Repository: spark Updated Branches: refs/heads/master b9a6f7499 -> dc8a6befa
[SPARK-24588][SS] streaming join should require HashClusteredPartitioning from children ## What changes were proposed in this pull request? In https://github.com/apache/spark/pull/19080 we simplified the distribution/partitioning framework, and make all the join-like operators require `HashClusteredDistribution` from children. Unfortunately streaming join operator was missed. This can cause wrong result. Think about ``` val input1 = MemoryStream[Int] val input2 = MemoryStream[Int] val df1 = input1.toDF.select('value as 'a, 'value * 2 as 'b) val df2 = input2.toDF.select('value as 'a, 'value * 2 as 'b).repartition('b) val joined = df1.join(df2, Seq("a", "b")).select('a) ``` The physical plan is ``` *(3) Project [a#5] +- StreamingSymmetricHashJoin [a#5, b#6], [a#10, b#11], Inner, condition = [ leftOnly = null, rightOnly = null, both = null, full = null ], state info [ checkpoint = <unknown>, runId = 54e31fce-f055-4686-b75d-fcd2b076f8d8, opId = 0, ver = 0, numPartitions = 5], 0, state cleanup [ left = null, right = null ] :- Exchange hashpartitioning(a#5, b#6, 5) : +- *(1) Project [value#1 AS a#5, (value#1 * 2) AS b#6] : +- StreamingRelation MemoryStream[value#1], [value#1] +- Exchange hashpartitioning(b#11, 5) +- *(2) Project [value#3 AS a#10, (value#3 * 2) AS b#11] +- StreamingRelation MemoryStream[value#3], [value#3] ``` The left table is hash partitioned by `a, b`, while the right table is hash partitioned by `b`. This means, we may have a matching record that is in different partitions, which should be in the output but not. ## How was this patch tested? N/A Author: Wenchen Fan <wenc...@databricks.com> Closes #21587 from cloud-fan/join. Project: http://git-wip-us.apache.org/repos/asf/spark/repo Commit: http://git-wip-us.apache.org/repos/asf/spark/commit/dc8a6bef Tree: http://git-wip-us.apache.org/repos/asf/spark/tree/dc8a6bef Diff: http://git-wip-us.apache.org/repos/asf/spark/diff/dc8a6bef Branch: refs/heads/master Commit: dc8a6befa5dad861a731b4d7865f3ccf37482ae0 Parents: b9a6f74 Author: Wenchen Fan <wenc...@databricks.com> Authored: Thu Jun 21 15:38:46 2018 -0700 Committer: Xiao Li <gatorsm...@gmail.com> Committed: Thu Jun 21 15:38:46 2018 -0700 ---------------------------------------------------------------------- .../catalyst/plans/physical/partitioning.scala | 53 +++-- .../spark/sql/catalyst/DistributionSuite.scala | 201 +++++++++++++++---- .../datasources/v2/DataSourcePartitioning.scala | 4 +- .../StreamingSymmetricHashJoinExec.scala | 4 +- .../sql/streaming/StreamingJoinSuite.scala | 14 ++ 5 files changed, 217 insertions(+), 59 deletions(-) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/spark/blob/dc8a6bef/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala ---------------------------------------------------------------------- diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala index 4d9a992..cc1a5e8 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala @@ -99,16 +99,19 @@ case class ClusteredDistribution( * This is a strictly stronger guarantee than [[ClusteredDistribution]]. Given a tuple and the * number of partitions, this distribution strictly requires which partition the tuple should be in. */ -case class HashClusteredDistribution(expressions: Seq[Expression]) extends Distribution { +case class HashClusteredDistribution( + expressions: Seq[Expression], + requiredNumPartitions: Option[Int] = None) extends Distribution { require( expressions != Nil, - "The expressions for hash of a HashPartitionedDistribution should not be Nil. " + + "The expressions for hash of a HashClusteredDistribution should not be Nil. " + "An AllTuples should be used to represent a distribution that only has " + "a single partition.") - override def requiredNumPartitions: Option[Int] = None - override def createPartitioning(numPartitions: Int): Partitioning = { + assert(requiredNumPartitions.isEmpty || requiredNumPartitions.get == numPartitions, + s"This HashClusteredDistribution requires ${requiredNumPartitions.get} partitions, but " + + s"the actual number of partitions is $numPartitions.") HashPartitioning(expressions, numPartitions) } } @@ -163,11 +166,22 @@ trait Partitioning { * i.e. the current dataset does not need to be re-partitioned for the `required` * Distribution (it is possible that tuples within a partition need to be reorganized). * + * A [[Partitioning]] can never satisfy a [[Distribution]] if its `numPartitions` does't match + * [[Distribution.requiredNumPartitions]]. + */ + final def satisfies(required: Distribution): Boolean = { + required.requiredNumPartitions.forall(_ == numPartitions) && satisfies0(required) + } + + /** + * The actual method that defines whether this [[Partitioning]] can satisfy the given + * [[Distribution]], after the `numPartitions` check. + * * By default a [[Partitioning]] can satisfy [[UnspecifiedDistribution]], and [[AllTuples]] if - * the [[Partitioning]] only have one partition. Implementations can overwrite this method with - * special logic. + * the [[Partitioning]] only have one partition. Implementations can also overwrite this method + * with special logic. */ - def satisfies(required: Distribution): Boolean = required match { + protected def satisfies0(required: Distribution): Boolean = required match { case UnspecifiedDistribution => true case AllTuples => numPartitions == 1 case _ => false @@ -186,9 +200,8 @@ case class RoundRobinPartitioning(numPartitions: Int) extends Partitioning case object SinglePartition extends Partitioning { val numPartitions = 1 - override def satisfies(required: Distribution): Boolean = required match { + override def satisfies0(required: Distribution): Boolean = required match { case _: BroadcastDistribution => false - case ClusteredDistribution(_, Some(requiredNumPartitions)) => requiredNumPartitions == 1 case _ => true } } @@ -205,16 +218,15 @@ case class HashPartitioning(expressions: Seq[Expression], numPartitions: Int) override def nullable: Boolean = false override def dataType: DataType = IntegerType - override def satisfies(required: Distribution): Boolean = { - super.satisfies(required) || { + override def satisfies0(required: Distribution): Boolean = { + super.satisfies0(required) || { required match { case h: HashClusteredDistribution => expressions.length == h.expressions.length && expressions.zip(h.expressions).forall { case (l, r) => l.semanticEquals(r) } - case ClusteredDistribution(requiredClustering, requiredNumPartitions) => - expressions.forall(x => requiredClustering.exists(_.semanticEquals(x))) && - (requiredNumPartitions.isEmpty || requiredNumPartitions.get == numPartitions) + case ClusteredDistribution(requiredClustering, _) => + expressions.forall(x => requiredClustering.exists(_.semanticEquals(x))) case _ => false } } @@ -246,15 +258,14 @@ case class RangePartitioning(ordering: Seq[SortOrder], numPartitions: Int) override def nullable: Boolean = false override def dataType: DataType = IntegerType - override def satisfies(required: Distribution): Boolean = { - super.satisfies(required) || { + override def satisfies0(required: Distribution): Boolean = { + super.satisfies0(required) || { required match { case OrderedDistribution(requiredOrdering) => val minSize = Seq(requiredOrdering.size, ordering.size).min requiredOrdering.take(minSize) == ordering.take(minSize) - case ClusteredDistribution(requiredClustering, requiredNumPartitions) => - ordering.map(_.child).forall(x => requiredClustering.exists(_.semanticEquals(x))) && - (requiredNumPartitions.isEmpty || requiredNumPartitions.get == numPartitions) + case ClusteredDistribution(requiredClustering, _) => + ordering.map(_.child).forall(x => requiredClustering.exists(_.semanticEquals(x))) case _ => false } } @@ -295,7 +306,7 @@ case class PartitioningCollection(partitionings: Seq[Partitioning]) * Returns true if any `partitioning` of this collection satisfies the given * [[Distribution]]. */ - override def satisfies(required: Distribution): Boolean = + override def satisfies0(required: Distribution): Boolean = partitionings.exists(_.satisfies(required)) override def toString: String = { @@ -310,7 +321,7 @@ case class PartitioningCollection(partitionings: Seq[Partitioning]) case class BroadcastPartitioning(mode: BroadcastMode) extends Partitioning { override val numPartitions: Int = 1 - override def satisfies(required: Distribution): Boolean = required match { + override def satisfies0(required: Distribution): Boolean = required match { case BroadcastDistribution(m) if m == mode => true case _ => false } http://git-wip-us.apache.org/repos/asf/spark/blob/dc8a6bef/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/DistributionSuite.scala ---------------------------------------------------------------------- diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/DistributionSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/DistributionSuite.scala index b47b8ad..3922810 100644 --- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/DistributionSuite.scala +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/DistributionSuite.scala @@ -41,34 +41,127 @@ class DistributionSuite extends SparkFunSuite { } } - test("HashPartitioning (with nullSafe = true) is the output partitioning") { - // Cases which do not need an exchange between two data properties. + test("UnspecifiedDistribution and AllTuples") { + // except `BroadcastPartitioning`, all other partitioning can satisfy UnspecifiedDistribution checkSatisfied( - HashPartitioning(Seq('a, 'b, 'c), 10), + UnknownPartitioning(-1), UnspecifiedDistribution, true) checkSatisfied( - HashPartitioning(Seq('a, 'b, 'c), 10), - ClusteredDistribution(Seq('a, 'b, 'c)), + RoundRobinPartitioning(10), + UnspecifiedDistribution, true) checkSatisfied( - HashPartitioning(Seq('b, 'c), 10), - ClusteredDistribution(Seq('a, 'b, 'c)), + SinglePartition, + UnspecifiedDistribution, + true) + + checkSatisfied( + HashPartitioning(Seq('a), 10), + UnspecifiedDistribution, + true) + + checkSatisfied( + RangePartitioning(Seq('a.asc), 10), + UnspecifiedDistribution, + true) + + checkSatisfied( + BroadcastPartitioning(IdentityBroadcastMode), + UnspecifiedDistribution, + false) + + // except `BroadcastPartitioning`, all other partitioning can satisfy AllTuples if they have + // only one partition. + checkSatisfied( + UnknownPartitioning(1), + AllTuples, + true) + + checkSatisfied( + UnknownPartitioning(10), + AllTuples, + false) + + checkSatisfied( + RoundRobinPartitioning(1), + AllTuples, + true) + + checkSatisfied( + RoundRobinPartitioning(10), + AllTuples, + false) + + checkSatisfied( + SinglePartition, + AllTuples, + true) + + checkSatisfied( + HashPartitioning(Seq('a), 1), + AllTuples, true) checkSatisfied( + HashPartitioning(Seq('a), 10), + AllTuples, + false) + + checkSatisfied( + RangePartitioning(Seq('a.asc), 1), + AllTuples, + true) + + checkSatisfied( + RangePartitioning(Seq('a.asc), 10), + AllTuples, + false) + + checkSatisfied( + BroadcastPartitioning(IdentityBroadcastMode), + AllTuples, + false) + } + + test("SinglePartition is the output partitioning") { + // SinglePartition can satisfy all the distributions except `BroadcastDistribution` + checkSatisfied( SinglePartition, ClusteredDistribution(Seq('a, 'b, 'c)), true) checkSatisfied( SinglePartition, + HashClusteredDistribution(Seq('a, 'b, 'c)), + true) + + checkSatisfied( + SinglePartition, OrderedDistribution(Seq('a.asc, 'b.asc, 'c.asc)), true) - // Cases which need an exchange between two data properties. + checkSatisfied( + SinglePartition, + BroadcastDistribution(IdentityBroadcastMode), + false) + } + + test("HashPartitioning is the output partitioning") { + // HashPartitioning can satisfy ClusteredDistribution iff its hash expressions are a subset of + // the required clustering expressions. + checkSatisfied( + HashPartitioning(Seq('a, 'b, 'c), 10), + ClusteredDistribution(Seq('a, 'b, 'c)), + true) + + checkSatisfied( + HashPartitioning(Seq('b, 'c), 10), + ClusteredDistribution(Seq('a, 'b, 'c)), + true) + checkSatisfied( HashPartitioning(Seq('a, 'b, 'c), 10), ClusteredDistribution(Seq('b, 'c)), @@ -79,37 +172,43 @@ class