[SPARK-17673] [SQL] Incorrect exchange reuse with RowDataSourceScan

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala

@@ -340,6 +340,8 @@ object DataSourceStrategy extends Strategy with Logging {
     // `Filter`s or cannot be handled by `relation`.
     val filterCondition = unhandledPredicates.reduceLeftOption(expressions.And)
 
+    // These metadata values make scan plans uniquely identifiable for equality checking.
+    // TODO(SPARK-17701) using strings for equality checking is brittle
     val metadata: Map[String, String] = {
       val pairs = ArrayBuffer.empty[(String, String)]
 
@@ -350,6 +352,8 @@ object DataSourceStrategy extends Strategy with Logging {
         }
         pairs += ("PushedFilters" -> markedFilters.mkString("[", ", ", "]"))
       }
+      pairs += ("ReadSchema" ->
+        StructType.fromAttributes(projects.map(_.toAttribute)).catalogString)
       pairs.toMap
     }
 

sql/core/src/test/scala/org/apache/spark/sql/jdbc/JDBCSuite.scala

@@ -791,4 +791,12 @@ class JDBCSuite extends SparkFunSuite
     val schema = JdbcUtils.schemaString(df, "jdbc:mysql://localhost:3306/temp")
     assert(schema.contains("`order` TEXT"))
   }
+
+  test("SPARK-17673: Exchange reuse respects differences in output schema") {
+    val df = sql("SELECT * FROM inttypes WHERE a IS NOT NULL")
+    val df1 = df.groupBy("a").agg("c" -> "min")
+    val df2 = df.groupBy("a").agg("d" -> "min")
+    val res = df1.union(df2)
+    assert(res.distinct().count() == 2)  // would be 1 if the exchange was incorrectly reused
+  }
 }