[SPARK-21657][SQL] optimize explode quadratic memory consumpation

## What changes were proposed in this pull request? The issue has been raised in two Jira tickets: [SPARK-21657](https://issues.apache.org/jira/browse/SPARK-21657), [SPARK-16998](https://issues.apache.org/jira/browse/SPARK-16998). Basically, what happens is that in collection generators like explode/inline we create many rows from each row. Currently each exploded row contains also the column on which it was created. This causes, for example, if we have a 10k array in one row that this array will get copy 10k times - to each of the row. this results a qudratic memory consumption. However, it is a common case that the original column gets projected out after the explode, so we can avoid duplicating it. In this solution we propose to identify this situation in the optimizer and turn on a flag for omitting the original column in the generation process. ## How was this patch tested? 1. We added a benchmark test to MiscBenchmark that shows x16 improvement in runtimes. 2. We ran some of the other tests in MiscBenchmark and they show 15% improvements. 3. We ran this code on a specific case from our production data with rows containing arrays of size ~200k and it reduced the runtime from 6 hours to 3 mins. Author: oraviv <[email protected]> Author: uzadude <[email protected]> Author: uzadude <[email protected]> Closes apache#19683 from uzadude/optimize_explode.
fjh100456 · fjh100456 · Jan 8, 2018 · Dec 23, 2017 · Dec 23, 2017 · Dec 23, 2017
commit fcf66a32760c74e601acb537c51b2311ece6e9d5
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
@@ -696,7 +696,7 @@ class Analyzer(
           (oldVersion, oldVersion.copy(aggregateExpressions = newAliases(aggregateExpressions)))
 
         case oldVersion: Generate
-            if oldVersion.generatedSet.intersect(conflictingAttributes).nonEmpty =>
+            if oldVersion.producedAttributes.intersect(conflictingAttributes).nonEmpty =>
           val newOutput = oldVersion.generatorOutput.map(_.newInstance())
           (oldVersion, oldVersion.copy(generatorOutput = newOutput))
 
@@ -1138,7 +1138,7 @@ class Analyzer(
           case g: Generate =>
             val maybeResolvedExprs = exprs.map(resolveExpression(_, g))
             val (newExprs, newChild) = resolveExprsAndAddMissingAttrs(maybeResolvedExprs, g.child)
-            (newExprs, g.copy(join = true, child = newChild))
+            (newExprs, g.copy(unrequiredChildIndex = Nil, child = newChild))
 
           // For `Distinct` and `SubqueryAlias`, we can't recursively resolve and add attributes
           // via its children.
@@ -1578,7 +1578,7 @@ class Analyzer(
             resolvedGenerator =
               Generate(
                 generator,
-                join = projectList.size > 1, // Only join if there are other expressions in SELECT.
+                unrequiredChildIndex = Nil,
                 outer = outer,
                 qualifier = None,
                 generatorOutput = ResolveGenerate.makeGeneratorOutput(generator, names),

diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/CheckAnalysis.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/CheckAnalysis.scala
@@ -608,8 +608,8 @@ trait CheckAnalysis extends PredicateHelper {
       // allows to have correlation under it
       // but must not host any outer references.
       // Note:
-      // Generator with join=false is treated as Category 4.
-      case g: Generate if g.join =>
+      // Generator with requiredChildOutput.isEmpty is treated as Category 4.
+      case g: Generate if g.requiredChildOutput.nonEmpty =>
         failOnInvalidOuterReference(g)
 
       // Category 4: Any other operators not in the above 3 categories

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
@@ -359,12 +359,12 @@ package object dsl {
 
       def generate(
         generator: Generator,
-        join: Boolean = false,
+        unrequiredChildIndex: Seq[Int] = Nil,
         outer: Boolean = false,
         alias: Option[String] = None,
         outputNames: Seq[String] = Nil): LogicalPlan =
-        Generate(generator, join = join, outer = outer, alias,
-          outputNames.map(UnresolvedAttribute(_)), logicalPlan)
+        Generate(generator, unrequiredChildIndex, outer,
+          alias, outputNames.map(UnresolvedAttribute(_)), logicalPlan)
 
       def insertInto(tableName: String, overwrite: Boolean = false): LogicalPlan =
         InsertIntoTable(