DistributionSuite extends SparkFunSuite { ClusteredDistribution(Seq('d, 'e)), false) + // HashPartitioning can satisfy HashClusteredDistribution iff its hash expressions are exactly + // same with the required hash clustering expressions. checkSatisfied( HashPartitioning(Seq('a, 'b, 'c), 10), - AllTuples, + HashClusteredDistribution(Seq('a, 'b, 'c)), + true) + + checkSatisfied( + HashPartitioning(Seq('c, 'b, 'a), 10), + HashClusteredDistribution(Seq('a, 'b, 'c)), false) checkSatisfied( + HashPartitioning(Seq('a, 'b), 10), + HashClusteredDistribution(Seq('a, 'b, 'c)), + false) + + // HashPartitioning cannot satisfy OrderedDistribution + checkSatisfied( HashPartitioning(Seq('a, 'b, 'c), 10), OrderedDistribution(Seq('a.asc, 'b.asc, 'c.asc)), false) checkSatisfied( - HashPartitioning(Seq('b, 'c), 10), + HashPartitioning(Seq('a, 'b, 'c), 1), OrderedDistribution(Seq('a.asc, 'b.asc, 'c.asc)), - false) + false) // TODO: this can be relaxed. - // TODO: We should check functional dependencies - /* checkSatisfied( - ClusteredDistribution(Seq('b)), - ClusteredDistribution(Seq('b + 1)), - true) - */ + HashPartitioning(Seq('b, 'c), 10), + OrderedDistribution(Seq('a.asc, 'b.asc, 'c.asc)), + false) } test("RangePartitioning is the output partitioning") { - // Cases which do not need an exchange between two data properties. - checkSatisfied( - RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), - UnspecifiedDistribution, - true) - + // RangePartitioning can satisfy OrderedDistribution iff its ordering is a prefix + // of the required ordering, or the required ordering is a prefix of its ordering. checkSatisfied( RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), OrderedDistribution(Seq('a.asc, 'b.asc, 'c.asc)), @@ -125,6 +224,27 @@ class DistributionSuite extends SparkFunSuite { OrderedDistribution(Seq('a.asc, 'b.asc, 'c.asc, 'd.desc)), true) + // TODO: We can have an optimization to first sort the dataset + // by a.asc and then sort b, and c in a partition. This optimization + // should tradeoff the benefit of a less number of Exchange operators + // and the parallelism. + checkSatisfied( + RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), + OrderedDistribution(Seq('a.asc, 'b.desc, 'c.asc)), + false) + + checkSatisfied( + RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), + OrderedDistribution(Seq('b.asc, 'a.asc)), + false) + + checkSatisfied( + RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), + OrderedDistribution(Seq('a.asc, 'b.asc, 'd.desc)), + false) + + // RangePartitioning can satisfy ClusteredDistribution iff its ordering expressions are a subset + // of the required clustering expressions. checkSatisfied( RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), ClusteredDistribution(Seq('a, 'b, 'c)), @@ -140,34 +260,47 @@ class DistributionSuite extends SparkFunSuite { ClusteredDistribution(Seq('b, 'c, 'a, 'd)), true) - // Cases which need an exchange between two data properties. - // TODO: We can have an optimization to first sort the dataset - // by a.asc and then sort b, and c in a partition. This optimization - // should tradeoff the benefit of a less number of Exchange operators - // and the parallelism. checkSatisfied( RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), - OrderedDistribution(Seq('a.asc, 'b.desc, 'c.asc)), + ClusteredDistribution(Seq('a, 'b)), false) checkSatisfied( RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), - OrderedDistribution(Seq('b.asc, 'a.asc)), + ClusteredDistribution(Seq('c, 'd)), false) + // RangePartitioning cannot satisfy HashClusteredDistribution checkSatisfied( RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), - ClusteredDistribution(Seq('a, 'b)), + HashClusteredDistribution(Seq('a, 'b, 'c)), false) + } + test("Partitioning.numPartitions must match Distribution.requiredNumPartitions to satisfy it") { checkSatisfied( - RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), - ClusteredDistribution(Seq('c, 'd)), + SinglePartition, + ClusteredDistribution(Seq('a, 'b, 