foldLeft(0)((a, b) => 31 * a + b) } } diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackend.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackend.scala index c3c546be6da15..cfb6592e14aa8 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackend.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackend.scala @@ -270,7 +270,8 @@ private[spark] class MesosSchedulerBackend( mesosTasks.foreach { case (slaveId, tasks) => slaveIdToWorkerOffer.get(slaveId).foreach(o => listenerBus.post(SparkListenerExecutorAdded(System.currentTimeMillis(), slaveId, - new ExecutorInfo(o.host, o.cores))) + // TODO: Add support for log urls for Mesos + new ExecutorInfo(o.host, o.cores, Map.empty))) ) d.launchTasks(Collections.singleton(slaveIdToOffer(slaveId).getId), tasks, filters) } diff --git a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsPage.scala b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsPage.scala index 363cb96de7998..956608d7c0cbe 100644 --- a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsPage.scala +++ b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsPage.scala @@ -26,7 +26,8 @@ import org.apache.spark.ui.{ToolTips, UIUtils, WebUIPage} import org.apache.spark.util.Utils /** Summary information about an executor to display in the UI. */ -private case class ExecutorSummaryInfo( +// Needs to be private[ui] because of a false positive MiMa failure. +private[ui] case class ExecutorSummaryInfo( id: String, hostPort: String, rddBlocks: Int, @@ -40,7 +41,8 @@ private case class ExecutorSummaryInfo( totalInputBytes: Long, totalShuffleRead: Long, totalShuffleWrite: Long, - maxMemory: Long) + maxMemory: Long, + executorLogs: Map[String, String]) private[ui] class ExecutorsPage( parent: ExecutorsTab, @@ -55,6 +57,7 @@ private[ui] class ExecutorsPage( val diskUsed = storageStatusList.map(_.diskUsed).sum val execInfo = for (statusId <- 0 until storageStatusList.size) yield getExecInfo(statusId) val execInfoSorted = execInfo.sortBy(_.id) + val logsExist = execInfo.filter(_.executorLogs.nonEmpty).nonEmpty val execTable = @@ -79,10 +82,11 @@ private[ui] class ExecutorsPage( Shuffle Write + {if (logsExist) else Seq.empty} {if (threadDumpEnabled) else Seq.empty} - {execInfoSorted.map(execRow)} + {execInfoSorted.map(execRow(_, logsExist))}
LogsThread Dump
@@ -107,7 +111,7 @@ private[ui] class ExecutorsPage( } /** Render an HTML row representing an executor */ - private def execRow(info: ExecutorSummaryInfo): Seq[Node] = { + private def execRow(info: ExecutorSummaryInfo, logsExist: Boolean): Seq[Node] = { val maximumMemory = info.maxMemory val memoryUsed = info.memoryUsed val diskUsed = info.diskUsed @@ -138,6 +142,21 @@ private[ui] class ExecutorsPage( {Utils.bytesToString(info.totalShuffleWrite)} + { + if (logsExist) { + + { + info.executorLogs.map { case (logName, logUrl) => + + } + } + + } + } { if (threadDumpEnabled) { val encodedId = URLEncoder.encode(info.id, "UTF-8") @@ -168,6 +187,7 @@ private[ui] class ExecutorsPage( val totalInputBytes = listener.executorToInputBytes.getOrElse(execId, 0L) val totalShuffleRead = listener.executorToShuffleRead.getOrElse(execId, 0L) val totalShuffleWrite = listener.executorToShuffleWrite.getOrElse(execId, 0L) + val executorLogs = listener.executorToLogUrls.getOrElse(execId, Map.empty) new ExecutorSummaryInfo( execId, @@ -183,7 +203,8 @@ private[ui] class ExecutorsPage( totalInputBytes, totalShuffleRead, totalShuffleWrite, - maxMem + maxMem, + executorLogs ) } } diff --git a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala index dd1c2b78c4094..a38cb75fdd8c6 100644 --- a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala +++ b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala @@ -51,9 +51,15 @@ class ExecutorsListener(storageStatusListener: StorageStatusListener) extends Sp val executorToOutputBytes = HashMap[String, Long]() val executorToShuffleRead = HashMap[String, Long]() val executorToShuffleWrite = HashMap[String, Long]() + val executorToLogUrls = HashMap[String, Map[String, String]]() def storageStatusList = storageStatusListener.storageStatusList + override def onExecutorAdded(executorAdded: SparkListenerExecutorAdded) = synchronized { + val eid = executorAdded.executorId + executorToLogUrls(eid) = executorAdded.executorInfo.logUrlMap + } + override def onTaskStart(taskStart: SparkListenerTaskStart) = synchronized { val eid = taskStart.taskInfo.executorId executorToTasksActive(eid) = executorToTasksActive.getOrElse(eid, 0) + 1 diff --git a/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala b/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala index 8e0e41ad3782e..c8407bbcb780b 100644 --- a/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala +++ b/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala @@ -383,7 +383,8 @@ private[spark] object JsonProtocol { def executorInfoToJson(executorInfo: ExecutorInfo): JValue = { ("Host" -> executorInfo.executorHost) ~ - ("Total Cores" -> executorInfo.totalCores) + ("Total Cores" -> executorInfo.totalCores) ~ + ("Log Urls" -> mapToJson(executorInfo.logUrlMap)) } /** ------------------------------ * @@ -792,7 +793,8 @@ private[spark] object JsonProtocol { def executorInfoFromJson(json: JValue): ExecutorInfo = { val executorHost = (json \ "Host").extract[String] val totalCores = (json \ "Total Cores").extract[Int] - new ExecutorInfo(executorHost, totalCores) + val logUrls = mapFromJson(json \ "Log Urls").toMap + new ExecutorInfo(executorHost, totalCores, logUrls) } /** -------------------------------- * diff --git a/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala b/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala index aa65f7e8915e6..ed02ca81e405c 100644 --- a/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala +++ b/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala @@ -117,7 +117,7 @@ class JsonProtocolSuite extends FunSuite { } def createExecutorRunner(): ExecutorRunner = { - new ExecutorRunner("appId", 123, createAppDesc(), 4, 1234, null, "workerId", "host", + new ExecutorRunner("appId", 123, createAppDesc(), 4, 1234, null, "workerId", "host", 123, new File("sparkHome"), new File("workDir"), "akka://worker", new SparkConf, Seq("localDir"), ExecutorState.RUNNING) } diff --git a/core/src/test/scala/org/apache/spark/deploy/LogUrlsStandaloneSuite.scala b/core/src/test/scala/org/apache/spark/deploy/LogUrlsStandaloneSuite.scala new file mode 100644 index 0000000000000..f33bdc73e40ac --- /dev/null +++ b/core/src/test/scala/org/apache/spark/deploy/LogUrlsStandaloneSuite.scala @@ -0,0 +1,59 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy + +import scala.collection.mutable + +import org.scalatest.{BeforeAndAfter, FunSuite} + +import org.apache.spark.scheduler.cluster.ExecutorInfo +import org.apache.spark.scheduler.{SparkListenerExecutorAdded, SparkListener} +import org.apache.spark.{SparkContext, LocalSparkContext} + +class LogUrlsStandaloneSuite extends FunSuite with LocalSparkContext with BeforeAndAfter { + + /** Length of time to wait while draining listener events. */ + val WAIT_TIMEOUT_MILLIS = 10000 + + before { + sc = new SparkContext("local-cluster[2,1,512]", "test") + } + + test("verify log urls get propagated from workers") { + val listener = new SaveExecutorInfo + sc.addSparkListener(listener) + + val rdd1 = sc.parallelize(1 to 100, 4) + val rdd2 = rdd1.map(_.toString) + rdd2.setName("Target RDD") + rdd2.count() + + assert(sc.listenerBus.waitUntilEmpty(WAIT_TIMEOUT_MILLIS)) + listener.addedExecutorInfos.values.foreach { info => + assert(info.logUrlMap.nonEmpty) + } + } + + private class SaveExecutorInfo extends SparkListener { + val addedExecutorInfos = mutable.Map[String, ExecutorInfo]() + + override def onExecutorAdded(executor: SparkListenerExecutorAdded) { + addedExecutorInfos(executor.executorId) = executor.executorInfo + } + } +} diff --git a/core/src/test/scala/org/apache/spark/deploy/worker/ExecutorRunnerTest.scala b/core/src/test/scala/org/apache/spark/deploy/worker/ExecutorRunnerTest.scala index 6f233d7cf97aa..76511699e5ac5 100644 --- a/core/src/test/scala/org/apache/spark/deploy/worker/ExecutorRunnerTest.scala +++ b/core/src/test/scala/org/apache/spark/deploy/worker/ExecutorRunnerTest.scala @@ -32,7 +32,7 @@ class ExecutorRunnerTest extends FunSuite { val sparkHome = sys.props.getOrElse("spark.test.home", fail("spark.test.home is not set!")) val appDesc = new ApplicationDescription("app name", Some(8), 500, Command("foo", Seq(appId), Map(), Seq(), Seq(), Seq()), "appUiUrl") - val er = new ExecutorRunner(appId, 1, appDesc, 8, 500, null, "blah", "worker321", + val er = new ExecutorRunner(appId, 1, appDesc, 8, 500, null, "blah", "worker321", 123, new File(sparkHome), new File("ooga"), "blah", new SparkConf, Seq("localDir"), ExecutorState.RUNNING) val builder = CommandUtils.buildProcessBuilder(appDesc.command, 512, sparkHome, er.substituteVariables) diff --git a/core/src/test/scala/org/apache/spark/scheduler/mesos/MesosSchedulerBackendSuite.scala b/core/src/test/scala/org/apache/spark/scheduler/mesos/MesosSchedulerBackendSuite.scala index f2ff98eb72daf..46ab02bfef780 100644 --- a/core/src/test/scala/org/apache/spark/scheduler/mesos/MesosSchedulerBackendSuite.scala +++ b/core/src/test/scala/org/apache/spark/scheduler/mesos/MesosSchedulerBackendSuite.scala @@ -43,7 +43,7 @@ class MesosSchedulerBackendSuite extends FunSuite with LocalSparkContext with Ea conf.set("spark.mesos.executor.home" , "/mesos-home") val listenerBus = EasyMock.createMock(classOf[LiveListenerBus]) - listenerBus.post(SparkListenerExecutorAdded(EasyMock.anyLong, "s1", new ExecutorInfo("host1", 2))) + listenerBus.post(SparkListenerExecutorAdded(EasyMock.anyLong, "s1", new ExecutorInfo("host1", 2, Map.empty))) EasyMock.replay(listenerBus) val sc = EasyMock.createMock(classOf[SparkContext]) @@ -88,7 +88,7 @@ class MesosSchedulerBackendSuite extends FunSuite with LocalSparkContext with Ea val taskScheduler = EasyMock.createMock(classOf[TaskSchedulerImpl]) val listenerBus = EasyMock.createMock(classOf[LiveListenerBus]) - listenerBus.post(SparkListenerExecutorAdded(EasyMock.anyLong, "s1", new ExecutorInfo("host1", 2))) + listenerBus.post(SparkListenerExecutorAdded(EasyMock.anyLong, "s1", new ExecutorInfo("host1", 2, Map.empty))) EasyMock.replay(listenerBus) val sc = EasyMock.createMock(classOf[SparkContext]) diff --git a/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala b/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala index 6577ebaa2e9a8..842f54529baf0 100644 --- a/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala +++ b/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala @@ -76,8 +76,9 @@ class JsonProtocolSuite extends FunSuite { val unpersistRdd = SparkListenerUnpersistRDD(12345) val applicationStart = SparkListenerApplicationStart("The winner of all", None, 42L, "Garfield") val applicationEnd = SparkListenerApplicationEnd(42L) + val logUrlMap = Map("stderr" -> "mystderr", "stdout" -> "mystdout").toMap val executorAdded = SparkListenerExecutorAdded(executorAddedTime, "exec1", - new ExecutorInfo("Hostee.awesome.com", 11)) + new ExecutorInfo("Hostee.awesome.com", 11, logUrlMap)) val executorRemoved = SparkListenerExecutorRemoved(executorRemovedTime, "exec2", "test reason") testEvent(stageSubmitted, stageSubmittedJsonString) @@ -100,13 +101,14 @@ class JsonProtocolSuite extends FunSuite { } test("Dependent Classes") { + val logUrlMap = Map("stderr" -> "mystderr", "stdout" -> "mystdout").toMap testRDDInfo(makeRddInfo(2, 3, 4, 5L, 6L)) testStageInfo(makeStageInfo(10, 20, 30, 40L, 50L)) testTaskInfo(makeTaskInfo(999L, 888, 55, 777L, false)) testTaskMetrics(makeTaskMetrics( 33333L, 44444L, 55555L, 66666L, 7, 8, hasHadoopInput = false, hasOutput = false)) testBlockManagerId(BlockManagerId("Hong", "Kong", 500)) - testExecutorInfo(new ExecutorInfo("host", 43)) + testExecutorInfo(new ExecutorInfo("host", 43, logUrlMap)) // StorageLevel testStorageLevel(StorageLevel.NONE) @@ -1463,7 +1465,11 @@ class JsonProtocolSuite extends FunSuite { | "Executor ID": "exec1", | "Executor Info": { | "Host": "Hostee.awesome.com", - | "Total Cores": 11 + | "Total Cores": 11, + | "Log Urls" : { + | "stderr" : "mystderr", + | "stdout" : "mystdout" + | } | } |} """ diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala index ee2002a35f523..408cf09b9bdfa 100644 --- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala +++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala @@ -56,7 +56,7 @@ class ExecutorRunnable( var rpc: YarnRPC = YarnRPC.create(conf) var nmClient: NMClient = _ val yarnConf: YarnConfiguration = new YarnConfiguration(conf) - lazy val env = prepareEnvironment + lazy val env = prepareEnvironment(container) def run = { logInfo("Starting Executor Container") @@ -254,7 +254,7 @@ class ExecutorRunnable( localResources } - private def prepareEnvironment: HashMap[String, String] = { + private def prepareEnvironment(container: Container): HashMap[String, String] = { val env = new HashMap[String, String]() val extraCp = sparkConf.getOption("spark.executor.extraClassPath") Client.populateClasspath(null, yarnConf, sparkConf, env, extraCp) @@ -270,6 +270,14 @@ class ExecutorRunnable( YarnSparkHadoopUtil.setEnvFromInputString(env, userEnvs) } + // Add log urls + sys.env.get("SPARK_USER").foreach { user => + val baseUrl = "http://%s/node/containerlogs/%s/%s" + .format(container.getNodeHttpAddress, ConverterUtils.toString(container.getId), user) + env("SPARK_LOG_URL_STDERR") = s"$baseUrl/stderr?start=0" + env("SPARK_LOG_URL_STDOUT") = s"$baseUrl/stdout?start=0" + } + System.getenv().filterKeys(_.startsWith("SPARK")).foreach { case (k, v) => env(k) = v } env } diff --git a/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala b/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala index 7165918e1bfcf..eda40efc4c77f 100644 --- a/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala +++ b/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala @@ -21,16 +21,17 @@ import java.io.File import java.util.concurrent.TimeUnit import scala.collection.JavaConversions._ +import scala.collection.mutable import com.google.common.base.Charsets import com.google.common.io.Files -import org.scalatest.{BeforeAndAfterAll, FunSuite, Matchers} - import org.apache.hadoop.yarn.conf.YarnConfiguration import org.apache.hadoop.yarn.server.MiniYARNCluster +import org.scalatest.{BeforeAndAfterAll, FunSuite, Matchers} import org.apache.spark.{Logging, SparkConf, SparkContext, SparkException} -import org.apache.spark.deploy.SparkHadoopUtil +import org.apache.spark.scheduler.cluster.ExecutorInfo +import org.apache.spark.scheduler.{SparkListener, SparkListenerExecutorAdded} import org.apache.spark.util.Utils class YarnClusterSuite extends FunSuite with BeforeAndAfterAll with Matchers with Logging { @@ -143,6 +144,11 @@ class YarnClusterSuite extends FunSuite with BeforeAndAfterAll with Matchers wit var result = File.createTempFile("result", null, tempDir) YarnClusterDriver.main(Array("yarn-client", result.getAbsolutePath())) checkResult(result) + + // verify log urls are present + YarnClusterDriver.listener.addedExecutorInfos.values.foreach { info => + assert(info.logUrlMap.nonEmpty) + } } test("run Spark in yarn-cluster mode") { @@ -156,6 +162,11 @@ class YarnClusterSuite extends FunSuite with BeforeAndAfterAll with Matchers wit "--num-executors", "1") Client.main(args) checkResult(result) + + // verify log urls are present. + YarnClusterDriver.listener.addedExecutorInfos.values.foreach { info => + assert(info.logUrlMap.nonEmpty) + } } test("run Spark in yarn-cluster mode unsuccessfully") { @@ -203,8 +214,19 @@ class YarnClusterSuite extends FunSuite with BeforeAndAfterAll with Matchers wit } +private class SaveExecutorInfo extends SparkListener { + val addedExecutorInfos = mutable.Map[String, ExecutorInfo]() + + override def onExecutorAdded(executor : SparkListenerExecutorAdded) { + addedExecutorInfos(executor.executorId) = executor.executorInfo + } +} + private object YarnClusterDriver extends Logging with Matchers { + val WAIT_TIMEOUT_MILLIS = 10000 + var listener: SaveExecutorInfo = null + def main(args: Array[String]) = { if (args.length != 2) { System.err.println( @@ -216,12 +238,15 @@ private object YarnClusterDriver extends Logging with Matchers { System.exit(1) } + listener = new SaveExecutorInfo val sc = new SparkContext(new SparkConf().setMaster(args(0)) .setAppName("yarn \"test app\" 'with quotes' and \\back\\slashes and $dollarSigns")) + sc.addSparkListener(listener) val status = new File(args(1)) var result = "failure" try { val data = sc.parallelize(1 to 4, 4).collect().toSet + assert(sc.listenerBus.waitUntilEmpty(WAIT_TIMEOUT_MILLIS)) data should be (Set(1, 2, 3, 4)) result = "success" } finally { From 0d74bd7fd7b2722d08eddc5c269b8b2b6cb47635 Mon Sep 17 00:00:00 2001 From: Josh Rosen Date: Fri, 6 Feb 2015 11:14:58 -0800 Subject: [PATCH 21/53] [SPARK-5555] Enable UISeleniumSuite tests This patch enables UISeleniumSuite, a set of tests for the Spark application web UI. These tests were previously disabled because they were slow, but I think we now have sufficient test time budget that the benefit of enabling them outweighs the time costs. Author: Josh Rosen Closes #4334 from JoshRosen/enable-uiseleniumsuite and squashes the following commits: 4ab9477 [Josh Rosen] Use BeforeAndAfterAll to cleanup WebDriver 71efc72 [Josh Rosen] Update broken UISeleniumSuite tests; use random port #. a5ab595 [Josh Rosen] Enable UISeleniumSuite tests. --- .../org/apache/spark/ui/UISeleniumSuite.scala | 29 +++++++++++++------ 1 file changed, 20 insertions(+), 9 deletions(-) diff --git a/core/src/test/scala/org/apache/spark/ui/UISeleniumSuite.scala b/core/src/test/scala/org/apache/spark/ui/UISeleniumSuite.scala index e85a436cdba17..6a972381faf14 100644 --- a/core/src/test/scala/org/apache/spark/ui/UISeleniumSuite.scala +++ b/core/src/test/scala/org/apache/spark/ui/UISeleniumSuite.scala @@ -32,12 +32,21 @@ import org.apache.spark.api.java.StorageLevels import org.apache.spark.shuffle.FetchFailedException /** - * Selenium tests for the Spark Web UI. These tests are not run by default - * because they're slow. + * Selenium tests for the Spark Web UI. */ -@DoNotDiscover -class UISeleniumSuite extends FunSuite with WebBrowser with Matchers { - implicit val webDriver: WebDriver = new HtmlUnitDriver +class UISeleniumSuite extends FunSuite with WebBrowser with Matchers with BeforeAndAfterAll { + + implicit var webDriver: WebDriver = _ + + override def beforeAll(): Unit = { + webDriver = new HtmlUnitDriver + } + + override def afterAll(): Unit = { + if (webDriver != null) { + webDriver.quit() + } + } /** * Create a test SparkContext with the SparkUI enabled. @@ -48,6 +57,7 @@ class UISeleniumSuite extends FunSuite with WebBrowser with Matchers { .setMaster("local") .setAppName("test") .set("spark.ui.enabled", "true") + .set("spark.ui.port", "0") val sc = new SparkContext(conf) assert(sc.ui.isDefined) sc @@ -93,7 +103,7 @@ class UISeleniumSuite extends FunSuite with WebBrowser with Matchers { } eventually(timeout(5 seconds), interval(50 milliseconds)) { go to (sc.ui.get.appUIAddress.stripSuffix("/") + "/stages") - find(id("active")).get.text should be("Active Stages (0)") + find(id("active")) should be(None) // Since we hide empty tables find(id("failed")).get.text should be("Failed Stages (1)") } @@ -105,7 +115,7 @@ class UISeleniumSuite extends FunSuite with WebBrowser with Matchers { } eventually(timeout(5 seconds), interval(50 milliseconds)) { go to (sc.ui.get.appUIAddress.stripSuffix("/") + "/stages") - find(id("active")).get.text should be("Active Stages (0)") + find(id("active")) should be(None) // Since we hide empty tables // The failure occurs before the stage becomes active, hence we should still show only one // failed stage, not two: find(id("failed")).get.text should be("Failed Stages (1)") @@ -167,13 +177,14 @@ class UISeleniumSuite extends FunSuite with WebBrowser with Matchers { test("job progress bars should handle stage / task failures") { withSpark(newSparkContext()) { sc => - val data = sc.parallelize(Seq(1, 2, 3)).map(identity).groupBy(identity) + val data = sc.parallelize(Seq(1, 2, 3), 1).map(identity).groupBy(identity) val shuffleHandle = data.dependencies.head.asInstanceOf[ShuffleDependency[_, _, _]].shuffleHandle // Simulate fetch failures: val mappedData = data.map { x => val taskContext = TaskContext.get - if (taskContext.attemptNumber == 0) { // Cause this stage to fail on its first attempt. + if (taskContext.taskAttemptId() == 1) { + // Cause the post-shuffle stage to fail on its first attempt with a single task failure val env = SparkEnv.get val bmAddress = env.blockManager.blockManagerId val shuffleId = shuffleHandle.shuffleId From 80f3bcb58f836cfe1829c85bdd349c10525c8a5e Mon Sep 17 00:00:00 2001 From: Liang-Chi Hsieh Date: Fri, 6 Feb 2015 11:22:11 -0800 Subject: [PATCH 22/53] [SPARK-5652][Mllib] Use broadcasted weights in LogisticRegressionModel `LogisticRegressionModel`'s `predictPoint` should directly use broadcasted weights. This pr also fixes the compilation errors of two unit test suite: `JavaLogisticRegressionSuite ` and `JavaLinearRegressionSuite`. Author: Liang-Chi Hsieh Closes #4429 from viirya/use_bcvalue and squashes the following commits: 5a797e5 [Liang-Chi Hsieh] Use broadcasted weights. Fix compilation error. --- .../spark/mllib/classification/LogisticRegression.scala | 8 ++++---- .../ml/classification/JavaLogisticRegressionSuite.java | 4 ++-- .../spark/ml/regression/JavaLinearRegressionSuite.java | 4 ++-- 3 files changed, 8 insertions(+), 8 deletions(-) diff --git a/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala index a668e7a7a359f..9a391bfff76a3 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala @@ -33,7 +33,7 @@ import org.apache.spark.rdd.RDD * * @param weights Weights computed for every feature. * @param intercept Intercept computed for this model. (Only used in Binary Logistic Regression. - * In Multinomial Logistic Regression, the intercepts will not be a single values, + * In Multinomial Logistic Regression, the intercepts will not be a single value, * so the intercepts will be part of the weights.) * @param numFeatures the dimension of the features. * @param numClasses the number of possible outcomes for k classes classification problem in @@ -107,7 +107,7 @@ class LogisticRegressionModel ( // If dataMatrix and weightMatrix have the same dimension, it's binary logistic regression. if (numClasses == 2) { require(numFeatures == weightMatrix.size) - val margin = dot(weights, dataMatrix) + intercept + val margin = dot(weightMatrix, dataMatrix) + intercept val score = 1.0 / (1.0 + math.exp(-margin)) threshold match { case Some(t) => if (score > t) 1.0 else 0.0 @@ -116,11 +116,11 @@ class LogisticRegressionModel ( } else { val dataWithBiasSize = weightMatrix.size / (numClasses - 1) - val weightsArray = weights match { + val weightsArray = weightMatrix match { case dv: DenseVector => dv.values case _ => throw new IllegalArgumentException( - s"weights only supports dense vector but got type ${weights.getClass}.") + s"weights only supports dense vector but got type ${weightMatrix.getClass}.") } val margins = (0 until numClasses - 1).map { i => diff --git a/mllib/src/test/java/org/apache/spark/ml/classification/JavaLogisticRegressionSuite.java b/mllib/src/test/java/org/apache/spark/ml/classification/JavaLogisticRegressionSuite.java index 26284023b0f69..d4b664479255d 100644 --- a/mllib/src/test/java/org/apache/spark/ml/classification/JavaLogisticRegressionSuite.java +++ b/mllib/src/test/java/org/apache/spark/ml/classification/JavaLogisticRegressionSuite.java @@ -84,7 +84,7 @@ public void logisticRegressionWithSetters() { .setThreshold(0.6) .setProbabilityCol("myProbability"); LogisticRegressionModel model = lr.fit(dataset); - assert(model.fittingParamMap().apply(lr.maxIter()) == 10); + assert(model.fittingParamMap().apply(lr.maxIter()).equals(10)); assert(model.fittingParamMap().apply(lr.regParam()).equals(1.0)); assert(model.fittingParamMap().apply(lr.threshold()).equals(0.6)); assert(model.getThreshold() == 0.6); @@ -109,7 +109,7 @@ public void logisticRegressionWithSetters() { // Call fit() with new params, and check as many params as we can. LogisticRegressionModel model2 = lr.fit(dataset, lr.maxIter().w(5), lr.regParam().w(0.1), lr.threshold().w(0.4), lr.probabilityCol().w("theProb")); - assert(model2.fittingParamMap().apply(lr.maxIter()) == 5); + assert(model2.fittingParamMap().apply(lr.maxIter()).equals(5)); assert(model2.fittingParamMap().apply(lr.regParam()).equals(0.1)); assert(model2.fittingParamMap().apply(lr.threshold()).equals(0.4)); assert(model2.getThreshold() == 0.4); diff --git a/mllib/src/test/java/org/apache/spark/ml/regression/JavaLinearRegressionSuite.java b/mllib/src/test/java/org/apache/spark/ml/regression/JavaLinearRegressionSuite.java index 5bd616e74d86c..40d5a92bb32af 100644 --- a/mllib/src/test/java/org/apache/spark/ml/regression/JavaLinearRegressionSuite.java +++ b/mllib/src/test/java/org/apache/spark/ml/regression/JavaLinearRegressionSuite.java @@ -76,13 +76,13 @@ public void linearRegressionWithSetters() { .setMaxIter(10) .setRegParam(1.0); LinearRegressionModel model = lr.fit(dataset); - assert(model.fittingParamMap().apply(lr.maxIter()) == 10); + assert(model.fittingParamMap().apply(lr.maxIter()).equals(10)); assert(model.fittingParamMap().apply(lr.regParam()).equals(1.0)); // Call fit() with new params, and check as many params as we can. LinearRegressionModel model2 = lr.fit(dataset, lr.maxIter().w(5), lr.regParam().w(0.1), lr.predictionCol().w("thePred")); - assert(model2.fittingParamMap().apply(lr.maxIter()) == 5); + assert(model2.fittingParamMap().apply(lr.maxIter()).equals(5)); assert(model2.fittingParamMap().apply(lr.regParam()).equals(0.1)); assert(model2.getPredictionCol().equals("thePred")); } From fb6c0cbac414f3c43dec2ab886ca8a1097b781f7 Mon Sep 17 00:00:00 2001 From: Josh Rosen Date: Fri, 6 Feb 2015 11:47:32 -0800 Subject: [PATCH 23/53] [HOTFIX] Fix test build break in ExecutorAllocationManagerSuite. This was caused because #3486 added a new field to ExecutorInfo and #4369 added new tests that created ExecutorInfos. These patches were merged in quick succession and were never tested together, hence the compilation error. --- .../spark/ExecutorAllocationManagerSuite.scala | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala b/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala index c2869628afbb3..9eb87f016068d 100644 --- a/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala +++ b/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala @@ -145,7 +145,7 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { // Verify that running a task reduces the cap sc.listenerBus.postToAll(SparkListenerStageSubmitted(createStageInfo(1, 3))) sc.listenerBus.postToAll(SparkListenerExecutorAdded( - 0L, "executor-1", new ExecutorInfo("host1", 1))) + 0L, "executor-1", new ExecutorInfo("host1", 1, Map.empty))) sc.listenerBus.postToAll(SparkListenerTaskStart(1, 0, createTaskInfo(0, 0, "executor-1"))) assert(numExecutorsPending(manager) === 4) assert(addExecutors(manager) === 1) @@ -579,13 +579,13 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { // New executors have registered sc.listenerBus.postToAll(SparkListenerExecutorAdded( - 0L, "executor-1", new ExecutorInfo("host1", 1))) + 0L, "executor-1", new ExecutorInfo("host1", 1, Map.empty))) assert(executorIds(manager).size === 1) assert(executorIds(manager).contains("executor-1")) assert(removeTimes(manager).size === 1) assert(removeTimes(manager).contains("executor-1")) sc.listenerBus.postToAll(SparkListenerExecutorAdded( - 0L, "executor-2", new ExecutorInfo("host2", 1))) + 0L, "executor-2", new ExecutorInfo("host2", 1, Map.empty))) assert(executorIds(manager).size === 2) assert(executorIds(manager).contains("executor-2")) assert(removeTimes(manager).size === 2) @@ -612,7 +612,7 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { sc.listenerBus.postToAll(SparkListenerTaskStart(0, 0, createTaskInfo(0, 0, "executor-1"))) sc.listenerBus.postToAll(SparkListenerExecutorAdded( - 0L, "executor-1", new ExecutorInfo("host1", 1))) + 0L, "executor-1", new ExecutorInfo("host1", 1, Map.empty))) assert(executorIds(manager).size === 1) assert(executorIds(manager).contains("executor-1")) assert(removeTimes(manager).size === 0) @@ -624,7 +624,7 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { assert(executorIds(manager).isEmpty) assert(removeTimes(manager).isEmpty) sc.listenerBus.postToAll(SparkListenerExecutorAdded( - 0L, "executor-1", new ExecutorInfo("host1", 1))) + 0L, "executor-1", new ExecutorInfo("host1", 1, Map.empty))) sc.listenerBus.postToAll(SparkListenerTaskStart(0, 0, createTaskInfo(0, 0, "executor-1"))) assert(executorIds(manager).size === 1) @@ -632,7 +632,7 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { assert(removeTimes(manager).size === 0) sc.listenerBus.postToAll(SparkListenerExecutorAdded( - 0L, "executor-2", new ExecutorInfo("host1", 1))) + 0L, "executor-2", new ExecutorInfo("host1", 1, Map.empty))) assert(executorIds(manager).size === 2) assert(executorIds(manager).contains("executor-2")) assert(removeTimes(manager).size === 1) From af2a2a263ac5d890e84d012b75fcb50e02c9ede8 Mon Sep 17 00:00:00 2001 From: zsxwing Date: Fri, 6 Feb 2015 11:50:20 -0800 Subject: [PATCH 24/53] [SPARK-4361][Doc] Add more docs for Hadoop Configuration I'm trying to point out reusing a Configuration in these APIs is dangerous. Any better idea? Author: zsxwing Closes #3225 from zsxwing/SPARK-4361 and squashes the following commits: fe4e3d5 [zsxwing] Add more docs for Hadoop Configuration --- .../scala/org/apache/spark/SparkContext.scala | 20 +++++++++++-- .../spark/api/java/JavaSparkContext.scala | 28 +++++++++++++++++++ 2 files changed, 46 insertions(+), 2 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/SparkContext.scala b/core/src/main/scala/org/apache/spark/SparkContext.scala index 24490fddc5c6a..5623587c36fa6 100644 --- a/core/src/main/scala/org/apache/spark/SparkContext.scala +++ b/core/src/main/scala/org/apache/spark/SparkContext.scala @@ -288,7 +288,12 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli // the bound port to the cluster manager properly ui.foreach(_.bind()) - /** A default Hadoop Configuration for the Hadoop code (e.g. file systems) that we reuse. */ + /** + * A default Hadoop Configuration for the Hadoop code (e.g. file systems) that we reuse. + * + * '''Note:''' As it will be reused in all Hadoop RDDs, it's better not to modify it unless you + * plan to set some global configurations for all Hadoop RDDs. + */ val hadoopConfiguration = SparkHadoopUtil.get.newConfiguration(conf) // Add each JAR given through the constructor @@ -694,7 +699,10 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli * necessary info (e.g. file name for a filesystem-based dataset, table name for HyperTable), * using the older MapReduce API (`org.apache.hadoop.mapred`). * - * @param conf JobConf for setting up the dataset + * @param conf JobConf for setting up the dataset. Note: This will be put into a Broadcast. + * Therefore if you plan to reuse this conf to create multiple RDDs, you need to make + * sure you won't modify the conf. A safe approach is always creating a new conf for + * a new RDD. * @param inputFormatClass Class of the InputFormat * @param keyClass Class of the keys * @param valueClass Class of the values @@ -830,6 +838,14 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli * Get an RDD for a given Hadoop file with an arbitrary new API InputFormat * and extra configuration options to pass to the input format. * + * @param conf Configuration for setting up the dataset. Note: This will be put into a Broadcast. + * Therefore if you plan to reuse this conf to create multiple RDDs, you need to make + * sure you won't modify the conf. A safe approach is always creating a new conf for + * a new RDD. + * @param fClass Class of the InputFormat + * @param kClass Class of the keys + * @param vClass Class of the values + * * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each * record, directly caching the returned RDD or directly passing it to an aggregation or shuffle * operation will create many references to the same object. diff --git a/core/src/main/scala/org/apache/spark/api/java/JavaSparkContext.scala b/core/src/main/scala/org/apache/spark/api/java/JavaSparkContext.scala index 97f5c9f257e09..6d6ed693be752 100644 --- a/core/src/main/scala/org/apache/spark/api/java/JavaSparkContext.scala +++ b/core/src/main/scala/org/apache/spark/api/java/JavaSparkContext.scala @@ -373,6 +373,15 @@ class JavaSparkContext(val sc: SparkContext) * other necessary info (e.g. file name for a filesystem-based dataset, table name for HyperTable, * etc). * + * @param conf JobConf for setting up the dataset. Note: This will be put into a Broadcast. + * Therefore if you plan to reuse this conf to create multiple RDDs, you need to make + * sure you won't modify the conf. A safe approach is always creating a new conf for + * a new RDD. + * @param inputFormatClass Class of the InputFormat + * @param keyClass Class of the keys + * @param valueClass Class of the values + * @param minPartitions Minimum number of Hadoop Splits to generate. + * * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each * record, directly caching the returned RDD will create many references to the same object. * If you plan to directly cache Hadoop writable objects, you should first copy them using @@ -395,6 +404,14 @@ class JavaSparkContext(val sc: SparkContext) * Get an RDD for a Hadoop-readable dataset from a Hadooop JobConf giving its InputFormat and any * other necessary info (e.g. file name for a filesystem-based dataset, table name for HyperTable, * + * @param conf JobConf for setting up the dataset. Note: This will be put into a Broadcast. + * Therefore if you plan to reuse this conf to create multiple RDDs, you need to make + * sure you won't modify the conf. A safe approach is always creating a new conf for + * a new RDD. + * @param inputFormatClass Class of the InputFormat + * @param keyClass Class of the keys + * @param valueClass Class of the values + * * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each * record, directly caching the returned RDD will create many references to the same object. * If you plan to directly cache Hadoop writable objects, you should first copy them using @@ -476,6 +493,14 @@ class JavaSparkContext(val sc: SparkContext) * Get an RDD for a given Hadoop file with an arbitrary new API InputFormat * and extra configuration options to pass to the input format. * + * @param conf Configuration for setting up the dataset. Note: This will be put into a Broadcast. + * Therefore if you plan to reuse this conf to create multiple RDDs, you need to make + * sure you won't modify the conf. A safe approach is always creating a new conf for + * a new RDD. + * @param fClass Class of the InputFormat + * @param kClass Class of the keys + * @param vClass Class of the values + * * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each * record, directly caching the returned RDD will create many references to the same object. * If you plan to directly cache Hadoop writable objects, you should first copy them using @@ -675,6 +700,9 @@ class JavaSparkContext(val sc: SparkContext) /** * Returns the Hadoop configuration used for the Hadoop code (e.g. file systems) we reuse. + * + * '''Note:''' As it will be reused in all Hadoop RDDs, it's better not to modify it unless you + * plan to set some global configurations for all Hadoop RDDs. */ def hadoopConfiguration(): Configuration = { sc.hadoopConfiguration From d34f79c8db79ae461fadae190446ebc19091bec9 Mon Sep 17 00:00:00 2001 From: WangTaoTheTonic Date: Fri, 6 Feb 2015 11:57:02 -0800 Subject: [PATCH 25/53] [SPARK-2945][YARN][Doc]add doc for spark.executor.instances https://issues.apache.org/jira/browse/SPARK-2945 spark.executor.instances works. As this JIRA recommended, we should add docs for this common config. Author: WangTaoTheTonic Closes #4350 from WangTaoTheTonic/SPARK-2945 and squashes the following commits: 4c3913a [WangTaoTheTonic] not compatible with dynamic allocation 5fa9c46 [WangTaoTheTonic] add doc for spark.executor.instances --- docs/running-on-yarn.md | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/docs/running-on-yarn.md b/docs/running-on-yarn.md index 68ab127bcf087..2b93eef6c26ed 100644 --- a/docs/running-on-yarn.md +++ b/docs/running-on-yarn.md @@ -104,6 +104,13 @@ Most of the configs are the same for Spark on YARN as for other deployment modes Comma-separated list of files to be placed