Search in sources :

Example 1 with WaitForAllVerticesToBeRunningOrFinished

use of org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished in project flink by apache.

the class JobManagerTest method testRequestPartitionStateUnregisteredExecution.

/**
	 * Tests the JobManager response when the execution is not registered with
	 * the ExecutionGraph.
	 */
@Test
public void testRequestPartitionStateUnregisteredExecution() throws Exception {
    new JavaTestKit(system) {

        {
            new Within(duration("15 seconds")) {

                @Override
                protected void run() {
                    // Setup
                    TestingCluster cluster = null;
                    try {
                        cluster = startTestingCluster(4, 1, DEFAULT_AKKA_ASK_TIMEOUT());
                        final IntermediateDataSetID rid = new IntermediateDataSetID();
                        // Create a task
                        final JobVertex sender = new JobVertex("Sender");
                        sender.setParallelism(1);
                        // just finish
                        sender.setInvokableClass(NoOpInvokable.class);
                        sender.createAndAddResultDataSet(rid, PIPELINED);
                        final JobVertex sender2 = new JobVertex("Blocking Sender");
                        sender2.setParallelism(1);
                        // just block
                        sender2.setInvokableClass(BlockingNoOpInvokable.class);
                        sender2.createAndAddResultDataSet(new IntermediateDataSetID(), PIPELINED);
                        final JobGraph jobGraph = new JobGraph("Fast finishing producer test job", sender, sender2);
                        final JobID jid = jobGraph.getJobID();
                        final ActorGateway jobManagerGateway = cluster.getLeaderGateway(TestingUtils.TESTING_DURATION());
                        // we can set the leader session ID to None because we don't use this gateway to send messages
                        final ActorGateway testActorGateway = new AkkaActorGateway(getTestActor(), null);
                        // Submit the job and wait for all vertices to be running
                        jobManagerGateway.tell(new SubmitJob(jobGraph, ListeningBehaviour.EXECUTION_RESULT), testActorGateway);
                        expectMsgClass(JobSubmitSuccess.class);
                        jobManagerGateway.tell(new WaitForAllVerticesToBeRunningOrFinished(jid), testActorGateway);
                        expectMsgClass(AllVerticesRunning.class);
                        Future<Object> egFuture = jobManagerGateway.ask(new RequestExecutionGraph(jobGraph.getJobID()), remaining());
                        ExecutionGraphFound egFound = (ExecutionGraphFound) Await.result(egFuture, remaining());
                        ExecutionGraph eg = (ExecutionGraph) egFound.executionGraph();
                        ExecutionVertex vertex = eg.getJobVertex(sender.getID()).getTaskVertices()[0];
                        while (vertex.getExecutionState() != ExecutionState.FINISHED) {
                            Thread.sleep(1);
                        }
                        IntermediateResultPartition partition = vertex.getProducedPartitions().values().iterator().next();
                        ResultPartitionID partitionId = new ResultPartitionID(partition.getPartitionId(), vertex.getCurrentExecutionAttempt().getAttemptId());
                        // Producer finished, request state
                        Object request = new RequestPartitionProducerState(jid, rid, partitionId);
                        Future<ExecutionState> producerStateFuture = jobManagerGateway.ask(request, getRemainingTime()).mapTo(ClassTag$.MODULE$.<ExecutionState>apply(ExecutionState.class));
                        assertEquals(ExecutionState.FINISHED, Await.result(producerStateFuture, getRemainingTime()));
                    } catch (Exception e) {
                        e.printStackTrace();
                        fail(e.getMessage());
                    } finally {
                        if (cluster != null) {
                            cluster.shutdown();
                        }
                    }
                }
            };
        }
    };
}
Also used : AkkaActorGateway(org.apache.flink.runtime.instance.AkkaActorGateway) RequestPartitionProducerState(org.apache.flink.runtime.messages.JobManagerMessages.RequestPartitionProducerState) ExecutionState(org.apache.flink.runtime.execution.ExecutionState) ExecutionVertex(org.apache.flink.runtime.executiongraph.ExecutionVertex) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) WaitForAllVerticesToBeRunningOrFinished(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished) TestingUtils.startTestingCluster(org.apache.flink.runtime.testingUtils.TestingUtils.startTestingCluster) TestingCluster(org.apache.flink.runtime.testingUtils.TestingCluster) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) IntermediateResultPartition(org.apache.flink.runtime.executiongraph.IntermediateResultPartition) ActorGateway(org.apache.flink.runtime.instance.ActorGateway) AkkaActorGateway(org.apache.flink.runtime.instance.AkkaActorGateway) ExecutionGraph(org.apache.flink.runtime.executiongraph.ExecutionGraph) RequestExecutionGraph(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph) IntermediateDataSetID(org.apache.flink.runtime.jobgraph.IntermediateDataSetID) RequestExecutionGraph(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph) ExecutionGraphFound(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.ExecutionGraphFound) ResultPartitionID(org.apache.flink.runtime.io.network.partition.ResultPartitionID) SubmitJob(org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob) JavaTestKit(akka.testkit.JavaTestKit) JobID(org.apache.flink.api.common.JobID) Test(org.junit.Test)