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
@@ -456,12 +456,15 @@ object ColumnPruning extends Rule[LogicalPlan] {
       f.copy(child = prunedChild(child, f.references))
     case e @ Expand(_, _, child) if (child.outputSet -- e.references).nonEmpty =>
       e.copy(child = prunedChild(child, e.references))
-    case g: Generate if !g.join && (g.child.outputSet -- g.references).nonEmpty =>
-      g.copy(child = prunedChild(g.child, g.references))
 
-    // Turn off `join` for Generate if no column from it's child is used
-    case p @ Project(_, g: Generate) if g.join && p.references.subsetOf(g.generatedSet) =>
-      p.copy(child = g.copy(join = false))
+    // prune unrequired references
+    case p @ Project(_, g: Generate) if p.references != g.outputSet =>
+      val requiredAttrs = p.references -- g.producedAttributes ++ g.generator.references
+      val newChild = prunedChild(g.child, requiredAttrs)
+      val unrequired = g.generator.references -- p.references
+      val unrequiredIndices = newChild.output.zipWithIndex.filter(t => unrequired.contains(t._1))
+        .map(_._2)
+      p.copy(child = g.copy(child = newChild, unrequiredChildIndex = unrequiredIndices))
 
     // Eliminate unneeded attributes from right side of a Left Existence Join.
     case j @ Join(_, right, LeftExistence(_), _) =>

diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala
@@ -623,7 +623,7 @@ class AstBuilder(conf: SQLConf) extends SqlBaseBaseVisitor[AnyRef] with Logging
     val expressions = expressionList(ctx.expression)
     Generate(
       UnresolvedGenerator(visitFunctionName(ctx.qualifiedName), expressions),
-      join = true,
+      unrequiredChildIndex = Nil,
       outer = ctx.OUTER != null,
       Some(ctx.tblName.getText.toLowerCase),
       ctx.colName.asScala.map(_.getText).map(UnresolvedAttribute.apply),

diff --git a/...st/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicLogicalOperators.scala b/...st/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicLogicalOperators.scala
@@ -73,8 +73,13 @@ case class Project(projectList: Seq[NamedExpression], child: LogicalPlan) extend
  * their output.
  *
  * @param generator the generator expression
- * @param join  when true, each output row is implicitly joined with the input tuple that produced
- *              it.
+ * @param unrequiredChildIndex this paramter starts as Nil and gets filled by the Optimizer.
+ *                             It's used as an optimization for omitting data generation that will
+ *                             be discarded next by a projection.
+ *                             A common use case is when we explode(array(..)) and are interested
+ *                             only in the exploded data and not in the original array. before this
+ *                             optimization the array got duplicated for each of its elements,
+ *                             causing O(n^^2) memory consumption. (see [SPARK-21657])
  * @param outer when true, each input row will be output at least once, even if the output of the
  *              given `generator` is empty.
  * @param qualifier Qualifier for the attributes of generator(UDTF)
@@ -83,15 +88,17 @@ case class Project(projectList: Seq[NamedExpression], child: LogicalPlan) extend
  */
 case class Generate(
     generator: Generator,
-    join: Boolean,
+    unrequiredChildIndex: Seq[Int],
     outer: Boolean,
     qualifier: Option[String],
     generatorOutput: Seq[Attribute],
     child: LogicalPlan)
   extends UnaryNode {
 
-  /** The set of all attributes produced by this node. */
-  def generatedSet: AttributeSet = AttributeSet(generatorOutput)
+  lazy val requiredChildOutput: Seq[Attribute] = {
+    val unrequiredSet = unrequiredChildIndex.toSet
+    child.output.zipWithIndex.filterNot(t => unrequiredSet.contains(t._2)).map(_._1)
+  }
 
   override lazy val resolved: Boolean = {
     generator.resolved &&
@@ -114,9 +121,7 @@ case class Generate(
     nullableOutput
   }
 
-  def output: Seq[Attribute] = {
-    if (join) child.output ++ qualifiedGeneratorOutput else qualifiedGeneratorOutput
-  }
+  def output: Seq[Attribute] = requiredChildOutput ++ qualifiedGeneratorOutput
 }
 
 case class Filter(condition: Expression, child: LogicalPlan)

diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ColumnPruningSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ColumnPruningSuite.scala
@@ -38,54 +38,64 @@ class ColumnPruningSuite extends PlanTest {
       CollapseProject) :: Nil
   }
 
-  test("Column pruning for Generate when Generate.join = false") {
-    val input = LocalRelation('a.int, 'b.array(StringType))
+  test("Column pruning for Generate when Generate.unrequiredChildIndex = child.output") {
+    val input = LocalRelation('a.int, 'b.int, 'c.array(StringType))
 