in the working directory of each executor. + + spark.executor.instances + 2 + + The number of executors. Note that this property is incompatible with spark.dynamicAllocation.enabled. + + spark.yarn.executor.memoryOverhead executorMemory * 0.07, with minimum of 384 From 70e5b030a78ddfdcc8c9eee568009f277dee0872 Mon Sep 17 00:00:00 2001 From: Nicholas Chammas Date: Fri, 6 Feb 2015 12:08:22 -0800 Subject: [PATCH 26/53] [SPARK-5628] Add version option to spark-ec2 Every proper command line tool should include a `--version` option or something similar. This PR adds this to `spark-ec2` using the standard functionality provided by `optparse`. One thing we don't do here is follow the Python convention of setting `__version__`, since it seems awkward given how `spark-ec2` is laid out. Author: Nicholas Chammas Closes #4414 from nchammas/spark-ec2-show-version and squashes the following commits: 914cab5 [Nicholas Chammas] add version info --- ec2/spark_ec2.py | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/ec2/spark_ec2.py b/ec2/spark_ec2.py index abab209a05ba0..0de4a62e203fd 100755 --- a/ec2/spark_ec2.py +++ b/ec2/spark_ec2.py @@ -39,6 +39,9 @@ from optparse import OptionParser from sys import stderr +SPARK_EC2_VERSION = "1.2.0" +SPARK_EC2_DIR = os.path.dirname(os.path.realpath(__file__)) + VALID_SPARK_VERSIONS = set([ "0.7.3", "0.8.0", @@ -54,9 +57,8 @@ "1.2.0", ]) -DEFAULT_SPARK_VERSION = "1.2.0" +DEFAULT_SPARK_VERSION = SPARK_EC2_VERSION DEFAULT_SPARK_GITHUB_REPO = "https://github.com/apache/spark" -SPARK_EC2_DIR = os.path.dirname(os.path.realpath(__file__)) MESOS_SPARK_EC2_BRANCH = "branch-1.3" # A URL prefix from which to fetch AMI information @@ -103,12 +105,10 @@ class UsageError(Exception): # Configure and parse our command-line arguments def parse_args(): parser = OptionParser( - usage="spark-ec2 [options] " - + "\n\n can be: launch, destroy, login, stop, start, get-master, reboot-slaves", - add_help_option=False) - parser.add_option( - "-h", "--help", action="help", - help="Show this help message and exit") + prog="spark-ec2", + version="%prog {v}".format(v=SPARK_EC2_VERSION), + usage="%prog [options] \n\n" + + " can be: launch, destroy, login, stop, start, get-master, reboot-slaves") parser.add_option( "-s", "--slaves", type="int", default=1, help="Number of slaves to launch (default: %default)") From d433816157bb3ae1f0fbe44efec43a0c906d9f82 Mon Sep 17 00:00:00 2001 From: Liang-Chi Hsieh Date: Fri, 6 Feb 2015 12:13:44 -0800 Subject: [PATCH 27/53] [SPARK-5650][SQL] Support optional 'FROM' clause In Hive, 'FROM' clause is optional. This pr supports it. Author: Liang-Chi Hsieh Closes #4426 from viirya/optional_from and squashes the following commits: fe81f31 [Liang-Chi Hsieh] Support optional 'FROM' clause. --- .../org/apache/spark/sql/hive/HiveQl.scala | 19 ++++++++++++++----- ... clause-0-b42b408a87b258921240058f880a721a | 1 + .../sql/hive/execution/HiveQuerySuite.scala | 3 +++ 3 files changed, 18 insertions(+), 5 deletions(-) create mode 100644 sql/hive/src/test/resources/golden/no from clause-0-b42b408a87b258921240058f880a721a diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala index c19a091719190..6265c61a1dabe 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala @@ -556,9 +556,14 @@ https://cwiki.apache.org/confluence/display/Hive/Enhanced+Aggregation%2C+Cube%2C case Token("TOK_TRUNCATETABLE", Token("TOK_TABLE_PARTITION",table)::Nil) => NativePlaceholder - case Token("TOK_QUERY", - Token("TOK_FROM", fromClause :: Nil) :: - insertClauses) => + case Token("TOK_QUERY", queryArgs) + if Seq("TOK_FROM", "TOK_INSERT").contains(queryArgs.head.getText) => + + val (fromClause: Option[ASTNode], insertClauses) = queryArgs match { + case Token("TOK_FROM", args: Seq[ASTNode]) :: insertClauses => + (Some(args.head), insertClauses) + case Token("TOK_INSERT", _) :: Nil => (None, queryArgs) + } // Return one query for each insert clause. val queries = insertClauses.map { case Token("TOK_INSERT", singleInsert) => @@ -599,8 +604,12 @@ https://cwiki.apache.org/confluence/display/Hive/Enhanced+Aggregation%2C+Cube%2C "TOK_LATERAL_VIEW"), singleInsert) } - - val relations = nodeToRelation(fromClause) + + val relations = fromClause match { + case Some(f) => nodeToRelation(f) + case None => NoRelation + } + val withWhere = whereClause.map { whereNode => val Seq(whereExpr) = whereNode.getChildren.toSeq Filter(nodeToExpr(whereExpr), relations) diff --git a/sql/hive/src/test/resources/golden/no from clause-0-b42b408a87b258921240058f880a721a b/sql/hive/src/test/resources/golden/no from clause-0-b42b408a87b258921240058f880a721a new file mode 100644 index 0000000000000..390d344ecb9d3 --- /dev/null +++ b/sql/hive/src/test/resources/golden/no from clause-0-b42b408a87b258921240058f880a721a @@ -0,0 +1 @@ +1 1 -1 diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala index bb73ff1ea7e43..d16a1e0b73faa 100644 --- a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala +++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala @@ -202,6 +202,9 @@ class HiveQuerySuite extends HiveComparisonTest with BeforeAndAfter { createQueryTest("having no references", "SELECT key FROM src GROUP BY key HAVING COUNT(*) > 1") + createQueryTest("no from clause", + "SELECT 1, +1, -1") + createQueryTest("boolean = number", """ |SELECT From 500dc2b4b3136029457e708859fe27da93b1f9e8 Mon Sep 17 00:00:00 2001 From: Tobias Schlatter Date: Fri, 6 Feb 2015 12:15:02 -0800 Subject: [PATCH 28/53] [SPARK-5640] Synchronize ScalaReflection where necessary Author: Tobias Schlatter Closes #4431 from gzm0/sync-scala-refl and squashes the following commits: c5da21e [Tobias Schlatter] [SPARK-5640] Synchronize ScalaReflection where necessary --- .../org/apache/spark/sql/catalyst/ScalaReflection.scala | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala index 0445f3aa07912..5d9c331ca5178 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala @@ -103,10 +103,11 @@ trait ScalaReflection { } /** Returns a catalyst DataType and its nullability for the given Scala Type using reflection. */ - def schemaFor[T: TypeTag]: Schema = schemaFor(typeOf[T]) + def schemaFor[T: TypeTag]: Schema = + ScalaReflectionLock.synchronized { schemaFor(typeOf[T]) } /** Returns a catalyst DataType and its nullability for the given Scala Type using reflection. */ - def schemaFor(tpe: `Type`): Schema = { + def schemaFor(tpe: `Type`): Schema = ScalaReflectionLock.synchronized { val className: String = tpe.erasure.typeSymbol.asClass.fullName tpe match { case t if Utils.classIsLoadable(className) && From a958d60975147fb1afc76fcbd80f65ac8d78759a Mon Sep 17 00:00:00 2001 From: q00251598 Date: Fri, 6 Feb 2015 12:29:26 -0800 Subject: [PATCH 29/53] [SPARK-5619][SQL] Support 'show roles' in HiveContext Author: q00251598 Closes #4397 from watermen/SPARK-5619 and squashes the following commits: f819b6c [q00251598] Support show roles in HiveContext. --- sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala | 1 + 1 file changed, 1 insertion(+) diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala index 6265c61a1dabe..be63aa1a93aee 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala @@ -68,6 +68,7 @@ private[hive] object HiveQl { "TOK_SHOWLOCKS", "TOK_UNLOCKTABLE", + "TOK_SHOW_ROLES", "TOK_CREATEROLE", "TOK_DROPROLE", "TOK_GRANT", From 0b7eb3f3b700080bf6cb810d092709a8a468e5db Mon Sep 17 00:00:00 2001 From: OopsOutOfMemory Date: Fri, 6 Feb 2015 12:33:20 -0800 Subject: [PATCH 30/53] [SPARK-5324][SQL] Results of describe can't be queried Make below code works. ``` sql("DESCRIBE test").registerTempTable("describeTest") sql("SELECT * FROM describeTest").collect() ``` Author: OopsOutOfMemory Author: Sheng, Li Closes #4249 from OopsOutOfMemory/desc_query and squashes the following commits: 6fee13d [OopsOutOfMemory] up-to-date e71430a [Sheng, Li] Update HiveOperatorQueryableSuite.scala 3ba1058 [OopsOutOfMemory] change to default argument aac7226 [OopsOutOfMemory] Merge branch 'master' into desc_query 68eb6dd [OopsOutOfMemory] Merge branch 'desc_query' of github.com:OopsOutOfMemory/spark into desc_query 354ad71 [OopsOutOfMemory] query describe command d541a35 [OopsOutOfMemory] refine test suite e1da481 [OopsOutOfMemory] refine test suite a780539 [OopsOutOfMemory] Merge branch 'desc_query' of github.com:OopsOutOfMemory/spark into desc_query 0015f82 [OopsOutOfMemory] code style dd0aaef [OopsOutOfMemory] code style c7d606d [OopsOutOfMemory] rename test suite 75f2342 [OopsOutOfMemory] refine code and test suite f942c9b [OopsOutOfMemory] initial 11559ae [OopsOutOfMemory] code style c5fdecf [OopsOutOfMemory] code style aeaea5f [OopsOutOfMemory] rename test suite ac2c3bb [OopsOutOfMemory] refine code and test suite 544573e [OopsOutOfMemory] initial --- .../org/apache/spark/sql/sources/ddl.scala | 11 ++-- .../org/apache/spark/sql/hive/HiveQl.scala | 5 +- .../HiveOperatorQueryableSuite.scala | 51 +++++++++++++++++++ .../sql/hive/execution/HiveQuerySuite.scala | 2 +- 4 files changed, 61 insertions(+), 8 deletions(-) create mode 100644 sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveOperatorQueryableSuite.scala diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/ddl.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/ddl.scala index 9c37e0169ff85..a692ef51b31ed 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/sources/ddl.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/ddl.scala @@ -296,11 +296,14 @@ private[sql] case class ResolvedDataSource(provider: Class[_], relation: BaseRel private[sql] case class DescribeCommand( table: LogicalPlan, isExtended: Boolean) extends Command { - override def output = Seq( + override val output = Seq( // Column names are based on Hive. - AttributeReference("col_name", StringType, nullable = false)(), - AttributeReference("data_type", StringType, nullable = false)(), - AttributeReference("comment", StringType, nullable = false)()) + AttributeReference("col_name", StringType, nullable = false, + new MetadataBuilder().putString("comment", "name of the column").build())(), + AttributeReference("data_type", StringType, nullable = false, + new MetadataBuilder().putString("comment", "data type of the column").build())(), + AttributeReference("comment", StringType, nullable = false, + new MetadataBuilder().putString("comment", "comment of the column").build())()) } private[sql] case class CreateTableUsing( diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala index be63aa1a93aee..4b7fa06532963 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala @@ -497,15 +497,14 @@ https://cwiki.apache.org/confluence/display/Hive/Enhanced+Aggregation%2C+Cube%2C // TODO: Actually, a user may mean tableName.columnName. Need to resolve this issue. val tableIdent = extractTableIdent(nameParts.head) DescribeCommand( - UnresolvedRelation(tableIdent, None), extended.isDefined) + UnresolvedRelation(tableIdent, None), isExtended = extended.isDefined) case Token(".", dbName :: tableName :: colName :: Nil) => // It is describing a column with the format like "describe db.table column". NativePlaceholder case tableName => // It is describing a table with the format like "describe table". DescribeCommand( - UnresolvedRelation(Seq(tableName.getText), None), - extended.isDefined) + UnresolvedRelation(Seq(tableName.getText), None), isExtended = extended.isDefined) } } // All other cases. diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveOperatorQueryableSuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveOperatorQueryableSuite.scala new file mode 100644 index 0000000000000..efbef68cd4447 --- /dev/null +++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveOperatorQueryableSuite.scala @@ -0,0 +1,51 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.hive.execution + +import org.apache.spark.sql.{Row, QueryTest} +import org.apache.spark.sql.hive.test.TestHive._ + +/** + * A set of tests that validates commands can also be queried by like a table + */ +class HiveOperatorQueryableSuite extends QueryTest { + test("SPARK-5324 query result of describe command") { + loadTestTable("src") + + // register a describe command to be a temp table + sql("desc src").registerTempTable("mydesc") + checkAnswer( + sql("desc mydesc"), + Seq( + Row("col_name", "string", "name of the column"), + Row("data_type", "string", "data type of the column"), + Row("comment", "string", "comment of the column"))) + + checkAnswer( + sql("select * from mydesc"), + Seq( + Row("key", "int", null), + Row("value", "string", null))) + + checkAnswer( + sql("select col_name, data_type, comment from mydesc"), + Seq( + Row("key", "int", null), + Row("value", "string", null))) + } +} diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala index d16a1e0b73faa..27047ce4b1b0b 100644 --- a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala +++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala @@ -59,7 +59,7 @@ class HiveQuerySuite extends HiveComparisonTest with BeforeAndAfter { Locale.setDefault(originalLocale) } - test("SPARK-4908: concurent hive native commands") { + test("SPARK-4908: concurrent hive native commands") { (1 to 100).par.map { _ => sql("USE default") sql("SHOW TABLES") From 3eccf29ce061559c86e6f7338851932fc89a9afd Mon Sep 17 00:00:00 2001 From: Yin Huai Date: Fri, 6 Feb 2015 12:38:07 -0800 Subject: [PATCH 31/53] [SPARK-5595][SPARK-5603][SQL] Add a rule to do PreInsert type casting and field renaming and invalidating in memory cache after INSERT This PR adds a rule to Analyzer that will add preinsert data type casting and field renaming to the select clause in an `INSERT INTO/OVERWRITE` statement. Also, with the change of this PR, we always invalidate our in memory data cache after inserting into a BaseRelation. cc marmbrus liancheng Author: Yin Huai Closes #4373 from yhuai/insertFollowUp and squashes the following commits: 08237a7 [Yin Huai] Merge remote-tracking branch 'upstream/master' into insertFollowUp 316542e [Yin Huai] Doc update. c9ccfeb [Yin Huai] Revert a unnecessary change. 84aecc4 [Yin Huai] Address comments. 1951fe1 [Yin Huai] Merge remote-tracking branch 'upstream/master' c18da34 [Yin Huai] Invalidate cache after insert. 727f21a [Yin Huai] Preinsert casting and renaming. --- .../org/apache/spark/sql/SQLContext.scala | 6 +- .../apache/spark/sql/json/JSONRelation.scala | 12 ++- .../sql/sources/DataSourceStrategy.scala | 2 +- .../apache/spark/sql/sources/commands.scala | 10 ++- .../apache/spark/sql/sources/interfaces.scala | 16 ++++ .../org/apache/spark/sql/sources/rules.scala | 76 ++++++++++++++++++ .../org/apache/spark/sql/json/JsonSuite.scala | 25 ++++++ .../spark/sql/sources/DataSourceTest.scala | 6 +- .../spark/sql/sources/InsertIntoSuite.scala | 80 +++++++++++++++++++ .../apache/spark/sql/hive/HiveContext.scala | 1 + 10 files changed, 227 insertions(+), 7 deletions(-) create mode 100644 sql/core/src/main/scala/org/apache/spark/sql/sources/rules.scala diff --git a/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala b/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala index 706ef6ad4f174..bf3990671029e 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala @@ -91,7 +91,11 @@ class SQLContext(@transient val sparkContext: SparkContext) @transient protected[sql] lazy val analyzer: Analyzer = - new Analyzer(catalog, functionRegistry, caseSensitive = true) + new Analyzer(catalog, functionRegistry, caseSensitive = true) { + override val extendedRules = + sources.PreInsertCastAndRename :: + Nil + } @transient protected[sql] lazy val optimizer: Optimizer = DefaultOptimizer diff --git a/sql/core/src/main/scala/org/apache/spark/sql/json/JSONRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/json/JSONRelation.scala index f27585d05a986..c4e14c6c92908 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/json/JSONRelation.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/json/JSONRelation.scala @@ -72,7 +72,6 @@ private[sql] case class JSONRelation( userSpecifiedSchema: Option[StructType])( @transient val sqlContext: SQLContext) extends TableScan with InsertableRelation { - // TODO: Support partitioned JSON relation. private def baseRDD = sqlContext.sparkContext.textFile(path) @@ -99,10 +98,21 @@ private[sql] case class JSONRelation( s"Unable to clear output directory ${filesystemPath.toString} prior" + s" to INSERT OVERWRITE a JSON table:\n${e.toString}") } + // Write the data. data.toJSON.saveAsTextFile(path) + // Right now, we assume that the schema is not changed. We will not update the schema. + // schema = data.schema } else { // TODO: Support INSERT INTO sys.error("JSON table only support INSERT OVERWRITE for now.") } } + + override def hashCode(): Int = 41 * (41 + path.hashCode) + schema.hashCode() + + override def equals(other: Any): Boolean = other match { + case that: JSONRelation => + (this.path == that.path) && (this.schema == that.schema) + case _ => false + } } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/DataSourceStrategy.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/DataSourceStrategy.scala index d23ffb8b7a960..624369afe87b5 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/sources/DataSourceStrategy.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/DataSourceStrategy.scala @@ -59,7 +59,7 @@ private[sql] object DataSourceStrategy extends Strategy { if (partition.nonEmpty) { sys.error(s"Insert into a partition is not allowed because $l is not partitioned.") } - execution.ExecutedCommand(InsertIntoRelation(t, query, overwrite)) :: Nil + execution.ExecutedCommand(InsertIntoDataSource(l, query, overwrite)) :: Nil case _ => Nil } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/commands.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/commands.scala index d7942dc30934b..c9cd0e6e93829 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/sources/commands.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/commands.scala @@ -19,17 +19,21 @@ package org.apache.spark.sql.sources import org.apache.spark.sql.{DataFrame, SQLContext} import org.apache.spark.sql.catalyst.expressions._ import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan -import org.apache.spark.sql.execution.RunnableCommand +import org.apache.spark.sql.execution.{LogicalRDD, RunnableCommand} -private[sql] case class InsertIntoRelation( - relation: InsertableRelation, +private[sql] case class InsertIntoDataSource( + logicalRelation: LogicalRelation, query: LogicalPlan, overwrite: Boolean) extends RunnableCommand { override def run(sqlContext: SQLContext) = { + val relation = logicalRelation.relation.asInstanceOf[InsertableRelation] relation.insert(DataFrame(sqlContext, query), overwrite) + // Invalidate the cache. + sqlContext.cacheManager.invalidateCache(logicalRelation) + Seq.empty[Row] } } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala index 40fc1f2aa2724..a640ba57e0885 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala @@ -158,6 +158,22 @@ trait CatalystScan extends BaseRelation { } @DeveloperApi +/** + * ::DeveloperApi:: + * A BaseRelation that can be used to insert data into it through the insert method. + * If overwrite in insert method is true, the old data in the relation should be overwritten with + * the new data. If overwrite in insert method is false, the new data should be appended. + * + * InsertableRelation has the following three assumptions. + * 1. It assumes that the data (Rows in the DataFrame) provided to the insert method + * exactly matches the ordinal of fields in the schema of the BaseRelation. + * 2. It assumes that the schema of this relation will not be changed. + * Even if the insert method updates the schema (e.g. a relation of JSON or Parquet data may have a + * schema update after an insert operation), the new schema will not be used. + * 3. It assumes that fields of the data provided in the insert method are nullable. + * If a data source needs to check the actual nullability of a field, it needs to do it in the + * insert method. + */ trait InsertableRelation extends BaseRelation { def insert(data: DataFrame, overwrite: Boolean): Unit } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/rules.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/rules.scala new file mode 100644 index 0000000000000..4ed22d363da5b --- /dev/null +++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/rules.scala @@ -0,0 +1,76 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.sources + +import org.apache.spark.sql.catalyst.expressions.{Attribute, Cast, Alias} +import org.apache.spark.sql.catalyst.plans.logical.{InsertIntoTable, LogicalPlan, Project} +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.types.DataType + +/** + * A rule to do pre-insert data type casting and field renaming. Before we insert into + * an [[InsertableRelation]], we will use this rule to make sure that + * the columns to be inserted have the correct data type and fields have the correct names. + * @param resolver The resolver used by the Analyzer. + */ +private[sql] object PreInsertCastAndRename extends Rule[LogicalPlan] { + def apply(plan: LogicalPlan): LogicalPlan = { + plan.transform { + // Wait until children are resolved. + case p: LogicalPlan if !p.childrenResolved => p + + // We are inserting into an InsertableRelation. + case i @ InsertIntoTable( + l @ LogicalRelation(r: InsertableRelation), partition, child, overwrite) => { + // First, make sure the data to be inserted have the same number of fields with the + // schema of the relation. + if (l.output.size != child.output.size) { + sys.error( + s"$l requires that the query in the SELECT clause of the INSERT INTO/OVERWRITE " + + s"statement generates the same number of columns as its schema.") + } + castAndRenameChildOutput(i, l.output, child) + } + } + } + + /** If necessary, cast data types and rename fields to the expected types and names. */ + def castAndRenameChildOutput( + insertInto: InsertIntoTable, + expectedOutput: Seq[Attribute], + child: LogicalPlan) = { + val newChildOutput = expectedOutput.zip(child.output).map { + case (expected, actual) => + val needCast = !DataType.equalsIgnoreNullability(expected.dataType, actual.dataType) + // We want to make sure the filed names in the data to be inserted exactly match + // names in the schema. + val needRename = expected.name != actual.name + (needCast, needRename) match { + case (true, _) => Alias(Cast(actual, expected.dataType), expected.name)() + case (false, true) => Alias(actual, expected.name)() + case (_, _) => actual + } + } + + if (newChildOutput == child.output) { + insertInto + } else { + insertInto.copy(child = Project(newChildOutput, child)) + } + } +} diff --git a/sql/core/src/test/scala/org/apache/spark/sql/json/JsonSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/json/JsonSuite.scala index 1396c6b7246d1..926ba68828ee8 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/json/JsonSuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/json/JsonSuite.scala @@ -23,6 +23,7 @@ import org.apache.spark.sql.TestData._ import org.apache.spark.sql.catalyst.util._ import org.apache.spark.sql.Dsl._ import org.apache.spark.sql.json.JsonRDD.{compatibleType, enforceCorrectType} +import org.apache.spark.sql.sources.LogicalRelation import org.apache.spark.sql.test.TestSQLContext import org.apache.spark.sql.test.TestSQLContext._ import org.apache.spark.sql.types._ @@ -923,6 +924,30 @@ class JsonSuite extends QueryTest { sql("select structWithArrayFields.field1[1], structWithArrayFields.field2[3] from complexTable"), Row(5, null) ) + } + test("JSONRelation equality test") { + val relation1 = + JSONRelation("path", 1.0, Some(StructType(StructField("a", IntegerType, true) :: Nil)))(null) + val logicalRelation1 = LogicalRelation(relation1) + val relation2 = + JSONRelation("path", 0.5, Some(StructType(StructField("a", IntegerType, true) :: Nil)))( + org.apache.spark.sql.test.TestSQLContext) + val logicalRelation2 = LogicalRelation(relation2) + val relation3 = + JSONRelation("path", 1.0, Some(StructType(StructField("b", StringType, true) :: Nil)))(null) + val logicalRelation3 = LogicalRelation(relation3) + + assert(relation1 === relation2) + assert(logicalRelation1.sameResult(logicalRelation2), + s"$logicalRelation1 and $logicalRelation2 should be considered having the same result.") + + assert(relation1 !== relation3) + assert(!logicalRelation1.sameResult(logicalRelation3), + s"$logicalRelation1 and $logicalRelation3 should be considered not having the same result.") + + assert(relation2 !== relation3) + assert(!logicalRelation2.sameResult(logicalRelation3), + s"$logicalRelation2 and $logicalRelation3 should be considered not having the same result.") } } diff --git a/sql/core/src/test/scala/org/apache/spark/sql/sources/DataSourceTest.scala b/sql/core/src/test/scala/org/apache/spark/sql/sources/DataSourceTest.scala index 9626252e742e5..53f5f7426e9e6 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/sources/DataSourceTest.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/sources/DataSourceTest.scala @@ -28,7 +28,11 @@ abstract class DataSourceTest extends QueryTest with BeforeAndAfter { implicit val caseInsensisitiveContext = new SQLContext(TestSQLContext.sparkContext) { @transient override protected[sql] lazy val analyzer: Analyzer = - new Analyzer(catalog, functionRegistry, caseSensitive = false) + new Analyzer(catalog, functionRegistry, caseSensitive = false) { + override val extendedRules = + PreInsertCastAndRename :: + Nil + } } } diff --git a/sql/core/src/test/scala/org/apache/spark/sql/sources/InsertIntoSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/sources/InsertIntoSuite.scala index f91cea6a37060..36e504e759152 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/sources/InsertIntoSuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/sources/InsertIntoSuite.scala @@ -63,6 +63,41 @@ class InsertIntoSuite extends DataSourceTest with BeforeAndAfterAll { ) } + test("PreInsert casting and renaming") { + sql( + s""" + |INSERT OVERWRITE TABLE jsonTable SELECT a * 2, a * 4 FROM jt + """.stripMargin) + + checkAnswer( + sql("SELECT a, b FROM jsonTable"), + (1 to 10).map(i => Row(i * 2, s"${i * 4}")) + ) + + sql( + s""" + |INSERT OVERWRITE TABLE jsonTable SELECT a * 4 AS A, a * 6 as c FROM jt + """.stripMargin) + + checkAnswer( + sql("SELECT a, b FROM jsonTable"), + (1 to 10).map(i => Row(i * 4, s"${i * 6}")) + ) + } + + test("SELECT clause generating a different number of columns is not allowed.") { + val message = intercept[RuntimeException] { + sql( + s""" + |INSERT OVERWRITE TABLE jsonTable SELECT a FROM jt + """.stripMargin) + }.getMessage + assert( + message.contains("generates the same number of columns as its schema"), + "SELECT clause generating a different number of columns should not be not allowed." + ) + } + test("INSERT OVERWRITE a JSONRelation multiple times") { sql( s""" @@ -93,4 +128,49 @@ class InsertIntoSuite extends DataSourceTest with BeforeAndAfterAll { """.stripMargin) } } + + test("Caching") { + // Cached Query Execution + cacheTable("jsonTable") + assertCached(sql("SELECT * FROM jsonTable")) + checkAnswer( + sql("SELECT * FROM jsonTable"), + (1 to 10).map(i => Row(i, s"str$i"))) + + assertCached(sql("SELECT a FROM jsonTable")) + checkAnswer( + sql("SELECT a FROM jsonTable"), + (1 to 10).map(Row(_)).toSeq) + + assertCached(sql("SELECT a FROM jsonTable WHERE a < 5")) + checkAnswer( + sql("SELECT a FROM jsonTable WHERE a < 5"), + (1 to 4).map(Row(_)).toSeq) + + assertCached(sql("SELECT a * 2 FROM jsonTable")) + checkAnswer( + sql("SELECT a * 2 FROM jsonTable"), + (1 to 10).map(i => Row(i * 2)).toSeq) + + assertCached(sql("SELECT x.a, y.a FROM jsonTable x JOIN jsonTable y ON x.a = y.a + 1"), 2) + checkAnswer( + sql("SELECT x.a, y.a FROM jsonTable x JOIN jsonTable y ON x.a = y.a + 1"), + (2 to 10).map(i => Row(i, i - 1)).toSeq) + + // Insert overwrite and keep the same schema. + sql( + s""" + |INSERT OVERWRITE TABLE jsonTable SELECT a * 2, b FROM jt + """.stripMargin) + // jsonTable should be recached. + assertCached(sql("SELECT * FROM jsonTable")) + // The cached data is the new data. + checkAnswer( + sql("SELECT a, b FROM jsonTable"), + sql("SELECT a * 2, b FROM jt").collect()) + + // Verify uncaching + uncacheTable("jsonTable") + assertCached(sql("SELECT * FROM jsonTable"), 0) + } } diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala index d2371d4a5519e..ad37b7d0e6f59 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala @@ -324,6 +324,7 @@ class HiveContext(sc: SparkContext) extends SQLContext(sc) { catalog.PreInsertionCasts :: ExtractPythonUdfs :: ResolveUdtfsAlias :: + sources.PreInsertCastAndRename :: Nil } From b62c35245a6c9bee6a335aa5dcfcca2547e85406 Mon Sep 17 00:00:00 2001 From: OopsOutOfMemory Date: Fri, 6 Feb 2015 12:41:28 -0800 Subject: [PATCH 32/53] [SQL][HiveConsole][DOC] HiveConsole `correct hiveconsole imports` Sorry for that PR #4330 has some mistakes. I correct it.... so it works correctly now. Author: OopsOutOfMemory Closes #4389 from OopsOutOfMemory/doc and squashes the following commits: 843eed9 [OopsOutOfMemory] correct hiveconsole imports --- project/SparkBuild.scala | 1 + sql/README.md | 9 ++++++++- 2 files changed, 9 insertions(+), 1 deletion(-) diff --git a/project/SparkBuild.scala b/project/SparkBuild.scala index f63f9c1982bb5..5e3051c091030 100644 --- a/project/SparkBuild.scala +++ b/project/SparkBuild.scala @@ -276,6 +276,7 @@ object Hive { |import org.apache.spark.sql.catalyst.plans.logical._ |import org.apache.spark.sql.catalyst.rules._ |import org.apache.spark.sql.catalyst.util._ + |import org.apache.spark.sql.Dsl._ |import org.apache.spark.sql.execution |import org.apache.spark.sql.hive._ |import org.apache.spark.sql.hive.test.TestHive._ diff --git a/sql/README.md b/sql/README.md index 254ab8eb6376f..a79249965ee67 100644 --- a/sql/README.md +++ b/sql/README.md @@ -29,12 +29,19 @@ From here you can execute queries with HiveQl and manipulate DataFrame by using catalyst$ build/sbt hive/console [info] Starting scala interpreter... +import org.apache.spark.sql.catalyst.analysis._ +import org.apache.spark.sql.catalyst.dsl._ +import org.apache.spark.sql.catalyst.errors._ +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.catalyst.plans.logical._ +import org.apache.spark.sql.catalyst.rules._ +import org.apache.spark.sql.catalyst.util._ import org.apache.spark.sql.Dsl._ +import org.apache.spark.sql.execution import org.apache.spark.sql.hive._ import org.apache.spark.sql.hive.test.TestHive._ import org.apache.spark.sql.types._ import org.apache.spark.sql.parquet.ParquetTestData -Welcome to Scala version 2.10.4 (Java HotSpot(TM) 64-Bit Server VM, Java 1.7.0_45). Type in expressions to have them evaluated. Type :help for more information. From bc36356080e3b52aaf61fc1e6b204146ab96d29f Mon Sep 17 00:00:00 2001 From: wangfei Date: Fri, 6 Feb 2015 12:42:23 -0800 Subject: [PATCH 33/53] [SQL][Minor] Remove cache keyword in SqlParser Since cache keyword already defined in `SparkSQLParser` and `SqlParser` of catalyst is a more general parser which should not cover keywords related to underlying compute engine, to remove cache keyword in `SqlParser`. Author: wangfei Closes #4393 from scwf/remove-cache-keyword and squashes the following commits: 10ade16 [wangfei] remove cache keyword in sql parser --- .../src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala | 1 - 1 file changed, 1 deletion(-) diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala index a9bd079c7049d..1c588ee2f7176 100755 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala @@ -57,7 +57,6 @@ class SqlParser extends AbstractSparkSQLParser { protected val AVG = Keyword("AVG") protected val BETWEEN = Keyword("BETWEEN") protected val BY = Keyword("BY") - protected val CACHE = Keyword("CACHE") protected val CASE = Keyword("CASE") protected val CAST = Keyword("CAST") protected val COALESCE = Keyword("COALESCE") From 4793c8402a19afe4df51129a7f99e07494a76af2 Mon Sep 17 00:00:00 2001 From: Wenchen Fan Date: Fri, 6 Feb 2015 13:08:09 -0800 Subject: [PATCH 34/53] [SPARK-5278][SQL] Introduce UnresolvedGetField and complete the check of ambiguous reference to fields When the `GetField` chain(`a.b.c.d.....`) is interrupted by `GetItem` like `a.b[0].c.d....`, then the check of ambiguous reference to fields is broken. The reason is that: for something like `a.b[0].c.d`, we first parse it to `GetField(GetField(GetItem(Unresolved("a.b"), 0), "c"), "d")`. Then in `LogicalPlan#resolve`, we resolve `"a.b"` and build a `GetField` chain from bottom(the relation). But for the 2 outer `GetFiled`, we have to resolve them in `Analyzer` or do it in `GetField` lazily, check data type of child, search needed field, etc. which is similar to what we have done in `LogicalPlan#resolve`. So in this PR, the fix is just copy the same logic in `LogicalPlan#resolve` to `Analyzer`, which is simple and quick, but I do suggest introduce `UnresolvedGetFiled` like I explained in https://github.com/apache/spark/pull/2405. Author: Wenchen Fan Closes #4068 from cloud-fan/simple and squashes the following commits: a6857b5 [Wenchen Fan] fix import order 8411c40 [Wenchen Fan] use UnresolvedGetField --- .../apache/spark/sql/catalyst/SqlParser.scala | 2 +- .../sql/catalyst/analysis/Analyzer.scala | 34 +++++++++++++------ .../sql/catalyst/analysis/unresolved.scala | 12 +++++++ .../spark/sql/catalyst/dsl/package.scala | 4 +-- .../catalyst/expressions/complexTypes.scala | 24 ++----------- .../sql/catalyst/optimizer/Optimizer.scala | 2 +- .../catalyst/plans/logical/LogicalPlan.scala | 33 ++---------------- .../ExpressionEvaluationSuite.scala | 25 ++++++++++---- .../optimizer/ConstantFoldingSuite.scala | 4 +-- .../scala/org/apache/spark/sql/Column.scala | 3 +- .../org/apache/spark/sql/hive/HiveQl.scala | 2 +- .../hive/execution/HiveResolutionSuite.scala | 27 +++++++++------ 12 files changed, 84 insertions(+), 88 deletions(-) diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala index 1c588ee2f7176..124f083669358 100755 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala @@ -372,7 +372,7 @@ class SqlParser extends AbstractSparkSQLParser { | expression ~ ("[" ~> expression <~ "]") ^^ { case base ~ ordinal => GetItem(base, ordinal) } | (expression <~ ".") ~ ident ^^ - { case base ~ fieldName => GetField(base, fieldName) } + { case base ~ fieldName => UnresolvedGetField(base, fieldName) } | cast | "(" ~> expression <~ ")" | function diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala index ae1aee02c64a5..0b59ed1739566 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala @@ -285,7 +285,7 @@ class Analyzer(catalog: Catalog, case q: LogicalPlan => logTrace(s"Attempting to resolve ${q.simpleString}") - q transformExpressions { + q transformExpressionsUp { case u @ UnresolvedAttribute(name) if resolver(name, VirtualColumn.groupingIdName) && q.isInstanceOf[GroupingAnalytics] => // Resolve the virtual column GROUPING__ID for the operator GroupingAnalytics @@ -295,15 +295,8 @@ class Analyzer(catalog: Catalog, val result = q.resolveChildren(name, resolver).getOrElse(u) logDebug(s"Resolving $u to $result") result - - // Resolve field names using the resolver. - case f @ GetField(child, fieldName) if !f.resolved && child.resolved => - child.dataType match { - case StructType(fields) => - val resolvedFieldName = fields.map(_.name).find(resolver(_, fieldName)) - resolvedFieldName.map(n => f.copy(fieldName = n)).getOrElse(f) - case _ => f - } + case UnresolvedGetField(child, fieldName) if child.resolved => + resolveGetField(child, fieldName) } } @@ -312,6 +305,27 @@ class Analyzer(catalog: Catalog, */ protected def containsStar(exprs: Seq[Expression]): Boolean = exprs.exists(_.collect { case _: Star => true }.nonEmpty) + + /** + * Returns the resolved `GetField`, and report error if no desired field or over one + * desired fields are found. + */ + protected def resolveGetField(expr: Expression, fieldName: String): Expression = { + expr.dataType match { + case StructType(fields) => + val actualField = fields.filter(f => resolver(f.name, fieldName)) + if (actualField.length == 0) { + sys.error( + s"No such struct field $fieldName in ${fields.map(_.name).mkString(", ")}") + } else if (actualField.length == 1) { + val field = actualField(0) + GetField(expr, field, fields.indexOf(field)) + } else { + sys.error(s"Ambiguous reference to fields ${actualField.mkString(", ")}") + } + case otherType => sys.error(s"GetField is not valid on fields of type $otherType") + } + } } /** diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/unresolved.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/unresolved.scala index f35921e2a772c..f959a50564011 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/unresolved.