[SPARK-20335] [SQL] [BACKPORT-2.1] Children expressions of Hive UDF impacts the determinism of Hive UDF

sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveUDFs.scala

@@ -42,7 +42,7 @@ private[hive] case class HiveSimpleUDF(
     name: String, funcWrapper: HiveFunctionWrapper, children: Seq[Expression])
   extends Expression with HiveInspectors with CodegenFallback with Logging {
 
-  override def deterministic: Boolean = isUDFDeterministic
+  override def deterministic: Boolean = isUDFDeterministic && children.forall(_.deterministic)
 
   override def nullable: Boolean = true
 
@@ -120,7 +120,7 @@ private[hive] case class HiveGenericUDF(
 
   override def nullable: Boolean = true
 
-  override def deterministic: Boolean = isUDFDeterministic
+  override def deterministic: Boolean = isUDFDeterministic && children.forall(_.deterministic)
 
   override def foldable: Boolean =
     isUDFDeterministic && returnInspector.isInstanceOf[ConstantObjectInspector]

sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/AggregationQuerySuite.scala

@@ -509,6 +509,19 @@ abstract class AggregationQuerySuite extends QueryTest with SQLTestUtils with Te
         Row(null, null, 110.0, null, null, 10.0) :: Nil)
   }
 
+  test("non-deterministic children expressions of UDAF") {
+    val e = intercept[AnalysisException] {
+      spark.sql(
+        """
+          |SELECT mydoublesum(value + 1.5 * key + rand())
+          |FROM agg1
+          |GROUP BY key
+        """.stripMargin)
+    }.getMessage
+    assert(Seq("nondeterministic expression",
+      "should not appear in the arguments of an aggregate function").forall(e.contains))
+  }
+
   test("interpreted aggregate function") {
     checkAnswer(
       spark.sql(

sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveUDFSuite.scala

@@ -31,6 +31,7 @@ import org.apache.hadoop.hive.serde2.objectinspector.primitive.PrimitiveObjectIn
 import org.apache.hadoop.io.{LongWritable, Writable}
 
 import org.apache.spark.sql.{AnalysisException, QueryTest, Row}
+import org.apache.spark.sql.catalyst.plans.logical.Project
 import org.apache.spark.sql.functions.max
 import org.apache.spark.sql.hive.test.TestHiveSingleton
 import org.apache.spark.sql.test.SQLTestUtils
@@ -338,6 +339,35 @@ class HiveUDFSuite extends QueryTest with TestHiveSingleton with SQLTestUtils {
     hiveContext.reset()
   }
 
+  test("non-deterministic children of UDF") {
+    withUserDefinedFunction("testStringStringUDF" -> true, "testGenericUDFHash" -> true) {
+      // HiveSimpleUDF
+      sql(s"CREATE TEMPORARY FUNCTION testStringStringUDF AS '${classOf[UDFStringString].getName}'")
+      val df1 = sql("SELECT testStringStringUDF(rand(), \"hello\")")
+      assert(!df1.logicalPlan.asInstanceOf[Project].projectList.forall(_.deterministic))
+
+      // HiveGenericUDF
+      sql(s"CREATE TEMPORARY FUNCTION testGenericUDFHash AS '${classOf[GenericUDFHash].getName}'")
+      val df2 = sql("SELECT testGenericUDFHash(rand())")
+      assert(!df2.logicalPlan.asInstanceOf[Project].projectList.forall(_.deterministic))
+    }
+  }
+
+  test("non-deterministic children expressions of UDAF") {
+    withTempView("view1") {
+      spark.range(1).selectExpr("id as x", "id as y").createTempView("view1")
+      withUserDefinedFunction("testUDAFPercentile" -> true) {
+        // non-deterministic children of Hive UDAF
+        sql(s"CREATE TEMPORARY FUNCTION testUDAFPercentile AS '${classOf[UDAFPercentile].getName}'")
+        val e1 = intercept[AnalysisException] {
+          sql("SELECT testUDAFPercentile(x, rand()) from view1 group by y")
+        }.getMessage
+        assert(Seq("nondeterministic expression",
+          "should not appear in the arguments of an aggregate function").forall(e1.contains))
+      }
+    }
+  }
+
   test("Hive UDFs with insufficient number of input arguments should trigger an analysis error") {
     Seq((1, 2)).toDF("a", "b").createOrReplaceTempView("testUDF")