-    val query = input.generate(Explode('b), join = false).analyze
+    val query =
+      input
+        .generate(Explode('c), outputNames = "explode" :: Nil)
+        .select('c, 'explode)
+        .analyze
 
     val optimized = Optimize.execute(query)
 
-    val correctAnswer = input.select('b).generate(Explode('b), join = false).analyze
+    val correctAnswer =
+      input
+        .select('c)
+        .generate(Explode('c), outputNames = "explode" :: Nil)
+        .analyze
 
     comparePlans(optimized, correctAnswer)
   }
 
-  test("Column pruning for Generate when Generate.join = true") {
-    val input = LocalRelation('a.int, 'b.int, 'c.array(StringType))
+  test("Fill Generate.unrequiredChildIndex if possible") {
+    val input = LocalRelation('b.array(StringType))
 
     val query =
       input
-        .generate(Explode('c), join = true, outputNames = "explode" :: Nil)
-        .select('a, 'explode)
+        .generate(Explode('b), outputNames = "explode" :: Nil)
+        .select(('explode + 1).as("result"))
         .analyze
 
     val optimized = Optimize.execute(query)
 
     val correctAnswer =
       input
-        .select('a, 'c)
-        .generate(Explode('c), join = true, outputNames = "explode" :: Nil)
-        .select('a, 'explode)
+        .generate(Explode('b), unrequiredChildIndex = input.output.zipWithIndex.map(_._2),
+          outputNames = "explode" :: Nil)
+         .select(('explode + 1).as("result"))
         .analyze
 
     comparePlans(optimized, correctAnswer)
   }
 
-  test("Turn Generate.join to false if possible") {
-    val input = LocalRelation('b.array(StringType))
+  test("Another fill Generate.unrequiredChildIndex if possible") {
+    val input = LocalRelation('a.int, 'b.int, 'c1.string, 'c2.string)
 
     val query =
       input
-        .generate(Explode('b), join = true, outputNames = "explode" :: Nil)
-        .select(('explode + 1).as("result"))
+        .generate(Explode(CreateArray(Seq('c1, 'c2))), outputNames = "explode" :: Nil)
+        .select('a, 'c1, 'explode)
         .analyze
 
     val optimized = Optimize.execute(query)
 
     val correctAnswer =
       input
-        .generate(Explode('b), join = false, outputNames = "explode" :: Nil)
-        .select(('explode + 1).as("result"))
+        .select('a, 'c1, 'c2)
+        .generate(Explode(CreateArray(Seq('c1, 'c2))),
+          unrequiredChildIndex = Seq(2),
+          outputNames = "explode" :: Nil)
         .analyze
 
     comparePlans(optimized, correctAnswer)

diff --git a/...catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/FilterPushdownSuite.scala b/...catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/FilterPushdownSuite.scala
@@ -624,14 +624,14 @@ class FilterPushdownSuite extends PlanTest {
   test("generate: predicate referenced no generated column") {
     val originalQuery = {
       testRelationWithArrayType
-        .generate(Explode('c_arr), true, false, Some("arr"))
+        .generate(Explode('c_arr), alias = Some("arr"))
         .where(('b >= 5) && ('a > 6))
     }
     val optimized = Optimize.execute(originalQuery.analyze)
     val correctAnswer = {
       testRelationWithArrayType
         .where(('b >= 5) && ('a > 6))
-        .generate(Explode('c_arr), true, false, Some("arr")).analyze
+        .generate(Explode('c_arr), alias = Some("arr")).analyze
     }
 
     comparePlans(optimized, correctAnswer)
@@ -640,14 +640,14 @@ class FilterPushdownSuite extends PlanTest {
   test("generate: non-deterministic predicate referenced no generated column") {
     val originalQuery = {
       testRelationWithArrayType
-        .generate(Explode('c_arr), true, false, Some("arr"))
+        .generate(Explode('c_arr), alias = Some("arr"))
         .where(('b >= 5) && ('a + Rand(10).as("rnd") > 6) && ('col > 6))
     }
     val optimized = Optimize.execute(originalQuery.analyze)
     val correctAnswer = {
       testRelationWithArrayType
         .where('b >= 5)
-        .generate(Explode('c_arr), true, false, Some("arr"))
+        .generate(Explode('c_arr), alias = Some("arr"))
         .where('a + Rand(10).as("rnd") > 6 && 'col > 6)
         .analyze
     }
@@ -659,14 +659,14 @@ class FilterPushdownSuite extends PlanTest {
     val generator = Explode('c_arr)
     val originalQuery = {
       testRelationWithArrayType
-        .generate(generator, true, false, Some("arr"))
+        .generate(generator, alias = Some("arr"))
         .where(('b >= 5) && ('c > 6))
     }
     val optimized = Optimize.execute(originalQuery.analyze)
     val referenceResult = {
       testRelationWithArrayType
         .where('b >= 5)
-        .generate(generator, true, false, Some("arr"))
+        .generate(generator, alias = Some("arr"))
         .where('c > 6).analyze
     }
 