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/unresolved.scala @@ -177,3 +177,15 @@ case class ResolvedStar(expressions: Seq[NamedExpression]) extends Star { override def expand(input: Seq[Attribute], resolver: Resolver): Seq[NamedExpression] = expressions override def toString = expressions.mkString("ResolvedStar(", ", ", ")") } + +case class UnresolvedGetField(child: Expression, fieldName: String) extends UnaryExpression { + override def dataType = throw new UnresolvedException(this, "dataType") + override def foldable = throw new UnresolvedException(this, "foldable") + override def nullable = throw new UnresolvedException(this, "nullable") + override lazy val resolved = false + + override def eval(input: Row = null): EvaluatedType = + throw new TreeNodeException(this, s"No function to evaluate expression. type: ${this.nodeName}") + + override def toString = s"$child.$fieldName" +} diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/dsl/package.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/dsl/package.scala index 417659eed5957..51a09ac0e1249 100755 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/dsl/package.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/dsl/package.scala @@ -22,7 +22,7 @@ import java.sql.{Date, Timestamp} import scala.language.implicitConversions import scala.reflect.runtime.universe.{TypeTag, typeTag} -import org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute +import org.apache.spark.sql.catalyst.analysis.{UnresolvedGetField, UnresolvedAttribute} import org.apache.spark.sql.catalyst.expressions._ import org.apache.spark.sql.catalyst.plans.logical._ import org.apache.spark.sql.catalyst.plans.{Inner, JoinType} @@ -101,7 +101,7 @@ package object dsl { def isNotNull = IsNotNull(expr) def getItem(ordinal: Expression) = GetItem(expr, ordinal) - def getField(fieldName: String) = GetField(expr, fieldName) + def getField(fieldName: String) = UnresolvedGetField(expr, fieldName) def cast(to: DataType) = Cast(expr, to) diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypes.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypes.scala index 1bc34f71441fe..66e2e5c4bafce 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypes.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypes.scala @@ -73,39 +73,19 @@ case class GetItem(child: Expression, ordinal: Expression) extends Expression { /** * Returns the value of fields in the Struct `child`. */ -case class GetField(child: Expression, fieldName: String) extends UnaryExpression { +case class GetField(child: Expression, field: StructField, ordinal: Int) extends UnaryExpression { type EvaluatedType = Any def dataType = field.dataType override def nullable = child.nullable || field.nullable override def foldable = child.foldable - protected def structType = child.dataType match { - case s: StructType => s - case otherType => sys.error(s"GetField is not valid on fields of type $otherType") - } - - lazy val field = - structType.fields - .find(_.name == fieldName) - .getOrElse(sys.error(s"No such field $fieldName in ${child.dataType}")) - - lazy val ordinal = structType.fields.indexOf(field) - - override lazy val resolved = childrenResolved && fieldResolved - - /** Returns true only if the fieldName is found in the child struct. */ - private def fieldResolved = child.dataType match { - case StructType(fields) => fields.map(_.name).contains(fieldName) - case _ => false - } - override def eval(input: Row): Any = { val baseValue = child.eval(input).asInstanceOf[Row] if (baseValue == null) null else baseValue(ordinal) } - override def toString = s"$child.$fieldName" + override def toString = s"$child.${field.name}" } /** diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala index 376a9f36568a7..8c8f2896eb99b 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala @@ -206,7 +206,7 @@ object NullPropagation extends Rule[LogicalPlan] { case e @ IsNotNull(c) if !c.nullable => Literal(true, BooleanType) case e @ GetItem(Literal(null, _), _) => Literal(null, e.dataType) case e @ GetItem(_, Literal(null, _)) => Literal(null, e.dataType) - case e @ GetField(Literal(null, _), _) => Literal(null, e.dataType) + case e @ GetField(Literal(null, _), _, _) => Literal(null, e.dataType) case e @ EqualNullSafe(Literal(null, _), r) => IsNull(r) case e @ EqualNullSafe(l, Literal(null, _)) => IsNull(l) case e @ Count(expr) if !expr.nullable => Count(Literal(1)) diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala index 65ae066e4b4b5..8d30528328946 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala @@ -18,7 +18,7 @@ package org.apache.spark.sql.catalyst.plans.logical import org.apache.spark.Logging -import org.apache.spark.sql.catalyst.analysis.Resolver +import org.apache.spark.sql.catalyst.analysis.{UnresolvedGetField, Resolver} import org.apache.spark.sql.catalyst.errors.TreeNodeException import org.apache.spark.sql.catalyst.expressions._ import org.apache.spark.sql.catalyst.plans.QueryPlan @@ -160,11 +160,7 @@ abstract class LogicalPlan extends QueryPlan[LogicalPlan] with Logging { // One match, but we also need to extract the requested nested field. case Seq((a, nestedFields)) => - val aliased = - Alias( - resolveNesting(nestedFields, a, resolver), - nestedFields.last)() // Preserve the case of the user's field access. - Some(aliased) + Some(Alias(nestedFields.foldLeft(a: Expression)(UnresolvedGetField), nestedFields.last)()) // No matches. case Seq() => @@ -177,31 +173,6 @@ abstract class LogicalPlan extends QueryPlan[LogicalPlan] with Logging { this, s"Ambiguous references to $name: ${ambiguousReferences.mkString(",")}") } } - - /** - * Given a list of successive nested field accesses, and a based expression, attempt to resolve - * the actual field lookups on this expression. - */ - private def resolveNesting( - nestedFields: List[String], - expression: Expression, - resolver: Resolver): Expression = { - - (nestedFields, expression.dataType) match { - case (Nil, _) => expression - case (requestedField :: rest, StructType(fields)) => - val actualField = fields.filter(f => resolver(f.name, requestedField)) - if (actualField.length == 0) { - sys.error( - s"No such struct field $requestedField in ${fields.map(_.name).mkString(", ")}") - } else if (actualField.length == 1) { - resolveNesting(rest, GetField(expression, actualField(0).name), resolver) - } else { - sys.error(s"Ambiguous reference to fields ${actualField.mkString(", ")}") - } - case (_, dt) => sys.error(s"Can't access nested field in type $dt") - } - } } /** diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala index 25d1c105a00a6..7cf6c80194f6c 100644 --- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala @@ -26,6 +26,7 @@ import org.scalatest.FunSuite import org.scalatest.Matchers._ import org.apache.spark.sql.catalyst.dsl.expressions._ +import org.apache.spark.sql.catalyst.analysis.UnresolvedGetField import org.apache.spark.sql.types._ @@ -846,23 +847,33 @@ class ExpressionEvaluationSuite extends FunSuite { checkEvaluation(GetItem(BoundReference(4, typeArray, true), Literal(null, IntegerType)), null, row) - checkEvaluation(GetField(BoundReference(2, typeS, nullable = true), "a"), "aa", row) - checkEvaluation(GetField(Literal(null, typeS), "a"), null, row) + def quickBuildGetField(expr: Expression, fieldName: String) = { + expr.dataType match { + case StructType(fields) => + val field = fields.find(_.name == fieldName).get + GetField(expr, field, fields.indexOf(field)) + } + } + + def quickResolve(u: UnresolvedGetField) = quickBuildGetField(u.child, u.fieldName) + + checkEvaluation(quickBuildGetField(BoundReference(2, typeS, nullable = true), "a"), "aa", row) + checkEvaluation(quickBuildGetField(Literal(null, typeS), "a"), null, row) val typeS_notNullable = StructType( StructField("a", StringType, nullable = false) :: StructField("b", StringType, nullable = false) :: Nil ) - assert(GetField(BoundReference(2,typeS, nullable = true), "a").nullable === true) - assert(GetField(BoundReference(2, typeS_notNullable, nullable = false), "a").nullable === false) + assert(quickBuildGetField(BoundReference(2,typeS, nullable = true), "a").nullable === true) + assert(quickBuildGetField(BoundReference(2, typeS_notNullable, nullable = false), "a").nullable === false) - assert(GetField(Literal(null, typeS), "a").nullable === true) - assert(GetField(Literal(null, typeS_notNullable), "a").nullable === true) + assert(quickBuildGetField(Literal(null, typeS), "a").nullable === true) + assert(quickBuildGetField(Literal(null, typeS_notNullable), "a").nullable === true) checkEvaluation('c.map(typeMap).at(3).getItem("aa"), "bb", row) checkEvaluation('c.array(typeArray.elementType).at(4).getItem(1), "bb", row) - checkEvaluation('c.struct(typeS).at(2).getField("a"), "aa", row) + checkEvaluation(quickResolve('c.struct(typeS).at(2).getField("a")), "aa", row) } test("arithmetic") { diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ConstantFoldingSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ConstantFoldingSuite.scala index 9fdf3efa02bb6..e22c62505860a 100644 --- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ConstantFoldingSuite.scala +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ConstantFoldingSuite.scala @@ -17,7 +17,7 @@ package org.apache.spark.sql.catalyst.optimizer -import org.apache.spark.sql.catalyst.analysis.EliminateAnalysisOperators +import org.apache.spark.sql.catalyst.analysis.{UnresolvedGetField, EliminateAnalysisOperators} import org.apache.spark.sql.catalyst.expressions._ import org.apache.spark.sql.catalyst.plans.logical.{LocalRelation, LogicalPlan} import org.apache.spark.sql.catalyst.plans.PlanTest @@ -184,7 +184,7 @@ class ConstantFoldingSuite extends PlanTest { GetItem(Literal(null, ArrayType(IntegerType)), 1) as 'c3, GetItem(Literal(Seq(1), ArrayType(IntegerType)), Literal(null, IntegerType)) as 'c4, - GetField( + UnresolvedGetField( Literal(null, StructType(Seq(StructField("a", IntegerType, true)))), "a") as 'c5, diff --git a/sql/core/src/main/scala/org/apache/spark/sql/Column.scala b/sql/core/src/main/scala/org/apache/spark/sql/Column.scala index 4c2aeadae9492..878b2b0556de7 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/Column.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/Column.scala @@ -23,6 +23,7 @@ import scala.language.implicitConversions import org.apache.spark.sql.Dsl.lit import org.apache.spark.sql.catalyst.expressions._ import org.apache.spark.sql.catalyst.plans.logical.{Subquery, Project, LogicalPlan} +import org.apache.spark.sql.catalyst.analysis.UnresolvedGetField import org.apache.spark.sql.types._ @@ -505,7 +506,7 @@ trait Column extends DataFrame { /** * An expression that gets a field by name in a [[StructField]]. */ - def getField(fieldName: String): Column = exprToColumn(GetField(expr, fieldName)) + def getField(fieldName: String): Column = exprToColumn(UnresolvedGetField(expr, fieldName)) /** * An expression that returns a substring. diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala index 4b7fa06532963..2a4b88092179f 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala @@ -1038,7 +1038,7 @@ https://cwiki.apache.org/confluence/display/Hive/Enhanced+Aggregation%2C+Cube%2C nodeToExpr(qualifier) match { case UnresolvedAttribute(qualifierName) => UnresolvedAttribute(qualifierName + "." + cleanIdentifier(attr)) - case other => GetField(other, attr) + case other => UnresolvedGetField(other, attr) } /* Stars (*) */ diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveResolutionSuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveResolutionSuite.scala index 178ece447758e..ff8130ae5f6bc 100644 --- a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveResolutionSuite.scala +++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveResolutionSuite.scala @@ -17,8 +17,7 @@ package org.apache.spark.sql.hive.execution -import org.apache.spark.sql.hive.test.TestHive -import org.apache.spark.sql.hive.test.TestHive.{sparkContext, sql} +import org.apache.spark.sql.hive.test.TestHive.{sparkContext, jsonRDD, sql} import org.apache.spark.sql.hive.test.TestHive.implicits._ case class Nested(a: Int, B: Int) @@ -29,16 +28,24 @@ case class Data(a: Int, B: Int, n: Nested, nestedArray: Seq[Nested]) */ class HiveResolutionSuite extends HiveComparisonTest { - case class NestedData(a: Seq[NestedData2], B: NestedData2) - case class NestedData2(a: NestedData3, B: NestedData3) - case class NestedData3(a: Int, B: Int) - test("SPARK-3698: case insensitive test for nested data") { - sparkContext.makeRDD(Seq.empty[NestedData]).registerTempTable("nested") + jsonRDD(sparkContext.makeRDD( + """{"a": [{"a": {"a": 1}}]}""" :: Nil)).registerTempTable("nested") // This should be successfully analyzed sql("SELECT a[0].A.A from nested").queryExecution.analyzed } + test("SPARK-5278: check ambiguous reference to fields") { + jsonRDD(sparkContext.makeRDD( + """{"a": [{"b": 1, "B": 2}]}""" :: Nil)).registerTempTable("nested") + + // there are 2 filed matching field name "b", we should report Ambiguous reference error + val exception = intercept[RuntimeException] { + sql("SELECT a[0].b from nested").queryExecution.analyzed + } + assert(exception.getMessage.contains("Ambiguous reference to fields")) + } + createQueryTest("table.attr", "SELECT src.key FROM src ORDER BY key LIMIT 1") @@ -68,7 +75,7 @@ class HiveResolutionSuite extends HiveComparisonTest { test("case insensitivity with scala reflection") { // Test resolution with Scala Reflection - TestHive.sparkContext.parallelize(Data(1, 2, Nested(1,2), Seq(Nested(1,2))) :: Nil) + sparkContext.parallelize(Data(1, 2, Nested(1,2), Seq(Nested(1,2))) :: Nil) .registerTempTable("caseSensitivityTest") val query = sql("SELECT a, b, A, B, n.a, n.b, n.A, n.B FROM caseSensitivityTest") @@ -79,14 +86,14 @@ class HiveResolutionSuite extends HiveComparisonTest { ignore("case insensitivity with scala reflection joins") { // Test resolution with Scala Reflection - TestHive.sparkContext.parallelize(Data(1, 2, Nested(1,2), Seq(Nested(1,2))) :: Nil) + sparkContext.parallelize(Data(1, 2, Nested(1,2), Seq(Nested(1,2))) :: Nil) .registerTempTable("caseSensitivityTest") sql("SELECT * FROM casesensitivitytest a JOIN casesensitivitytest b ON a.a = b.a").collect() } test("nested repeated resolution") { - TestHive.sparkContext.parallelize(Data(1, 2, Nested(1,2), Seq(Nested(1,2))) :: Nil) + sparkContext.parallelize(Data(1, 2, Nested(1,2), Seq(Nested(1,2))) :: Nil) .registerTempTable("nestedRepeatedTest") assert(sql("SELECT nestedArray[0].a FROM nestedRepeatedTest").collect().head(0) === 1) } From 3d3ecd77411bfdd1c72be51616c46a6caf839be2 Mon Sep 17 00:00:00 2001 From: OopsOutOfMemory Date: Fri, 6 Feb 2015 13:20:10 -0800 Subject: [PATCH 35/53] [SPARK-5586][Spark Shell][SQL] Make `sqlContext` available in spark shell Result is like this ``` 15/02/05 13:41:22 INFO SparkILoop: Created spark context.. Spark context available as sc. 15/02/05 13:41:22 INFO SparkILoop: Created sql context.. SQLContext available as sqlContext. scala> sq sql sqlContext sqlParser sqrt ``` Author: OopsOutOfMemory Closes #4387 from OopsOutOfMemory/sqlContextInShell and squashes the following commits: c7f5203 [OopsOutOfMemory] auto-import sql() function e160697 [OopsOutOfMemory] Merge branch 'sqlContextInShell' of https://github.com/OopsOutOfMemory/spark into sqlContextInShell 37c0a16 [OopsOutOfMemory] auto detect hive support a9c59d9 [OopsOutOfMemory] rename and reduce range of imports 6b9e309 [OopsOutOfMemory] Merge branch 'master' into sqlContextInShell cae652f [OopsOutOfMemory] make sqlContext available in spark shell --- .../org/apache/spark/repl/SparkILoop.scala | 19 +++++++++++++++++++ .../apache/spark/repl/SparkILoopInit.scala | 10 ++++++++++ .../scala/org/apache/spark/repl/Main.scala | 18 ++++++++++++++++++ .../org/apache/spark/repl/SparkILoop.scala | 12 +++++++++++- 4 files changed, 58 insertions(+), 1 deletion(-) diff --git a/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoop.scala b/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoop.scala index 72c1a989999b4..b4db3df795177 100644 --- a/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoop.scala +++ b/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoop.scala @@ -45,6 +45,7 @@ import scala.reflect.api.{Mirror, TypeCreator, Universe => ApiUniverse} import org.apache.spark.Logging import org.apache.spark.SparkConf import org.apache.spark.SparkContext +import org.apache.spark.sql.SQLContext import org.apache.spark.util.Utils /** The Scala interactive shell. It provides a read-eval-print loop @@ -130,6 +131,7 @@ class SparkILoop( // NOTE: Must be public for visibility @DeveloperApi var sparkContext: SparkContext = _ + var sqlContext: SQLContext = _ override def echoCommandMessage(msg: String) { intp.reporter printMessage msg @@ -1016,6 +1018,23 @@ class SparkILoop( sparkContext } + @DeveloperApi + def createSQLContext(): SQLContext = { + val name = "org.apache.spark.sql.hive.HiveContext" + val loader = Utils.getContextOrSparkClassLoader + try { + sqlContext = loader.loadClass(name).getConstructor(classOf[SparkContext]) + .newInstance(sparkContext).asInstanceOf[SQLContext] + logInfo("Created sql context (with Hive support)..") + } + catch { + case cnf: java.lang.ClassNotFoundException => + sqlContext = new SQLContext(sparkContext) + logInfo("Created sql context..") + } + sqlContext + } + private def getMaster(): String = { val master = this.master match { case Some(m) => m diff --git a/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoopInit.scala b/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoopInit.scala index 99bd777c04fdb..0cf2de6d399b0 100644 --- a/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoopInit.scala +++ b/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoopInit.scala @@ -127,7 +127,17 @@ private[repl] trait SparkILoopInit { _sc } """) + command(""" + @transient val sqlContext = { + val _sqlContext = org.apache.spark.repl.Main.interp.createSQLContext() + println("SQL context available as sqlContext.") + _sqlContext + } + """) command("import org.apache.spark.SparkContext._") + command("import sqlContext.implicits._") + command("import sqlContext.sql") + command("import org.apache.spark.sql.Dsl._") } } diff --git a/repl/scala-2.11/src/main/scala/org/apache/spark/repl/Main.scala b/repl/scala-2.11/src/main/scala/org/apache/spark/repl/Main.scala index 69e44d4f916e1..dc25692749aad 100644 --- a/repl/scala-2.11/src/main/scala/org/apache/spark/repl/Main.scala +++ b/repl/scala-2.11/src/main/scala/org/apache/spark/repl/Main.scala @@ -19,6 +19,7 @@ package org.apache.spark.repl import org.apache.spark.util.Utils import org.apache.spark._ +import org.apache.spark.sql.SQLContext import scala.tools.nsc.Settings import scala.tools.nsc.interpreter.SparkILoop @@ -34,6 +35,7 @@ object Main extends Logging { "-Yrepl-outdir", s"${outputDir.getAbsolutePath}", "-Yrepl-sync"), true) val classServer = new HttpServer(conf, outputDir, new SecurityManager(conf)) var sparkContext: SparkContext = _ + var sqlContext: SQLContext = _ var interp = new SparkILoop // this is a public var because tests reset it. def main(args: Array[String]) { @@ -74,6 +76,22 @@ object Main extends Logging { sparkContext } + def createSQLContext(): SQLContext = { + val name = "org.apache.spark.sql.hive.HiveContext" + val loader = Utils.getContextOrSparkClassLoader + try { + sqlContext = loader.loadClass(name).getConstructor(classOf[SparkContext]) + .newInstance(sparkContext).asInstanceOf[SQLContext] + logInfo("Created sql context (with Hive support)..") + } + catch { + case cnf: java.lang.ClassNotFoundException => + sqlContext = new SQLContext(sparkContext) + logInfo("Created sql context..") + } + sqlContext + } + private def getMaster: String = { val master = { val envMaster = sys.env.get("MASTER") diff --git a/repl/scala-2.11/src/main/scala/org/apache/spark/repl/SparkILoop.scala b/repl/scala-2.11/src/main/scala/org/apache/spark/repl/SparkILoop.scala index 250727305970d..1bd2a6991404b 100644 --- a/repl/scala-2.11/src/main/scala/org/apache/spark/repl/SparkILoop.scala +++ b/repl/scala-2.11/src/main/scala/org/apache/spark/repl/SparkILoop.scala @@ -66,8 +66,18 @@ class SparkILoop(in0: Option[BufferedReader], protected val out: JPrintWriter) println("Spark context available as sc.") _sc } - """) + """) + command( """ + @transient val sqlContext = { + val _sqlContext = org.apache.spark.repl.Main.createSQLContext() + println("SQL context available as sqlContext.") + _sqlContext + } + """) command("import org.apache.spark.SparkContext._") + command("import sqlContext.implicits._") + command("import sqlContext.sql") + command("import org.apache.spark.sql.Dsl._") } } From 0f3a36071a44e986d97981032d5b192477b38bbd Mon Sep 17 00:00:00 2001 From: GenTang Date: Fri, 6 Feb 2015 13:27:34 -0800 Subject: [PATCH 36/53] [SPARK-4983] Insert waiting time before tagging EC2 instances The boto API doesn't support tag EC2 instances in the same call that launches them. We add a five-second wait so EC2 has enough time to propagate the information so that the tagging can succeed. Author: GenTang Author: Gen TANG Closes #3986 from GenTang/spark-4983 and squashes the following commits: 13e257d [Gen TANG] modification of comments 47f06755 [GenTang] print the information ab7a931 [GenTang] solve the issus spark-4983 by inserting waiting time 3179737 [GenTang] Revert "handling exceptions about adding tags to ec2" 6a8b53b [GenTang] Revert "the improvement of exception handling" 13e97a6 [GenTang] Revert "typo" 63fd360 [GenTang] typo 692fc2b [GenTang] the improvement of exception handling 6adcf6d [GenTang] handling exceptions about adding tags to ec2 --- ec2/spark_ec2.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/ec2/spark_ec2.py b/ec2/spark_ec2.py index 0de4a62e203fd..7371558274d78 100755 --- a/ec2/spark_ec2.py +++ b/ec2/spark_ec2.py @@ -569,6 +569,9 @@ def launch_cluster(conn, opts, cluster_name): master_nodes = master_res.instances print "Launched master in %s, regid = %s" % (zone, master_res.id) + # This wait time corresponds to SPARK-4983 + print "Waiting for AWS to propagate instance metadata..." + time.sleep(5) # Give the instances descriptive names for master in master_nodes: master.add_tag( From 65181b7512c2192311705dfdfad588a80d3a6fde Mon Sep 17 00:00:00 2001 From: Xiangrui Meng Date: Fri, 6 Feb 2015 13:52:35 -0800 Subject: [PATCH 37/53] [HOTFIX][MLLIB] fix a compilation error with java 6 Author: Xiangrui Meng Closes #4442 from mengxr/java6-fix and squashes the following commits: 2098500 [Xiangrui Meng] fix a compilation error with java 6 --- .../org/apache/spark/examples/ml/JavaDeveloperApiExample.java | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java index 42d4d7d0bef26..4d9dad9f23038 100644 --- a/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java +++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java @@ -116,7 +116,7 @@ class MyJavaLogisticRegression */ IntParam maxIter = new IntParam(this, "maxIter", "max number of iterations"); - int getMaxIter() { return (int)get(maxIter); } + int getMaxIter() { return (Integer) get(maxIter); } public MyJavaLogisticRegression() { setMaxIter(100); @@ -124,7 +124,7 @@ public MyJavaLogisticRegression() { // The parameter setter is in this class since it should return type MyJavaLogisticRegression. MyJavaLogisticRegression setMaxIter(int value) { - return (MyJavaLogisticRegression)set(maxIter, value); + return (MyJavaLogisticRegression) set(maxIter, value); } // This method is used by fit(). From b3872e00d155939e40366debda635fc3fb12cc73 Mon Sep 17 00:00:00 2001 From: Vladimir Vladimirov Date: Fri, 6 Feb 2015 13:55:02 -0800 Subject: [PATCH 38/53] SPARK-5633 pyspark saveAsTextFile support for compression codec See https://issues.apache.org/jira/browse/SPARK-5633 for details Author: Vladimir Vladimirov Closes #4403 from smartkiwi/master and squashes the following commits: 94c014e [Vladimir Vladimirov] SPARK-5633 pyspark saveAsTextFile support for compression codec --- python/pyspark/rdd.py | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-) diff --git a/python/pyspark/rdd.py b/python/pyspark/rdd.py index 6e029bf7f13fc..bd4f16e058045 100644 --- a/python/pyspark/rdd.py +++ b/python/pyspark/rdd.py @@ -1366,10 +1366,14 @@ def saveAsPickleFile(self, path, batchSize=10): ser = BatchedSerializer(PickleSerializer(), batchSize) self._reserialize(ser)._jrdd.saveAsObjectFile(path) - def saveAsTextFile(self, path): + def saveAsTextFile(self, path, compressionCodecClass=None): """ Save this RDD as a text file, using string representations of elements. + @param path: path to text file + @param compressionCodecClass: (None by default) string i.e. + "org.apache.hadoop.io.compress.GzipCodec" + >>> tempFile = NamedTemporaryFile(delete=True) >>> tempFile.close() >>> sc.parallelize(range(10)).saveAsTextFile(tempFile.name) @@ -1385,6 +1389,16 @@ def saveAsTextFile(self, path): >>> sc.parallelize(['', 'foo', '', 'bar', '']).saveAsTextFile(tempFile2.name) >>> ''.join(sorted(input(glob(tempFile2.name + "/part-0000*")))) '\\n\\n\\nbar\\nfoo\\n' + + Using compressionCodecClass + + >>> tempFile3 = NamedTemporaryFile(delete=True) + >>> tempFile3.close() + >>> codec = "org.apache.hadoop.io.compress.GzipCodec" + >>> sc.parallelize(['foo', 'bar']).saveAsTextFile(tempFile3.name, codec) + >>> from fileinput import input, hook_compressed + >>> ''.join(sorted(input(glob(tempFile3.name + "/part*.gz"), openhook=hook_compressed))) + 'bar\\nfoo\\n' """ def func(split, iterator): for x in iterator: @@ -1395,7 +1409,11 @@ def func(split, iterator): yield x keyed = self.mapPartitionsWithIndex(func) keyed._bypass_serializer = True - keyed._jrdd.map(self.ctx._jvm.BytesToString()).saveAsTextFile(path) + if compressionCodecClass: + compressionCodec = self.ctx._jvm.java.lang.Class.forName(compressionCodecClass) + keyed._jrdd.map(self.ctx._jvm.BytesToString()).saveAsTextFile(path, compressionCodec) + else: + keyed._jrdd.map(self.ctx._jvm.BytesToString()).saveAsTextFile(path) # Pair functions From ca66159a4f30d65fa4cd32dbf3ff23978cb7f99b Mon Sep 17 00:00:00 2001 From: Kashish Jain Date: Fri, 6 Feb 2015 13:47:23 -0800 Subject: [PATCH 39/53] SPARK-5613: Catch the ApplicationNotFoundException exception to avoid thread from getting killed on yarn restart. [SPARK-5613] Added a catch block to catch the ApplicationNotFoundException. Without this catch block the thread gets killed on occurrence of this exception. This Exception occurs when yarn restarts and tries to find an application id for a spark job which got interrupted due to yarn getting stopped. See the stacktrace in the bug for more details. Author: Kashish Jain Closes #4392 from kasjain/branch-1.2 and squashes the following commits: 4831000 [Kashish Jain] SPARK-5613: Catch the ApplicationNotFoundException exception to avoid thread from getting killed on yarn restart. --- .../cluster/YarnClientSchedulerBackend.scala | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala b/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala index 690f927e938c3..f1b5aafac4066 100644 --- a/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala +++ b/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala @@ -20,6 +20,7 @@ package org.apache.spark.scheduler.cluster import scala.collection.mutable.ArrayBuffer import org.apache.hadoop.yarn.api.records.{ApplicationId, YarnApplicationState} +import org.apache.hadoop.yarn.exceptions.ApplicationNotFoundException import org.apache.spark.{SparkException, Logging, SparkContext} import org.apache.spark.deploy.yarn.{Client, ClientArguments} @@ -133,8 +134,14 @@ private[spark] class YarnClientSchedulerBackend( val t = new Thread { override def run() { while (!stopping) { - val report = client.getApplicationReport(appId) - val state = report.getYarnApplicationState() + var state: YarnApplicationState = null + try { + val report = client.getApplicationReport(appId) + state = report.getYarnApplicationState() + } catch { + case e: ApplicationNotFoundException => + state = YarnApplicationState.KILLED + } if (state == YarnApplicationState.FINISHED || state == YarnApplicationState.KILLED || state == YarnApplicationState.FAILED) { From 5687bab8fdfdc5345b8c5b9be8d4595299005fc8 Mon Sep 17 00:00:00 2001 From: Marcelo Vanzin Date: Fri, 6 Feb 2015 14:23:09 -0800 Subject: [PATCH 40/53] [SPARK-5600] [core] Clean up FsHistoryProvider test, fix app sort order. Clean up some test setup code to remove duplicate instantiation of the provider. Also make sure unfinished apps are sorted correctly. Author: Marcelo Vanzin Closes #4370 from vanzin/SPARK-5600 and squashes the following commits: 0d048d5 [Marcelo Vanzin] Cleanup test code a bit. 2585119 [Marcelo Vanzin] Review feedback. 8b97544 [Marcelo Vanzin] Merge branch 'master' into SPARK-5600 be979e9 [Marcelo Vanzin] Merge branch 'master' into SPARK-5600 298371c [Marcelo Vanzin] [SPARK-5600] [core] Clean up FsHistoryProvider test, fix app sort order. --- .../deploy/history/FsHistoryProvider.scala | 15 +++- .../history/FsHistoryProviderSuite.scala | 85 +++++++++---------- 2 files changed, 52 insertions(+), 48 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala b/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala index 92125f2df7d10..868c63d30a202 100644 --- a/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala +++ b/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala @@ -194,7 +194,7 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis None } } - .sortBy { info => (-info.endTime, -info.startTime) } + .sortWith(compareAppInfo) lastModifiedTime = newLastModifiedTime @@ -214,7 +214,7 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis val newIterator = logInfos.iterator.buffered val oldIterator = applications.values.iterator.buffered while (newIterator.hasNext && oldIterator.hasNext) { - if (newIterator.head.endTime > oldIterator.head.endTime) { + if (compareAppInfo(newIterator.head, oldIterator.head)) { addIfAbsent(newIterator.next) } else { addIfAbsent(oldIterator.next) @@ -230,6 +230,17 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis } } + /** + * Comparison function that defines the sort order for the application listing. + * + * @return Whether `i1` should precede `i2`. + */ + private def compareAppInfo( + i1: FsApplicationHistoryInfo, + i2: FsApplicationHistoryInfo): Boolean = { + if (i1.endTime != i2.endTime) i1.endTime >= i2.endTime else i1.startTime >= i2.startTime + } + /** * Replays the events in the specified log file and returns information about the associated * application. diff --git a/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala b/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala index 1d95432258111..85939eaadccc7 100644 --- a/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala +++ b/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala @@ -37,13 +37,8 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers private var testDir: File = null - private var provider: FsHistoryProvider = null - before { testDir = Utils.createTempDir() - provider = new FsHistoryProvider(new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - .set("spark.history.fs.updateInterval", "0")) } after { @@ -51,40 +46,41 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers } test("Parse new and old application logs") { - val conf = new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - .set("spark.history.fs.updateInterval", "0") - val provider = new FsHistoryProvider(conf) + val provider = new FsHistoryProvider(createTestConf()) // Write a new-style application log. - val logFile1 = new File(testDir, "new1") - writeFile(logFile1, true, None, - SparkListenerApplicationStart("app1-1", None, 1L, "test"), - SparkListenerApplicationEnd(2L) + val newAppComplete = new File(testDir, "new1") + writeFile(newAppComplete, true, None, + SparkListenerApplicationStart("new-app-complete", None, 1L, "test"), + SparkListenerApplicationEnd(4L) ) // Write an unfinished app, new-style. - val logFile2 = new File(testDir, "new2" + EventLoggingListener.IN_PROGRESS) - writeFile(logFile2, true, None, - SparkListenerApplicationStart("app2-2", None, 1L, "test") + val newAppIncomplete = new File(testDir, "new2" + EventLoggingListener.IN_PROGRESS) + writeFile(newAppIncomplete, true, None, + SparkListenerApplicationStart("new-app-incomplete", None, 1L, "test") ) // Write an old-style application log. - val oldLog = new File(testDir, "old1") - oldLog.mkdir() - createEmptyFile(new File(oldLog, provider.SPARK_VERSION_PREFIX + "1.0")) - writeFile(new File(oldLog, provider.LOG_PREFIX + "1"), false, None, - SparkListenerApplicationStart("app3", None, 2L, "test"), + val oldAppComplete = new File(testDir, "old1") + oldAppComplete.mkdir() + createEmptyFile(new File(oldAppComplete, provider.SPARK_VERSION_PREFIX + "1.0")) + writeFile(new File(oldAppComplete, provider.LOG_PREFIX + "1"), false, None, + SparkListenerApplicationStart("old-app-complete", None, 2L, "test"), SparkListenerApplicationEnd(3L) ) - createEmptyFile(new File(oldLog, provider.APPLICATION_COMPLETE)) + createEmptyFile(new File(oldAppComplete, provider.APPLICATION_COMPLETE)) + + // Check for logs so that we force the older unfinished app to be loaded, to make + // sure unfinished apps are also sorted correctly. + provider.checkForLogs() // Write an unfinished app, old-style. - val oldLog2 = new File(testDir, "old2") - oldLog2.mkdir() - createEmptyFile(new File(oldLog2, provider.SPARK_VERSION_PREFIX + "1.0")) - writeFile(new File(oldLog2, provider.LOG_PREFIX + "1"), false, None, - SparkListenerApplicationStart("app4", None, 2L, "test") + val oldAppIncomplete = new File(testDir, "old2") + oldAppIncomplete.mkdir() + createEmptyFile(new File(oldAppIncomplete, provider.SPARK_VERSION_PREFIX + "1.0")) + writeFile(new File(oldAppIncomplete, provider.LOG_PREFIX + "1"), false, None, + SparkListenerApplicationStart("old-app-incomplete", None, 2L, "test") ) // Force a reload of data from the log directory, and check that both logs are loaded. @@ -96,14 +92,14 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers list.size should be (4) list.count(e => e.completed) should be (2) - list(0) should be (ApplicationHistoryInfo(oldLog.getName(), "app3", 2L, 3L, - oldLog.lastModified(), "test", true)) - list(1) should be (ApplicationHistoryInfo(logFile1.getName(), "app1-1", 1L, 2L, - logFile1.lastModified(), "test", true)) - list(2) should be (ApplicationHistoryInfo(oldLog2.getName(), "app4", 2L, -1L, - oldLog2.lastModified(), "test", false)) - list(3) should be (ApplicationHistoryInfo(logFile2.getName(), "app2-2", 1L, -1L, - logFile2.lastModified(), "test", false)) + list(0) should be (ApplicationHistoryInfo(newAppComplete.getName(), "new-app-complete", 1L, 4L, + newAppComplete.lastModified(), "test", true)) + list(1) should be (ApplicationHistoryInfo(oldAppComplete.getName(), "old-app-complete", 2L, 3L, + oldAppComplete.lastModified(), "test", true)) + list(2) should be (ApplicationHistoryInfo(oldAppIncomplete.getName(), "old-app-incomplete", 2L, + -1L, oldAppIncomplete.lastModified(), "test", false)) + list(3) should be (ApplicationHistoryInfo(newAppIncomplete.getName(), "new-app-incomplete", 1L, + -1L, newAppIncomplete.lastModified(), "test", false)) // Make sure the UI can be rendered. list.foreach { case info => @@ -113,6 +109,7 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers } test("Parse legacy logs with compression codec set") { + val provider = new FsHistoryProvider(createTestConf()) val testCodecs = List((classOf[LZFCompressionCodec].getName(), true), (classOf[SnappyCompressionCodec].getName(), true), ("invalid.codec", false)) @@ -156,10 +153,7 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers ) logFile2.setReadable(false, false) - val conf = new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - .set("spark.history.fs.updateInterval", "0") - val provider = new FsHistoryProvider(conf) + val provider = new FsHistoryProvider(createTestConf()) provider.checkForLogs() val list = provider.getListing().toSeq @@ -168,10 +162,7 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers } test("history file is renamed from inprogress to completed") { - val conf = new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - .set("spark.testing", "true") - val provider = new FsHistoryProvider(conf) + val provider = new FsHistoryProvider(createTestConf()) val logFile1 = new File(testDir, "app1" + EventLoggingListener.IN_PROGRESS) writeFile(logFile1, true, None, @@ -191,9 +182,7 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers } test("SPARK-5582: empty log directory") { - val conf = new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - val provider = new FsHistoryProvider(conf) + val provider = new FsHistoryProvider(createTestConf()) val logFile1 = new File(testDir, "app1" + EventLoggingListener.IN_PROGRESS) writeFile(logFile1, true, None, @@ -229,4 +218,8 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers new FileOutputStream(file).close() } + private def createTestConf(): SparkConf = { + new SparkConf().set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) + } + } From 57961567ef104efb3174e67d762c5d9d6263b800 Mon Sep 17 00:00:00 2001 From: Michael Armbrust Date: Fri, 6 Feb 2015 14:27:06 -0800 Subject: [PATCH 41/53] [HOTFIX] Fix the maven build after adding sqlContext to spark-shell Follow up to #4387 to fix the build break. Author: Michael Armbrust Closes #4443 from marmbrus/fixMaven and squashes the following commits: 1eeba7d [Michael Armbrust] try again 7f5fb15 [Michael Armbrust] [HOTFIX] Fix the maven build after adding sqlContext to spark-shell --- repl/pom.xml | 1 - 1 file changed, 1 deletion(-) diff --git a/repl/pom.xml b/repl/pom.xml index bd39b90fd8714..3d4adf8fd5b03 100644 --- a/repl/pom.xml +++ b/repl/pom.xml @@ -66,7 +66,6 @@ org.apache.spark spark-sql_${scala.binary.version} ${project.version} - test org.scala-lang From dcd1e42d6b6ac08d2c0736bf61a15f515a1f222b Mon Sep 17 00:00:00 2001 From: Kostas Sakellis Date: Fri, 6 Feb 2015 14:31:20 -0800 Subject: [PATCH 42/53] [SPARK-4874] [CORE] Collect record count metrics Collects record counts for both Input/Output and Shuffle Metrics. For the input/output metrics, it just appends the counter every time the iterators get accessed. For shuffle on the write side, we count the metrics post aggregation (after a map side combine) and on the read side we count the metrics pre aggregation. This allows both the bytes read/written metrics and the records read/written to line up. For backwards compatibility, if we deserialize an older event that doesn't have record metrics, we set the metric to -1. Author: Kostas Sakellis Closes #4067 from ksakellis/kostas-spark-4874 and squashes the following commits: bd919be [Kostas Sakellis] Changed 'Records Read' in shuffleReadMetrics json output to 'Total Records Read' dad4d57 [Kostas Sakellis] Add a comment and check to BlockObjectWriter so that it cannot be reopend. 