'c), Some(10)), + false) + + checkSatisfied( + SinglePartition, + HashClusteredDistribution(Seq('a, 'b, 'c), Some(10)), + false) + + checkSatisfied( + HashPartitioning(Seq('a, 'b, 'c), 10), + ClusteredDistribution(Seq('a, 'b, 'c), Some(5)), + false) + + checkSatisfied( + HashPartitioning(Seq('a, 'b, 'c), 10), + HashClusteredDistribution(Seq('a, 'b, 'c), Some(5)), false) checkSatisfied( RangePartitioning(Seq('a.asc, 'b.asc, 'c.asc), 10), - AllTuples, + ClusteredDistribution(Seq('a, 'b, 'c), Some(5)), false) } } http://git-wip-us.apache.org/repos/asf/spark/blob/dc8a6bef/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/DataSourcePartitioning.scala ---------------------------------------------------------------------- diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/DataSourcePartitioning.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/DataSourcePartitioning.scala index 017a673..33079d5 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/DataSourcePartitioning.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/DataSourcePartitioning.scala @@ -30,8 +30,8 @@ class DataSourcePartitioning( override val numPartitions: Int = partitioning.numPartitions() - override def satisfies(required: physical.Distribution): Boolean = { - super.satisfies(required) || { + override def satisfies0(required: physical.Distribution): Boolean = { + super.satisfies0(required) || { required match { case d: physical.ClusteredDistribution if isCandidate(d.clustering) => val attrs = d.clustering.map(_.asInstanceOf[Attribute]) http://git-wip-us.apache.org/repos/asf/spark/blob/dc8a6bef/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/StreamingSymmetricHashJoinExec.scala ---------------------------------------------------------------------- diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/StreamingSymmetricHashJoinExec.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/StreamingSymmetricHashJoinExec.scala index afa664e..50cf971 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/StreamingSymmetricHashJoinExec.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/StreamingSymmetricHashJoinExec.scala @@ -167,8 +167,8 @@ case class StreamingSymmetricHashJoinExec( val nullRight = new GenericInternalRow(right.output.map(_.withNullability(true)).length) override def requiredChildDistribution: Seq[Distribution] = - ClusteredDistribution(leftKeys, stateInfo.map(_.numPartitions)) :: - ClusteredDistribution(rightKeys, stateInfo.map(_.numPartitions)) :: Nil + HashClusteredDistribution(leftKeys, stateInfo.map(_.numPartitions)) :: + HashClusteredDistribution(rightKeys, stateInfo.map(_.numPartitions)) :: Nil override def output: Seq[Attribute] = joinType match { case _: InnerLike => left.output ++ right.output http://git-wip-us.apache.org/repos/asf/spark/blob/dc8a6bef/sql/core/src/test/scala/org/apache/spark/sql/streaming/StreamingJoinSuite.scala ---------------------------------------------------------------------- diff --git a/sql/core/src/test/scala/org/apache/spark/sql/streaming/StreamingJoinSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/streaming/StreamingJoinSuite.scala index 1f62357..c5cc8df 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/streaming/StreamingJoinSuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/streaming/StreamingJoinSuite.scala @@ -404,6 +404,20 @@ class StreamingInnerJoinSuite extends StreamTest with StateStoreMetricsTest with AddData(input3, 5, 10), CheckNewAnswer((5, 10, 5, 15, 5, 25))) } + + test("streaming join should require HashClusteredDistribution from children") { + val input1 = MemoryStream[Int] + val input2 = MemoryStream[Int] + + val df1 = input1.toDF.select('value as 'a, 'value * 2 as 'b) + val df2 = input2.toDF.select('value as 'a, 'value * 2 as 'b).repartition('b) + val joined = df1.join(df2, Seq("a", "b")).select('a) + + testStream(joined)( + AddData(input1, 1.to(1000): _*), + AddData(input2, 1.to(1000): _*), + CheckAnswer(1.to(1000): _*)) + } } --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@spark.apache.org For additional commands, e-mail: commits-h...@spark.apache.org