[SPARK-20356][SQL] Pruned InMemoryTableScanExec should have correct output partitioning and ordering

## What changes were proposed in this pull request?

The output of `InMemoryTableScanExec` can be pruned and mismatch with `InMemoryRelation` and its child plan's output. This causes wrong output partitioning and ordering.

## How was this patch tested?

Jenkins tests.

Please review http://spark.apache.org/contributing.html before opening a pull request.

Author: Liang-Chi Hsieh <viirya@gmail.com>

Closes #17679 from viirya/SPARK-20356.

1 files changed, 15 insertions, 0 deletions

diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/columnar/InMemoryColumnarQuerySuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/columnar/InMemoryColumnarQuerySuite.scala
index 1e6a6a8ba3..109b1d9db6 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/execution/columnar/InMemoryColumnarQuerySuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/columnar/InMemoryColumnarQuerySuite.scala
@@ -414,4 +414,19 @@ class InMemoryColumnarQuerySuite extends QueryTest with SharedSQLContext {
       assert(partitionedAttrs.subsetOf(inMemoryScan.outputSet))
     }
   }
+
+  test("SPARK-20356: pruned InMemoryTableScanExec should have correct ordering and partitioning") {
+    withSQLConf("spark.sql.shuffle.partitions" -> "200") {
+      val df1 = Seq(("a", 1), ("b", 1), ("c", 2)).toDF("item", "group")
+      val df2 = Seq(("a", 1), ("b", 2), ("c", 3)).toDF("item", "id")
+      val df3 = df1.join(df2, Seq("item")).select($"id", $"group".as("item")).distinct()
+
+      df3.unpersist()
+      val agg_without_cache = df3.groupBy($"item").count()
+
+      df3.cache()
+      val agg_with_cache = df3.groupBy($"item").count()
+      checkAnswer(agg_without_cache, agg_with_cache)
+    }
+  }
 }