6f236a1 [Kostas Sakellis] Renamed _recordsWritten in ShuffleWriteMetrics to be more consistent 70620a0 [Kostas Sakellis] CR Feedback 17faa3a [Kostas Sakellis] Removed AtomicLong in favour of using Long b6f9923 [Kostas Sakellis] Merge AfterNextInterceptingIterator with InterruptableIterator to save a function call 46c8186 [Kostas Sakellis] Combined Bytes and # records into one column 57551c1 [Kostas Sakellis] Conforms to SPARK-3288 6cdb44e [Kostas Sakellis] Removed the generic InterceptingIterator and repalced it with specific implementation 1aa273c [Kostas Sakellis] CR Feedback 1bb78b1 [Kostas Sakellis] [SPARK-4874] [CORE] Collect record count metrics --- .../scala/org/apache/spark/CacheManager.scala | 11 +- .../apache/spark/executor/TaskMetrics.scala | 54 +++-- .../org/apache/spark/rdd/HadoopRDD.scala | 6 +- .../org/apache/spark/rdd/NewHadoopRDD.scala | 6 +- .../apache/spark/rdd/PairRDDFunctions.scala | 12 +- .../hash/BlockStoreShuffleFetcher.scala | 8 +- .../apache/spark/storage/BlockManager.scala | 2 +- .../spark/storage/BlockObjectWriter.scala | 24 ++- .../scala/org/apache/spark/ui/ToolTips.scala | 9 +- .../apache/spark/ui/exec/ExecutorsTab.scala | 6 + .../apache/spark/ui/jobs/ExecutorTable.scala | 86 ++++++-- .../spark/ui/jobs/JobProgressListener.scala | 24 +++ .../org/apache/spark/ui/jobs/StagePage.scala | 148 +++++++++----- .../org/apache/spark/ui/jobs/UIData.scala | 14 ++ .../org/apache/spark/util/JsonProtocol.scala | 19 +- .../util/collection/ExternalSorter.scala | 1 + .../metrics/InputOutputMetricsSuite.scala | 186 ++++++++++++++++-- .../storage/BlockObjectWriterSuite.scala | 21 ++ .../ui/jobs/JobProgressListenerSuite.scala | 2 +- .../apache/spark/util/JsonProtocolSuite.scala | 55 +++++- 20 files changed, 548 insertions(+), 146 deletions(-) diff --git a/core/src/main/scala/org/apache/spark/CacheManager.scala b/core/src/main/scala/org/apache/spark/CacheManager.scala index a0c0372b7f0ef..a96d754744a05 100644 --- a/core/src/main/scala/org/apache/spark/CacheManager.scala +++ b/core/src/main/scala/org/apache/spark/CacheManager.scala @@ -47,10 +47,15 @@ private[spark] class CacheManager(blockManager: BlockManager) extends Logging { val inputMetrics = blockResult.inputMetrics val existingMetrics = context.taskMetrics .getInputMetricsForReadMethod(inputMetrics.readMethod) - existingMetrics.addBytesRead(inputMetrics.bytesRead) - - new InterruptibleIterator(context, blockResult.data.asInstanceOf[Iterator[T]]) + existingMetrics.incBytesRead(inputMetrics.bytesRead) + val iter = blockResult.data.asInstanceOf[Iterator[T]] + new InterruptibleIterator[T](context, iter) { + override def next(): T = { + existingMetrics.incRecordsRead(1) + delegate.next() + } + } case None => // Acquire a lock for loading this partition // If another thread already holds the lock, wait for it to finish return its results diff --git a/core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala b/core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala index 97912c68c5982..d05659193b334 100644 --- a/core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala +++ b/core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala @@ -194,18 +194,19 @@ class TaskMetrics extends Serializable { /** * Aggregates shuffle read metrics for all registered dependencies into shuffleReadMetrics. */ - private[spark] def updateShuffleReadMetrics() = synchronized { + private[spark] def updateShuffleReadMetrics(): Unit = synchronized { val merged = new ShuffleReadMetrics() for (depMetrics <- depsShuffleReadMetrics) { merged.incFetchWaitTime(depMetrics.fetchWaitTime) merged.incLocalBlocksFetched(depMetrics.localBlocksFetched) merged.incRemoteBlocksFetched(depMetrics.remoteBlocksFetched) merged.incRemoteBytesRead(depMetrics.remoteBytesRead) + merged.incRecordsRead(depMetrics.recordsRead) } _shuffleReadMetrics = Some(merged) } - private[spark] def updateInputMetrics() = synchronized { + private[spark] def updateInputMetrics(): Unit = synchronized { inputMetrics.foreach(_.updateBytesRead()) } } @@ -242,27 +243,31 @@ object DataWriteMethod extends Enumeration with Serializable { @DeveloperApi case class InputMetrics(readMethod: DataReadMethod.Value) { - private val _bytesRead: AtomicLong = new AtomicLong() + /** + * This is volatile so that it is visible to the updater thread. + */ + @volatile @transient var bytesReadCallback: Option[() => Long] = None /** * Total bytes read. */ - def bytesRead: Long = _bytesRead.get() - @volatile @transient var bytesReadCallback: Option[() => Long] = None + private var _bytesRead: Long = _ + def bytesRead: Long = _bytesRead + def incBytesRead(bytes: Long) = _bytesRead += bytes /** - * Adds additional bytes read for this read method. + * Total records read. */ - def addBytesRead(bytes: Long) = { - _bytesRead.addAndGet(bytes) - } + private var _recordsRead: Long = _ + def recordsRead: Long = _recordsRead + def incRecordsRead(records: Long) = _recordsRead += records /** * Invoke the bytesReadCallback and mutate bytesRead. */ def updateBytesRead() { bytesReadCallback.foreach { c => - _bytesRead.set(c()) + _bytesRead = c() } } @@ -287,6 +292,13 @@ case class OutputMetrics(writeMethod: DataWriteMethod.Value) { private var _bytesWritten: Long = _ def bytesWritten = _bytesWritten private[spark] def setBytesWritten(value : Long) = _bytesWritten = value + + /** + * Total records written + */ + private var _recordsWritten: Long = 0L + def recordsWritten = _recordsWritten + private[spark] def setRecordsWritten(value: Long) = _recordsWritten = value } /** @@ -301,7 +313,7 @@ class ShuffleReadMetrics extends Serializable { private var _remoteBlocksFetched: Int = _ def remoteBlocksFetched = _remoteBlocksFetched private[spark] def incRemoteBlocksFetched(value: Int) = _remoteBlocksFetched += value - private[spark] def defRemoteBlocksFetched(value: Int) = _remoteBlocksFetched -= value + private[spark] def decRemoteBlocksFetched(value: Int) = _remoteBlocksFetched -= value /** * Number of local blocks fetched in this shuffle by this task @@ -309,8 +321,7 @@ class ShuffleReadMetrics extends Serializable { private var _localBlocksFetched: Int = _ def localBlocksFetched = _localBlocksFetched private[spark] def incLocalBlocksFetched(value: Int) = _localBlocksFetched += value - private[spark] def defLocalBlocksFetched(value: Int) = _localBlocksFetched -= value - + private[spark] def decLocalBlocksFetched(value: Int) = _localBlocksFetched -= value /** * Time the task spent waiting for remote shuffle blocks. This only includes the time @@ -334,6 +345,14 @@ class ShuffleReadMetrics extends Serializable { * Number of blocks fetched in this shuffle by this task (remote or local) */ def totalBlocksFetched = _remoteBlocksFetched + _localBlocksFetched + + /** + * Total number of records read from the shuffle by this task + */ + private var _recordsRead: Long = _ + def recordsRead = _recordsRead + private[spark] def incRecordsRead(value: Long) = _recordsRead += value + private[spark] def decRecordsRead(value: Long) = _recordsRead -= value } /** @@ -358,5 +377,12 @@ class ShuffleWriteMetrics extends Serializable { private[spark] def incShuffleWriteTime(value: Long) = _shuffleWriteTime += value private[spark] def decShuffleWriteTime(value: Long) = _shuffleWriteTime -= value - + /** + * Total number of records written to the shuffle by this task + */ + @volatile private var _shuffleRecordsWritten: Long = _ + def shuffleRecordsWritten = _shuffleRecordsWritten + private[spark] def incShuffleRecordsWritten(value: Long) = _shuffleRecordsWritten += value + private[spark] def decShuffleRecordsWritten(value: Long) = _shuffleRecordsWritten -= value + private[spark] def setShuffleRecordsWritten(value: Long) = _shuffleRecordsWritten = value } diff --git a/core/src/main/scala/org/apache/spark/rdd/HadoopRDD.scala b/core/src/main/scala/org/apache/spark/rdd/HadoopRDD.scala index 89adddcf0ac36..486e86ce1bb19 100644 --- a/core/src/main/scala/org/apache/spark/rdd/HadoopRDD.scala +++ b/core/src/main/scala/org/apache/spark/rdd/HadoopRDD.scala @@ -247,7 +247,9 @@ class HadoopRDD[K, V]( case eof: EOFException => finished = true } - + if (!finished) { + inputMetrics.incRecordsRead(1) + } (key, value) } @@ -261,7 +263,7 @@ class HadoopRDD[K, V]( // If we can't get the bytes read from the FS stats, fall back to the split size, // which may be inaccurate. try { - inputMetrics.addBytesRead(split.inputSplit.value.getLength) + inputMetrics.incBytesRead(split.inputSplit.value.getLength) } catch { case e: java.io.IOException => logWarning("Unable to get input size to set InputMetrics for task", e) diff --git a/core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala b/core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala index 44b9ffd2a53fd..7fb94840df99c 100644 --- a/core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala +++ b/core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala @@ -151,7 +151,9 @@ class NewHadoopRDD[K, V]( throw new java.util.NoSuchElementException("End of stream") } havePair = false - + if (!finished) { + inputMetrics.incRecordsRead(1) + } (reader.getCurrentKey, reader.getCurrentValue) } @@ -165,7 +167,7 @@ class NewHadoopRDD[K, V]( // If we can't get the bytes read from the FS stats, fall back to the split size, // which may be inaccurate. try { - inputMetrics.addBytesRead(split.serializableHadoopSplit.value.getLength) + inputMetrics.incBytesRead(split.serializableHadoopSplit.value.getLength) } catch { case e: java.io.IOException => logWarning("Unable to get input size to set InputMetrics for task", e) diff --git a/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala b/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala index 49b88a90ab5af..955b42c3baaa1 100644 --- a/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala +++ b/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala @@ -34,7 +34,7 @@ import org.apache.hadoop.io.SequenceFile.CompressionType import org.apache.hadoop.io.compress.CompressionCodec import org.apache.hadoop.mapred.{FileOutputCommitter, FileOutputFormat, JobConf, OutputFormat} import org.apache.hadoop.mapreduce.{Job => NewAPIHadoopJob, OutputFormat => NewOutputFormat, -RecordWriter => NewRecordWriter} + RecordWriter => NewRecordWriter} import org.apache.spark._ import org.apache.spark.Partitioner.defaultPartitioner @@ -993,8 +993,8 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) val (outputMetrics, bytesWrittenCallback) = initHadoopOutputMetrics(context) val writer = format.getRecordWriter(hadoopContext).asInstanceOf[NewRecordWriter[K,V]] + var recordsWritten = 0L try { - var recordsWritten = 0L while (iter.hasNext) { val pair = iter.next() writer.write(pair._1, pair._2) @@ -1008,6 +1008,7 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) } committer.commitTask(hadoopContext) bytesWrittenCallback.foreach { fn => outputMetrics.setBytesWritten(fn()) } + outputMetrics.setRecordsWritten(recordsWritten) 1 } : Int @@ -1065,8 +1066,8 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) writer.setup(context.stageId, context.partitionId, taskAttemptId) writer.open() + var recordsWritten = 0L try { - var recordsWritten = 0L while (iter.hasNext) { val record = iter.next() writer.write(record._1.asInstanceOf[AnyRef], record._2.asInstanceOf[AnyRef]) @@ -1080,6 +1081,7 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) } writer.commit() bytesWrittenCallback.foreach { fn => outputMetrics.setBytesWritten(fn()) } + outputMetrics.setRecordsWritten(recordsWritten) } self.context.runJob(self, writeToFile) @@ -1097,9 +1099,9 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) private def maybeUpdateOutputMetrics(bytesWrittenCallback: Option[() => Long], outputMetrics: OutputMetrics, recordsWritten: Long): Unit = { - if (recordsWritten % PairRDDFunctions.RECORDS_BETWEEN_BYTES_WRITTEN_METRIC_UPDATES == 0 - && bytesWrittenCallback.isDefined) { + if (recordsWritten % PairRDDFunctions.RECORDS_BETWEEN_BYTES_WRITTEN_METRIC_UPDATES == 0) { bytesWrittenCallback.foreach { fn => outputMetrics.setBytesWritten(fn()) } + outputMetrics.setRecordsWritten(recordsWritten) } } diff --git a/core/src/main/scala/org/apache/spark/shuffle/hash/BlockStoreShuffleFetcher.scala b/core/src/main/scala/org/apache/spark/shuffle/hash/BlockStoreShuffleFetcher.scala index e3e7434df45b0..7a2c5ae32d98b 100644 --- a/core/src/main/scala/org/apache/spark/shuffle/hash/BlockStoreShuffleFetcher.scala +++ b/core/src/main/scala/org/apache/spark/shuffle/hash/BlockStoreShuffleFetcher.scala @@ -86,6 +86,12 @@ private[hash] object BlockStoreShuffleFetcher extends Logging { context.taskMetrics.updateShuffleReadMetrics() }) - new InterruptibleIterator[T](context, completionIter) + new InterruptibleIterator[T](context, completionIter) { + val readMetrics = context.taskMetrics.createShuffleReadMetricsForDependency() + override def next(): T = { + readMetrics.incRecordsRead(1) + delegate.next() + } + } } } diff --git a/core/src/main/scala/org/apache/spark/storage/BlockManager.scala b/core/src/main/scala/org/apache/spark/storage/BlockManager.scala index 8bc5a1cd18b64..86dbd89f0ffb8 100644 --- a/core/src/main/scala/org/apache/spark/storage/BlockManager.scala +++ b/core/src/main/scala/org/apache/spark/storage/BlockManager.scala @@ -53,7 +53,7 @@ private[spark] class BlockResult( readMethod: DataReadMethod.Value, bytes: Long) { val inputMetrics = new InputMetrics(readMethod) - inputMetrics.addBytesRead(bytes) + inputMetrics.incBytesRead(bytes) } /** diff --git a/core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala b/core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala index 3198d766fca37..81164178b9e8e 100644 --- a/core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala +++ b/core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala @@ -29,7 +29,8 @@ import org.apache.spark.executor.ShuffleWriteMetrics * appending data to an existing block, and can guarantee atomicity in the case of faults * as it allows the caller to revert partial writes. * - * This interface does not support concurrent writes. + * This interface does not support concurrent writes. Also, once the writer has + * been opened, it cannot be reopened again. */ private[spark] abstract class BlockObjectWriter(val blockId: BlockId) { @@ -95,6 +96,7 @@ private[spark] class DiskBlockObjectWriter( private var ts: TimeTrackingOutputStream = null private var objOut: SerializationStream = null private var initialized = false + private var hasBeenClosed = false /** * Cursors used to represent positions in the file. @@ -115,11 +117,16 @@ private[spark] class DiskBlockObjectWriter( private var finalPosition: Long = -1 private var reportedPosition = initialPosition - /** Calling channel.position() to update the write metrics can be a little bit expensive, so we - * only call it every N writes */ - private var writesSinceMetricsUpdate = 0 + /** + * Keep track of number of records written and also use this to periodically + * output bytes written since the latter is expensive to do for each record. + */ + private var numRecordsWritten = 0 override def open(): BlockObjectWriter = { + if (hasBeenClosed) { + throw new IllegalStateException("Writer already closed. Cannot be reopened.") + } fos = new FileOutputStream(file, true) ts = new TimeTrackingOutputStream(fos) channel = fos.getChannel() @@ -145,6 +152,7 @@ private[spark] class DiskBlockObjectWriter( ts = null objOut = null initialized = false + hasBeenClosed = true } } @@ -168,6 +176,7 @@ private[spark] class DiskBlockObjectWriter( override def revertPartialWritesAndClose() { try { writeMetrics.decShuffleBytesWritten(reportedPosition - initialPosition) + writeMetrics.decShuffleRecordsWritten(numRecordsWritten) if (initialized) { objOut.flush() @@ -193,12 +202,11 @@ private[spark] class DiskBlockObjectWriter( } objOut.writeObject(value) + numRecordsWritten += 1 + writeMetrics.incShuffleRecordsWritten(1) - if (writesSinceMetricsUpdate == 32) { - writesSinceMetricsUpdate = 0 + if (numRecordsWritten % 32 == 0) { updateBytesWritten() - } else { - writesSinceMetricsUpdate += 1 } } diff --git a/core/src/main/scala/org/apache/spark/ui/ToolTips.scala b/core/src/main/scala/org/apache/spark/ui/ToolTips.scala index 4307029d44fbb..3a15e603b1969 100644 --- a/core/src/main/scala/org/apache/spark/ui/ToolTips.scala +++ b/core/src/main/scala/org/apache/spark/ui/ToolTips.scala @@ -29,14 +29,15 @@ private[spark] object ToolTips { val SHUFFLE_READ_BLOCKED_TIME = "Time that the task spent blocked waiting for shuffle data to be read from remote machines." - val INPUT = "Bytes read from Hadoop or from Spark storage." + val INPUT = "Bytes and records read from Hadoop or from Spark storage." - val OUTPUT = "Bytes written to Hadoop." + val OUTPUT = "Bytes and records written to Hadoop." - val SHUFFLE_WRITE = "Bytes written to disk in order to be read by a shuffle in a future stage." + val SHUFFLE_WRITE = + "Bytes and records written to disk in order to be read by a shuffle in a future stage." val SHUFFLE_READ = - """Bytes read from remote executors. Typically less than shuffle write bytes + """Bytes and records read from remote executors. Typically less than shuffle write bytes because this does not include shuffle data read locally.""" val GETTING_RESULT_TIME = diff --git a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala index a38cb75fdd8c6..3afd7ef07d7c9 100644 --- a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala +++ b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala @@ -48,7 +48,9 @@ class ExecutorsListener(storageStatusListener: StorageStatusListener) extends Sp val executorToTasksFailed = HashMap[String, Int]() val executorToDuration = HashMap[String, Long]() val executorToInputBytes = HashMap[String, Long]() + val executorToInputRecords = HashMap[String, Long]() val executorToOutputBytes = HashMap[String, Long]() + val executorToOutputRecords = HashMap[String, Long]() val executorToShuffleRead = HashMap[String, Long]() val executorToShuffleWrite = HashMap[String, Long]() val executorToLogUrls = HashMap[String, Map[String, String]]() @@ -84,10 +86,14 @@ class ExecutorsListener(storageStatusListener: StorageStatusListener) extends Sp metrics.inputMetrics.foreach { inputMetrics => executorToInputBytes(eid) = executorToInputBytes.getOrElse(eid, 0L) + inputMetrics.bytesRead + executorToInputRecords(eid) = + executorToInputRecords.getOrElse(eid, 0L) + inputMetrics.recordsRead } metrics.outputMetrics.foreach { outputMetrics => executorToOutputBytes(eid) = executorToOutputBytes.getOrElse(eid, 0L) + outputMetrics.bytesWritten + executorToOutputRecords(eid) = + executorToOutputRecords.getOrElse(eid, 0L) + outputMetrics.recordsWritten } metrics.shuffleReadMetrics.foreach { shuffleRead => executorToShuffleRead(eid) = diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala b/core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala index 9836d11a6d85f..1f8536d1b7195 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala @@ -36,6 +36,20 @@ private[ui] class ExecutorTable(stageId: Int, stageAttemptId: Int, parent: Stage /** Special table which merges two header cells. */ private def executorTable[T](): Seq[Node] = { + val stageData = listener.stageIdToData.get((stageId, stageAttemptId)) + var hasInput = false + var hasOutput = false + var hasShuffleWrite = false + var hasShuffleRead = false + var hasBytesSpilled = false + stageData.foreach(data => { + hasInput = data.hasInput + hasOutput = data.hasOutput + hasShuffleRead = data.hasShuffleRead + hasShuffleWrite = data.hasShuffleWrite + hasBytesSpilled = data.hasBytesSpilled + }) + @@ -44,12 +58,32 @@ private[ui] class ExecutorTable(stageId: Int, stageAttemptId: Int, parent: Stage - - - - - - + {if (hasInput) { + + }} + {if (hasOutput) { + + }} + {if (hasShuffleRead) { + + }} + {if (hasShuffleWrite) { + + }} + {if (hasBytesSpilled) { + + + }} {createExecutorTable()} @@ -76,18 +110,34 @@ private[ui] class ExecutorTable(stageId: Int, stageAttemptId: Int, parent: Stage - - - - - - + {if (stageData.hasInput) { + + }} + {if (stageData.hasOutput) { + + }} + {if (stageData.hasShuffleRead) { + + }} + {if (stageData.hasShuffleWrite) { + + }} + {if (stageData.hasBytesSpilled) { + + + }} } case None => diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala b/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala index 4d200eeda86b9..f463f8d7c7215 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala @@ -394,24 +394,48 @@ class JobProgressListener(conf: SparkConf) extends SparkListener with Logging { stageData.shuffleWriteBytes += shuffleWriteDelta execSummary.shuffleWrite += shuffleWriteDelta + val shuffleWriteRecordsDelta = + (taskMetrics.shuffleWriteMetrics.map(_.shuffleRecordsWritten).getOrElse(0L) + - oldMetrics.flatMap(_.shuffleWriteMetrics).map(_.shuffleRecordsWritten).getOrElse(0L)) + stageData.shuffleWriteRecords += shuffleWriteRecordsDelta + execSummary.shuffleWriteRecords += shuffleWriteRecordsDelta + val shuffleReadDelta = (taskMetrics.shuffleReadMetrics.map(_.remoteBytesRead).getOrElse(0L) - oldMetrics.flatMap(_.shuffleReadMetrics).map(_.remoteBytesRead).getOrElse(0L)) stageData.shuffleReadBytes += shuffleReadDelta execSummary.shuffleRead += shuffleReadDelta + val shuffleReadRecordsDelta = + (taskMetrics.shuffleReadMetrics.map(_.recordsRead).getOrElse(0L) + - oldMetrics.flatMap(_.shuffleReadMetrics).map(_.recordsRead).getOrElse(0L)) + stageData.shuffleReadRecords += shuffleReadRecordsDelta + execSummary.shuffleReadRecords += shuffleReadRecordsDelta + val inputBytesDelta = (taskMetrics.inputMetrics.map(_.bytesRead).getOrElse(0L) - oldMetrics.flatMap(_.inputMetrics).map(_.bytesRead).getOrElse(0L)) stageData.inputBytes += inputBytesDelta execSummary.inputBytes += inputBytesDelta + val inputRecordsDelta = + (taskMetrics.inputMetrics.map(_.recordsRead).getOrElse(0L) + - oldMetrics.flatMap(_.inputMetrics).map(_.recordsRead).getOrElse(0L)) + stageData.inputRecords += inputRecordsDelta + execSummary.inputRecords += inputRecordsDelta + val outputBytesDelta = (taskMetrics.outputMetrics.map(_.bytesWritten).getOrElse(0L) - oldMetrics.flatMap(_.outputMetrics).map(_.bytesWritten).getOrElse(0L)) stageData.outputBytes += outputBytesDelta execSummary.outputBytes += outputBytesDelta + val outputRecordsDelta = + (taskMetrics.outputMetrics.map(_.recordsWritten).getOrElse(0L) + - oldMetrics.flatMap(_.outputMetrics).map(_.recordsWritten).getOrElse(0L)) + stageData.outputRecords += outputRecordsDelta + execSummary.outputRecords += outputRecordsDelta + val diskSpillDelta = taskMetrics.diskBytesSpilled - oldMetrics.map(_.diskBytesSpilled).getOrElse(0L) stageData.diskBytesSpilled += diskSpillDelta diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala b/core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala index d8be1b20b3acd..02a3cc3e43c25 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala @@ -56,11 +56,6 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { val numCompleted = tasks.count(_.taskInfo.finished) val accumulables = listener.stageIdToData((stageId, stageAttemptId)).accumulables val hasAccumulators = accumulables.size > 0 - val hasInput = stageData.inputBytes > 0 - val hasOutput = stageData.outputBytes > 0 - val hasShuffleRead = stageData.shuffleReadBytes > 0 - val hasShuffleWrite = stageData.shuffleWriteBytes > 0 - val hasBytesSpilled = stageData.memoryBytesSpilled > 0 && stageData.diskBytesSpilled > 0 val summary =
@@ -69,31 +64,33 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { Total task time across all tasks: {UIUtils.formatDuration(stageData.executorRunTime)} - {if (hasInput) { + {if (stageData.hasInput) {
  • - Input: - {Utils.bytesToString(stageData.inputBytes)} + Input Size / Records: + {s"${Utils.bytesToString(stageData.inputBytes)} / ${stageData.inputRecords}"}
    • }} - {if (hasOutput) { + {if (stageData.hasOutput) {
  • Output: - {Utils.bytesToString(stageData.outputBytes)} + {s"${Utils.bytesToString(stageData.outputBytes)} / ${stageData.outputRecords}"}
    • }} - {if (hasShuffleRead) { + {if (stageData.hasShuffleRead) {
  • Shuffle read: - {Utils.bytesToString(stageData.shuffleReadBytes)} + {s"${Utils.bytesToString(stageData.shuffleReadBytes)} / " + + s"${stageData.shuffleReadRecords}"}
    • }} - {if (hasShuffleWrite) { + {if (stageData.hasShuffleWrite) {
  • Shuffle write: - {Utils.bytesToString(stageData.shuffleWriteBytes)} + {s"${Utils.bytesToString(stageData.shuffleWriteBytes)} / " + + s"${stageData.shuffleWriteRecords}"}
    • }} - {if (hasBytesSpilled) { + {if (stageData.hasBytesSpilled) {
  • Shuffle spill (memory): {Utils.bytesToString(stageData.memoryBytesSpilled)} @@ -132,7 +129,7 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { Task Deserialization Time
    • - {if (hasShuffleRead) { + {if (stageData.hasShuffleRead) {
  • @@ -174,25 +171,32 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { ("Result Serialization Time", TaskDetailsClassNames.RESULT_SERIALIZATION_TIME), ("Getting Result Time", TaskDetailsClassNames.GETTING_RESULT_TIME)) ++ {if (hasAccumulators) Seq(("Accumulators", "")) else Nil} ++ - {if (hasInput) Seq(("Input", "")) else Nil} ++ - {if (hasOutput) Seq(("Output", "")) else Nil} ++ - {if (hasShuffleRead) { + {if (stageData.hasInput) Seq(("Input Size / Records", "")) else Nil} ++ + {if (stageData.hasOutput) Seq(("Output Size / Records", "")) else Nil} ++ + {if (stageData.hasShuffleRead) { Seq(("Shuffle Read Blocked Time", TaskDetailsClassNames.SHUFFLE_READ_BLOCKED_TIME), - ("Shuffle Read", "")) + ("Shuffle Read Size / Records", "")) + } else { + Nil + }} ++ + {if (stageData.hasShuffleWrite) { + Seq(("Write Time", ""), ("Shuffle Write Size / Records", "")) + } else { + Nil + }} ++ + {if (stageData.hasBytesSpilled) { + Seq(("Shuffle Spill (Memory)", ""), ("Shuffle Spill (Disk)", "")) } else { Nil }} ++ - {if (hasShuffleWrite) Seq(("Write Time", ""), ("Shuffle Write", "")) else Nil} ++ - {if (hasBytesSpilled) Seq(("Shuffle Spill (Memory)", ""), ("Shuffle Spill (Disk)", "")) - else Nil} ++ Seq(("Errors", "")) val unzipped = taskHeadersAndCssClasses.unzip val taskTable = UIUtils.listingTable( unzipped._1, - taskRow(hasAccumulators, hasInput, hasOutput, hasShuffleRead, hasShuffleWrite, - hasBytesSpilled), + taskRow(hasAccumulators, stageData.hasInput, stageData.hasOutput, + stageData.hasShuffleRead, stageData.hasShuffleWrite, stageData.hasBytesSpilled), tasks, headerClasses = unzipped._2) // Excludes tasks which failed and have incomplete metrics @@ -203,8 +207,11 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { None } else { + def getDistributionQuantiles(data: Seq[Double]): IndexedSeq[Double] = + Distribution(data).get.getQuantiles() + def getFormattedTimeQuantiles(times: Seq[Double]): Seq[Node] = { - Distribution(times).get.getQuantiles().map { millis => + getDistributionQuantiles(times).map { millis =>
    • } } @@ -273,17 +280,36 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { getFormattedTimeQuantiles(schedulerDelays) def getFormattedSizeQuantiles(data: Seq[Double]) = - Distribution(data).get.getQuantiles().map(d => ) + getDistributionQuantiles(data).map(d => ) + + def getFormattedSizeQuantilesWithRecords(data: Seq[Double], records: Seq[Double]) = { + val recordDist = getDistributionQuantiles(records).iterator + getDistributionQuantiles(data).map(d => + + ) + } val inputSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.inputMetrics.map(_.bytesRead).getOrElse(0L).toDouble } - val inputQuantiles = +: getFormattedSizeQuantiles(inputSizes) + + val inputRecords = validTasks.map { case TaskUIData(_, metrics, _) => + metrics.get.inputMetrics.map(_.recordsRead).getOrElse(0L).toDouble + } + + val inputQuantiles = +: + getFormattedSizeQuantilesWithRecords(inputSizes, inputRecords) val outputSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.outputMetrics.map(_.bytesWritten).getOrElse(0L).toDouble } - val outputQuantiles = +: getFormattedSizeQuantiles(outputSizes) + + val outputRecords = validTasks.map { case TaskUIData(_, metrics, _) => + metrics.get.outputMetrics.map(_.recordsWritten).getOrElse(0L).toDouble + } + + val outputQuantiles = +: + getFormattedSizeQuantilesWithRecords(outputSizes, outputRecords) val shuffleReadBlockedTimes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.shuffleReadMetrics.map(_.fetchWaitTime).getOrElse(0L).toDouble @@ -294,14 +320,24 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { val shuffleReadSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.shuffleReadMetrics.map(_.remoteBytesRead).getOrElse(0L).toDouble } - val shuffleReadQuantiles = +: - getFormattedSizeQuantiles(shuffleReadSizes) + + val shuffleReadRecords = validTasks.map { case TaskUIData(_, metrics, _) => + metrics.get.shuffleReadMetrics.map(_.recordsRead).getOrElse(0L).toDouble + } + + val shuffleReadQuantiles = +: + getFormattedSizeQuantilesWithRecords(shuffleReadSizes, shuffleReadRecords) val shuffleWriteSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.shuffleWriteMetrics.map(_.shuffleBytesWritten).getOrElse(0L).toDouble } - val shuffleWriteQuantiles = +: - getFormattedSizeQuantiles(shuffleWriteSizes) + + val shuffleWriteRecords = validTasks.map { case TaskUIData(_, metrics, _) => + metrics.get.shuffleWriteMetrics.map(_.shuffleRecordsWritten).getOrElse(0L).toDouble + } + + val shuffleWriteQuantiles = +: + getFormattedSizeQuantilesWithRecords(shuffleWriteSizes, shuffleWriteRecords) val memoryBytesSpilledSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.memoryBytesSpilled.toDouble @@ -326,9 +362,9 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { {serializationQuantiles} , {gettingResultQuantiles}, - if (hasInput) {inputQuantiles} else Nil, - if (hasOutput) {outputQuantiles} else Nil, - if (hasShuffleRead) { + if (stageData.hasInput) {inputQuantiles} else Nil, + if (stageData.hasOutput) {outputQuantiles} else Nil, + if (stageData.hasShuffleRead) { {shuffleReadBlockedQuantiles} @@ -336,9 +372,9 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { } else { Nil }, - if (hasShuffleWrite) {shuffleWriteQuantiles} else Nil, - if (hasBytesSpilled) {memoryBytesSpilledQuantiles} else Nil, - if (hasBytesSpilled) {diskBytesSpilledQuantiles} else Nil) + if (stageData.hasShuffleWrite) {shuffleWriteQuantiles} else Nil, + if (stageData.hasBytesSpilled) {memoryBytesSpilledQuantiles} else Nil, + if (stageData.hasBytesSpilled) {diskBytesSpilledQuantiles} else Nil) val quantileHeaders = Seq("Metric", "Min", "25th percentile", "Median", "75th percentile", "Max") @@ -397,26 +433,32 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { val inputReadable = maybeInput .map(m => s"${Utils.bytesToString(m.bytesRead)} (${m.readMethod.toString.toLowerCase()})") .getOrElse("") + val inputRecords = maybeInput.map(_.recordsRead.toString).getOrElse("") val maybeOutput = metrics.flatMap(_.outputMetrics) val outputSortable = maybeOutput.map(_.bytesWritten.toString).getOrElse("") val outputReadable = maybeOutput .map(m => s"${Utils.bytesToString(m.bytesWritten)}") .getOrElse("") + val outputRecords = maybeOutput.map(_.recordsWritten.toString).getOrElse("") - val maybeShuffleReadBlockedTime = metrics.flatMap(_.shuffleReadMetrics).map(_.fetchWaitTime) - val shuffleReadBlockedTimeSortable = maybeShuffleReadBlockedTime.map(_.toString).getOrElse("") + val maybeShuffleRead = metrics.flatMap(_.shuffleReadMetrics) + val shuffleReadBlockedTimeSortable = maybeShuffleRead + .map(_.fetchWaitTime.toString).getOrElse("") val shuffleReadBlockedTimeReadable = - maybeShuffleReadBlockedTime.map(ms => UIUtils.formatDuration(ms)).getOrElse("") + maybeShuffleRead.map(ms => UIUtils.formatDuration(ms.fetchWaitTime)).getOrElse("") - val maybeShuffleRead = metrics.flatMap(_.shuffleReadMetrics).map(_.remoteBytesRead) - val shuffleReadSortable = maybeShuffleRead.map(_.toString).getOrElse("") - val shuffleReadReadable = maybeShuffleRead.map(Utils.bytesToString).getOrElse("") + val shuffleReadSortable = maybeShuffleRead.map(_.remoteBytesRead.toString).getOrElse("") + val shuffleReadReadable = maybeShuffleRead + .map(m => s"${Utils.bytesToString(m.remoteBytesRead)}").getOrElse("") + val shuffleReadRecords = maybeShuffleRead.map(_.recordsRead.toString).getOrElse("") - val maybeShuffleWrite = - metrics.flatMap(_.shuffleWriteMetrics).map(_.shuffleBytesWritten) - val shuffleWriteSortable = maybeShuffleWrite.map(_.toString).getOrElse("") - val shuffleWriteReadable = maybeShuffleWrite.map(Utils.bytesToString).getOrElse("") + val maybeShuffleWrite = metrics.flatMap(_.shuffleWriteMetrics) + val shuffleWriteSortable = maybeShuffleWrite.map(_.shuffleBytesWritten.toString).getOrElse("") + val shuffleWriteReadable = maybeShuffleWrite + .map(m => s"${Utils.bytesToString(m.shuffleBytesWritten)}").getOrElse("") + val shuffleWriteRecords = maybeShuffleWrite + .map(_.shuffleRecordsWritten.toString).getOrElse("") val maybeWriteTime = metrics.flatMap(_.shuffleWriteMetrics).map(_.shuffleWriteTime) val writeTimeSortable = maybeWriteTime.map(_.toString).getOrElse("") @@ -472,12 +514,12 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { }} {if (hasInput) { }} {if (hasOutput) { }} {if (hasShuffleRead) { @@ -486,7 +528,7 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { {shuffleReadBlockedTimeReadable} }} {if (hasShuffleWrite) { @@ -494,7 +536,7 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { {writeTimeReadable} }} {if (hasBytesSpilled) { diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala b/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala index 01f7e23212c3d..69aac6c862de5 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala @@ -31,9 +31,13 @@ private[jobs] object UIData { var failedTasks : Int = 0 var succeededTasks : Int = 0 var inputBytes : Long = 0 + var inputRecords : Long = 0 var outputBytes : Long = 0 + var outputRecords : Long = 0 var shuffleRead : Long = 0 + var shuffleReadRecords : Long = 0 var shuffleWrite : Long = 0 + var shuffleWriteRecords : Long = 0 var memoryBytesSpilled : Long = 0 var diskBytesSpilled : Long = 0 } @@ -73,9 +77,13 @@ private[jobs] object UIData { var executorRunTime: Long = _ var inputBytes: Long = _ + var inputRecords: Long = _ var outputBytes: Long = _ + var outputRecords: Long = _ var shuffleReadBytes: Long = _ + var shuffleReadRecords : Long = _ var shuffleWriteBytes: Long = _ + var shuffleWriteRecords: Long = _ var memoryBytesSpilled: Long = _ var diskBytesSpilled: Long = _ @@ -85,6 +93,12 @@ private[jobs] object UIData { var accumulables = new HashMap[Long, AccumulableInfo] var taskData = new HashMap[Long, TaskUIData] var executorSummary = new HashMap[String, ExecutorSummary] + + def hasInput = inputBytes > 0 + def hasOutput = outputBytes > 0 + def hasShuffleRead = shuffleReadBytes > 0 + def hasShuffleWrite = shuffleWriteBytes > 0 + def hasBytesSpilled = memoryBytesSpilled > 0 && diskBytesSpilled > 0 } /** diff --git a/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala b/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala index c8407bbcb780b..b0b545640f5aa 100644 --- a/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala +++ b/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala @@ -293,22 +293,26 @@ private[spark] object JsonProtocol { ("Remote Blocks Fetched" -> shuffleReadMetrics.remoteBlocksFetched) ~ ("Local Blocks Fetched" -> shuffleReadMetrics.localBlocksFetched) ~ ("Fetch Wait Time" -> shuffleReadMetrics.fetchWaitTime) ~ - ("Remote Bytes Read" -> shuffleReadMetrics.remoteBytesRead) + ("Remote Bytes Read" -> shuffleReadMetrics.remoteBytesRead) ~ + ("Total Records Read" -> shuffleReadMetrics.recordsRead) } def shuffleWriteMetricsToJson(shuffleWriteMetrics: ShuffleWriteMetrics): JValue = { ("Shuffle Bytes Written" -> shuffleWriteMetrics.shuffleBytesWritten) ~ - ("Shuffle Write Time" -> shuffleWriteMetrics.shuffleWriteTime) + ("Shuffle Write Time" -> shuffleWriteMetrics.shuffleWriteTime) ~ + ("Shuffle Records Written" -> shuffleWriteMetrics.shuffleRecordsWritten) } def inputMetricsToJson(inputMetrics: InputMetrics): JValue = { ("Data Read Method" -> inputMetrics.readMethod.toString) ~ - ("Bytes Read" -> inputMetrics.bytesRead) + ("Bytes Read" -> inputMetrics.bytesRead) ~ + ("Records Read" -> inputMetrics.recordsRead) } def outputMetricsToJson(outputMetrics: OutputMetrics): JValue = { ("Data Write Method" -> outputMetrics.writeMethod.toString) ~ - ("Bytes Written" -> outputMetrics.bytesWritten) + ("Bytes Written" -> outputMetrics.bytesWritten) ~ + ("Records Written" -> outputMetrics.recordsWritten) } def taskEndReasonToJson(taskEndReason: TaskEndReason): JValue = { @@ -670,6 +674,7 @@ private[spark] object JsonProtocol { metrics.incLocalBlocksFetched((json \ "Local Blocks Fetched").extract[Int]) metrics.incFetchWaitTime((json \ "Fetch Wait Time").extract[Long]) metrics.incRemoteBytesRead((json \ "Remote Bytes Read").extract[Long]) + metrics.incRecordsRead((json \ "Total Records Read").extractOpt[Long].getOrElse(0)) metrics } @@ -677,13 +682,16 @@ private[spark] object JsonProtocol { val metrics = new ShuffleWriteMetrics metrics.incShuffleBytesWritten((json \ "Shuffle Bytes Written").extract[Long]) metrics.incShuffleWriteTime((json \ "Shuffle Write Time").extract[Long]) + metrics.setShuffleRecordsWritten((json \ "Shuffle Records Written") + .extractOpt[Long].getOrElse(0)) metrics } def inputMetricsFromJson(json: JValue): InputMetrics = { val metrics = new InputMetrics( DataReadMethod.withName((json \ "Data Read Method").extract[String])) - metrics.addBytesRead((json \ "Bytes Read").extract[Long]) + metrics.incBytesRead((json \ "Bytes Read").extract[Long]) + metrics.incRecordsRead((json \ "Records Read").extractOpt[Long].getOrElse(0)) metrics } @@ -691,6 +699,7 @@ private[spark] object JsonProtocol { val metrics = new OutputMetrics( DataWriteMethod.withName((json \ "Data Write Method").extract[String])) metrics.setBytesWritten((json \ "Bytes Written").extract[Long]) + metrics.setRecordsWritten((json \ "Records Written").extractOpt[Long].getOrElse(0)) metrics } diff --git a/core/src/main/scala/org/apache/spark/util/collection/ExternalSorter.scala b/core/src/main/scala/org/apache/spark/util/collection/ExternalSorter.scala index 6ba03841f746b..eaec5a71e6819 100644 --- a/core/src/main/scala/org/apache/spark/util/collection/ExternalSorter.scala +++ b/core/src/main/scala/org/apache/spark/util/collection/ExternalSorter.scala @@ -763,6 +763,7 @@ private[spark] class ExternalSorter[K, V, C]( if (curWriteMetrics != null) { m.incShuffleBytesWritten(curWriteMetrics.shuffleBytesWritten) m.incShuffleWriteTime(curWriteMetrics.shuffleWriteTime) + m.incShuffleRecordsWritten(curWriteMetrics.shuffleRecordsWritten) } } diff --git a/core/src/test/scala/org/apache/spark/metrics/InputOutputMetricsSuite.scala b/core/src/test/scala/org/apache/spark/metrics/InputOutputMetricsSuite.scala index 81db66ae17464..78fa98a3b9065 100644 --- a/core/src/test/scala/org/apache/spark/metrics/InputOutputMetricsSuite.scala +++ b/core/src/test/scala/org/apache/spark/metrics/InputOutputMetricsSuite.scala @@ -21,44 +21,46 @@ import java.io.