Skip to content

[SPARK-23614][SQL] Fix incorrect reuse exchange when caching is used #20831

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
viirya wants to merge 4 commits into from
Closed

[SPARK-23614][SQL] Fix incorrect reuse exchange when caching is used #20831

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
e1f28e2
Fix incorrect reuse exchange when caching is used.
viirya Mar 15, 2018
a3f86b2
Address comment.
viirya Mar 15, 2018
fb0f949
Fix error.
viirya Mar 15, 2018
a592882
Make few variables as lazy.
viirya Mar 21, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 10 additions & 0 deletions sql/core/src/main/scala/org/apache/spark/sql/execution/columnar/InMemoryRelation.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,7 @@ import org.apache.spark.rdd.RDD
import org.apache.spark.sql.catalyst.InternalRow
import org.apache.spark.sql.catalyst.analysis.MultiInstanceRelation
import org.apache.spark.sql.catalyst.expressions._
import org.apache.spark.sql.catalyst.plans.QueryPlan
import org.apache.spark.sql.catalyst.plans.logical
import org.apache.spark.sql.catalyst.plans.logical.{HintInfo, Statistics}
import org.apache.spark.sql.execution.SparkPlan
Expand Down Expand Up @@ -68,6 +69,15 @@ case class InMemoryRelation(

override protected def innerChildren: Seq[SparkPlan] = Seq(child)

override def doCanonicalize(): logical.LogicalPlan =
copy(output = output.map(QueryPlan.normalizeExprId(_, child.output)),
storageLevel = StorageLevel.NONE,
Copy link
Contributor

@cloud-fan cloud-fan Mar 20, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can we follow the parameter order in the constructor?

Copy link
Member Author

@viirya viirya Mar 21, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is followed. I just ignored useCompression, batchSize as they are just primitives and don't need to be canonicalized here.

child = child.canonicalized,
tableName = None)(
_cachedColumnBuffers,
sizeInBytesStats,
statsOfPlanToCache)
Copy link
Member

@maropu maropu Mar 15, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We need to copy these cached data?

Copy link
Member Author

@viirya viirya Mar 15, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

_cachedColumnBuffers can't be null. If it is null, copy will trigger buildBuffers.

Copy link
Member

@maropu maropu Mar 16, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

statsOfPlanToCache and sizeInBytesStats, too? For instance, ResolveHint drops hints in canonicalization:

spark/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/hints.scala

Line 44 in 3675af7

override def doCanonicalize(): LogicalPlan = child.canonicalized

Copy link
Member Author

@viirya viirya Mar 16, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

cachedColumnBuffers, sizeInBytesStats, statsOfPlanToCache won't be considered when comparing two InMemoryRelation. So instead of create empty instances of statistics, I just use the original values.

Copy link
Member

@maropu maropu Mar 16, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

aha, ok. Thanks!


override def producedAttributes: AttributeSet = outputSet

@transient val partitionStatistics = new PartitionStatistics(output)
Expand Down
7 changes: 6 additions & 1 deletion sql/core/src/main/scala/org/apache/spark/sql/execution/columnar/InMemoryTableScanExec.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,7 @@ import org.apache.spark.sql.catalyst.dsl.expressions._
import org.apache.spark.sql.catalyst.expressions._
import org.apache.spark.sql.catalyst.plans.QueryPlan
import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
import org.apache.spark.sql.execution.{ColumnarBatchScan, LeafExecNode, WholeStageCodegenExec}
import org.apache.spark.sql.execution.{ColumnarBatchScan, LeafExecNode, SparkPlan, WholeStageCodegenExec}
import org.apache.spark.sql.execution.vectorized._
import org.apache.spark.sql.types._
import org.apache.spark.sql.vectorized.{ColumnarBatch, ColumnVector}
Expand All @@ -38,6 +38,11 @@ case class InMemoryTableScanExec(

override protected def innerChildren: Seq[QueryPlan[_]] = Seq(relation) ++ super.innerChildren

override def doCanonicalize(): SparkPlan =
copy(attributes = attributes.map(QueryPlan.normalizeExprId(_, relation.output)),
predicates = predicates.map(QueryPlan.normalizeExprId(_, relation.output)),
relation = relation.canonicalized.asInstanceOf[InMemoryRelation])

override def vectorTypes: Option[Seq[String]] =
Option(Seq.fill(attributes.length)(
if (!conf.offHeapColumnVectorEnabled) {
Expand Down
9 changes: 9 additions & 0 deletions sql/core/src/test/scala/org/apache/spark/sql/DatasetSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1446,8 +1446,17 @@ class DatasetSuite extends QueryTest with SharedSQLContext {
val data = Seq(("a", null))
checkDataset(data.toDS(), data: _*)
}

test("SPARK-23614: Union produces incorrect results when caching is used") {
val cached = spark.createDataset(Seq(TestDataUnion(1, 2, 3), TestDataUnion(4, 5, 6))).cache()
val group1 = cached.groupBy("x").agg(min(col("y")) as "value")
val group2 = cached.groupBy("x").agg(min(col("z")) as "value")
checkAnswer(group1.union(group2), Row(4, 5) :: Row(1, 2) :: Row(4, 6) :: Row(1, 3) :: Nil)
}
}

case class TestDataUnion(x: Int, y: Int, z: Int)

case class SingleData(id: Int)
case class DoubleData(id: Int, val1: String)
case class TripleData(id: Int, val1: String, val2: Long)
Expand Down
7 changes: 7 additions & 0 deletions sql/core/src/test/scala/org/apache/spark/sql/execution/ExchangeSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,4 +125,11 @@ class ExchangeSuite extends SparkPlanTest with SharedSQLContext {
assertConsistency(spark.range(10000).map(i => Random.nextInt(1000).toLong))
}
}

test("SPARK-23614: Fix incorrect reuse exchange when caching is used") {
val cached = spark.createDataset(Seq((1, 2, 3), (4, 5, 6))).cache()
val projection1 = cached.select("_1", "_2").queryExecution.executedPlan
val projection2 = cached.select("_1", "_3").queryExecution.executedPlan
assert(!projection1.sameResult(projection2))
}
}