[SPARK-35303][PYTHON] Enable pinned thread mode by default

core/src/main/scala/org/apache/spark/api/python/Py4JServer.scala

@@ -34,7 +34,7 @@ private[spark] class Py4JServer(sparkConf: SparkConf) extends Logging {
   // Java system properties and such
   private val localhost = InetAddress.getLoopbackAddress()
   private[spark] val server = if (sys.env.getOrElse(
-      "PYSPARK_PIN_THREAD", "false").toLowerCase(Locale.ROOT) == "true") {
+      "PYSPARK_PIN_THREAD", "true").toLowerCase(Locale.ROOT) == "true") {
     new py4j.ClientServer.ClientServerBuilder()
       .authToken(secret)
       .javaPort(0)

docs/job-scheduling.md

@@ -297,10 +297,6 @@ in each corresponding JVM thread. Due to this limitation, it is unable to set a
 via `sc.setJobGroup` in a separate PVM thread, which also disallows to cancel the job via `sc.cancelJobGroup`
 later.
 
-In order to synchronize PVM threads with JVM threads, you should set `PYSPARK_PIN_THREAD` environment variable
-to `true`. This pinned thread mode allows one PVM thread has one corresponding JVM thread. With this mode,
 `pyspark.InheritableThread` is recommended to use together for a PVM thread to inherit the inheritable attributes
- such as local properties in a JVM thread.
-
-Note that `PYSPARK_PIN_THREAD` is currently experimental and not recommended for use in production.
+ such as local properties in a JVM thread, and to avoid resource leak.
 

python/docs/source/migration_guide/pyspark_3.1_to_3.2.rst

@@ -23,3 +23,9 @@ Upgrading from PySpark 3.1 to 3.2
 * In Spark 3.2, the PySpark methods from sql, ml, spark_on_pandas modules raise the ``TypeError`` instead of ``ValueError`` when are applied to an param of inappropriate type.
 
 * In Spark 3.2, the traceback from Python UDFs, pandas UDFs and pandas function APIs are simplified by default without the traceback from the internal Python workers. In Spark 3.1 or earlier, the traceback from Python workers was printed out. To restore the behavior before Spark 3.2, you can set ``spark.sql.execution.pyspark.udf.simplifiedTraceback.enabled`` to ``false``.
+
+* In Spark 3.2, pinned thread mode is enabled by default to map each Python thread to the corresponding JVM thread. Previously,
+  one JVM thread could be reused for multiple Python threads, which resulted in one JVM thread local being shared to multiple Python threads.
+  Also, note that now ``pyspark.InheritableThread`` or ``pyspark.inheritable_thread_target`` is recommended to use together for a Python thread
+  to properly inherit the inheritable attributes such as local properties in a JVM thread, and to avoid a potential resource leak issue.
+  To restore the behavior before Spark 3.2, you can set ``PYSPARK_PIN_THREAD`` environment variable to ``false``.

python/docs/source/reference/pyspark.rst

@@ -41,7 +41,7 @@ Public Classes
     BarrierTaskContext
     BarrierTaskInfo
     InheritableThread
-
+    util.VersionUtils
 
 Spark Context APIs
 ------------------
@@ -239,6 +239,7 @@ Management
 .. autosummary::
     :toctree: api/
 
+    inheritable_thread_target
     SparkConf.contains
     SparkConf.get
     SparkConf.getAll
@@ -279,3 +280,5 @@ Management
     BarrierTaskContext.resources
     BarrierTaskContext.stageId
     BarrierTaskContext.taskAttemptId
+    util.VersionUtils.majorMinorVersion
+

python/pyspark/__init__.py

@@ -53,7 +53,7 @@
 from pyspark.rdd import RDD, RDDBarrier
 from pyspark.files import SparkFiles
 from pyspark.status import StatusTracker, SparkJobInfo, SparkStageInfo
-from pyspark.util import InheritableThread
+from pyspark.util import InheritableThread, inheritable_thread_target
 from pyspark.storagelevel import StorageLevel
 from pyspark.accumulators import Accumulator, AccumulatorParam
 from pyspark.broadcast import Broadcast
@@ -125,5 +125,5 @@ def wrapper(self, *args, **kwargs):
     "Accumulator", "AccumulatorParam", "MarshalSerializer", "PickleSerializer",
     "StatusTracker", "SparkJobInfo", "SparkStageInfo", "Profiler", "BasicProfiler", "TaskContext",
     "RDDBarrier", "BarrierTaskContext", "BarrierTaskInfo", "InheritableThread",
-    "__version__",
+    "inheritable_thread_target", "__version__",
 ]

python/pyspark/__init__.pyi

@@ -51,7 +51,10 @@ from pyspark.taskcontext import (  # noqa: F401
     BarrierTaskInfo as BarrierTaskInfo,
     TaskContext as TaskContext,
 )
-from pyspark.util import InheritableThread as InheritableThread  # noqa: F401
+from pyspark.util import (
+    InheritableThread as InheritableThread,  # noqa: F401
+    inheritable_thread_target as inheritable_thread_target,  # noqa: F401
+)
 from pyspark.version import __version__ as __version__
 
 # Compatibility imports

python/pyspark/context.py

@@ -1104,13 +1104,8 @@ def setJobGroup(self, groupId, description, interruptOnCancel=False):
         ensure that the tasks are actually stopped in a timely manner, but is off by default due
         to HDFS-1208, where HDFS may respond to Thread.interrupt() by marking nodes as dead.
 