{File, FileWriter, PrintWriter} import scala.collection.mutable.ArrayBuffer -import org.scalatest.FunSuite - +import org.apache.commons.lang.math.RandomUtils import org.apache.hadoop.conf.Configuration import org.apache.hadoop.fs.{FileSystem, Path} import org.apache.hadoop.io.{LongWritable, Text} -import org.apache.hadoop.mapred.{FileSplit => OldFileSplit, InputSplit => OldInputSplit, JobConf, - LineRecordReader => OldLineRecordReader, RecordReader => OldRecordReader, Reporter, - TextInputFormat => OldTextInputFormat} import org.apache.hadoop.mapred.lib.{CombineFileInputFormat => OldCombineFileInputFormat, - CombineFileSplit => OldCombineFileSplit, CombineFileRecordReader => OldCombineFileRecordReader} -import org.apache.hadoop.mapreduce.{InputSplit => NewInputSplit, RecordReader => NewRecordReader, - TaskAttemptContext} + CombineFileRecordReader => OldCombineFileRecordReader, CombineFileSplit => OldCombineFileSplit} +import org.apache.hadoop.mapred.{JobConf, Reporter, FileSplit => OldFileSplit, + InputSplit => OldInputSplit, LineRecordReader => OldLineRecordReader, + RecordReader => OldRecordReader, TextInputFormat => OldTextInputFormat} import org.apache.hadoop.mapreduce.lib.input.{CombineFileInputFormat => NewCombineFileInputFormat, CombineFileRecordReader => NewCombineFileRecordReader, CombineFileSplit => NewCombineFileSplit, FileSplit => NewFileSplit, TextInputFormat => NewTextInputFormat} +import org.apache.hadoop.mapreduce.lib.output.{TextOutputFormat => NewTextOutputFormat} +import org.apache.hadoop.mapreduce.{TaskAttemptContext, InputSplit => NewInputSplit, + RecordReader => NewRecordReader} +import org.scalatest.{BeforeAndAfter, FunSuite} import org.apache.spark.SharedSparkContext import org.apache.spark.deploy.SparkHadoopUtil import org.apache.spark.scheduler.{SparkListener, SparkListenerTaskEnd} import org.apache.spark.util.Utils -class InputOutputMetricsSuite extends FunSuite with SharedSparkContext { +class InputOutputMetricsSuite extends FunSuite with SharedSparkContext + with BeforeAndAfter { @transient var tmpDir: File = _ @transient var tmpFile: File = _ @transient var tmpFilePath: String = _ + @transient val numRecords: Int = 100000 + @transient val numBuckets: Int = 10 - override def beforeAll() { - super.beforeAll() - + before { tmpDir = Utils.createTempDir() val testTempDir = new File(tmpDir, "test") testTempDir.mkdir() tmpFile = new File(testTempDir, getClass.getSimpleName + ".txt") val pw = new PrintWriter(new FileWriter(tmpFile)) - for (x <- 1 to 1000000) { - pw.println("s") + for (x <- 1 to numRecords) { + pw.println(RandomUtils.nextInt(numBuckets)) } pw.close() @@ -66,8 +68,7 @@ class InputOutputMetricsSuite extends FunSuite with SharedSparkContext { tmpFilePath = "file://" + tmpFile.getAbsolutePath } - override def afterAll() { - super.afterAll() + after { Utils.deleteRecursively(tmpDir) } @@ -155,6 +156,101 @@ class InputOutputMetricsSuite extends FunSuite with SharedSparkContext { assert(bytesRead >= tmpFile.length()) } + test("input metrics on records read - simple") { + val records = runAndReturnRecordsRead { + sc.textFile(tmpFilePath, 4).count() + } + assert(records == numRecords) + } + + test("input metrics on records read - more stages") { + val records = runAndReturnRecordsRead { + sc.textFile(tmpFilePath, 4) + .map(key => (key.length, 1)) + .reduceByKey(_ + _) + .count() + } + assert(records == numRecords) + } + + test("input metrics on records - New Hadoop API") { + val records = runAndReturnRecordsRead { + sc.newAPIHadoopFile(tmpFilePath, classOf[NewTextInputFormat], classOf[LongWritable], + classOf[Text]).count() + } + assert(records == numRecords) + } + + test("input metrics on recordsd read with cache") { + // prime the cache manager + val rdd = sc.textFile(tmpFilePath, 4).cache() + rdd.collect() + + val records = runAndReturnRecordsRead { + rdd.count() + } + + assert(records == numRecords) + } + + test("shuffle records read metrics") { + val recordsRead = runAndReturnShuffleRecordsRead { + sc.textFile(tmpFilePath, 4) + .map(key => (key, 1)) + .groupByKey() + .collect() + } + assert(recordsRead == numRecords) + } + + test("shuffle records written metrics") { + val recordsWritten = runAndReturnShuffleRecordsWritten { + sc.textFile(tmpFilePath, 4) + .map(key => (key, 1)) + .groupByKey() + .collect() + } + assert(recordsWritten == numRecords) + } + + /** + * Tests the metrics from end to end. + * 1) reading a hadoop file + * 2) shuffle and writing to a hadoop file. + * 3) writing to hadoop file. + */ + test("input read/write and shuffle read/write metrics all line up") { + var inputRead = 0L + var outputWritten = 0L + var shuffleRead = 0L + var shuffleWritten = 0L + sc.addSparkListener(new SparkListener() { + override def onTaskEnd(taskEnd: SparkListenerTaskEnd) { + val metrics = taskEnd.taskMetrics + metrics.inputMetrics.foreach(inputRead += _.recordsRead) + metrics.outputMetrics.foreach(outputWritten += _.recordsWritten) + metrics.shuffleReadMetrics.foreach(shuffleRead += _.recordsRead) + metrics.shuffleWriteMetrics.foreach(shuffleWritten += _.shuffleRecordsWritten) + } + }) + + val tmpFile = new File(tmpDir, getClass.getSimpleName) + + sc.textFile(tmpFilePath, 4) + .map(key => (key, 1)) + .reduceByKey(_+_) + .saveAsTextFile("file://" + tmpFile.getAbsolutePath) + + sc.listenerBus.waitUntilEmpty(500) + assert(inputRead == numRecords) + + // Only supported on newer Hadoop + if (SparkHadoopUtil.get.getFSBytesWrittenOnThreadCallback().isDefined) { + assert(outputWritten == numBuckets) + } + assert(shuffleRead == shuffleWritten) + } + test("input metrics with interleaved reads") { val numPartitions = 2 val cartVector = 0 to 9 @@ -193,18 +289,66 @@ class InputOutputMetricsSuite extends FunSuite with SharedSparkContext { assert(cartesianBytes == firstSize * numPartitions + (cartVector.length * secondSize)) } - private def runAndReturnBytesRead(job : => Unit): Long = { - val taskBytesRead = new ArrayBuffer[Long]() + private def runAndReturnBytesRead(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.inputMetrics.map(_.bytesRead)) + } + + private def runAndReturnRecordsRead(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.inputMetrics.map(_.recordsRead)) + } + + private def runAndReturnRecordsWritten(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.outputMetrics.map(_.recordsWritten)) + } + + private def runAndReturnShuffleRecordsRead(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.shuffleReadMetrics.map(_.recordsRead)) + } + + private def runAndReturnShuffleRecordsWritten(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.shuffleWriteMetrics.map(_.shuffleRecordsWritten)) + } + + private def runAndReturnMetrics(job: => Unit, + collector: (SparkListenerTaskEnd) => Option[Long]): Long = { + val taskMetrics = new ArrayBuffer[Long]() sc.addSparkListener(new SparkListener() { override def onTaskEnd(taskEnd: SparkListenerTaskEnd) { - taskBytesRead += taskEnd.taskMetrics.inputMetrics.get.bytesRead + collector(taskEnd).foreach(taskMetrics += _) } }) job sc.listenerBus.waitUntilEmpty(500) - taskBytesRead.sum + taskMetrics.sum + } + + test("output metrics on records written") { + // Only supported on newer Hadoop + if (SparkHadoopUtil.get.getFSBytesWrittenOnThreadCallback().isDefined) { + val file = new File(tmpDir, getClass.getSimpleName) + val filePath = "file://" + file.getAbsolutePath + + val records = runAndReturnRecordsWritten { + sc.parallelize(1 to numRecords).saveAsTextFile(filePath) + } + assert(records == numRecords) + } + } + + test("output metrics on records written - new Hadoop API") { + // Only supported on newer Hadoop + if (SparkHadoopUtil.get.getFSBytesWrittenOnThreadCallback().isDefined) { + val file = new File(tmpDir, getClass.getSimpleName) + val filePath = "file://" + file.getAbsolutePath + + val records = runAndReturnRecordsWritten { + sc.parallelize(1 to numRecords).map(key => (key.toString, key.toString)) + .saveAsNewAPIHadoopFile[NewTextOutputFormat[String, String]](filePath) + } + assert(records == numRecords) + } } test("output metrics when writing text file") { @@ -318,4 +462,4 @@ class NewCombineTextRecordReaderWrapper( override def getCurrentValue(): Text = delegate.getCurrentValue override def getProgress(): Float = delegate.getProgress override def close(): Unit = delegate.close() -} \ No newline at end of file +} diff --git a/core/src/test/scala/org/apache/spark/storage/BlockObjectWriterSuite.scala b/core/src/test/scala/org/apache/spark/storage/BlockObjectWriterSuite.scala index bbc7e1357b90d..c21c92b63ad13 100644 --- a/core/src/test/scala/org/apache/spark/storage/BlockObjectWriterSuite.scala +++ b/core/src/test/scala/org/apache/spark/storage/BlockObjectWriterSuite.scala @@ -31,6 +31,8 @@ class BlockObjectWriterSuite extends FunSuite { new JavaSerializer(new SparkConf()), 1024, os => os, true, writeMetrics) writer.write(Long.box(20)) + // Record metrics update on every write + assert(writeMetrics.shuffleRecordsWritten === 1) // Metrics don't update on every write assert(writeMetrics.shuffleBytesWritten == 0) // After 32 writes, metrics should update @@ -39,6 +41,7 @@ class BlockObjectWriterSuite extends FunSuite { writer.write(Long.box(i)) } assert(writeMetrics.shuffleBytesWritten > 0) + assert(writeMetrics.shuffleRecordsWritten === 33) writer.commitAndClose() assert(file.length() == writeMetrics.shuffleBytesWritten) } @@ -51,6 +54,8 @@ class BlockObjectWriterSuite extends FunSuite { new JavaSerializer(new SparkConf()), 1024, os => os, true, writeMetrics) writer.write(Long.box(20)) + // Record metrics update on every write + assert(writeMetrics.shuffleRecordsWritten === 1) // Metrics don't update on every write assert(writeMetrics.shuffleBytesWritten == 0) // After 32 writes, metrics should update @@ -59,7 +64,23 @@ class BlockObjectWriterSuite extends FunSuite { writer.write(Long.box(i)) } assert(writeMetrics.shuffleBytesWritten > 0) + assert(writeMetrics.shuffleRecordsWritten === 33) writer.revertPartialWritesAndClose() assert(writeMetrics.shuffleBytesWritten == 0) + assert(writeMetrics.shuffleRecordsWritten == 0) + } + + test("Reopening a closed block writer") { + val file = new File("somefile") + file.deleteOnExit() + val writeMetrics = new ShuffleWriteMetrics() + val writer = new DiskBlockObjectWriter(new TestBlockId("0"), file, + new JavaSerializer(new SparkConf()), 1024, os => os, true, writeMetrics) + + writer.open() + writer.close() + intercept[IllegalStateException] { + writer.open() + } } } diff --git a/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala b/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala index 68074ae32a672..e8405baa8e3ea 100644 --- a/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala +++ b/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala @@ -234,7 +234,7 @@ class JobProgressListenerSuite extends FunSuite with LocalSparkContext with Matc taskMetrics.incMemoryBytesSpilled(base + 6) val inputMetrics = new InputMetrics(DataReadMethod.Hadoop) taskMetrics.setInputMetrics(Some(inputMetrics)) - inputMetrics.addBytesRead(base + 7) + inputMetrics.incBytesRead(base + 7) val outputMetrics = new OutputMetrics(DataWriteMethod.Hadoop) taskMetrics.outputMetrics = Some(outputMetrics) outputMetrics.setBytesWritten(base + 8) diff --git a/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala b/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala index 842f54529baf0..f3017dc42cd5c 100644 --- a/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala +++ b/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala @@ -189,6 +189,34 @@ class JsonProtocolSuite extends FunSuite { assert(newMetrics.inputMetrics.isEmpty) } + test("Input/Output records backwards compatibility") { + // records read were added after 1.2 + val metrics = makeTaskMetrics(1L, 2L, 3L, 4L, 5, 6, + hasHadoopInput = true, hasOutput = true, hasRecords = false) + assert(metrics.inputMetrics.nonEmpty) + assert(metrics.outputMetrics.nonEmpty) + val newJson = JsonProtocol.taskMetricsToJson(metrics) + val oldJson = newJson.removeField { case (field, _) => field == "Records Read" } + .removeField { case (field, _) => field == "Records Written" } + val newMetrics = JsonProtocol.taskMetricsFromJson(oldJson) + assert(newMetrics.inputMetrics.get.recordsRead == 0) + assert(newMetrics.outputMetrics.get.recordsWritten == 0) + } + + test("Shuffle Read/Write records backwards compatibility") { + // records read were added after 1.2 + val metrics = makeTaskMetrics(1L, 2L, 3L, 4L, 5, 6, + hasHadoopInput = false, hasOutput = false, hasRecords = false) + assert(metrics.shuffleReadMetrics.nonEmpty) + assert(metrics.shuffleWriteMetrics.nonEmpty) + val newJson = JsonProtocol.taskMetricsToJson(metrics) + val oldJson = newJson.removeField { case (field, _) => field == "Total Records Read" } + .removeField { case (field, _) => field == "Shuffle Records Written" } + val newMetrics = JsonProtocol.taskMetricsFromJson(oldJson) + assert(newMetrics.shuffleReadMetrics.get.recordsRead == 0) + assert(newMetrics.shuffleWriteMetrics.get.shuffleRecordsWritten == 0) + } + test("OutputMetrics backward compatibility") { // OutputMetrics were added after 1.1 val metrics = makeTaskMetrics(1L, 2L, 3L, 4L, 5, 6, hasHadoopInput = false, hasOutput = true) @@ -644,7 +672,8 @@ class JsonProtocolSuite extends FunSuite { e: Int, f: Int, hasHadoopInput: Boolean, - hasOutput: Boolean) = { + hasOutput: Boolean, + hasRecords: Boolean = true) = { val t = new TaskMetrics t.setHostname("localhost") t.setExecutorDeserializeTime(a) @@ -656,7 +685,8 @@ class JsonProtocolSuite extends FunSuite { if (hasHadoopInput) { val inputMetrics = new InputMetrics(DataReadMethod.Hadoop) - inputMetrics.addBytesRead(d + e + f) + inputMetrics.incBytesRead(d + e + f) + inputMetrics.incRecordsRead(if (hasRecords) (d + e + f) / 100 else -1) t.setInputMetrics(Some(inputMetrics)) } else { val sr = new ShuffleReadMetrics @@ -664,16 +694,19 @@ class JsonProtocolSuite extends FunSuite { sr.incLocalBlocksFetched(e) sr.incFetchWaitTime(a + d) sr.incRemoteBlocksFetched(f) + sr.incRecordsRead(if (hasRecords) (b + d) / 100 else -1) t.setShuffleReadMetrics(Some(sr)) } if (hasOutput) { val outputMetrics = new OutputMetrics(DataWriteMethod.Hadoop) outputMetrics.setBytesWritten(a + b + c) + outputMetrics.setRecordsWritten(if (hasRecords) (a + b + c)/100 else -1) t.outputMetrics = Some(outputMetrics) } else { val sw = new ShuffleWriteMetrics sw.incShuffleBytesWritten(a + b + c) sw.incShuffleWriteTime(b + c + d) + sw.setShuffleRecordsWritten(if (hasRecords) (a + b + c) / 100 else -1) t.shuffleWriteMetrics = Some(sw) } // Make at most 6 blocks @@ -907,11 +940,13 @@ class JsonProtocolSuite extends FunSuite { | "Remote Blocks Fetched": 800, | "Local Blocks Fetched": 700, | "Fetch Wait Time": 900, - | "Remote Bytes Read": 1000 + | "Remote Bytes Read": 1000, + | "Total Records Read" : 10 | }, | "Shuffle Write Metrics": { | "Shuffle Bytes Written": 1200, - | "Shuffle Write Time": 1500 + | "Shuffle Write Time": 1500, + | "Shuffle Records Written": 12 | }, | "Updated Blocks": [ | { @@ -988,11 +1023,13 @@ class JsonProtocolSuite extends FunSuite { | "Disk Bytes Spilled": 0, | "Shuffle Write Metrics": { | "Shuffle Bytes Written": 1200, - | "Shuffle Write Time": 1500 + | "Shuffle Write Time": 1500, + | "Shuffle Records Written": 12 | }, | "Input Metrics": { | "Data Read Method": "Hadoop", - | "Bytes Read": 2100 + | "Bytes Read": 2100, + | "Records Read": 21 | }, | "Updated Blocks": [ | { @@ -1069,11 +1106,13 @@ class JsonProtocolSuite extends FunSuite { | "Disk Bytes Spilled": 0, | "Input Metrics": { | "Data Read Method": "Hadoop", - | "Bytes Read": 2100 + | "Bytes Read": 2100, + | "Records Read": 21 | }, | "Output Metrics": { | "Data Write Method": "Hadoop", - | "Bytes Written": 1200 + | "Bytes Written": 1200, + | "Records Written": 12 | }, | "Updated Blocks": [ | { From 2bda1c1d376afd8abe6a04be345461752f3fb1b6 Mon Sep 17 00:00:00 2001 From: huangzhaowei Date: Fri, 6 Feb 2015 14:35:29 -0800 Subject: [PATCH 43/53] [SPARK-5444][Network]Add a retry to deal with the conflict port in netty server. If the `spark.blockMnager.port` had conflicted with a specific port, Spark will throw an exception and exit. So add a retry to avoid this situation. Author: huangzhaowei Closes #4240 from SaintBacchus/NettyPortConflict and squashes the following commits: cc926d2 [huangzhaowei] Add a retry to deal with the conflict port in netty server. --- .../spark/network/server/TransportServer.java | 36 +++++++++++++++++-- .../spark/network/util/TransportConf.java | 7 ++++ 2 files changed, 41 insertions(+), 2 deletions(-) diff --git a/network/common/src/main/java/org/apache/spark/network/server/TransportServer.java b/network/common/src/main/java/org/apache/spark/network/server/TransportServer.java index 625c3257d764e..ef209991804b4 100644 --- a/network/common/src/main/java/org/apache/spark/network/server/TransportServer.java +++ b/network/common/src/main/java/org/apache/spark/network/server/TransportServer.java @@ -100,8 +100,7 @@ protected void initChannel(SocketChannel ch) throws Exception { } }); - channelFuture = bootstrap.bind(new InetSocketAddress(portToBind)); - channelFuture.syncUninterruptibly(); + bindRightPort(portToBind); port = ((InetSocketAddress) channelFuture.channel().localAddress()).getPort(); logger.debug("Shuffle server started on port :" + port); @@ -123,4 +122,37 @@ public void close() { bootstrap = null; } + /** + * Attempt to bind to the specified port up to a fixed number of retries. + * If all attempts fail after the max number of retries, exit. + */ + private void bindRightPort(int portToBind) { + int maxPortRetries = conf.portMaxRetries(); + + for (int i = 0; i <= maxPortRetries; i++) { + int tryPort = -1; + if (0 == portToBind) { + // Do not increment port if tryPort is 0, which is treated as a special port + tryPort = 0; + } else { + // If the new port wraps around, do not try a privilege port + tryPort = ((portToBind + i - 1024) % (65536 - 1024)) + 1024; + } + try { + channelFuture = bootstrap.bind(new InetSocketAddress(tryPort)); + channelFuture.syncUninterruptibly(); + return; + } catch (Exception e) { + logger.warn("Netty service could not bind on port " + tryPort + + ". Attempting the next port."); + if (i >= maxPortRetries) { + logger.error(e.getMessage() + ": Netty server failed after " + + maxPortRetries + " retries."); + + // If it can't find a right port, it should exit directly. + System.exit(-1); + } + } + } + } } diff --git a/network/common/src/main/java/org/apache/spark/network/util/TransportConf.java b/network/common/src/main/java/org/apache/spark/network/util/TransportConf.java index 6c9178688693f..2eaf3b71d9a49 100644 --- a/network/common/src/main/java/org/apache/spark/network/util/TransportConf.java +++ b/network/common/src/main/java/org/apache/spark/network/util/TransportConf.java @@ -98,4 +98,11 @@ public int memoryMapBytes() { public boolean lazyFileDescriptor() { return conf.getBoolean("spark.shuffle.io.lazyFD", true); } + + /** + * Maximum number of retries when binding to a port before giving up. + */ + public int portMaxRetries() { + return conf.getInt("spark.port.maxRetries", 16); + } } From 61073f832128845a76469fc37376483b784c927b Mon Sep 17 00:00:00 2001 From: lianhuiwang Date: Fri, 6 Feb 2015 14:47:52 -0800 Subject: [PATCH 44/53] [SPARK-4994][network]Cleanup removed executors' ShuffleInfo in yarn shuffle service when the application is completed, yarn's nodemanager can remove application's local-dirs.but all executors' metadata of completed application havenot be removed. now it lets yarn ShuffleService to have much more memory to store Executors' ShuffleInfo. so these metadata need to be removed. Author: lianhuiwang Closes #3828 from lianhuiwang/SPARK-4994 and squashes the following commits: f3ba1d2 [lianhuiwang] Cleanup removed executors' ShuffleInfo --- .../org/apache/spark/network/yarn/YarnShuffleService.java | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/network/yarn/src/main/java/org/apache/spark/network/yarn/YarnShuffleService.java b/network/yarn/src/main/java/org/apache/spark/network/yarn/YarnShuffleService.java index a34aabe9e78a6..63b21222e7b77 100644 --- a/network/yarn/src/main/java/org/apache/spark/network/yarn/YarnShuffleService.java +++ b/network/yarn/src/main/java/org/apache/spark/network/yarn/YarnShuffleService.java @@ -76,6 +76,9 @@ public class YarnShuffleService extends AuxiliaryService { // The actual server that serves shuffle files private TransportServer shuffleServer = null; + // Handles registering executors and opening shuffle blocks + private ExternalShuffleBlockHandler blockHandler; + public YarnShuffleService() { super("spark_shuffle"); logger.info("Initializing YARN shuffle service for Spark"); @@ -99,7 +102,8 @@ protected void serviceInit(Configuration conf) { // If authentication is enabled, set up the shuffle server to use a // special RPC handler that filters out unauthenticated fetch requests boolean authEnabled = conf.getBoolean(SPARK_AUTHENTICATE_KEY, DEFAULT_SPARK_AUTHENTICATE); - RpcHandler rpcHandler = new ExternalShuffleBlockHandler(transportConf); + blockHandler = new ExternalShuffleBlockHandler(transportConf); + RpcHandler rpcHandler = blockHandler; if (authEnabled) { secretManager = new ShuffleSecretManager(); rpcHandler = new SaslRpcHandler(rpcHandler, secretManager); @@ -136,6 +140,7 @@ public void stopApplication(ApplicationTerminationContext context) { if (isAuthenticationEnabled()) { secretManager.unregisterApp(appId); } + blockHandler.applicationRemoved(appId, false /* clean up local dirs */); } catch (Exception e) { logger.error("Exception when stopping application {}", appId, e); } From 76c4bf59f6544b9c02d99fd18436427cbce632e6 Mon Sep 17 00:00:00 2001 From: Michael Armbrust Date: Fri, 6 Feb 2015 15:11:02 -0800 Subject: [PATCH 45/53] [SQL] Use TestSQLContext in Java tests Sometimes tests were failing due to the creation of multiple `SparkContext`s in a single JVM. Author: Michael Armbrust Closes #4441 from marmbrus/javaTests and squashes the following commits: 657b1e0 [Michael Armbrust] [SQL] Use TestSQLContext in Java tests --- .../java/org/apache/spark/sql/api/java/JavaAPISuite.java | 7 +++---- .../apache/spark/sql/api/java/JavaApplySchemaSuite.java | 6 +++--- 2 files changed, 6 insertions(+), 7 deletions(-) diff --git a/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaAPISuite.java b/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaAPISuite.java index e5588938ea162..a21a15409080c 100644 --- a/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaAPISuite.java +++ b/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaAPISuite.java @@ -19,6 +19,7 @@ import java.io.Serializable; +import org.apache.spark.sql.test.TestSQLContext$; import org.junit.After; import org.junit.Before; import org.junit.Test; @@ -37,14 +38,12 @@ public class JavaAPISuite implements Serializable { @Before public void setUp() { - sc = new JavaSparkContext("local", "JavaAPISuite"); - sqlContext = new SQLContext(sc); + sqlContext = TestSQLContext$.MODULE$; + sc = new JavaSparkContext(sqlContext.sparkContext()); } @After public void tearDown() { - sc.stop(); - sc = null; } @SuppressWarnings("unchecked") diff --git a/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaApplySchemaSuite.java b/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaApplySchemaSuite.java index 8510bac499092..2e6e977fdc752 100644 --- a/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaApplySchemaSuite.java +++ b/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaApplySchemaSuite.java @@ -22,6 +22,7 @@ import java.util.Arrays; import java.util.List; +import org.apache.spark.sql.test.TestSQLContext$; import org.junit.After; import org.junit.Assert; import org.junit.Before; @@ -42,13 +43,12 @@ public class JavaApplySchemaSuite implements Serializable { @Before public void setUp() { - javaCtx = new JavaSparkContext("local", "JavaApplySchemaSuite"); - javaSqlCtx = new SQLContext(javaCtx); + javaSqlCtx = TestSQLContext$.MODULE$; + javaCtx = new JavaSparkContext(javaSqlCtx.sparkContext()); } @After public void tearDown() { - javaCtx.stop(); javaCtx = null; javaSqlCtx = null; } From c4021401e326cd5a412a70425f5c75405284880e Mon Sep 17 00:00:00 2001 From: Cheng Lian Date: Fri, 6 Feb 2015 15:23:42 -0800 Subject: [PATCH 46/53] [SQL] [Minor] HiveParquetSuite was disabled by mistake, re-enable them [Review on Reviewable](https://reviewable.io/reviews/apache/spark/4440) Author: Cheng Lian Closes #4440 from liancheng/parquet-oops and squashes the following commits: f21ede4 [Cheng Lian] HiveParquetSuite was disabled by mistake, re-enable them. --- .../spark/sql/parquet/ParquetQuerySuite.scala | 1 - .../apache/spark/sql/parquet/HiveParquetSuite.scala | 13 ++++++++++--- 2 files changed, 10 insertions(+), 4 deletions(-) diff --git a/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetQuerySuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetQuerySuite.scala index 48c7598343e55..cba06835f9a61 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetQuerySuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetQuerySuite.scala @@ -35,7 +35,6 @@ class ParquetQuerySuite extends QueryTest with ParquetTest { } } - // TODO Re-enable this after data source insertion API is merged test(s"$prefix: appending") { val data = (0 until 10).map(i => (i, i.toString)) withParquetTable(data, "t") { diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/parquet/HiveParquetSuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/parquet/HiveParquetSuite.scala index eae69af5864aa..e89b4489f15d1 100644 --- a/sql/hive/src/test/scala/org/apache/spark/sql/parquet/HiveParquetSuite.scala +++ b/sql/hive/src/test/scala/org/apache/spark/sql/parquet/HiveParquetSuite.scala @@ -17,7 +17,7 @@ package org.apache.spark.sql.parquet -import org.apache.spark.sql.QueryTest +import org.apache.spark.sql.{SQLConf, QueryTest} import org.apache.spark.sql.catalyst.expressions.Row import org.apache.spark.sql.hive.test.TestHive @@ -64,8 +64,7 @@ class HiveParquetSuite extends QueryTest with ParquetTest { } } - // TODO Re-enable this after data source insertion API is merged - ignore(s"$prefix: INSERT OVERWRITE TABLE Parquet table") { + test(s"$prefix: INSERT OVERWRITE TABLE Parquet table") { withParquetTable((1 to 10).map(i => (i, s"val_$i")), "t") { withTempPath { file => sql("SELECT * FROM t LIMIT 1").saveAsParquetFile(file.getCanonicalPath) @@ -81,4 +80,12 @@ class HiveParquetSuite extends QueryTest with ParquetTest { } } } + + withSQLConf(SQLConf.PARQUET_USE_DATA_SOURCE_API -> "true") { + run("Parquet data source enabled") + } + + withSQLConf(SQLConf.PARQUET_USE_DATA_SOURCE_API -> "false") { + run("Parquet data source disabled") + } } From 0e23ca9f805b46d9b3472330676e5c8db926b8f5 Mon Sep 17 00:00:00 2001 From: Xiangrui Meng Date: Fri, 6 Feb 2015 15:42:59 -0800 Subject: [PATCH 47/53] [SPARK-5601][MLLIB] make streaming linear algorithms Java-friendly Overload `trainOn`, `predictOn`, and `predictOnValues`. CC freeman-lab Author: Xiangrui Meng Closes #4432 from mengxr/streaming-java and squashes the following commits: 6a79b85 [Xiangrui Meng] add java test for streaming logistic regression 2d7b357 [Xiangrui Meng] organize imports 1f662b3 [Xiangrui Meng] make streaming linear algorithms Java-friendly --- .../regression/StreamingLinearAlgorithm.scala | 20 ++++- .../JavaStreamingLogisticRegressionSuite.java | 82 +++++++++++++++++++ .../JavaStreamingLinearRegressionSuite.java | 80 ++++++++++++++++++ 3 files changed, 181 insertions(+), 1 deletion(-) create mode 100644 mllib/src/test/java/org/apache/spark/ml/classification/JavaStreamingLogisticRegressionSuite.java create mode 100644 mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java diff --git a/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearAlgorithm.scala b/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearAlgorithm.scala index c854f124451cf..ce95c063db970 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearAlgorithm.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearAlgorithm.scala @@ -21,7 +21,9 @@ import scala.reflect.ClassTag import org.apache.spark.Logging import org.apache.spark.annotation.DeveloperApi +import org.apache.spark.api.java.JavaSparkContext.fakeClassTag import org.apache.spark.mllib.linalg.{Vector, Vectors} +import org.apache.spark.streaming.api.java.{JavaDStream, JavaPairDStream} import org.apache.spark.streaming.dstream.DStream /** @@ -76,7 +78,7 @@ abstract class StreamingLinearAlgorithm[ * * @param data DStream containing labeled data */ - def trainOn(data: DStream[LabeledPoint]) { + def trainOn(data: DStream[LabeledPoint]): Unit = { if (model.isEmpty) { throw new IllegalArgumentException("Model must be initialized before starting training.") } @@ -99,6 +101,9 @@ abstract class StreamingLinearAlgorithm[ } } + /** Java-friendly version of `trainOn`. */ + def trainOn(data: JavaDStream[LabeledPoint]): Unit = trainOn(data.dstream) + /** * Use the model to make predictions on batches of data from a DStream * @@ -112,6 +117,11 @@ abstract class StreamingLinearAlgorithm[ data.map(model.get.predict) } + /** Java-friendly version of `predictOn`. */ + def predictOn(data: JavaDStream[Vector]): JavaDStream[java.lang.Double] = { + JavaDStream.fromDStream(predictOn(data.dstream).asInstanceOf[DStream[java.lang.Double]]) + } + /** * Use the model to make predictions on the values of a DStream and carry over its keys. * @param data DStream containing feature vectors @@ -124,4 +134,12 @@ abstract class StreamingLinearAlgorithm[ } data.mapValues(model.get.predict) } + + + /** Java-friendly version of `predictOnValues`. */ + def predictOnValues[K](data: JavaPairDStream[K, Vector]): JavaPairDStream[K, java.lang.Double] = { + implicit val tag = fakeClassTag[K] + JavaPairDStream.fromPairDStream( + predictOnValues(data.dstream).asInstanceOf[DStream[(K, java.lang.Double)]]) + } } diff --git a/mllib/src/test/java/org/apache/spark/ml/classification/JavaStreamingLogisticRegressionSuite.java b/mllib/src/test/java/org/apache/spark/ml/classification/JavaStreamingLogisticRegressionSuite.java new file mode 100644 index 0000000000000..ac945ba6f23c1 --- /dev/null +++ b/mllib/src/test/java/org/apache/spark/ml/classification/JavaStreamingLogisticRegressionSuite.java @@ -0,0 +1,82 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.classification; + +import java.io.Serializable; +import java.util.List; + +import scala.Tuple2; + +import com.google.common.collect.Lists; +import org.junit.After; +import org.junit.Before; +import org.junit.Test; + +import org.apache.spark.SparkConf; +import org.apache.spark.mllib.classification.StreamingLogisticRegressionWithSGD; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.mllib.regression.LabeledPoint; +import org.apache.spark.streaming.Duration; +import org.apache.spark.streaming.api.java.JavaDStream; +import org.apache.spark.streaming.api.java.JavaPairDStream; +import org.apache.spark.streaming.api.java.JavaStreamingContext; +import static org.apache.spark.streaming.JavaTestUtils.*; + +public class JavaStreamingLogisticRegressionSuite implements Serializable { + + protected transient JavaStreamingContext ssc; + + @Before + public void setUp() { + SparkConf conf = new SparkConf() + .setMaster("local[2]") + .setAppName("test") + .set("spark.streaming.clock", "org.apache.spark.streaming.util.ManualClock"); + ssc = new JavaStreamingContext(conf, new Duration(1000)); + ssc.checkpoint("checkpoint"); + } + + @After + public void tearDown() { + ssc.stop(); + ssc = null; + } + + @Test + @SuppressWarnings("unchecked") + public void javaAPI() { + List trainingBatch = Lists.newArrayList( + new LabeledPoint(1.0, Vectors.dense(1.0)), + new LabeledPoint(0.0, Vectors.dense(0.0))); + JavaDStream training = + attachTestInputStream(ssc, Lists.newArrayList(trainingBatch, trainingBatch), 2); + List> testBatch = Lists.newArrayList( + new Tuple2(10, Vectors.dense(1.0)), + new Tuple2(11, Vectors.dense(0.0))); + JavaPairDStream test = JavaPairDStream.fromJavaDStream( + attachTestInputStream(ssc, Lists.newArrayList(testBatch, testBatch), 2)); + StreamingLogisticRegressionWithSGD slr = new StreamingLogisticRegressionWithSGD() + .setNumIterations(2) + .setInitialWeights(Vectors.dense(0.0)); + slr.trainOn(training); + JavaPairDStream prediction = slr.predictOnValues(test); + attachTestOutputStream(prediction.count()); + runStreams(ssc, 2, 2); + } +} diff --git a/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java b/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java new file mode 100644 index 0000000000000..a4dd1ac39a3c8 --- /dev/null +++ b/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java @@ -0,0 +1,80 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.mllib.regression; + +import java.io.Serializable; +import java.util.List; + +import scala.Tuple2; + +import com.google.common.collect.Lists; +import org.junit.After; +import org.junit.Before; +import org.junit.Test; + +import org.apache.spark.SparkConf; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.streaming.Duration; +import org.apache.spark.streaming.api.java.JavaDStream; +import org.apache.spark.streaming.api.java.JavaPairDStream; +import org.apache.spark.streaming.api.java.JavaStreamingContext; +import static org.apache.spark.streaming.JavaTestUtils.