@@ -687,7 +687,7 @@ class FilterPushdownSuite extends PlanTest {
   test("generate: all conjuncts referenced generated column") {
     val originalQuery = {
       testRelationWithArrayType
-        .generate(Explode('c_arr), true, false, Some("arr"))
+        .generate(Explode('c_arr), alias = Some("arr"))
         .where(('col > 6) || ('b > 5)).analyze
     }
     val optimized = Optimize.execute(originalQuery)

diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/parser/PlanParserSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/parser/PlanParserSuite.scala
@@ -276,7 +276,7 @@ class PlanParserSuite extends AnalysisTest {
     assertEqual(
       "select * from t lateral view explode(x) expl as x",
       table("t")
-        .generate(explode, join = true, outer = false, Some("expl"), Seq("x"))
+        .generate(explode, alias = Some("expl"), outputNames = Seq("x"))
         .select(star()))
 
     // Multiple lateral views
@@ -286,12 +286,12 @@ class PlanParserSuite extends AnalysisTest {
         |lateral view explode(x) expl
         |lateral view outer json_tuple(x, y) jtup q, z""".stripMargin,
       table("t")
-        .generate(explode, join = true, outer = false, Some("expl"), Seq.empty)
-        .generate(jsonTuple, join = true, outer = true, Some("jtup"), Seq("q", "z"))
+        .generate(explode, alias = Some("expl"))
+        .generate(jsonTuple, outer = true, alias = Some("jtup"), outputNames = Seq("q", "z"))
         .select(star()))
 
     // Multi-Insert lateral views.
-    val from = table("t1").generate(explode, join = true, outer = false, Some("expl"), Seq("x"))
+    val from = table("t1").generate(explode, alias = Some("expl"), outputNames = Seq("x"))
     assertEqual(
       """from t1
         |lateral view explode(x) expl as x
@@ -303,7 +303,7 @@ class PlanParserSuite extends AnalysisTest {
         |where s < 10
       """.stripMargin,
       Union(from
-        .generate(jsonTuple, join = true, outer = false, Some("jtup"), Seq("q", "z"))
+        .generate(jsonTuple, alias = Some("jtup"), outputNames = Seq("q", "z"))
         .select(star())
         .insertInto("t2"),
         from.where('s < 10).select(star()).insertInto("t3")))
@@ -312,10 +312,8 @@ class PlanParserSuite extends AnalysisTest {
     val expected = table("t")
       .generate(
         UnresolvedGenerator(FunctionIdentifier("posexplode"), Seq('x)),
-        join = true,
-        outer = false,
-        Some("posexpl"),
-        Seq("x", "y"))
+        alias = Some("posexpl"),
+        outputNames = Seq("x", "y"))
       .select(star())
     assertEqual(
       "select * from t lateral view posexplode(x) posexpl as x, y",

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala b/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala
@@ -2095,7 +2095,7 @@ class Dataset[T] private[sql](
     val generator = UserDefinedGenerator(elementSchema, rowFunction, input.map(_.expr))
 
     withPlan {
-      Generate(generator, join = true, outer = false,
+      Generate(generator, unrequiredChildIndex = Nil, outer = false,
         qualifier = None, generatorOutput = Nil, planWithBarrier)
     }
   }
@@ -2136,7 +2136,7 @@ class Dataset[T] private[sql](
     val generator = UserDefinedGenerator(elementSchema, rowFunction, apply(inputColumn).expr :: Nil)
 
     withPlan {
-      Generate(generator, join = true, outer = false,
+      Generate(generator, unrequiredChildIndex = Nil, outer = false,
         qualifier = None, generatorOutput = Nil, planWithBarrier)
     }
   }