-        Currently, setting a group ID (set to local properties) with multiple threads
-        does not properly work. Internally threads on PVM and JVM are not synced, and JVM
-        thread can be reused for multiple threads on PVM, which fails to isolate local
-        properties for each thread on PVM.
-
-        To avoid this, enable the pinned thread mode by setting ``PYSPARK_PIN_THREAD``
-        environment variable to ``true`` and uses :class:`pyspark.InheritableThread`.
+        If you run jobs in parallel, use :class:`pyspark.InheritableThread` for thread
+        local inheritance, and preventing resource leak.
 
         Examples
         --------
@@ -1148,13 +1143,8 @@ def setLocalProperty(self, key, value):
 
         Notes
         -----
-        Currently, setting a local property with multiple threads does not properly work.
-        Internally threads on PVM and JVM are not synced, and JVM thread
-        can be reused for multiple threads on PVM, which fails to isolate local properties
-        for each thread on PVM.
-
-        To avoid this, enable the pinned thread mode by setting ``PYSPARK_PIN_THREAD``
-        environment variable to ``true`` and uses :class:`pyspark.InheritableThread`.
+        If you run jobs in parallel, use :class:`pyspark.InheritableThread` for thread
+        local inheritance, and preventing resource leak.
         """
         self._jsc.setLocalProperty(key, value)
 
@@ -1171,13 +1161,8 @@ def setJobDescription(self, value):
 
         Notes
         -----
-        Currently, setting a job description (set to local properties) with multiple
-        threads does not properly work. Internally threads on PVM and JVM are not synced,
-        and JVM thread can be reused for multiple threads on PVM, which fails to isolate
-        local properties for each thread on PVM.
-
-        To avoid this, enable the pinned thread mode by setting ``PYSPARK_PIN_THREAD``
-        environment variable to ``true`` and uses :class:`pyspark.InheritableThread`.
+        If you run jobs in parallel, use :class:`pyspark.InheritableThread` for thread
+        local inheritance, and preventing resource leak.
         """
         self._jsc.setJobDescription(value)
 

python/pyspark/java_gateway.py

@@ -130,7 +130,7 @@ def killChild():
             atexit.register(killChild)
 
     # Connect to the gateway (or client server to pin the thread between JVM and Python)
-    if os.environ.get("PYSPARK_PIN_THREAD", "false").lower() == "true":
+    if os.environ.get("PYSPARK_PIN_THREAD", "true").lower() == "true":
         gateway = ClientServer(
             java_parameters=JavaParameters(
                 port=gateway_port,

python/pyspark/serializers.py

@@ -65,7 +65,7 @@
 pickle_protocol = pickle.HIGHEST_PROTOCOL
 
 from pyspark import cloudpickle
-from pyspark.util import print_exec
+from pyspark.util import print_exec  # type: ignore
 
 
 __all__ = ["PickleSerializer", "MarshalSerializer", "UTF8Deserializer"]

python/pyspark/shuffle.py

@@ -28,7 +28,7 @@
 import heapq
 from pyspark.serializers import BatchedSerializer, PickleSerializer, FlattenedValuesSerializer, \
     CompressedSerializer, AutoBatchedSerializer
-from pyspark.util import fail_on_stopiteration
+from pyspark.util import fail_on_stopiteration  # type: ignore
 
 
 try:

python/pyspark/tests/test_context.py

@@ -220,7 +220,7 @@ def test_progress_api(self):
             def run():
                 # When thread is pinned, job group should be set for each thread for now.
                 # Local properties seem not being inherited like Scala side does.
-                if os.environ.get("PYSPARK_PIN_THREAD", "false").lower() == "true":
+                if os.environ.get("PYSPARK_PIN_THREAD", "true").lower() == "true":
                     sc.setJobGroup('test_progress_api', '', True)
                 try:
                     rdd.count()

python/pyspark/util.py

@@ -276,13 +276,13 @@ def inheritable_thread_target(f):
 
     When the pinned thread mode is off, it return the original ``f``.
 
+    .. versionadded:: 3.2.0
+
     Parameters
     ----------
     f : function
         the original thread target.
 
-    .. versionadded:: 3.2.0
-
     Notes
     -----
     This API is experimental.

python/pyspark/util.pyi

@@ -0,0 +1,29 @@
+#
+# 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.
+
+import threading
+from typing import Callable
+
+class VersionUtils(object):
+    @staticmethod
+    def majorMinorVersion(sparkVersion: str): ...
+
+def inheritable_thread_target(f: Callable) -> Callable: ...
+
+class InheritableThread(threading.Thread):
+    pass

python/pyspark/worker.py

@@ -45,7 +45,7 @@
 from pyspark.sql.pandas.serializers import ArrowStreamPandasUDFSerializer, CogroupUDFSerializer
 from pyspark.sql.pandas.types import to_arrow_type
 from pyspark.sql.types import StructType
-from pyspark.util import fail_on_stopiteration, try_simplify_traceback
+from pyspark.util import fail_on_stopiteration, try_simplify_traceback  # type: ignore
 from pyspark import shuffle
 
 pickleSer = PickleSerializer()