[GitHub] [spark] alismess-db commented on a change in pull request #28912: [SPARK-32057][SQL][test-hive1.2] ExecuteStatement: cancel and close should not transiently ERROR

[GitBox <gi...@apache.org>]

alismess-db commented on a change in pull request #28912:
URL: https://github.com/apache/spark/pull/28912#discussion_r448252971



##########
File path: sql/hive-thriftserver/src/test/scala/org/apache/spark/sql/hive/thriftserver/SparkExecuteStatementOperationSuite.scala
##########
@@ -42,4 +57,65 @@ class SparkExecuteStatementOperationSuite extends SparkFunSuite {
     assert(columns.get(1).getType().getName == "INT")
     assert(columns.get(1).getComment() == "")
   }
+
+  Seq(
+    (OperationState.CANCELED, (_: SparkExecuteStatementOperation).cancel()),
+    (OperationState.CLOSED, (_: SparkExecuteStatementOperation).close())
+  ).foreach { case (finalState, transition) =>
+    test("SPARK-32057 SparkExecuteStatementOperation should not transiently become ERROR " +
+      s"before being set to $finalState") {
+      val hiveSession = new HiveSessionImpl(ThriftserverShimUtils.testedProtocolVersions.head,
+      "username", "password", new HiveConf, "ip address")
+      hiveSession.open(new util.HashMap)
+
+      HiveThriftServer2.eventManager = mock(classOf[HiveThriftServer2EventManager])
+
+      val spySqlContext = spy(sqlContext)
+
+      // When cancel() is called on the operation, cleanup causes an exception to be thrown inside
+      // of execute(). This should not cause the state to become ERROR. The exception here will be
+      // triggered in our custom cleanup().
+      val signal = new Semaphore(0)
+      val dataFrame = mock(classOf[DataFrame], RETURNS_DEEP_STUBS)
+      when(dataFrame.collect()).thenAnswer((_: InvocationOnMock) => {
+        signal.acquire()
+        throw new RuntimeException("Operation was cancelled by test cleanup.")
+      })
+      val statement = "stmt"
+      doReturn(dataFrame, Nil: _*).when(spySqlContext).sql(statement)
+
+      val executeStatementOperation = new MySparkExecuteStatementOperation(spySqlContext,
+        hiveSession, statement, signal, finalState)
+
+      val run = new Thread() {
+        override def run(): Unit = executeStatementOperation.runInternal()
+      }
+      assert(executeStatementOperation.getStatus.getState === OperationState.INITIALIZED)
+      run.start()
+      eventually(timeout(5.seconds)) {
+        assert(executeStatementOperation.getStatus.getState === OperationState.RUNNING)
+      }
+      transition(executeStatementOperation)
+      run.join()
+      assert(executeStatementOperation.getStatus.getState === finalState)
+    }
+  }
+  private class MySparkExecuteStatementOperation(
+      sqlContext: SQLContext,
+      hiveSession: HiveSession,
+      statement: String,
+      signal: Semaphore,
+      finalState: OperationState)
+    extends SparkExecuteStatementOperation(sqlContext, hiveSession, statement,
+      new util.HashMap, false) {
+
+    override def cleanup(): Unit = {
+      super.cleanup()
+      signal.release()
+      // Allow time for the exception to propagate.
+      Thread.sleep(1000)

Review comment:
       In the case where the state is already `finalState` when cleanup() is called, then the following would pass:
   
   ```
   signal.release()
   eventually(timeout(5 seconds)) {
     assert(getStatus.getState === finalState)
   }
   ```
   
   because we have not given the exception time to propagate in the other thread. Eventually will execute the first iteration (whose assert is true) then exit immediately. By sleeping, we check if ERROR does not surface even after having become `finalState`.




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org