*; + +public class JavaStreamingLinearRegressionSuite implements Serializable { + + protected transient JavaStreamingContext ssc; + + @Before + public void setUp() { + SparkConf conf = new SparkConf() + .setMaster("local[2]") + .setAppName("test") + .set("spark.streaming.clock", "org.apache.spark.streaming.util.ManualClock"); + ssc = new JavaStreamingContext(conf, new Duration(1000)); + ssc.checkpoint("checkpoint"); + } + + @After + public void tearDown() { + ssc.stop(); + ssc = null; + } + + @Test + @SuppressWarnings("unchecked") + public void javaAPI() { + List trainingBatch = Lists.newArrayList( + new LabeledPoint(1.0, Vectors.dense(1.0)), + new LabeledPoint(0.0, Vectors.dense(0.0))); + JavaDStream training = + attachTestInputStream(ssc, Lists.newArrayList(trainingBatch, trainingBatch), 2); + List> testBatch = Lists.newArrayList( + new Tuple2(10, Vectors.dense(1.0)), + new Tuple2(11, Vectors.dense(0.0))); + JavaPairDStream test = JavaPairDStream.fromJavaDStream( + attachTestInputStream(ssc, Lists.newArrayList(testBatch, testBatch), 2)); + StreamingLinearRegressionWithSGD slr = new StreamingLinearRegressionWithSGD() + .setNumIterations(2) + .setInitialWeights(Vectors.dense(0.0)); + slr.trainOn(training); + JavaPairDStream prediction = slr.predictOnValues(test); + attachTestOutputStream(prediction.count()); + runStreams(ssc, 2, 2); + } +} From e772b4e4e1b790199dd000bd096a8917cb8def24 Mon Sep 17 00:00:00 2001 From: Grzegorz Dubicki Date: Fri, 6 Feb 2015 15:43:58 -0800 Subject: [PATCH 48/53] SPARK-5403: Ignore UserKnownHostsFile in SSH calls See https://issues.apache.org/jira/browse/SPARK-5403 Author: Grzegorz Dubicki Closes #4196 from grzegorz-dubicki/SPARK-5403 and squashes the following commits: a7d863f [Grzegorz Dubicki] Resolve start command hanging issue --- ec2/spark_ec2.py | 1 + 1 file changed, 1 insertion(+) diff --git a/ec2/spark_ec2.py b/ec2/spark_ec2.py index 7371558274d78..3f7242a53d6fd 100755 --- a/ec2/spark_ec2.py +++ b/ec2/spark_ec2.py @@ -899,6 +899,7 @@ def stringify_command(parts): def ssh_args(opts): parts = ['-o', 'StrictHostKeyChecking=no'] + parts += ['-o', 'UserKnownHostsFile=/dev/null'] if opts.identity_file is not None: parts += ['-i', opts.identity_file] return parts From 1390e56fa86d07b588f0c3a1a2cc25ab5bb0db67 Mon Sep 17 00:00:00 2001 From: Andrew Or Date: Fri, 6 Feb 2015 15:57:06 -0800 Subject: [PATCH 49/53] [SPARK-5388] Provide a stable application submission gateway for standalone cluster mode The goal is to provide a stable, REST-based application submission gateway that is not inherently based on Akka, which is unstable across versions. This PR targets standalone cluster mode, but is implemented in a general enough manner that can be potentially extended to other modes in the future. Client mode is currently not included in the changes here because there are many more Akka messages exchanged there. As of the changes here, the Master will advertise two ports, 7077 and 6066. We need to keep around the old one (7077) for client mode and older versions of Spark submit. However, all new versions of Spark submit will use the REST gateway (6066). By the way this includes ~700 lines of tests and ~200 lines of license. Author: Andrew Or Closes #4216 from andrewor14/rest and squashes the following commits: 8d7ce07 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest 6f0c597 [Andrew Or] Use nullable fields for integer and boolean values dfe4bd7 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest b9e2a08 [Andrew Or] Minor comments 02b5cea [Andrew Or] Fix tests d2b1ef8 [Andrew Or] Comment changes + minor code refactoring across the board 9c82a36 [Andrew Or] Minor comment and wording updates b4695e7 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest c9a8ad7 [Andrew Or] Do not include appResource and mainClass as properties 6fc7670 [Andrew Or] Report REST server response back to the user 40e6095 [Andrew Or] Pass submit parameters through system properties cbd670b [Andrew Or] Include unknown fields, if any, in server response 9fee16f [Andrew Or] Include server protocol version on mismatch 09f873a [Andrew Or] Fix style 8188e61 [Andrew Or] Upgrade Jackson from 2.3.0 to 2.4.4 37538e0 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest 9165ae8 [Andrew Or] Fall back to Akka if endpoint was not REST 252d53c [Andrew Or] Clean up server error handling behavior further c643f64 [Andrew Or] Fix style bbbd329 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest 792e112 [Andrew Or] Use specific HTTP response codes on error f98660b [Andrew Or] Version the protocol and include it in REST URL 721819f [Andrew Or] Provide more REST-like interface for submit/kill/status 581f7bf [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest 9e0d1af [Andrew Or] Move some classes around to reduce number of files (minor) 42e5de4 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest 1f1c03f [Andrew Or] Use Jackson's DefaultScalaModule to simplify messages 9229433 [Andrew Or] Reduce duplicate naming in REST field ade28fd [Andrew Or] Clean up REST response output in Spark submit b2fef8b [Andrew Or] Abstract the success field to the general response 6c57b4b [Andrew Or] Increase timeout in end-to-end tests bf696ff [Andrew Or] Add checks for enabling REST when using kill/status 7ee6737 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest e2f7f5f [Andrew Or] Provide more safeguard against missing fields 9581df7 [Andrew Or] Clean up uses of exceptions 914fdff [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest e2104e6 [Andrew Or] stable -> rest 3db7379 [Andrew Or] Fix comments and name fields for better error messages 8d43486 [Andrew Or] Replace SubmitRestProtocolAction with class name df90e8b [Andrew Or] Use Jackson for JSON de/serialization d7a1f9f [Andrew Or] Fix local cluster tests efa5e18 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest e42c131 [Andrew Or] Add end-to-end tests for standalone REST protocol 837475b [Andrew Or] Show the REST port on the Master UI d8d3717 [Andrew Or] Use a daemon thread pool for REST server 6568ca5 [Andrew Or] Merge branch 'master' of github.com:apache/spark into rest 77774ba [Andrew Or] Minor fixes 206cae4 [Andrew Or] Refactor and add tests for the REST protocol 63c05b3 [Andrew Or] Remove MASTER as a field (minor) 9e21b72 [Andrew Or] Action -> SparkSubmitAction (minor) 51c5ca6 [Andrew Or] Distinguish client and server side Spark versions b44e103 [Andrew Or] Implement status requests + fix validation behavior 120ab9d [Andrew Or] Support kill and request driver status through SparkSubmit 544de1d [Andrew Or] Major clean ups in code and comments e958cae [Andrew Or] Supported nested values in messages 484bd21 [Andrew Or] Specify an ordering for fields in SubmitDriverRequestMessage 6ff088d [Andrew Or] Rename classes to generalize REST protocol af9d9cb [Andrew Or] Integrate REST protocol in standalone mode 53e7c0e [Andrew Or] Initial client, server, and all the messages --- core/pom.xml | 8 + .../org/apache/spark/ui/static/webui.css | 6 + .../scala/org/apache/spark/SparkContext.scala | 2 +- .../apache/spark/deploy/ClientArguments.scala | 20 +- .../apache/spark/deploy/DeployMessage.scala | 15 +- .../spark/deploy/LocalSparkCluster.scala | 12 +- .../org/apache/spark/deploy/SparkSubmit.scala | 142 ++++-- .../spark/deploy/SparkSubmitArguments.scala | 81 +++- .../apache/spark/deploy/master/Master.scala | 46 +- .../spark/deploy/master/MasterMessages.scala | 4 +- .../spark/deploy/master/ui/MasterPage.scala | 18 +- .../deploy/rest/StandaloneRestClient.scala | 307 ++++++++++++ .../deploy/rest/StandaloneRestServer.scala | 449 ++++++++++++++++++ .../rest/SubmitRestProtocolException.scala | 36 ++ .../rest/SubmitRestProtocolMessage.scala | 146 ++++++ .../rest/SubmitRestProtocolRequest.scala | 78 +++ .../rest/SubmitRestProtocolResponse.scala | 85 ++++ .../spark/deploy/JsonProtocolSuite.scala | 3 +- .../spark/deploy/SparkSubmitSuite.scala | 57 ++- .../rest/StandaloneRestSubmitSuite.scala | 265 +++++++++++ .../deploy/rest/SubmitRestProtocolSuite.scala | 324 +++++++++++++ .../KryoSerializerDistributedSuite.scala | 6 +- pom.xml | 11 + 23 files changed, 2027 insertions(+), 94 deletions(-) create mode 100644 core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestClient.scala create mode 100644 core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestServer.scala create mode 100644 core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolException.scala create mode 100644 core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolMessage.scala create mode 100644 core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolRequest.scala create mode 100644 core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolResponse.scala create mode 100644 core/src/test/scala/org/apache/spark/deploy/rest/StandaloneRestSubmitSuite.scala create mode 100644 core/src/test/scala/org/apache/spark/deploy/rest/SubmitRestProtocolSuite.scala diff --git a/core/pom.xml b/core/pom.xml index 4daaf88147142..66180035e61f1 100644 --- a/core/pom.xml +++ b/core/pom.xml @@ -243,6 +243,14 @@ io.dropwizard.metrics metrics-graphite + + com.fasterxml.jackson.core + jackson-databind + + + com.fasterxml.jackson.module + jackson-module-scala_2.10 + org.apache.derby derby diff --git a/core/src/main/resources/org/apache/spark/ui/static/webui.css b/core/src/main/resources/org/apache/spark/ui/static/webui.css index f23ba9dba167f..68b33b5f0d7c7 100644 --- a/core/src/main/resources/org/apache/spark/ui/static/webui.css +++ b/core/src/main/resources/org/apache/spark/ui/static/webui.css @@ -103,6 +103,12 @@ span.expand-details { float: right; } +span.rest-uri { + font-size: 10pt; + font-style: italic; + color: gray; +} + pre { font-size: 0.8em; } diff --git a/core/src/main/scala/org/apache/spark/SparkContext.scala b/core/src/main/scala/org/apache/spark/SparkContext.scala index 5623587c36fa6..71bdbc9b38ddb 100644 --- a/core/src/main/scala/org/apache/spark/SparkContext.scala +++ b/core/src/main/scala/org/apache/spark/SparkContext.scala @@ -2110,7 +2110,7 @@ object SparkContext extends Logging { val scheduler = new TaskSchedulerImpl(sc) val localCluster = new LocalSparkCluster( - numSlaves.toInt, coresPerSlave.toInt, memoryPerSlaveInt) + numSlaves.toInt, coresPerSlave.toInt, memoryPerSlaveInt, sc.conf) val masterUrls = localCluster.start() val backend = new SparkDeploySchedulerBackend(scheduler, sc, masterUrls) scheduler.initialize(backend) diff --git a/core/src/main/scala/org/apache/spark/deploy/ClientArguments.scala b/core/src/main/scala/org/apache/spark/deploy/ClientArguments.scala index e5873ce724b9f..415bd50591692 100644 --- a/core/src/main/scala/org/apache/spark/deploy/ClientArguments.scala +++ b/core/src/main/scala/org/apache/spark/deploy/ClientArguments.scala @@ -29,8 +29,7 @@ import org.apache.spark.util.{IntParam, MemoryParam} * Command-line parser for the driver client. */ private[spark] class ClientArguments(args: Array[String]) { - val defaultCores = 1 - val defaultMemory = 512 + import ClientArguments._ var cmd: String = "" // 'launch' or 'kill' var logLevel = Level.WARN @@ -39,9 +38,9 @@ private[spark] class ClientArguments(args: Array[String]) { var master: String = "" var jarUrl: String = "" var mainClass: String = "" - var supervise: Boolean = false - var memory: Int = defaultMemory - var cores: Int = defaultCores + var supervise: Boolean = DEFAULT_SUPERVISE + var memory: Int = DEFAULT_MEMORY + var cores: Int = DEFAULT_CORES private var _driverOptions = ListBuffer[String]() def driverOptions = _driverOptions.toSeq @@ -50,7 +49,7 @@ private[spark] class ClientArguments(args: Array[String]) { parse(args.toList) - def parse(args: List[String]): Unit = args match { + private def parse(args: List[String]): Unit = args match { case ("--cores" | "-c") :: IntParam(value) :: tail => cores = value parse(tail) @@ -106,9 +105,10 @@ private[spark] class ClientArguments(args: Array[String]) { |Usage: DriverClient kill | |Options: - | -c CORES, --cores CORES Number of cores to request (default: $defaultCores) - | -m MEMORY, --memory MEMORY Megabytes of memory to request (default: $defaultMemory) + | -c CORES, --cores CORES Number of cores to request (default: $DEFAULT_CORES) + | -m MEMORY, --memory MEMORY Megabytes of memory to request (default: $DEFAULT_MEMORY) | -s, --supervise Whether to restart the driver on failure + | (default: $DEFAULT_SUPERVISE) | -v, --verbose Print more debugging output """.stripMargin System.err.println(usage) @@ -117,6 +117,10 @@ private[spark] class ClientArguments(args: Array[String]) { } object ClientArguments { + private[spark] val DEFAULT_CORES = 1 + private[spark] val DEFAULT_MEMORY = 512 // MB + private[spark] val DEFAULT_SUPERVISE = false + def isValidJarUrl(s: String): Boolean = { try { val uri = new URI(s) diff --git a/core/src/main/scala/org/apache/spark/deploy/DeployMessage.scala b/core/src/main/scala/org/apache/spark/deploy/DeployMessage.scala index 243d8edb72ed3..7f600d89604a2 100644 --- a/core/src/main/scala/org/apache/spark/deploy/DeployMessage.scala +++ b/core/src/main/scala/org/apache/spark/deploy/DeployMessage.scala @@ -148,15 +148,22 @@ private[deploy] object DeployMessages { // Master to MasterWebUI - case class MasterStateResponse(host: String, port: Int, workers: Array[WorkerInfo], - activeApps: Array[ApplicationInfo], completedApps: Array[ApplicationInfo], - activeDrivers: Array[DriverInfo], completedDrivers: Array[DriverInfo], - status: MasterState) { + case class MasterStateResponse( + host: String, + port: Int, + restPort: Option[Int], + workers: Array[WorkerInfo], + activeApps: Array[ApplicationInfo], + completedApps: Array[ApplicationInfo], + activeDrivers: Array[DriverInfo], + completedDrivers: Array[DriverInfo], + status: MasterState) { Utils.checkHost(host, "Required hostname") assert (port > 0) def uri = "spark://" + host + ":" + port + def restUri: Option[String] = restPort.map { p => "spark://" + host + ":" + p } } // WorkerWebUI to Worker diff --git a/core/src/main/scala/org/apache/spark/deploy/LocalSparkCluster.scala b/core/src/main/scala/org/apache/spark/deploy/LocalSparkCluster.scala index 9a7a113c95715..0401b15446a7b 100644 --- a/core/src/main/scala/org/apache/spark/deploy/LocalSparkCluster.scala +++ b/core/src/main/scala/org/apache/spark/deploy/LocalSparkCluster.scala @@ -33,7 +33,11 @@ import org.apache.spark.util.Utils * fault recovery without spinning up a lot of processes. */ private[spark] -class LocalSparkCluster(numWorkers: Int, coresPerWorker: Int, memoryPerWorker: Int) +class LocalSparkCluster( + numWorkers: Int, + coresPerWorker: Int, + memoryPerWorker: Int, + conf: SparkConf) extends Logging { private val localHostname = Utils.localHostName() @@ -43,9 +47,11 @@ class LocalSparkCluster(numWorkers: Int, coresPerWorker: Int, memoryPerWorker: I def start(): Array[String] = { logInfo("Starting a local Spark cluster with " + numWorkers + " workers.") + // Disable REST server on Master in this mode unless otherwise specified + val _conf = conf.clone().setIfMissing("spark.master.rest.enabled", "false") + /* Start the Master */ - val conf = new SparkConf(false) - val (masterSystem, masterPort, _) = Master.startSystemAndActor(localHostname, 0, 0, conf) + val (masterSystem, masterPort, _, _) = Master.startSystemAndActor(localHostname, 0, 0, _conf) masterActorSystems += masterSystem val masterUrl = "spark://" + localHostname + ":" + masterPort val masters = Array(masterUrl) diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala index 9d25e647f1703..6d213926f3d7b 100644 --- a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala +++ b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala @@ -18,25 +18,35 @@ package org.apache.spark.deploy import java.io.{File, PrintStream} -import java.lang.reflect.{Modifier, InvocationTargetException} +import java.lang.reflect.{InvocationTargetException, Modifier} import java.net.URL + import scala.collection.mutable.{ArrayBuffer, HashMap, Map} import org.apache.hadoop.fs.Path import org.apache.ivy.Ivy import org.apache.ivy.core.LogOptions -import org.apache.ivy.core.module.descriptor.{DefaultExcludeRule, DefaultDependencyDescriptor, DefaultModuleDescriptor} -import org.apache.ivy.core.module.id.{ModuleId, ArtifactId, ModuleRevisionId} +import org.apache.ivy.core.module.descriptor._ +import org.apache.ivy.core.module.id.{ArtifactId, ModuleId, ModuleRevisionId} import org.apache.ivy.core.report.ResolveReport -import org.apache.ivy.core.resolve.{IvyNode, ResolveOptions} +import org.apache.ivy.core.resolve.ResolveOptions import org.apache.ivy.core.retrieve.RetrieveOptions import org.apache.ivy.core.settings.IvySettings import org.apache.ivy.plugins.matcher.GlobPatternMatcher import org.apache.ivy.plugins.resolver.{ChainResolver, IBiblioResolver} -import org.apache.spark.executor.ExecutorURLClassLoader + +import org.apache.spark.deploy.rest._ +import org.apache.spark.executor._ import org.apache.spark.util.Utils -import org.apache.spark.executor.ChildExecutorURLClassLoader -import org.apache.spark.executor.MutableURLClassLoader + +/** + * Whether to submit, kill, or request the status of an application. + * The latter two operations are currently supported only for standalone cluster mode. + */ +private[spark] object SparkSubmitAction extends Enumeration { + type SparkSubmitAction = Value + val SUBMIT, KILL, REQUEST_STATUS = Value +} /** * Main gateway of launching a Spark application. @@ -83,21 +93,74 @@ object SparkSubmit { if (appArgs.verbose) { printStream.println(appArgs) } - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) - launch(childArgs, classpath, sysProps, mainClass, appArgs.verbose) + appArgs.action match { + case SparkSubmitAction.SUBMIT => submit(appArgs) + case SparkSubmitAction.KILL => kill(appArgs) + case SparkSubmitAction.REQUEST_STATUS => requestStatus(appArgs) + } + } + + /** Kill an existing submission using the REST protocol. Standalone cluster mode only. */ + private def kill(args: SparkSubmitArguments): Unit = { + new StandaloneRestClient() + .killSubmission(args.master, args.submissionToKill) } /** - * @return a tuple containing - * (1) the arguments for the child process, - * (2) a list of classpath entries for the child, - * (3) a list of system properties and env vars, and - * (4) the main class for the child + * Request the status of an existing submission using the REST protocol. + * Standalone cluster mode only. */ - private[spark] def createLaunchEnv(args: SparkSubmitArguments) - : (ArrayBuffer[String], ArrayBuffer[String], Map[String, String], String) = { + private def requestStatus(args: SparkSubmitArguments): Unit = { + new StandaloneRestClient() + .requestSubmissionStatus(args.master, args.submissionToRequestStatusFor) + } - // Values to return + /** + * Submit the application using the provided parameters. + * + * This runs in two steps. First, we prepare the launch environment by setting up + * the appropriate classpath, system properties, and application arguments for + * running the child main class based on the cluster manager and the deploy mode. + * Second, we use this launch environment to invoke the main method of the child + * main class. + */ + private[spark] def submit(args: SparkSubmitArguments): Unit = { + val (childArgs, childClasspath, sysProps, childMainClass) = prepareSubmitEnvironment(args) + // In standalone cluster mode, there are two submission gateways: + // (1) The traditional Akka gateway using o.a.s.deploy.Client as a wrapper + // (2) The new REST-based gateway introduced in Spark 1.3 + // The latter is the default behavior as of Spark 1.3, but Spark submit will fail over + // to use the legacy gateway if the master endpoint turns out to be not a REST server. + if (args.isStandaloneCluster && args.useRest) { + try { + printStream.println("Running Spark using the REST application submission protocol.") + runMain(childArgs, childClasspath, sysProps, childMainClass) + } catch { + // Fail over to use the legacy submission gateway + case e: SubmitRestConnectionException => + printWarning(s"Master endpoint ${args.master} was not a REST server. " + + "Falling back to legacy submission gateway instead.") + args.useRest = false + submit(args) + } + // In all other modes, just run the main class as prepared + } else { + runMain(childArgs, childClasspath, sysProps, childMainClass) + } + } + + /** + * Prepare the environment for submitting an application. + * This returns a 4-tuple: + * (1) the arguments for the child process, + * (2) a list of classpath entries for the child, + * (3) a map of system properties, and + * (4) the main class for the child + * Exposed for testing. + */ + private[spark] def prepareSubmitEnvironment(args: SparkSubmitArguments) + : (Seq[String], Seq[String], Map[String, String], String) = { + // Return values val childArgs = new ArrayBuffer[String]() val childClasspath = new ArrayBuffer[String]() val sysProps = new HashMap[String, String]() @@ -235,10 +298,13 @@ object SparkSubmit { sysProp = "spark.driver.extraLibraryPath"), // Standalone cluster only + // Do not set CL arguments here because there are multiple possibilities for the main class OptionAssigner(args.jars, STANDALONE, CLUSTER, sysProp = "spark.jars"), OptionAssigner(args.ivyRepoPath, STANDALONE, CLUSTER, sysProp = "spark.jars.ivy"), - OptionAssigner(args.driverMemory, STANDALONE, CLUSTER, clOption = "--memory"), - OptionAssigner(args.driverCores, STANDALONE, CLUSTER, clOption = "--cores"), + OptionAssigner(args.driverMemory, STANDALONE, CLUSTER, sysProp = "spark.driver.memory"), + OptionAssigner(args.driverCores, STANDALONE, CLUSTER, sysProp = "spark.driver.cores"), + OptionAssigner(args.supervise.toString, STANDALONE, CLUSTER, + sysProp = "spark.driver.supervise"), // Yarn client only OptionAssigner(args.queue, YARN, CLIENT, sysProp = "spark.yarn.queue"), @@ -279,7 +345,6 @@ object SparkSubmit { if (args.childArgs != null) { childArgs ++= args.childArgs } } - // Map all arguments to command-line options or system properties for our chosen mode for (opt <- options) { if (opt.value != null && @@ -301,14 +366,21 @@ object SparkSubmit { sysProps.put("spark.jars", jars.mkString(",")) } - // In standalone-cluster mode, use Client as a wrapper around the user class - if (clusterManager == STANDALONE && deployMode == CLUSTER) { - childMainClass = "org.apache.spark.deploy.Client" - if (args.supervise) { - childArgs += "--supervise" + // In standalone cluster mode, use the REST client to submit the application (Spark 1.3+). + // All Spark parameters are expected to be passed to the client through system properties. + if (args.isStandaloneCluster) { + if (args.useRest) { + childMainClass = "org.apache.spark.deploy.rest.StandaloneRestClient" + childArgs += (args.primaryResource, args.mainClass) + } else { + // In legacy standalone cluster mode, use Client as a wrapper around the user class + childMainClass = "org.apache.spark.deploy.Client" + if (args.supervise) { childArgs += "--supervise" } + Option(args.driverMemory).foreach { m => childArgs += ("--memory", m) } + Option(args.driverCores).foreach { c => childArgs += ("--cores", c) } + childArgs += "launch" + childArgs += (args.master, args.primaryResource, args.mainClass) } - childArgs += "launch" - childArgs += (args.master, args.primaryResource, args.mainClass) if (args.childArgs != null) { childArgs ++= args.childArgs } @@ -345,7 +417,7 @@ object SparkSubmit { // Ignore invalid spark.driver.host in cluster modes. if (deployMode == CLUSTER) { - sysProps -= ("spark.driver.host") + sysProps -= "spark.driver.host" } // Resolve paths in certain spark properties @@ -374,9 +446,15 @@ object SparkSubmit { (childArgs, childClasspath, sysProps, childMainClass) } - private def launch( - childArgs: ArrayBuffer[String], - childClasspath: ArrayBuffer[String], + /** + * Run the main method of the child class using the provided launch environment. + * + * Note that this main class will not be the one provided by the user if we're + * running cluster deploy mode or python applications. + */ + private def runMain( + childArgs: Seq[String], + childClasspath: Seq[String], sysProps: Map[String, String], childMainClass: String, verbose: Boolean = false) { @@ -697,7 +775,7 @@ private[spark] object SparkSubmitUtils { * Provides an indirection layer for passing arguments as system properties or flags to * the user's driver program or to downstream launcher tools. */ -private[spark] case class OptionAssigner( +private case class OptionAssigner( value: String, clusterManager: Int, deployMode: Int, diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala b/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala index 5cadc534f4baa..bd0ae26fd8210 100644 --- a/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala +++ b/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala @@ -22,6 +22,7 @@ import java.util.jar.JarFile import scala.collection.mutable.{ArrayBuffer, HashMap} +import org.apache.spark.deploy.SparkSubmitAction._ import org.apache.spark.util.Utils /** @@ -39,8 +40,6 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St var driverExtraClassPath: String = null var driverExtraLibraryPath: String = null var driverExtraJavaOptions: String = null - var driverCores: String = null - var supervise: Boolean = false var queue: String = null var numExecutors: String = null var files: String = null @@ -56,8 +55,16 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St var verbose: Boolean = false var isPython: Boolean = false var pyFiles: String = null + var action: SparkSubmitAction = null val sparkProperties: HashMap[String, String] = new HashMap[String, String]() + // Standalone cluster mode only + var supervise: Boolean = false + var driverCores: String = null + var submissionToKill: String = null + var submissionToRequestStatusFor: String = null + var useRest: Boolean = true // used internally + /** Default properties present in the currently defined defaults file. */ lazy val defaultSparkProperties: HashMap[String, String] = { val defaultProperties = new HashMap[String, String]() @@ -82,7 +89,7 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St // Use `sparkProperties` map along with env vars to fill in any missing parameters loadEnvironmentArguments() - checkRequiredArguments() + validateArguments() /** * Merge values from the default properties file with those specified through --conf. @@ -107,6 +114,15 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St .orElse(sparkProperties.get("spark.master")) .orElse(env.get("MASTER")) .orNull + driverExtraClassPath = Option(driverExtraClassPath) + .orElse(sparkProperties.get("spark.driver.extraClassPath")) + .orNull + driverExtraJavaOptions = Option(driverExtraJavaOptions) + .orElse(sparkProperties.get("spark.driver.extraJavaOptions")) + .orNull + driverExtraLibraryPath = Option(driverExtraLibraryPath) + .orElse(sparkProperties.get("spark.driver.extraLibraryPath")) + .orNull driverMemory = Option(driverMemory) .orElse(sparkProperties.get("spark.driver.memory")) .orElse(env.get("SPARK_DRIVER_MEMORY")) @@ -166,10 +182,21 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St if (name == null && primaryResource != null) { name = Utils.stripDirectory(primaryResource) } + + // Action should be SUBMIT unless otherwise specified + action = Option(action).getOrElse(SUBMIT) } /** Ensure that required fields exists. Call this only once all defaults are loaded. */ - private def checkRequiredArguments(): Unit = { + private def validateArguments(): Unit = { + action match { + case SUBMIT => validateSubmitArguments() + case KILL => validateKillArguments() + case REQUEST_STATUS => validateStatusRequestArguments() + } + } + + private def validateSubmitArguments(): Unit = { if (args.length == 0) { printUsageAndExit(-1) } @@ -192,6 +219,29 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St } } + private def validateKillArguments(): Unit = { + if (!master.startsWith("spark://")) { + SparkSubmit.printErrorAndExit("Killing submissions is only supported in standalone mode!") + } + if (submissionToKill == null) { + SparkSubmit.printErrorAndExit("Please specify a submission to kill.") + } + } + + private def validateStatusRequestArguments(): Unit = { + if (!master.startsWith("spark://")) { + SparkSubmit.printErrorAndExit( + "Requesting submission statuses is only supported in standalone mode!") + } + if (submissionToRequestStatusFor == null) { + SparkSubmit.printErrorAndExit("Please specify a submission to request status for.") + } + } + + def isStandaloneCluster: Boolean = { + master.startsWith("spark://") && deployMode == "cluster" + } + override def toString = { s"""Parsed arguments: | master $master @@ -300,6 +350,22 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St propertiesFile = value parse(tail) + case ("--kill") :: value :: tail => + submissionToKill = value + if (action != null) { + SparkSubmit.printErrorAndExit(s"Action cannot be both $action and $KILL.") + } + action = KILL + parse(tail) + + case ("--status") :: value :: tail => + submissionToRequestStatusFor = value + if (action != null) { + SparkSubmit.printErrorAndExit(s"Action cannot be both $action and $REQUEST_STATUS.") + } + action = REQUEST_STATUS + parse(tail) + case ("--supervise") :: tail => supervise = true parse(tail) @@ -372,7 +438,10 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St outStream.println("Unknown/unsupported param " + unknownParam) } outStream.println( - """Usage: spark-submit [options] [app options] + """Usage: spark-submit [options] [app arguments] + |Usage: spark-submit --kill [submission ID] --master [spark://...] + |Usage: spark-submit --status [submission ID] --master [spark://...] + | |Options: | --master MASTER_URL spark://host:port, mesos://host:port, yarn, or local. | --deploy-mode DEPLOY_MODE Whether to launch the driver program locally ("client") or @@ -413,6 +482,8 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St | Spark standalone with cluster deploy mode only: | --driver-cores NUM Cores for driver (Default: 1). | --supervise If given, restarts the driver on failure. + | --kill SUBMISSION_ID If given, kills the driver specified. + | --status SUBMISSION_ID If given, requests the status of the driver specified. | | Spark standalone and Mesos only: | --total-executor-cores NUM Total cores for all executors. diff --git a/core/src/main/scala/org/apache/spark/deploy/master/Master.scala b/core/src/main/scala/org/apache/spark/deploy/master/Master.scala index 5eeb9fe526248..b8b1a25abff2e 100644 --- a/core/src/main/scala/org/apache/spark/deploy/master/Master.scala +++ b/core/src/main/scala/org/apache/spark/deploy/master/Master.scala @@ -43,6 +43,7 @@ import org.apache.spark.deploy.history.HistoryServer import org.apache.spark.deploy.master.DriverState.DriverState import org.apache.spark.deploy.master.MasterMessages._ import org.apache.spark.deploy.master.ui.MasterWebUI +import org.apache.spark.deploy.rest.StandaloneRestServer import org.apache.spark.metrics.MetricsSystem import org.apache.spark.scheduler.{EventLoggingListener, ReplayListenerBus} import org.apache.spark.ui.SparkUI @@ -52,12 +53,12 @@ private[spark] class Master( host: String, port: Int, webUiPort: Int, - val securityMgr: SecurityManager) + val securityMgr: SecurityManager, + val conf: SparkConf) extends Actor with ActorLogReceive with Logging with LeaderElectable { import context.dispatcher // to use Akka's scheduler.schedule() - val conf = new SparkConf val hadoopConf = SparkHadoopUtil.get.newConfiguration(conf) def createDateFormat = new SimpleDateFormat("yyyyMMddHHmmss") // For application IDs @@ -121,6 +122,17 @@ private[spark] class Master( throw new SparkException("spark.deploy.defaultCores must be positive") } + // Alternative application submission gateway that is stable across Spark versions + private val restServerEnabled = conf.getBoolean("spark.master.rest.enabled", true) + private val restServer = + if (restServerEnabled) { + val port = conf.getInt("spark.master.rest.port", 6066) + Some(new StandaloneRestServer(host, port, self, masterUrl, conf)) + } else { + None + } + private val restServerBoundPort = restServer.map(_.start()) + override def preStart() { logInfo("Starting Spark master at " + masterUrl) logInfo(s"Running Spark version ${org.apache.spark.SPARK_VERSION}") @@ -174,6 +186,7 @@ private[spark] class Master( recoveryCompletionTask.cancel() } webUi.stop() + restServer.foreach(_.stop()) masterMetricsSystem.stop() applicationMetricsSystem.stop() persistenceEngine.close() @@ -421,7 +434,9 @@ private[spark] class Master( } case RequestMasterState => { - sender ! MasterStateResponse(host, port, workers.toArray, apps.toArray, completedApps.toArray, + sender ! MasterStateResponse( + host, port, restServerBoundPort, + workers.toArray, apps.toArray, completedApps.toArray, drivers.toArray, completedDrivers.toArray, state) } @@ -429,8 +444,8 @@ private[spark] class Master( timeOutDeadWorkers() } - case RequestWebUIPort => { - sender ! WebUIPortResponse(webUi.boundPort) + case BoundPortsRequest => { + sender ! BoundPortsResponse(port, webUi.boundPort, restServerBoundPort) } } @@ -851,7 +866,7 @@ private[spark] object Master extends Logging { SignalLogger.register(log) val conf = new SparkConf val args = new MasterArguments(argStrings, conf) - val (actorSystem, _, _) = startSystemAndActor(args.host, args.port, args.webUiPort, conf) + val (actorSystem, _, _, _) = startSystemAndActor(args.host, args.port, args.webUiPort, conf) actorSystem.awaitTermination() } @@ -875,19 +890,26 @@ private[spark] object Master extends Logging { Address(protocol, systemName, host, port) } + /** + * Start the Master and return a four tuple of: + * (1) The Master actor system + * (2) The bound port + * (3) The web UI bound port + * (4) The REST server bound port, if any + */ def startSystemAndActor( host: String, port: Int, webUiPort: Int, - conf: SparkConf): (ActorSystem, Int, Int) = { + conf: SparkConf): (ActorSystem, Int, Int, Option[Int]) = { val securityMgr = new SecurityManager(conf) val (actorSystem, boundPort) = AkkaUtils.createActorSystem(systemName, host, port, conf = conf, securityManager = securityMgr) - val actor = actorSystem.actorOf(Props(classOf[Master], host, boundPort, webUiPort, - securityMgr), actorName) + val actor = actorSystem.actorOf( + Props(classOf[Master], host, boundPort, webUiPort, securityMgr, conf), actorName) val timeout = AkkaUtils.askTimeout(conf) - val respFuture = actor.ask(RequestWebUIPort)(timeout) - val resp = Await.result(respFuture, timeout).asInstanceOf[WebUIPortResponse] - (actorSystem, boundPort, resp.webUIBoundPort) + val portsRequest = actor.ask(BoundPortsRequest)(timeout) + val portsResponse = Await.result(portsRequest, timeout).asInstanceOf[BoundPortsResponse] + (actorSystem, boundPort, portsResponse.webUIPort, portsResponse.restPort) } } diff --git a/core/src/main/scala/org/apache/spark/deploy/master/MasterMessages.scala b/core/src/main/scala/org/apache/spark/deploy/master/MasterMessages.scala index db72d8ae9bdaf..15c6296888f70 100644 --- a/core/src/main/scala/org/apache/spark/deploy/master/MasterMessages.scala +++ b/core/src/main/scala/org/apache/spark/deploy/master/MasterMessages.scala @@ -36,7 +36,7 @@ private[master] object MasterMessages { case object CompleteRecovery - case object RequestWebUIPort + case object BoundPortsRequest - case class WebUIPortResponse(webUIBoundPort: Int) + case class BoundPortsResponse(actorPort: Int, webUIPort: Int, restPort: Option[Int]) } diff --git a/core/src/main/scala/org/apache/spark/deploy/master/ui/MasterPage.scala b/core/src/main/scala/org/apache/spark/deploy/master/ui/MasterPage.scala index 7ca3b08a28728..b47a081053e77 100644 --- a/core/src/main/scala/org/apache/spark/deploy/master/ui/MasterPage.scala +++ b/core/src/main/scala/org/apache/spark/deploy/master/ui/MasterPage.scala @@ -46,19 +46,19 @@ private[spark] class MasterPage(parent: MasterWebUI) extends WebUIPage("") { val stateFuture = (master ? RequestMasterState)(timeout).mapTo[MasterStateResponse] val state = Await.result(stateFuture, timeout) - val workerHeaders = Seq("Id", "Address", "State", "Cores", "Memory") + val workerHeaders = Seq("Worker Id", "Address", "State", "Cores", "Memory") val workers = state.workers.sortBy(_.id) val workerTable = UIUtils.listingTable(workerHeaders, workerRow, workers) - val appHeaders = Seq("ID", "Name", "Cores", "Memory per Node", "Submitted Time", "User", - "State", "Duration") + val appHeaders = Seq("Application ID", "Name", "Cores", "Memory per Node", "Submitted Time", + "User", "State", "Duration") val activeApps = state.activeApps.sortBy(_.startTime).reverse val activeAppsTable = UIUtils.listingTable(appHeaders, appRow, activeApps) val completedApps = state.completedApps.sortBy(_.endTime).reverse val completedAppsTable = UIUtils.listingTable(appHeaders, appRow, completedApps) - val driverHeaders = Seq("ID", "Submitted Time", "Worker", "State", "Cores", "Memory", - "Main Class") + val driverHeaders = Seq("Submission ID", "Submitted Time", "Worker", "State", "Cores", + "Memory", "Main Class") val activeDrivers = state.activeDrivers.sortBy(_.startTime).reverse val activeDriversTable = UIUtils.listingTable(driverHeaders, driverRow, activeDrivers) val completedDrivers = state.completedDrivers.sortBy(_.startTime).reverse @@ -73,6 +73,14 @@ private[spark] class MasterPage(parent: MasterWebUI) extends WebUIPage("") {
