[SPARK-17424] Dataset job fails from unsound substitution in ScalaReflect

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.6.1, 2.0.0
• Fix Version/s: 2.0.3, 2.1.2, 2.2.0
• Component/s: Spark Core
• Assignee: Ryan Blue
• Reporter: Ryan Blue
• Created: 06/Sep/16 23:10
• Updated: 12/May/17 12:41
• Resolved: 12/May/17 12:40

Description

I have a job that uses datasets in 1.6.1 and is failing with this error:

16/09/02 17:02:56 ERROR Driver ApplicationMaster: User class threw exception: java.lang.AssertionError: assertion failed: Unsound substitution from List(type T, type U) to List()
java.lang.AssertionError: assertion failed: Unsound substitution from List(type T, type U) to List()
    at scala.reflect.internal.Types$SubstMap.<init>(Types.scala:4644)
    at scala.reflect.internal.Types$SubstTypeMap.<init>(Types.scala:4761)
    at scala.reflect.internal.Types$Type.subst(Types.scala:796)
    at scala.reflect.internal.Types$TypeApiImpl.substituteTypes(Types.scala:321)
    at scala.reflect.internal.Types$TypeApiImpl.substituteTypes(Types.scala:298)
    at org.apache.spark.sql.catalyst.ScalaReflection$$anonfun$getConstructorParameters$1.apply(ScalaReflection.scala:769)
    at org.apache.spark.sql.catalyst.ScalaReflection$$anonfun$getConstructorParameters$1.apply(ScalaReflection.scala:768)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.immutable.List.foreach(List.scala:318)
    at scala.collection.TraversableLike$class.map(TraversableLike.scala:244)
    at scala.collection.AbstractTraversable.map(Traversable.scala:105)
    at org.apache.spark.sql.catalyst.ScalaReflection$class.getConstructorParameters(ScalaReflection.scala:768)
    at org.apache.spark.sql.catalyst.ScalaReflection$.getConstructorParameters(ScalaReflection.scala:30)
    at org.apache.spark.sql.catalyst.ScalaReflection$.getConstructorParameters(ScalaReflection.scala:610)
    at org.apache.spark.sql.catalyst.trees.TreeNode.org$apache$spark$sql$catalyst$trees$TreeNode$$argNames$lzycompute(TreeNode.scala:418)
    at org.apache.spark.sql.catalyst.trees.TreeNode.org$apache$spark$sql$catalyst$trees$TreeNode$$argNames(TreeNode.scala:418)
    at org.apache.spark.sql.catalyst.trees.TreeNode$$anonfun$argsMap$1.apply(TreeNode.scala:415)
    at org.apache.spark.sql.catalyst.trees.TreeNode$$anonfun$argsMap$1.apply(TreeNode.scala:414)
    at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
    at scala.collection.Iterator$class.foreach(Iterator.scala:727)
    at scala.collection.AbstractIterator.foreach(Iterator.scala:1157)
    at scala.collection.TraversableOnce$class.toMap(TraversableOnce.scala:279)
    at scala.collection.AbstractIterator.toMap(Iterator.scala:1157)
    at org.apache.spark.sql.catalyst.trees.TreeNode.argsMap(TreeNode.scala:416)
    at org.apache.spark.sql.execution.SparkPlanInfo$.fromSparkPlan(SparkPlanInfo.scala:46)
    at org.apache.spark.sql.execution.SparkPlanInfo$$anonfun$2.apply(SparkPlanInfo.scala:44)
    at org.apache.spark.sql.execution.SparkPlanInfo$$anonfun$2.apply(SparkPlanInfo.scala:44)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.immutable.List.foreach(List.scala:318)
    at scala.collection.TraversableLike$class.map(TraversableLike.scala:244)
    at scala.collection.AbstractTraversable.map(Traversable.scala:105)
    at org.apache.spark.sql.execution.SparkPlanInfo$.fromSparkPlan(SparkPlanInfo.scala:44)
    at org.apache.spark.sql.execution.SparkPlanInfo$$anonfun$2.apply(SparkPlanInfo.scala:44)
    at org.apache.spark.sql.execution.SparkPlanInfo$$anonfun$2.apply(SparkPlanInfo.scala:44)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.immutable.List.foreach(List.scala:318)
    at scala.collection.TraversableLike$class.map(TraversableLike.scala:244)
    at scala.collection.AbstractTraversable.map(Traversable.scala:105)
    at org.apache.spark.sql.execution.SparkPlanInfo$.fromSparkPlan(SparkPlanInfo.scala:44)
    at org.apache.spark.sql.execution.SparkPlanInfo$$anonfun$2.apply(SparkPlanInfo.scala:44)
    at org.apache.spark.sql.execution.SparkPlanInfo$$anonfun$2.apply(SparkPlanInfo.scala:44)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.immutable.List.foreach(List.scala:318)
    at scala.collection.TraversableLike$class.map(TraversableLike.scala:244)
    at scala.collection.AbstractTraversable.map(Traversable.scala:105)
    at org.apache.spark.sql.execution.SparkPlanInfo$.fromSparkPlan(SparkPlanInfo.scala:44)
    at org.apache.spark.sql.execution.SQLExecution$.withNewExecutionId(SQLExecution.scala:51)
    at org.apache.spark.sql.execution.QueryExecution.toRdd$lzycompute(QueryExecution.scala:56)
    at org.apache.spark.sql.execution.QueryExecution.toRdd(QueryExecution.scala:55)
    at org.apache.spark.sql.DataFrameWriter.insertInto(DataFrameWriter.scala:193)
    at org.apache.spark.sql.DataFrameWriter.insertInto(DataFrameWriter.scala:166)
    at com.netflix.jobs.main(Processing.scala)
    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:498)
    at org.apache.spark.deploy.yarn.ApplicationMaster$$anon$2.run(ApplicationMaster.scala:557)

I think this is the same bug as SPARK-13067. It looks like that issue wasn't fixed, there was just a work-around added to get the test passing.

The problem is that the reflection code is trying to substitute concrete types for type parameters of MapPartitions[T, U], but the concrete types aren't known. So Spark ends up calling substituteTypes to substitute T and U with Nil (which gets shown as List()).

An easy fix that works for me is this:

    // if there are type variables to fill in, do the substitution (SomeClass[T] -> SomeClass[Int])
    if (actualTypeArgs.nonEmpty) {
      params.map { p =>
        p.name.toString -> p.typeSignature.substituteTypes(formalTypeArgs, actualTypeArgs)
      }
    } else {
      params.map { p =>
        p.name.toString -> p.typeSignature
      }
    }

Does this sound like a reasonable solution?

Edit: I think this affects 2.0.0 because the call to substituteTypes is unchanged

Issue Links

• supercedes: SPARK-13067 DataFrameSuite.simple explode fail locally
• links to: [Github] Pull Request #15062 (rdblue)