[SPARK-2393][SQL] Cost estimation optimization framework for Catalyst logical plans & sample usage. #1238
Conversation
|
Merged build triggered. |
|
Merged build started. |
|
Merged build triggered. |
|
Merged build started. |
|
Merged build finished. |
|
Refer to this link for build results: https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/16174/ |
|
Merged build finished. |
|
Refer to this link for build results: https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/16173/ |
| @@ -72,8 +72,9 @@ private[sql] abstract class SparkStrategies extends QueryPlanner[SparkPlan] { | |||
| condition, | |||
| left, | |||
| right @ PhysicalOperation(_, _, b: BaseRelation)) | |||
| if broadcastTables.contains(b.tableName) => | |||
| broadcastHashJoin(leftKeys, rightKeys, left, right, condition, BuildRight) | |||
| if broadcastTables.contains(b.tableName) | |||
There was a problem hiding this comment.
remove broadcastTables.contains(b.tableName) and then your if can fit in a single line, and you don't need to indent the next line ...
|
Merged build triggered. |
|
Merged build started. |
|
Merged build triggered. |
|
Merged build started. |
|
Merged build finished. |
|
Refer to this link for build results: https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/16316/ |
|
Jenkins, retest this please. |
|
Merged build triggered. |
|
Merged build started. |
|
Merged build finished. |
|
Refer to this link for build results: https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/16319/ |
|
Merged build finished. |
|
Refer to this link for build results: https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/16317/ |
|
Merged build triggered. |
|
Merged build started. |
Additionally, remove size estimate from ParquetRelation since the Hadoop FileSystem API calls can be expensive (e.g. S3FileSystem has a lot of RPCs).
Also cleanups & scaladoc fixes per review comments.
|
Rebased & addressed review comments. |
|
Jenkins test this please |
|
QA tests have started for PR 1238. This patch merges cleanly. |
|
QA results for PR 1238: |
|
Thanks! I've merged this into master. |
… logical plans & sample usage. The idea is that every Catalyst logical plan gets hold of a Statistics class, the usage of which provides useful estimations on various statistics. See the implementations of `MetastoreRelation`. This patch also includes several usages of the estimation interface in the planner. For instance, we now use physical table sizes from the estimate interface to convert an equi-join to a broadcast join (when doing so is beneficial, as determined by a size threshold). Finally, there are a couple minor accompanying changes including: - Remove the not-in-use `BaseRelation`. - Make SparkLogicalPlan take a `SQLContext` in the second param list. Author: Zongheng Yang <[email protected]> Closes apache#1238 from concretevitamin/estimates and squashes the following commits: 329071d [Zongheng Yang] Address review comments; turn config name from string to field in SQLConf. 8663e84 [Zongheng Yang] Use BigInt for stat; for logical leaves, by default throw an exception. 2f2fb89 [Zongheng Yang] Fix statistics for SparkLogicalPlan. 9951305 [Zongheng Yang] Remove childrenStats. 16fc60a [Zongheng Yang] Avoid calling statistics on plans if auto join conversion is disabled. 8bd2816 [Zongheng Yang] Add a note on performance of statistics. 6e594b8 [Zongheng Yang] Get size info from metastore for MetastoreRelation. 01b7a3e [Zongheng Yang] Update scaladoc for a field and move it to @param section. 549061c [Zongheng Yang] Remove numTuples in Statistics for now. 729a8e2 [Zongheng Yang] Update docs to be more explicit. 573e644 [Zongheng Yang] Remove singleton SQLConf and move back `settings` to the trait. 2d99eb5 [Zongheng Yang] {Cleanup, use synchronized in, enrich} StatisticsSuite. ca5b825 [Zongheng Yang] Inject SQLContext into SparkLogicalPlan, removing SQLConf mixin from it. 43d38a6 [Zongheng Yang] Revert optimization for BroadcastNestedLoopJoin (this fixes tests). 0ef9e5b [Zongheng Yang] Use multiplication instead of sum for default estimates. 4ef0d26 [Zongheng Yang] Make Statistics a case class. 3ba8f3e [Zongheng Yang] Add comment. e5bcf5b [Zongheng Yang] Fix optimization conditions & update scala docs to explain. 7d9216a [Zongheng Yang] Apply estimation to planning ShuffleHashJoin & BroadcastNestedLoopJoin. 73cde01 [Zongheng Yang] Move SQLConf back. Assign default sizeInBytes to SparkLogicalPlan. 73412be [Zongheng Yang] Move SQLConf to Catalyst & add default val for sizeInBytes. 7a60ab7 [Zongheng Yang] s/Estimates/Statistics, s/cardinality/numTuples. de3ae13 [Zongheng Yang] Add parquetAfter() properly in test. dcff9bd [Zongheng Yang] Cleanups. 84301a4 [Zongheng Yang] Refactors. 5bf5586 [Zongheng Yang] Typo. 56a8e6e [Zongheng Yang] Prototype impl of estimations for Catalyst logical plans.
* [CARMEL-6439] Add configuration to enable log column lineage * remove log * useless condition * skip empty lineage
The idea is that every Catalyst logical plan gets hold of a Statistics class, the usage of which provides useful estimations on various statistics. See the implementations of
MetastoreRelation.This patch also includes several usages of the estimation interface in the planner. For instance, we now use physical table sizes from the estimate interface to convert an equi-join to a broadcast join (when doing so is beneficial, as determined by a size threshold).
Finally, there are a couple minor accompanying changes including:
BaseRelation.SQLContextin the second param list.