    • URL: {state.uri}
      • + { + state.restUri.map { uri => +
    • + REST URL: {uri} + (cluster mode) +
      • + }.getOrElse { Seq.empty } + }
    • Workers: {state.workers.size}
    • Cores: {state.workers.map(_.cores).sum} Total, {state.workers.map(_.coresUsed).sum} Used
      • diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestClient.scala b/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestClient.scala new file mode 100644 index 0000000000000..115aa5278bb62 --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestClient.scala @@ -0,0 +1,307 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.io.{DataOutputStream, FileNotFoundException} +import java.net.{HttpURLConnection, SocketException, URL} + +import scala.io.Source + +import com.fasterxml.jackson.databind.JsonMappingException +import com.google.common.base.Charsets + +import org.apache.spark.{Logging, SparkConf, SPARK_VERSION => sparkVersion} + +/** + * A client that submits applications to the standalone Master using a REST protocol. + * This client is intended to communicate with the [[StandaloneRestServer]] and is + * currently used for cluster mode only. + * + * In protocol version v1, the REST URL takes the form http://[host:port]/v1/submissions/[action], + * where [action] can be one of create, kill, or status. Each type of request is represented in + * an HTTP message sent to the following prefixes: + * (1) submit - POST to /submissions/create + * (2) kill - POST /submissions/kill/[submissionId] + * (3) status - GET /submissions/status/[submissionId] + * + * In the case of (1), parameters are posted in the HTTP body in the form of JSON fields. + * Otherwise, the URL fully specifies the intended action of the client. + * + * Since the protocol is expected to be stable across Spark versions, existing fields cannot be + * added or removed, though new optional fields can be added. In the rare event that forward or + * backward compatibility is broken, Spark must introduce a new protocol version (e.g. v2). + * + * The client and the server must communicate using the same version of the protocol. If there + * is a mismatch, the server will respond with the highest protocol version it supports. A future + * implementation of this client can use that information to retry using the version specified + * by the server. + */ +private[spark] class StandaloneRestClient extends Logging { + import StandaloneRestClient._ + + /** + * Submit an application specified by the parameters in the provided request. + * + * If the submission was successful, poll the status of the submission and report + * it to the user. Otherwise, report the error message provided by the server. + */ + def createSubmission( + master: String, + request: CreateSubmissionRequest): SubmitRestProtocolResponse = { + logInfo(s"Submitting a request to launch an application in $master.") + validateMaster(master) + val url = getSubmitUrl(master) + val response = postJson(url, request.toJson) + response match { + case s: CreateSubmissionResponse => + reportSubmissionStatus(master, s) + handleRestResponse(s) + case unexpected => + handleUnexpectedRestResponse(unexpected) + } + response + } + + /** Request that the server kill the specified submission. */ + def killSubmission(master: String, submissionId: String): SubmitRestProtocolResponse = { + logInfo(s"Submitting a request to kill submission $submissionId in $master.") + validateMaster(master) + val response = post(getKillUrl(master, submissionId)) + response match { + case k: KillSubmissionResponse => handleRestResponse(k) + case unexpected => handleUnexpectedRestResponse(unexpected) + } + response + } + + /** Request the status of a submission from the server. */ + def requestSubmissionStatus( + master: String, + submissionId: String, + quiet: Boolean = false): SubmitRestProtocolResponse = { + logInfo(s"Submitting a request for the status of submission $submissionId in $master.") + validateMaster(master) + val response = get(getStatusUrl(master, submissionId)) + response match { + case s: SubmissionStatusResponse => if (!quiet) { handleRestResponse(s) } + case unexpected => handleUnexpectedRestResponse(unexpected) + } + response + } + + /** Construct a message that captures the specified parameters for submitting an application. */ + def constructSubmitRequest( + appResource: String, + mainClass: String, + appArgs: Array[String], + sparkProperties: Map[String, String], + environmentVariables: Map[String, String]): CreateSubmissionRequest = { + val message = new CreateSubmissionRequest + message.clientSparkVersion = sparkVersion + message.appResource = appResource + message.mainClass = mainClass + message.appArgs = appArgs + message.sparkProperties = sparkProperties + message.environmentVariables = environmentVariables + message.validate() + message + } + + /** Send a GET request to the specified URL. */ + private def get(url: URL): SubmitRestProtocolResponse = { + logDebug(s"Sending GET request to server at $url.") + val conn = url.openConnection().asInstanceOf[HttpURLConnection] + conn.setRequestMethod("GET") + readResponse(conn) + } + + /** Send a POST request to the specified URL. */ + private def post(url: URL): SubmitRestProtocolResponse = { + logDebug(s"Sending POST request to server at $url.") + val conn = url.openConnection().asInstanceOf[HttpURLConnection] + conn.setRequestMethod("POST") + readResponse(conn) + } + + /** Send a POST request with the given JSON as the body to the specified URL. */ + private def postJson(url: URL, json: String): SubmitRestProtocolResponse = { + logDebug(s"Sending POST request to server at $url:\n$json") + val conn = url.openConnection().asInstanceOf[HttpURLConnection] + conn.setRequestMethod("POST") + conn.setRequestProperty("Content-Type", "application/json") + conn.setRequestProperty("charset", "utf-8") + conn.setDoOutput(true) + val out = new DataOutputStream(conn.getOutputStream) + out.write(json.getBytes(Charsets.UTF_8)) + out.close() + readResponse(conn) + } + + /** + * Read the response from the server and return it as a validated [[SubmitRestProtocolResponse]]. + * If the response represents an error, report the embedded message to the user. + */ + private def readResponse(connection: HttpURLConnection): SubmitRestProtocolResponse = { + try { + val responseJson = Source.fromInputStream(connection.getInputStream).mkString + logDebug(s"Response from the server:\n$responseJson") + val response = SubmitRestProtocolMessage.fromJson(responseJson) + response.validate() + response match { + // If the response is an error, log the message + case error: ErrorResponse => + logError(s"Server responded with error:\n${error.message}") + error + // Otherwise, simply return the response + case response: SubmitRestProtocolResponse => response + case unexpected => + throw new SubmitRestProtocolException( + s"Message received from server was not a response:\n${unexpected.toJson}") + } + } catch { + case unreachable @ (_: FileNotFoundException | _: SocketException) => + throw new SubmitRestConnectionException( + s"Unable to connect to server ${connection.getURL}", unreachable) + case malformed @ (_: SubmitRestProtocolException | _: JsonMappingException) => + throw new SubmitRestProtocolException( + "Malformed response received from server", malformed) + } + } + + /** Return the REST URL for creating a new submission. */ + private def getSubmitUrl(master: String): URL = { + val baseUrl = getBaseUrl(master) + new URL(s"$baseUrl/create") + } + + /** Return the REST URL for killing an existing submission. */ + private def getKillUrl(master: String, submissionId: String): URL = { + val baseUrl = getBaseUrl(master) + new URL(s"$baseUrl/kill/$submissionId") + } + + /** Return the REST URL for requesting the status of an existing submission. */ + private def getStatusUrl(master: String, submissionId: String): URL = { + val baseUrl = getBaseUrl(master) + new URL(s"$baseUrl/status/$submissionId") + } + + /** Return the base URL for communicating with the server, including the protocol version. */ + private def getBaseUrl(master: String): String = { + val masterUrl = master.stripPrefix("spark://").stripSuffix("/") + s"http://$masterUrl/$PROTOCOL_VERSION/submissions" + } + + /** Throw an exception if this is not standalone mode. */ + private def validateMaster(master: String): Unit = { + if (!master.startsWith("spark://")) { + throw new IllegalArgumentException("This REST client is only supported in standalone mode.") + } + } + + /** Report the status of a newly created submission. */ + private def reportSubmissionStatus( + master: String, + submitResponse: CreateSubmissionResponse): Unit = { + if (submitResponse.success) { + val submissionId = submitResponse.submissionId + if (submissionId != null) { + logInfo(s"Submission successfully created as $submissionId. Polling submission state...") + pollSubmissionStatus(master, submissionId) + } else { + // should never happen + logError("Application successfully submitted, but submission ID was not provided!") + } + } else { + val failMessage = Option(submitResponse.message).map { ": " + _ }.getOrElse("") + logError("Application submission failed" + failMessage) + } + } + + /** + * Poll the status of the specified submission and log it. + * This retries up to a fixed number of times before giving up. + */ + private def pollSubmissionStatus(master: String, submissionId: String): Unit = { + (1 to REPORT_DRIVER_STATUS_MAX_TRIES).foreach { _ => + val response = requestSubmissionStatus(master, submissionId, quiet = true) + val statusResponse = response match { + case s: SubmissionStatusResponse => s + case _ => return // unexpected type, let upstream caller handle it + } + if (statusResponse.success) { + val driverState = Option(statusResponse.driverState) + val workerId = Option(statusResponse.workerId) + val workerHostPort = Option(statusResponse.workerHostPort) + val exception = Option(statusResponse.message) + // Log driver state, if present + driverState match { + case Some(state) => logInfo(s"State of driver $submissionId is now $state.") + case _ => logError(s"State of driver $submissionId was not found!") + } + // Log worker node, if present + (workerId, workerHostPort) match { + case (Some(id), Some(hp)) => logInfo(s"Driver is running on worker $id at $hp.") + case _ => + } + // Log exception stack trace, if present + exception.foreach { e => logError(e) } + return + } + Thread.sleep(REPORT_DRIVER_STATUS_INTERVAL) + } + logError(s"Error: Master did not recognize driver $submissionId.") + } + + /** Log the response sent by the server in the REST application submission protocol. */ + private def handleRestResponse(response: SubmitRestProtocolResponse): Unit = { + logInfo(s"Server responded with ${response.messageType}:\n${response.toJson}") + } + + /** Log an appropriate error if the response sent by the server is not of the expected type. */ + private def handleUnexpectedRestResponse(unexpected: SubmitRestProtocolResponse): Unit = { + logError(s"Error: Server responded with message of unexpected type ${unexpected.messageType}.") + } +} + +private[spark] object StandaloneRestClient { + val REPORT_DRIVER_STATUS_INTERVAL = 1000 + val REPORT_DRIVER_STATUS_MAX_TRIES = 10 + val PROTOCOL_VERSION = "v1" + + /** Submit an application, assuming Spark parameters are specified through system properties. */ + def main(args: Array[String]): Unit = { + if (args.size < 2) { + sys.error("Usage: StandaloneRestClient [app resource] [main class] [app args*]") + sys.exit(1) + } + val appResource = args(0) + val mainClass = args(1) + val appArgs = args.slice(2, args.size) + val conf = new SparkConf + val master = conf.getOption("spark.master").getOrElse { + throw new IllegalArgumentException("'spark.master' must be set.") + } + val sparkProperties = conf.getAll.toMap + val environmentVariables = sys.env.filter { case (k, _) => k.startsWith("SPARK_") } + val client = new StandaloneRestClient + val submitRequest = client.constructSubmitRequest( + appResource, mainClass, appArgs, sparkProperties, environmentVariables) + client.createSubmission(master, submitRequest) + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestServer.scala b/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestServer.scala new file mode 100644 index 0000000000000..2033d67e1f394 --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestServer.scala @@ -0,0 +1,449 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +* See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.io.{DataOutputStream, File} +import java.net.InetSocketAddress +import javax.servlet.http.{HttpServlet, HttpServletRequest, HttpServletResponse} + +import scala.io.Source + +import akka.actor.ActorRef +import com.fasterxml.jackson.databind.JsonMappingException +import com.google.common.base.Charsets +import org.eclipse.jetty.server.Server +import org.eclipse.jetty.servlet.{ServletHolder, ServletContextHandler} +import org.eclipse.jetty.util.thread.QueuedThreadPool +import org.json4s._ +import org.json4s.jackson.JsonMethods._ + +import org.apache.spark.{Logging, SparkConf, SPARK_VERSION => sparkVersion} +import org.apache.spark.util.{AkkaUtils, Utils} +import org.apache.spark.deploy.{Command, DeployMessages, DriverDescription} +import org.apache.spark.deploy.ClientArguments._ + +/** + * A server that responds to requests submitted by the [[StandaloneRestClient]]. + * This is intended to be embedded in the standalone Master and used in cluster mode only. + * + * This server responds with different HTTP codes depending on the situation: + * 200 OK - Request was processed successfully + * 400 BAD REQUEST - Request was malformed, not successfully validated, or of unexpected type + * 468 UNKNOWN PROTOCOL VERSION - Request specified a protocol this server does not understand + * 500 INTERNAL SERVER ERROR - Server throws an exception internally while processing the request + * + * The server always includes a JSON representation of the relevant [[SubmitRestProtocolResponse]] + * in the HTTP body. If an error occurs, however, the server will include an [[ErrorResponse]] + * instead of the one expected by the client. If the construction of this error response itself + * fails, the response will consist of an empty body with a response code that indicates internal + * server error. + * + * @param host the address this server should bind to + * @param requestedPort the port this server will attempt to bind to + * @param masterActor reference to the Master actor to which requests can be sent + * @param masterUrl the URL of the Master new drivers will attempt to connect to + * @param masterConf the conf used by the Master + */ +private[spark] class StandaloneRestServer( + host: String, + requestedPort: Int, + masterActor: ActorRef, + masterUrl: String, + masterConf: SparkConf) + extends Logging { + + import StandaloneRestServer._ + + private var _server: Option[Server] = None + private val baseContext = s"/$PROTOCOL_VERSION/submissions" + + // A mapping from servlets to the URL prefixes they are responsible for + private val servletToContext = Map[StandaloneRestServlet, String]( + new SubmitRequestServlet(masterActor, masterUrl, masterConf) -> s"$baseContext/create/*", + new KillRequestServlet(masterActor, masterConf) -> s"$baseContext/kill/*", + new StatusRequestServlet(masterActor, masterConf) -> s"$baseContext/status/*", + new ErrorServlet -> "/" // default handler + ) + + /** Start the server and return the bound port. */ + def start(): Int = { + val (server, boundPort) = Utils.startServiceOnPort[Server](requestedPort, doStart, masterConf) + _server = Some(server) + logInfo(s"Started REST server for submitting applications on port $boundPort") + boundPort + } + + /** + * Map the servlets to their corresponding contexts and attach them to a server. + * Return a 2-tuple of the started server and the bound port. + */ + private def doStart(startPort: Int): (Server, Int) = { + val server = new Server(new InetSocketAddress(host, startPort)) + val threadPool = new QueuedThreadPool + threadPool.setDaemon(true) + server.setThreadPool(threadPool) + val mainHandler = new ServletContextHandler + mainHandler.setContextPath("/") + servletToContext.foreach { case (servlet, prefix) => + mainHandler.addServlet(new ServletHolder(servlet), prefix) + } + server.setHandler(mainHandler) + server.start() + val boundPort = server.getConnectors()(0).getLocalPort + (server, boundPort) + } + + def stop(): Unit = { + _server.foreach(_.stop()) + } +} + +private object StandaloneRestServer { + val PROTOCOL_VERSION = StandaloneRestClient.PROTOCOL_VERSION + val SC_UNKNOWN_PROTOCOL_VERSION = 468 +} + +/** + * An abstract servlet for handling requests passed to the [[StandaloneRestServer]]. + */ +private abstract class StandaloneRestServlet extends HttpServlet with Logging { + + /** Service a request. If an exception is thrown in the process, indicate server error. */ + protected override def service( + request: HttpServletRequest, + response: HttpServletResponse): Unit = { + try { + super.service(request, response) + } catch { + case e: Exception => + logError("Exception while handling request", e) + response.setStatus(HttpServletResponse.SC_INTERNAL_SERVER_ERROR) + } + } + + /** + * Serialize the given response message to JSON and send it through the response servlet. + * This validates the response before sending it to ensure it is properly constructed. + */ + protected def sendResponse( + responseMessage: SubmitRestProtocolResponse, + responseServlet: HttpServletResponse): Unit = { + val message = validateResponse(responseMessage, responseServlet) + responseServlet.setContentType("application/json") + responseServlet.setCharacterEncoding("utf-8") + responseServlet.setStatus(HttpServletResponse.SC_OK) + val content = message.toJson.getBytes(Charsets.UTF_8) + val out = new DataOutputStream(responseServlet.getOutputStream) + out.write(content) + out.close() + } + + /** + * Return any fields in the client request message that the server does not know about. + * + * The mechanism for this is to reconstruct the JSON on the server side and compare the + * diff between this JSON and the one generated on the client side. Any fields that are + * only in the client JSON are treated as unexpected. + */ + protected def findUnknownFields( + requestJson: String, + requestMessage: SubmitRestProtocolMessage): Array[String] = { + val clientSideJson = parse(requestJson) + val serverSideJson = parse(requestMessage.toJson) + val Diff(_, _, unknown) = clientSideJson.diff(serverSideJson) + unknown match { + case j: JObject => j.obj.map { case (k, _) => k }.toArray + case _ => Array.empty[String] // No difference + } + } + + /** Return a human readable String representation of the exception. */ + protected def formatException(e: Throwable): String = { + val stackTraceString = e.getStackTrace.map { "\t" + _ }.mkString("\n") + s"$e\n$stackTraceString" + } + + /** Construct an error message to signal the fact that an exception has been thrown. */ + protected def handleError(message: String): ErrorResponse = { + val e = new ErrorResponse + e.serverSparkVersion = sparkVersion + e.message = message + e + } + + /** + * Validate the response to ensure that it is correctly constructed. + * + * If it is, simply return the message as is. Otherwise, return an error response instead + * to propagate the exception back to the client and set the appropriate error code. + */ + private def validateResponse( + responseMessage: SubmitRestProtocolResponse, + responseServlet: HttpServletResponse): SubmitRestProtocolResponse = { + try { + responseMessage.validate() + responseMessage + } catch { + case e: Exception => + responseServlet.setStatus(HttpServletResponse.SC_INTERNAL_SERVER_ERROR) + handleError("Internal server error: " + formatException(e)) + } + } +} + +/** + * A servlet for handling kill requests passed to the [[StandaloneRestServer]]. + */ +private class KillRequestServlet(masterActor: ActorRef, conf: SparkConf) + extends StandaloneRestServlet { + + /** + * If a submission ID is specified in the URL, have the Master kill the corresponding + * driver and return an appropriate response to the client. Otherwise, return error. + */ + protected override def doPost( + request: HttpServletRequest, + response: HttpServletResponse): Unit = { + val submissionId = request.getPathInfo.stripPrefix("/") + val responseMessage = + if (submissionId.nonEmpty) { + handleKill(submissionId) + } else { + response.setStatus(HttpServletResponse.SC_BAD_REQUEST) + handleError("Submission ID is missing in kill request.") + } + sendResponse(responseMessage, response) + } + + private def handleKill(submissionId: String): KillSubmissionResponse = { + val askTimeout = AkkaUtils.askTimeout(conf) + val response = AkkaUtils.askWithReply[DeployMessages.KillDriverResponse]( + DeployMessages.RequestKillDriver(submissionId), masterActor, askTimeout) + val k = new KillSubmissionResponse + k.serverSparkVersion = sparkVersion + k.message = response.message + k.submissionId = submissionId + k.success = response.success + k + } +} + +/** + * A servlet for handling status requests passed to the [[StandaloneRestServer]]. + */ +private class StatusRequestServlet(masterActor: ActorRef, conf: SparkConf) + extends StandaloneRestServlet { + + /** + * If a submission ID is specified in the URL, request the status of the corresponding + * driver from the Master and include it in the response. Otherwise, return error. + */ + protected override def doGet( + request: HttpServletRequest, + response: HttpServletResponse): Unit = { + val submissionId = request.getPathInfo.stripPrefix("/") + val responseMessage = + if (submissionId.nonEmpty) { + handleStatus(submissionId) + } else { + response.setStatus(HttpServletResponse.SC_BAD_REQUEST) + handleError("Submission ID is missing in status request.") + } + sendResponse(responseMessage, response) + } + + private def handleStatus(submissionId: String): SubmissionStatusResponse = { + val askTimeout = AkkaUtils.askTimeout(conf) + val response = AkkaUtils.askWithReply[DeployMessages.DriverStatusResponse]( + DeployMessages.RequestDriverStatus(submissionId), masterActor, askTimeout) + val message = response.exception.map { s"Exception from the cluster:\n" + formatException(_) } + val d = new SubmissionStatusResponse + d.serverSparkVersion = sparkVersion + d.submissionId = submissionId + d.success = response.found + d.driverState = response.state.map(_.toString).orNull + d.workerId = response.workerId.orNull + d.workerHostPort = response.workerHostPort.orNull + d.message = message.orNull + d + } +} + +/** + * A servlet for handling submit requests passed to the [[StandaloneRestServer]]. + */ +private class SubmitRequestServlet( + masterActor: ActorRef, + masterUrl: String, + conf: SparkConf) + extends StandaloneRestServlet { + + /** + * Submit an application to the Master with parameters specified in the request. + * + * The request is assumed to be a [[SubmitRestProtocolRequest]] in the form of JSON. + * If the request is successfully processed, return an appropriate response to the + * client indicating so. Otherwise, return error instead. + */ + protected override def doPost( + requestServlet: HttpServletRequest, + responseServlet: HttpServletResponse): Unit = { + val responseMessage = + try { + val requestMessageJson = Source.fromInputStream(requestServlet.getInputStream).mkString + val requestMessage = SubmitRestProtocolMessage.fromJson(requestMessageJson) + // The response should have already been validated on the client. + // In case this is not true, validate it ourselves to avoid potential NPEs. + requestMessage.validate() + handleSubmit(requestMessageJson, requestMessage, responseServlet) + } catch { + // The client failed to provide a valid JSON, so this is not our fault + case e @ (_: JsonMappingException | _: SubmitRestProtocolException) => + responseServlet.setStatus(HttpServletResponse.SC_BAD_REQUEST) + handleError("Malformed request: " + formatException(e)) + } + sendResponse(responseMessage, responseServlet) + } + + /** + * Handle the submit request and construct an appropriate response to return to the client. + * + * This assumes that the request message is already successfully validated. + * If the request message is not of the expected type, return error to the client. + */ + private def handleSubmit( + requestMessageJson: String, + requestMessage: SubmitRestProtocolMessage, + responseServlet: HttpServletResponse): SubmitRestProtocolResponse = { + requestMessage match { + case submitRequest: CreateSubmissionRequest => + val askTimeout = AkkaUtils.askTimeout(conf) + val driverDescription = buildDriverDescription(submitRequest) + val response = AkkaUtils.askWithReply[DeployMessages.SubmitDriverResponse]( + DeployMessages.RequestSubmitDriver(driverDescription), masterActor, askTimeout) + val submitResponse = new CreateSubmissionResponse + submitResponse.serverSparkVersion = sparkVersion + submitResponse.message = response.message + submitResponse.success = response.success + submitResponse.submissionId = response.driverId.orNull + val unknownFields = findUnknownFields(requestMessageJson, requestMessage) + if (unknownFields.nonEmpty) { + // If there are fields that the server does not know about, warn the client + submitResponse.unknownFields = unknownFields + } + submitResponse + case unexpected => + responseServlet.setStatus(HttpServletResponse.SC_BAD_REQUEST) + handleError(s"Received message of unexpected type ${unexpected.messageType}.") + } + } + + /** + * Build a driver description from the fields specified in the submit request. + * + * This involves constructing a command that takes into account memory, java options, + * classpath and other settings to launch the driver. This does not currently consider + * fields used by python applications since python is not supported in standalone + * cluster mode yet. + */ + private def buildDriverDescription(request: CreateSubmissionRequest): DriverDescription = { + // Required fields, including the main class because python is not yet supported + val appResource = Option(request.appResource).getOrElse { + throw new SubmitRestMissingFieldException("Application jar is missing.") + } + val mainClass = Option(request.mainClass).getOrElse { + throw new SubmitRestMissingFieldException("Main class is missing.") + } + + // Optional fields + val sparkProperties = request.sparkProperties + val driverMemory = sparkProperties.get("spark.driver.memory") + val driverCores = sparkProperties.get("spark.driver.cores") + val driverExtraJavaOptions = sparkProperties.get("spark.driver.extraJavaOptions") + val driverExtraClassPath = sparkProperties.get("spark.driver.extraClassPath") + val driverExtraLibraryPath = sparkProperties.get("spark.driver.extraLibraryPath") + val superviseDriver = sparkProperties.get("spark.driver.supervise") + val appArgs = request.appArgs + val environmentVariables = request.environmentVariables + + // Construct driver description + val conf = new SparkConf(false) + .setAll(sparkProperties) + .set("spark.master", masterUrl) + val extraClassPath = driverExtraClassPath.toSeq.flatMap(_.split(File.pathSeparator)) + val extraLibraryPath = driverExtraLibraryPath.toSeq.flatMap(_.split(File.pathSeparator)) + val extraJavaOpts = driverExtraJavaOptions.map(Utils.splitCommandString).getOrElse(Seq.empty) + val sparkJavaOpts = Utils.sparkJavaOpts(conf) + val javaOpts = sparkJavaOpts ++ extraJavaOpts + val command = new Command( + "org.apache.spark.deploy.worker.DriverWrapper", + Seq("{{WORKER_URL}}", mainClass) ++ appArgs, // args to the DriverWrapper + environmentVariables, extraClassPath, extraLibraryPath, javaOpts) + val actualDriverMemory = driverMemory.map(Utils.memoryStringToMb).getOrElse(DEFAULT_MEMORY) + val actualDriverCores = driverCores.map(_.toInt).getOrElse(DEFAULT_CORES) + val actualSuperviseDriver = superviseDriver.map(_.toBoolean).getOrElse(DEFAULT_SUPERVISE) + new DriverDescription( + appResource, actualDriverMemory, actualDriverCores, actualSuperviseDriver, command) + } +} + +/** + * A default servlet that handles error cases that are not captured by other servlets. + */ +private class ErrorServlet extends StandaloneRestServlet { + private val serverVersion = StandaloneRestServer.PROTOCOL_VERSION + + /** Service a faulty request by returning an appropriate error message to the client. */ + protected override def service( + request: HttpServletRequest, + response: HttpServletResponse): Unit = { + val path = request.getPathInfo + val parts = path.stripPrefix("/").split("/").toSeq + var versionMismatch = false + var msg = + parts match { + case Nil => + // http://host:port/ + "Missing protocol version." + case `serverVersion` :: Nil => + // http://host:port/correct-version + "Missing the /submissions prefix." + case `serverVersion` :: "submissions" :: Nil => + // http://host:port/correct-version/submissions + "Missing an action: please specify one of /create, /kill, or /status." + case unknownVersion :: _ => + // http://host:port/unknown-version/* + versionMismatch = true + s"Unknown protocol version '$unknownVersion'." + case _ => + // never reached + s"Malformed path $path." + } + msg += s" Please submit requests through http://[host]:[port]/$serverVersion/submissions/..." + val error = handleError(msg) + // If there is a version mismatch, include the highest protocol version that + // this server supports in case the client wants to retry with our version + if (versionMismatch) { + error.highestProtocolVersion = serverVersion + response.setStatus(StandaloneRestServer.SC_UNKNOWN_PROTOCOL_VERSION) + } else { + response.setStatus(HttpServletResponse.SC_BAD_REQUEST) + } + sendResponse(error, response) + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolException.scala b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolException.scala new file mode 100644 index 0000000000000..d7a0bdbe10778 --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolException.scala @@ -0,0 +1,36 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +/** + * An exception thrown in the REST application submission protocol. + */ +private[spark] class SubmitRestProtocolException(message: String, cause: Throwable = null) + extends Exception(message, cause) + +/** + * An exception thrown if a field is missing from a [[SubmitRestProtocolMessage]]. + */ +private[spark] class SubmitRestMissingFieldException(message: String) + extends SubmitRestProtocolException(message) + +/** + * An exception thrown if the REST client cannot reach the REST server. + */ +private[spark] class SubmitRestConnectionException(message: String, cause: Throwable) + extends SubmitRestProtocolException(message, cause) diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolMessage.scala b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolMessage.scala new file mode 100644 index 0000000000000..b877898231e3e --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolMessage.scala @@ -0,0 +1,146 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import scala.util.Try + +import com.fasterxml.jackson.annotation._ +import com.fasterxml.jackson.annotation.JsonAutoDetect.Visibility +import com.fasterxml.jackson.annotation.JsonInclude.Include +import com.fasterxml.jackson.databind.{DeserializationFeature, ObjectMapper, SerializationFeature} +import com.fasterxml.jackson.module.scala.DefaultScalaModule +import org.json4s.JsonAST._ +import org.json4s.jackson.JsonMethods._ + +import org.apache.spark.util.Utils + +/** + * An abstract message exchanged in the REST application submission protocol. + * + * This message is intended to be serialized to and deserialized from JSON in the exchange. + * Each message can either be a request or a response and consists of three common fields: + * (1) the action, which fully specifies the type of the message + * (2) the Spark version of the client / server + * (3) an optional message + */ +@JsonInclude(Include.NON_NULL) +@JsonAutoDetect(getterVisibility = Visibility.ANY, setterVisibility = Visibility.ANY) +@JsonPropertyOrder(alphabetic = true) +private[spark] abstract class SubmitRestProtocolMessage { + @JsonIgnore + val messageType = Utils.getFormattedClassName(this) + + val action: String = messageType + var message: String = null + + // For JSON deserialization + private def setAction(a: String): Unit = { } + + /** + * Serialize the message to JSON. + * This also ensures that the message is valid and its fields are in the expected format. + */ + def toJson: String = { + validate() + SubmitRestProtocolMessage.mapper.writeValueAsString(this) + } + + /** + * Assert the validity of the message. + * If the validation fails, throw a [[SubmitRestProtocolException]]. + */ + final def validate(): Unit = { + try { + doValidate() + } catch { + case e: Exception => + throw new SubmitRestProtocolException(s"Validation of message $messageType failed!", e) + } + } + + /** Assert the validity of the message */ + protected def doValidate(): Unit = { + if (action == null) { + throw new SubmitRestMissingFieldException(s"The action field is missing in $messageType") + } + } + + /** Assert that the specified field is set in this message. */ + protected def assertFieldIsSet[T](value: T, name: String): Unit = { + if (value == null) { + throw new SubmitRestMissingFieldException(s"'$name' is missing in message $messageType.") + } + } + + /** + * Assert a condition when validating this message. + * If the assertion fails, throw a [[SubmitRestProtocolException]]. + */ + protected def assert(condition: Boolean, failMessage: String): Unit = { + if (!condition) { throw new SubmitRestProtocolException(failMessage) } + } +} + +/** + * Helper methods to process serialized [[SubmitRestProtocolMessage]]s. + */ +private[spark] object SubmitRestProtocolMessage { + private val packagePrefix = this.getClass.getPackage.getName + private val mapper = new ObjectMapper() + .configure(DeserializationFeature.FAIL_ON_UNKNOWN_PROPERTIES, false) + .enable(SerializationFeature.INDENT_OUTPUT) + .registerModule(DefaultScalaModule) + + /** + * Parse the value of the action field from the given JSON. + * If the action field is not found, throw a [[SubmitRestMissingFieldException]]. + */ + def parseAction(json: String): String = { + parse(json).asInstanceOf[JObject].obj + .find { case (f, _) => f == "action" } + .map { case (_, v) => v.asInstanceOf[JString].s } + .getOrElse { + throw new SubmitRestMissingFieldException(s"Action field not found in JSON:\n$json") + } + } + + /** + * Construct a [[SubmitRestProtocolMessage]] from its JSON representation. + * + * This method first parses the action from the JSON and uses it to infer the message type. + * Note that the action must represent one of the [[SubmitRestProtocolMessage]]s defined in + * this package. Otherwise, a [[ClassNotFoundException]] will be thrown. + */ + def fromJson(json: String): SubmitRestProtocolMessage = { + val className = parseAction(json) + val clazz = Class.forName(packagePrefix + "." + className) + .asSubclass[SubmitRestProtocolMessage](classOf[SubmitRestProtocolMessage]) + fromJson(json, clazz) + } + + /** + * Construct a [[SubmitRestProtocolMessage]] from its JSON representation. + * + * This method determines the type of the message from the class provided instead of + * inferring it from the action field. This is useful for deserializing JSON that + * represents custom user-defined messages. + */ + def fromJson[T <: SubmitRestProtocolMessage](json: String, clazz: Class[T]): T = { + mapper.readValue(json, clazz) + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolRequest.scala b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolRequest.scala new file mode 100644 index 0000000000000..9e1fd8c40cabd --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolRequest.scala @@ -0,0 +1,78 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import scala.util.Try + +import org.apache.spark.util.Utils + +/** + * An abstract request sent from the client in the REST application submission protocol. + */ +private[spark] abstract class SubmitRestProtocolRequest extends SubmitRestProtocolMessage { + var clientSparkVersion: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(clientSparkVersion, "clientSparkVersion") + } +} + +/** + * A request to launch a new application in the REST application submission protocol. + */ +private[spark] class CreateSubmissionRequest extends SubmitRestProtocolRequest { + var appResource: String = null + var mainClass: String = null + var appArgs: Array[String] = null + var sparkProperties: Map[String, String] = null + var environmentVariables: Map[String, String] = null + + protected override def doValidate(): Unit = { + super.doValidate() + assert(sparkProperties != null, "No Spark properties set!") + assertFieldIsSet(appResource, "appResource") + assertPropertyIsSet("spark.app.name") + assertPropertyIsBoolean("spark.driver.supervise") + assertPropertyIsNumeric("spark.driver.cores") + assertPropertyIsNumeric("spark.cores.max") + assertPropertyIsMemory("spark.driver.memory") + assertPropertyIsMemory("spark.executor.memory") + } + + private def assertPropertyIsSet(key: String): Unit = + assertFieldIsSet(sparkProperties.getOrElse(key, null), key) + + private def assertPropertyIsBoolean(key: String): Unit = + assertProperty[Boolean](key, "boolean", _.toBoolean) + + private def assertPropertyIsNumeric(key: String): Unit = + assertProperty[Int](key, "numeric", _.toInt) + + private def assertPropertyIsMemory(key: String): Unit = + assertProperty[Int](key, "memory", Utils.memoryStringToMb) + + /** Assert that a Spark property can be converted to a certain type. */ + private def assertProperty[T](key: String, valueType: String, convert: (String => T)): Unit = { + sparkProperties.get(key).foreach { value => + Try(convert(value)).getOrElse { + throw new SubmitRestProtocolException( + s"Property '$key' expected $valueType value: actual was '$value'.") + } + } + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolResponse.scala b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolResponse.scala new file mode 100644 index 0000000000000..16dfe041d4bea --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolResponse.scala @@ -0,0 +1,85 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.lang.Boolean + +/** + * An abstract response sent from the server in the REST application submission protocol. + */ +private[spark] abstract class SubmitRestProtocolResponse extends SubmitRestProtocolMessage { + var serverSparkVersion: String = null + var success: Boolean = null + var unknownFields: Array[String] = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(serverSparkVersion, "serverSparkVersion") + } +} + +/** + * A response to a [[CreateSubmissionRequest]] in the REST application submission protocol. + */ +private[spark] class CreateSubmissionResponse extends SubmitRestProtocolResponse { + var submissionId: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(success, "success") + } +} + +/** + * A response to a kill request in the REST application submission protocol. + */ +private[spark] class KillSubmissionResponse extends SubmitRestProtocolResponse { + var submissionId: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(submissionId, "submissionId") + assertFieldIsSet(success, "success") + } +} + +/** + * A response to a status request in the REST application submission protocol. + */ +private[spark] class SubmissionStatusResponse extends SubmitRestProtocolResponse { + var submissionId: String = null + var driverState: String = null + var workerId: String = null + var workerHostPort: String = null + + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(submissionId, "submissionId") + assertFieldIsSet(success, "success") + } +} + +/** + * An error response message used in the REST application submission protocol. + */ +private[spark] class ErrorResponse extends SubmitRestProtocolResponse { + // The highest protocol version that the server knows about + // This is set when the client specifies an unknown version + var highestProtocolVersion: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(message, "message") + } +} diff --git a/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala b/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala index ed02ca81e405c..e955636cf5b59 100644 --- a/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala +++ b/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala @@ -68,7 +68,8 @@ class JsonProtocolSuite extends FunSuite { val completedApps = Array[ApplicationInfo]() val activeDrivers = Array(createDriverInfo()) val completedDrivers = Array(createDriverInfo()) - val stateResponse = new MasterStateResponse("host", 8080, workers, activeApps, completedApps, + val stateResponse = new MasterStateResponse( + "host", 8080, None, workers, activeApps, completedApps, activeDrivers, completedDrivers, RecoveryState.ALIVE) val output = JsonProtocol.writeMasterState(stateResponse) assertValidJson(output) diff --git a/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala b/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala index 3f1355f82893e..1ddccae1262bc 100644 --- a/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala +++ b/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala @@ -141,7 +141,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) val childArgsStr = childArgs.mkString(" ") childArgsStr should include ("--class org.SomeClass") childArgsStr should include ("--executor-memory 5g") @@ -180,7 +180,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) childArgs.mkString(" ") should be ("arg1 arg2") mainClass should be ("org.SomeClass") classpath should have length (4) @@ -201,6 +201,18 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties } test("handles standalone cluster mode") { + testStandaloneCluster(useRest = true) + } + + test("handles legacy standalone cluster mode") { + testStandaloneCluster(useRest = false) + } + + /** + * Test whether the launch environment is correctly set up in standalone cluster mode. + * @param useRest whether to use the REST submission gateway introduced in Spark 1.3 + */ + private def testStandaloneCluster(useRest: Boolean): Unit = { val clArgs = Seq( "--deploy-mode", "cluster", "--master", "spark://h:p", @@ -212,17 +224,26 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + appArgs.useRest = useRest + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) val childArgsStr = childArgs.mkString(" ") - childArgsStr should startWith ("--memory 4g --cores 5 --supervise") - childArgsStr should include regex ("launch spark://h:p .*thejar.jar org.SomeClass arg1 arg2") - mainClass should be ("org.apache.spark.deploy.Client") - classpath should have size (0) - sysProps should have size (5) + if (useRest) { + childArgsStr should endWith ("thejar.jar org.SomeClass arg1 arg2") + mainClass should be ("org.apache.spark.deploy.rest.StandaloneRestClient") + } else { + childArgsStr should startWith ("--supervise --memory 4g --cores 5") + childArgsStr should include regex "launch spark://h:p .*thejar.jar org.SomeClass arg1 arg2" + mainClass should be ("org.apache.spark.deploy.Client") + } + classpath should have size 0 + sysProps should have size 8 sysProps.keys should contain ("SPARK_SUBMIT") sysProps.keys should contain ("spark.master") sysProps.keys should contain ("spark.app.name") sysProps.keys should contain ("spark.jars") + sysProps.keys should contain ("spark.driver.memory") + sysProps.keys should contain ("spark.driver.cores") + sysProps.keys should contain ("spark.driver.supervise") sysProps.keys should contain ("spark.shuffle.spill") sysProps("spark.shuffle.spill") should be ("false") } @@ -239,7 +260,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) childArgs.mkString(" ") should be ("arg1 arg2") mainClass should be ("org.SomeClass") classpath should have length (1) @@ -261,7 +282,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) childArgs.mkString(" ") should be ("arg1 arg2") mainClass should be ("org.SomeClass") classpath should have length (1) @@ -281,7 +302,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (_, _, sysProps, mainClass) = createLaunchEnv(appArgs) + val (_, _, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) sysProps("spark.executor.memory") should be ("5g") sysProps("spark.master") should be ("yarn-cluster") mainClass should be ("org.apache.spark.deploy.yarn.Client") @@ -339,7 +360,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "--files", files, "thejar.jar") val appArgs = new SparkSubmitArguments(clArgs) - val sysProps = SparkSubmit.createLaunchEnv(appArgs)._3 + val sysProps = SparkSubmit.prepareSubmitEnvironment(appArgs)._3 appArgs.jars should be (Utils.resolveURIs(jars)) appArgs.files should be (Utils.resolveURIs(files)) sysProps("spark.jars") should be (Utils.resolveURIs(jars + ",thejar.jar")) @@ -354,7 +375,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar" ) val appArgs2 = new SparkSubmitArguments(clArgs2) - val sysProps2 = SparkSubmit.createLaunchEnv(appArgs2)._3 + val sysProps2 = SparkSubmit.prepareSubmitEnvironment(appArgs2)._3 appArgs2.files should be (Utils.resolveURIs(files)) appArgs2.archives should be (Utils.resolveURIs(archives)) sysProps2("spark.yarn.dist.files") should be (Utils.resolveURIs(files)) @@ -367,7 +388,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "mister.py" ) val appArgs3 = new SparkSubmitArguments(clArgs3) - val sysProps3 = SparkSubmit.createLaunchEnv(appArgs3)._3 + val sysProps3 = SparkSubmit.prepareSubmitEnvironment(appArgs3)._3 appArgs3.pyFiles should be (Utils.resolveURIs(pyFiles)) sysProps3("spark.submit.pyFiles") should be ( PythonRunner.formatPaths(Utils.resolveURIs(pyFiles)).mkString(",")) @@ -392,7 +413,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar" ) val appArgs = new SparkSubmitArguments(clArgs) - val sysProps = SparkSubmit.createLaunchEnv(appArgs)._3 + val sysProps = SparkSubmit.prepareSubmitEnvironment(appArgs)._3 sysProps("spark.jars") should be(Utils.resolveURIs(jars + ",thejar.jar")) sysProps("spark.files") should be(Utils.resolveURIs(files)) @@ -409,7 +430,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar" ) val appArgs2 = new SparkSubmitArguments(clArgs2) - val sysProps2 = SparkSubmit.createLaunchEnv(appArgs2)._3 + val sysProps2 = SparkSubmit.prepareSubmitEnvironment(appArgs2)._3 sysProps2("spark.yarn.dist.files") should be(Utils.resolveURIs(files)) sysProps2("spark.yarn.dist.archives") should be(Utils.resolveURIs(archives)) @@ -424,7 +445,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "mister.py" ) val appArgs3 = new SparkSubmitArguments(clArgs3) - val sysProps3 = SparkSubmit.createLaunchEnv(appArgs3)._3 + val sysProps3 = SparkSubmit.prepareSubmitEnvironment(appArgs3)._3 sysProps3("spark.submit.pyFiles") should be( PythonRunner.formatPaths(Utils.resolveURIs(pyFiles)).mkString(",")) } @@ -440,7 +461,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties val appArgs = new SparkSubmitArguments(args, Map("SPARK_CONF_DIR" -> path)) assert(appArgs.propertiesFile != null) assert(appArgs.propertiesFile.startsWith(path)) - appArgs.executorMemory should be ("2.3g") + appArgs.executorMemory should be ("2.3g") } } diff --git a/core/src/test/scala/org/apache/spark/deploy/rest/StandaloneRestSubmitSuite.scala b/core/src/test/scala/org/apache/spark/deploy/rest/StandaloneRestSubmitSuite.scala new file mode 100644 index 0000000000000..29aed89b67aa7 --- /dev/null +++ b/core/src/test/scala/org/apache/spark/deploy/rest/StandaloneRestSubmitSuite.scala @@ -0,0 +1,265 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +* See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.io.{File, FileInputStream, FileOutputStream, PrintWriter} +import java.util.jar.{JarEntry, JarOutputStream} +import java.util.zip.ZipEntry + +import scala.collection.mutable.ArrayBuffer +import scala.io.Source + +import akka.actor.ActorSystem +import com.google.common.io.ByteStreams +import org.scalatest.{BeforeAndAfterAll, BeforeAndAfterEach, FunSuite} +import org.scalatest.exceptions.TestFailedException + +import org.apache.spark._ +import org.apache.spark.util.Utils +import org.apache.spark.deploy.{SparkSubmit, SparkSubmitArguments} +import org.apache.spark.deploy.master.{DriverState, Master} +import org.apache.spark.deploy.worker.Worker + +/** + * End-to-end tests for the REST application submission protocol in standalone mode. + */ +class StandaloneRestSubmitSuite extends FunSuite with BeforeAndAfterAll with BeforeAndAfterEach { + private val systemsToStop = new ArrayBuffer[ActorSystem] + private val masterRestUrl = startLocalCluster() + private val client = new StandaloneRestClient + private val mainJar = StandaloneRestSubmitSuite.createJar() + private val mainClass = StandaloneRestApp.getClass.getName.stripSuffix("$") + + override def afterAll() { + systemsToStop.foreach(_.shutdown()) + } + + test("simple submit until completion") { + val resultsFile = File.createTempFile("test-submit", ".txt") + val numbers = Seq(1, 2, 3) + val size = 500 + val submissionId = submitApplication(resultsFile, numbers, size) + waitUntilFinished(submissionId) + validateResult(resultsFile, numbers, size) + } + + test("kill empty submission") { + val response = client.killSubmission(masterRestUrl, "submission-that-does-not-exist") + val killResponse = getKillResponse(response) + val killSuccess = killResponse.success + assert(!killSuccess) + } + + test("kill running submission") { + val resultsFile = File.createTempFile("test-kill", ".txt") + val numbers = Seq(1, 2, 3) + val size = 500 + val submissionId = submitApplication(resultsFile, numbers, size) + val response = client.killSubmission(masterRestUrl, submissionId) + val killResponse = getKillResponse(response) + val killSuccess = killResponse.success + waitUntilFinished(submissionId) + val response2 = client.requestSubmissionStatus(masterRestUrl, submissionId) + val statusResponse = getStatusResponse(response2) + val statusSuccess = statusResponse.success + val driverState = statusResponse.driverState + assert(killSuccess) + assert(statusSuccess) + assert(driverState === DriverState.KILLED.toString) + // we should not see the expected results because we killed the submission + intercept[TestFailedException] { validateResult(resultsFile, numbers, size) } + } + + test("request status for empty submission") { + val response = client.requestSubmissionStatus(masterRestUrl, "submission-that-does-not-exist") + val statusResponse = getStatusResponse(response) + val statusSuccess = statusResponse.success + assert(!statusSuccess) + } + + /** + * Start a local cluster containing one Master and a few Workers. + * Do not use [[org.apache.spark.deploy.LocalSparkCluster]] here because we want the REST URL. + * Return the Master's REST URL to which applications should be submitted. + */ + private def startLocalCluster(): String = { + val conf = new SparkConf(false) + .set("spark.master.rest.enabled", "true") + .set("spark.master.rest.port", "0") + val (numWorkers, coresPerWorker, memPerWorker) = (2, 1, 512) + val localHostName = Utils.localHostName() + val (masterSystem, masterPort, _, _masterRestPort) = + Master.startSystemAndActor(localHostName, 0, 0, conf) + val masterRestPort = _masterRestPort.getOrElse { fail("REST server not started on Master!") } + val masterUrl = "spark://" + localHostName + ":" + masterPort + val masterRestUrl = "spark://" + localHostName + ":" + masterRestPort + (1 to numWorkers).foreach { n => + val (workerSystem, _) = Worker.startSystemAndActor( + localHostName, 0, 0, coresPerWorker, memPerWorker, Array(masterUrl), null, Some(n)) + systemsToStop.append(workerSystem) + } + systemsToStop.append(masterSystem) + masterRestUrl + } + + /** Submit the [[StandaloneRestApp]] and return the corresponding submission ID. */ + private def submitApplication(resultsFile: File, numbers: Seq[Int], size: Int): String = { + val appArgs = Seq(resultsFile.getAbsolutePath) ++ numbers.map(_.toString) ++ Seq(size.toString) + val commandLineArgs = Array( + "--deploy-mode", "cluster", + "--master", masterRestUrl, + "--name", mainClass, + "--class", mainClass, + mainJar) ++ appArgs + val args = new SparkSubmitArguments(commandLineArgs) + val (_, _, sparkProperties, _) = SparkSubmit.prepareSubmitEnvironment(args) + val request = client.constructSubmitRequest( + mainJar, mainClass, appArgs.toArray, sparkProperties.toMap, Map.empty) + val response = client.createSubmission(masterRestUrl, request) + val submitResponse = getSubmitResponse(response) + val submissionId = submitResponse.submissionId + assert(submissionId != null, "Application submission was unsuccessful!") + submissionId + } + + /** Wait until the given submission has finished running up to the specified timeout. */ + private def waitUntilFinished(submissionId: String, maxSeconds: Int = 30): Unit = { + var finished = false + val expireTime = System.currentTimeMillis + maxSeconds * 1000 + while (!finished) { + val response = client.requestSubmissionStatus(masterRestUrl, submissionId) + val statusResponse = getStatusResponse(response) + val driverState = statusResponse.driverState + finished = + driverState != DriverState.SUBMITTED.toString && + driverState != DriverState.RUNNING.toString + if (System.currentTimeMillis > expireTime) { + fail(s"Driver $submissionId did not finish within $maxSeconds seconds.") + } + } + } + + /** Return the response as a submit response, or fail with error otherwise. */ + private def getSubmitResponse(response: SubmitRestProtocolResponse): CreateSubmissionResponse = { + response match { + case s: CreateSubmissionResponse => s + case e: ErrorResponse => fail(s"Server returned error: ${e.message}") + case r => fail(s"Expected submit response. Actual: ${r.toJson}") + } + } + + /** Return the response as a kill response, or fail with error otherwise. */ + private def getKillResponse(response: SubmitRestProtocolResponse): KillSubmissionResponse = { + response match { + case k: KillSubmissionResponse => k + case e: ErrorResponse => fail(s"Server returned error: ${e.message}") + case r => fail(s"Expected kill response. Actual: ${r.toJson}") + } + } + + /** Return the response as a status response, or fail with error otherwise. */ + private def getStatusResponse(response: SubmitRestProtocolResponse): SubmissionStatusResponse = { + response match { + case s: SubmissionStatusResponse => s + case e: ErrorResponse => fail(s"Server returned error: ${e.message}") + case r => fail(s"Expected status response. Actual: ${r.toJson}") + } + } + + /** Validate whether the application produced the corrupt output. */ + private def validateResult(resultsFile: File, numbers: Seq[Int], size: Int): Unit = { + val lines = Source.fromFile(resultsFile.getAbsolutePath).getLines().toSeq + val unexpectedContent = + if (lines.nonEmpty) { + "[\n" + lines.map { l => " " + l }.mkString("\n") + "\n]" + } else { + "[EMPTY]" + } + assert(lines.size === 2, s"Unexpected content in file: $unexpectedContent") + assert(lines(0).toInt === numbers.sum, s"Sum of ${numbers.mkString(",")} is incorrect") + assert(lines(1).toInt === (size / 2) + 1, "Result of Spark job is incorrect") + } +} + +private object StandaloneRestSubmitSuite { + private val pathPrefix = this.getClass.getPackage.getName.replaceAll("\\.", "/") + + /** + * Create a jar that contains all the class files needed for running the [[StandaloneRestApp]]. + * Return the absolute path to that jar. + */ + def createJar(): String = { + val jarFile = File.createTempFile("test-standalone-rest-protocol", ".jar") + val jarFileStream = new FileOutputStream(jarFile) + val jarStream = new JarOutputStream(jarFileStream, new java.util.jar.Manifest) + jarStream.putNextEntry(new ZipEntry(pathPrefix)) + getClassFiles.foreach { cf => + jarStream.putNextEntry(new JarEntry(pathPrefix + "/" + cf.getName)) + val in = new FileInputStream(cf) + ByteStreams.copy(in, jarStream) + in.close() + } + jarStream.close() + jarFileStream.close() + jarFile.getAbsolutePath + } + + /** + * Return a list of class files compiled for [[StandaloneRestApp]]. + * This includes all the anonymous classes used in the application. + */ + private def getClassFiles: Seq[File] = { + val className = Utils.getFormattedClassName(StandaloneRestApp) + val clazz = StandaloneRestApp.getClass + val basePath = clazz.getProtectionDomain.getCodeSource.getLocation.toURI.getPath + val baseDir = new File(basePath + "/" + pathPrefix) + baseDir.listFiles().filter(_.getName.contains(className)) + } +} + +/** + * Sample application to be submitted to the cluster using the REST gateway. + * All relevant classes will be packaged into a jar at run time. + */ +object StandaloneRestApp { + // Usage: [path to results file] [num1] [num2] [num3] [rddSize] + // The first line of the results file should be (num1 + num2 + num3) + // The second line should be (rddSize / 2) + 1 + def main(args: Array[String]) { + assert(args.size == 5, s"Expected exactly 5 arguments: ${args.mkString(",")}") + val resultFile = new File(args(0)) + val writer = new PrintWriter(resultFile) + try { + val conf = new SparkConf() + val sc = new SparkContext(conf) + val firstLine = args(1).toInt + args(2).toInt + args(3).toInt + val secondLine = sc.parallelize(1 to args(4).toInt) + .map { i => (i / 2, i) } + .reduceByKey(_ + _) + .count() + writer.println(firstLine) + writer.println(secondLine) + } catch { + case e: Exception => + writer.println(e) + e.getStackTrace.foreach { l => writer.println(" " + l) } + } finally { + writer.close() + } + } +} diff --git a/core/src/test/scala/org/apache/spark/deploy/rest/SubmitRestProtocolSuite.scala b/core/src/test/scala/org/apache/spark/deploy/rest/SubmitRestProtocolSuite.scala new file mode 100644 index 0000000000000..1d64ec201e647 --- /dev/null +++ b/core/src/test/scala/org/apache/spark/deploy/rest/SubmitRestProtocolSuite.scala @@ -0,0 +1,324 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.lang.Boolean +import java.lang.Integer + +import org.json4s.jackson.JsonMethods._ +import org.scalatest.FunSuite + +import org.apache.spark.SparkConf + +/** + * Tests for the REST application submission protocol. + */ +class SubmitRestProtocolSuite extends FunSuite { + + test("validate") { + val request = new DummyRequest + intercept[SubmitRestProtocolException] { request.validate() } // missing everything + request.clientSparkVersion = "1.2.3" + intercept[SubmitRestProtocolException] { request.validate() } // missing name and age + request.name = "something" + intercept[SubmitRestProtocolException] { request.validate() } // missing only age + request.age = 2 + intercept[SubmitRestProtocolException] { request.validate() } // age too low + request.age = 10 + request.validate() // everything is set properly + request.clientSparkVersion = null + intercept[SubmitRestProtocolException] { request.validate() } // missing only Spark version + request.clientSparkVersion = "1.2.3" + request.name = null + intercept[SubmitRestProtocolException] { request.validate() } // missing only name + request.message = "not-setting-name" + intercept[SubmitRestProtocolException] { request.validate() } // still missing name + } + + test("request to and from JSON") { + val request = new DummyRequest + intercept[SubmitRestProtocolException] { request.toJson } // implicit validation + request.clientSparkVersion = "1.2.3" + request.active = true + request.age = 25 + request.name = "jung" + val json = request.toJson + assertJsonEquals(json, dummyRequestJson) + val newRequest = SubmitRestProtocolMessage.fromJson(json, classOf[DummyRequest]) + assert(newRequest.clientSparkVersion === "1.2.3") + assert(newRequest.clientSparkVersion === "1.2.3") + assert(newRequest.active) + assert(newRequest.age === 25) + assert(newRequest.name === "jung") + assert(newRequest.message === null) + } + + test("response to and from JSON") { + val response = new DummyResponse + response.serverSparkVersion = "3.3.4" + response.success = true + val json = response.toJson + assertJsonEquals(json, dummyResponseJson) + val newResponse = SubmitRestProtocolMessage.fromJson(json, classOf[DummyResponse]) + assert(newResponse.serverSparkVersion === "3.3.4") + assert(newResponse.serverSparkVersion === "3.3.4") + assert(newResponse.success) + assert(newResponse.message === null) + } + + test("CreateSubmissionRequest") { + val message = new CreateSubmissionRequest + intercept[SubmitRestProtocolException] { message.validate() } + message.clientSparkVersion = "1.2.3" + message.appResource = "honey-walnut-cherry.jar" + message.mainClass = "org.apache.spark.examples.SparkPie" + val conf = new SparkConf(false) + conf.set("spark.app.name", "SparkPie") + message.sparkProperties = conf.getAll.toMap + message.validate() + // optional fields + conf.set("spark.jars", "mayonnaise.jar,ketchup.jar") + conf.set("spark.files", "fireball.png") + conf.set("spark.driver.memory", "512m") + conf.set("spark.driver.cores", "180") + conf.set("spark.driver.extraJavaOptions", " -Dslices=5 -Dcolor=mostly_red") + conf.set("spark.driver.extraClassPath", "food-coloring.jar") + conf.set("spark.driver.extraLibraryPath", "pickle.jar") + conf.set("spark.driver.supervise", "false") + conf.set("spark.executor.memory", "256m") + conf.set("spark.cores.max", "10000") + message.sparkProperties = conf.getAll.toMap + message.appArgs = Array("two slices", "a hint of cinnamon") + message.environmentVariables = Map("PATH" -> "/dev/null") + message.validate() + // bad fields + var badConf = conf.clone().set("spark.driver.cores", "one hundred feet") + message.sparkProperties = badConf.getAll.toMap + intercept[SubmitRestProtocolException] { message.validate() } + badConf = conf.clone().set("spark.driver.supervise", "nope, never") + message.sparkProperties = badConf.getAll.toMap + intercept[SubmitRestProtocolException] { message.validate() } + badConf = conf.clone().set("spark.cores.max", "two men") + message.sparkProperties = badConf.getAll.toMap + intercept[SubmitRestProtocolException] { message.validate() } + message.sparkProperties = conf.getAll.toMap + // test JSON + val json = message.toJson + assertJsonEquals(json, submitDriverRequestJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[CreateSubmissionRequest]) + assert(newMessage.clientSparkVersion === "1.2.3") + assert(newMessage.appResource === "honey-walnut-cherry.jar") + assert(newMessage.mainClass === "org.apache.spark.examples.SparkPie") + assert(newMessage.sparkProperties("spark.app.name") === "SparkPie") + assert(newMessage.sparkProperties("spark.jars") === "mayonnaise.jar,ketchup.jar") + assert(newMessage.sparkProperties("spark.files") === "fireball.png") + assert(newMessage.sparkProperties("spark.driver.memory") === "512m") + assert(newMessage.sparkProperties("spark.driver.cores") === "180") + assert(newMessage.sparkProperties("spark.driver.extraJavaOptions") === " -Dslices=5 -Dcolor=mostly_red") + assert(newMessage.sparkProperties("spark.driver.extraClassPath") === "food-coloring.jar") + assert(newMessage.sparkProperties("spark.driver.extraLibraryPath") === "pickle.jar") + assert(newMessage.sparkProperties("spark.driver.supervise") === "false") + assert(newMessage.sparkProperties("spark.executor.memory") === "256m") + assert(newMessage.sparkProperties("spark.cores.max") === "10000") + assert(newMessage.appArgs === message.appArgs) + assert(newMessage.sparkProperties === message.sparkProperties) + assert(newMessage.environmentVariables === message.environmentVariables) + } + + test("CreateSubmissionResponse") { + val message = new CreateSubmissionResponse + intercept[SubmitRestProtocolException] { message.validate() } + message.serverSparkVersion = "1.2.3" + message.submissionId = "driver_123" + message.success = true + message.validate() + // test JSON + val json = message.toJson + assertJsonEquals(json, submitDriverResponseJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[CreateSubmissionResponse]) + assert(newMessage.serverSparkVersion === "1.2.3") + assert(newMessage.submissionId === "driver_123") + assert(newMessage.success) + } + + test("KillSubmissionResponse") { + val message = new KillSubmissionResponse + intercept[SubmitRestProtocolException] { message.validate() } + message.serverSparkVersion = "1.2.3" + message.submissionId = "driver_123" + message.success = true + message.validate() + // test JSON + val json = message.toJson + assertJsonEquals(json, killDriverResponseJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[KillSubmissionResponse]) + assert(newMessage.serverSparkVersion === "1.2.3") + assert(newMessage.submissionId === "driver_123") + assert(newMessage.success) + } + + test("SubmissionStatusResponse") { + val message = new SubmissionStatusResponse + intercept[SubmitRestProtocolException] { message.validate() } + message.serverSparkVersion = "1.2.3" + message.submissionId = "driver_123" + message.success = true + message.validate() + // optional fields + message.driverState = "RUNNING" + message.workerId = "worker_123" + message.workerHostPort = "1.2.3.4:7780" + // test JSON + val json = message.toJson + assertJsonEquals(json, driverStatusResponseJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[SubmissionStatusResponse]) + assert(newMessage.serverSparkVersion === "1.2.3") + assert(newMessage.submissionId === "driver_123") + assert(newMessage.driverState === "RUNNING") + assert(newMessage.success) + assert(newMessage.workerId === "worker_123") + assert(newMessage.workerHostPort === "1.2.3.4:7780") + } + + test("ErrorResponse") { + val message = new ErrorResponse + intercept[SubmitRestProtocolException] { message.validate() } + message.serverSparkVersion = "1.2.3" + message.message = "Field not found in submit request: X" + message.validate() + // test JSON + val json = message.toJson + assertJsonEquals(json, errorJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[ErrorResponse]) + assert(newMessage.serverSparkVersion === "1.2.3") + assert(newMessage.message === "Field not found in submit request: X") + } + + private val dummyRequestJson = + """ + |{ + | "action" : "DummyRequest", + | "active" : true, + | "age" : 25, + | "clientSparkVersion" : "1.2.3", + | "name" : "jung" + |} + """.stripMargin + + private val dummyResponseJson = + """ + |{ + | "action" : "DummyResponse", + | "serverSparkVersion" : "3.3.4", + | "success": true + |} + """.stripMargin + + private val submitDriverRequestJson = + """ + |{ + | "action" : "CreateSubmissionRequest", + | "appArgs" : [ "two slices", "a hint of cinnamon" ], + | "appResource" : "honey-walnut-cherry.jar", + | "clientSparkVersion" : "1.2.3", + | "environmentVariables" : { + | "PATH" : "/dev/null" + | }, + | "mainClass" : "org.apache.spark.examples.SparkPie", + | "sparkProperties" : { + | "spark.driver.extraLibraryPath" : "pickle.jar", + | "spark.jars" : "mayonnaise.jar,ketchup.jar", + | "spark.driver.supervise" : "false", + | "spark.app.name" : "SparkPie", + | "spark.cores.max" : "10000", + | "spark.driver.memory" : "512m", + | "spark.files" : "fireball.png", + | "spark.driver.cores" : "180", + | "spark.driver.extraJavaOptions" : " -Dslices=5 -Dcolor=mostly_red", + | "spark.executor.memory" : "256m", + | "spark.driver.extraClassPath" : "food-coloring.jar" + | } + |} + """.stripMargin + + private val submitDriverResponseJson = + """ + |{ + | "action" : "CreateSubmissionResponse", + | "serverSparkVersion" : "1.2.3", + | "submissionId" : "driver_123", + | "success" : true + |} + """.stripMargin + + private val killDriverResponseJson = + """ + |{ + | "action" : "KillSubmissionResponse", + | "serverSparkVersion" : "1.2.3", + | "submissionId" : "driver_123", + | "success" : true + |} + """.stripMargin + + private val driverStatusResponseJson = + """ + |{ + | "action" : "SubmissionStatusResponse", + | "driverState" : "RUNNING", + | "serverSparkVersion" : "1.2.3", + | "submissionId" : "driver_123", + | "success" : true, + | "workerHostPort" : "1.2.3.4:7780", + | "workerId" : "worker_123" + |} + """.stripMargin + + private val errorJson = + """ + |{ + | "action" : "ErrorResponse", + | "message" : "Field not found in submit request: X", + | "serverSparkVersion" : "1.2.3" + |} + """.stripMargin + + /** Assert that the contents in the two JSON strings are equal after ignoring whitespace. */ + private def assertJsonEquals(jsonString1: String, jsonString2: String): Unit = { + val trimmedJson1 = jsonString1.trim + val trimmedJson2 = jsonString2.trim + val json1 = compact(render(parse(trimmedJson1))) + val json2 = compact(render(parse(trimmedJson2))) + // Put this on a separate line to avoid printing comparison twice when test fails + val equals = json1 == json2 + assert(equals, "\"[%s]\" did not equal \"[%s]\"".format(trimmedJson1, trimmedJson2)) + } +} + +private class DummyResponse extends SubmitRestProtocolResponse +private class DummyRequest extends SubmitRestProtocolRequest { + var active: Boolean = null + var age: Integer = null + var name: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(name, "name") + assertFieldIsSet(age, "age") + assert(age > 5, "Not old enough!") + } +} diff --git a/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala b/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala index 855f1b6276089..054a4c64897a9 100644 --- a/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala +++ b/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala @@ -29,9 +29,9 @@ class KryoSerializerDistributedSuite extends FunSuite { test("kryo objects are serialised consistently in different processes") { val conf = new SparkConf(false) - conf.set("spark.serializer", "org.apache.spark.serializer.KryoSerializer") - conf.set("spark.kryo.registrator", classOf[AppJarRegistrator].getName) - conf.set("spark.task.maxFailures", "1") + .set("spark.serializer", "org.apache.spark.serializer.KryoSerializer") + .set("spark.kryo.registrator", classOf[AppJarRegistrator].getName) + .set("spark.task.maxFailures", "1") val jar = TestUtils.createJarWithClasses(List(AppJarRegistrator.customClassName)) conf.setJars(List(jar.getPath)) diff --git a/pom.xml b/pom.xml index aef450ae63121..da8ee077ddf85 100644 --- a/pom.xml +++ b/pom.xml @@ -154,6 +154,7 @@ org.scala-lang 3.6.3 1.8.8 + 2.4.4 1.1.1.6
    Executor IDTotal Tasks Failed Tasks Succeeded TasksInputOutputShuffle ReadShuffle WriteShuffle Spill (Memory)Shuffle Spill (Disk) + Input Size / Records + + Output Size / Records + + + Shuffle Read Size / Records + + + Shuffle Write Size / Records + Shuffle Spill (Memory)Shuffle Spill (Disk)
    {v.failedTasks + v.succeededTasks} {v.failedTasks} {v.succeededTasks} - {Utils.bytesToString(v.inputBytes)} - {Utils.bytesToString(v.outputBytes)} - {Utils.bytesToString(v.shuffleRead)} - {Utils.bytesToString(v.shuffleWrite)} - {Utils.bytesToString(v.memoryBytesSpilled)} - {Utils.bytesToString(v.diskBytesSpilled)} + {s"${Utils.bytesToString(v.inputBytes)} / ${v.inputRecords}"} + + {s"${Utils.bytesToString(v.outputBytes)} / ${v.outputRecords}"} + + {s"${Utils.bytesToString(v.shuffleRead)} / ${v.shuffleReadRecords}"} + + {s"${Utils.bytesToString(v.shuffleWrite)} / ${v.shuffleWriteRecords}"} + + {Utils.bytesToString(v.memoryBytesSpilled)} + + {Utils.bytesToString(v.diskBytesSpilled)} +
    {UIUtils.formatDuration(millis.toLong)}{Utils.bytesToString(d.toLong)}{Utils.bytesToString(d.toLong)}{s"${Utils.bytesToString(d.toLong)} / ${recordDist.next().toLong}"}InputInput Size / RecordsOutputOutput Size / RecordsShuffle Read (Remote)Shuffle Read Size / Records (Remote)Shuffle WriteShuffle Write Size / Records
    - {inputReadable} + {s"$inputReadable / $inputRecords"} - {outputReadable} + {s"$outputReadable / $outputRecords"} - {shuffleReadReadable} + {s"$shuffleReadReadable / $shuffleReadRecords"} - {shuffleWriteReadable} + {s"$shuffleWriteReadable / $shuffleWriteRecords"}