Example 2 with WaitForAllVerticesToBeRunningOrFinished

use of org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished in project flink by apache.

the class JobManagerTest method testRequestPartitionState.

/**
	 * Tests responses to partition state requests.
	 */
@Test
public void testRequestPartitionState() throws Exception {
    new JavaTestKit(system) {

        {
            new Within(duration("15 seconds")) {

                @Override
                protected void run() {
                    // Setup
                    TestingCluster cluster = null;
                    try {
                        cluster = startTestingCluster(2, 1, DEFAULT_AKKA_ASK_TIMEOUT());
                        final IntermediateDataSetID rid = new IntermediateDataSetID();
                        // Create a task
                        final JobVertex sender = new JobVertex("Sender");
                        sender.setParallelism(1);
                        // just block
                        sender.setInvokableClass(BlockingNoOpInvokable.class);
                        sender.createAndAddResultDataSet(rid, PIPELINED);
                        final JobGraph jobGraph = new JobGraph("Blocking test job", sender);
                        final JobID jid = jobGraph.getJobID();
                        final ActorGateway jobManagerGateway = cluster.getLeaderGateway(TestingUtils.TESTING_DURATION());
                        // we can set the leader session ID to None because we don't use this gateway to send messages
                        final ActorGateway testActorGateway = new AkkaActorGateway(getTestActor(), null);
                        // Submit the job and wait for all vertices to be running
                        jobManagerGateway.tell(new SubmitJob(jobGraph, ListeningBehaviour.EXECUTION_RESULT), testActorGateway);
                        expectMsgClass(JobSubmitSuccess.class);
                        jobManagerGateway.tell(new WaitForAllVerticesToBeRunningOrFinished(jid), testActorGateway);
                        expectMsgClass(AllVerticesRunning.class);
                        // This is the mock execution ID of the task requesting the state of the partition
                        final ExecutionAttemptID receiver = new ExecutionAttemptID();
                        // Request the execution graph to get the runtime info
                        jobManagerGateway.tell(new RequestExecutionGraph(jid), testActorGateway);
                        final ExecutionGraph eg = (ExecutionGraph) expectMsgClass(ExecutionGraphFound.class).executionGraph();
                        final ExecutionVertex vertex = eg.getJobVertex(sender.getID()).getTaskVertices()[0];
                        final IntermediateResultPartition partition = vertex.getProducedPartitions().values().iterator().next();
                        final ResultPartitionID partitionId = new ResultPartitionID(partition.getPartitionId(), vertex.getCurrentExecutionAttempt().getAttemptId());
                        // - The test ----------------------------------------------------------------------
                        // 1. All execution states
                        RequestPartitionProducerState request = new RequestPartitionProducerState(jid, rid, partitionId);
                        for (ExecutionState state : ExecutionState.values()) {
                            ExecutionGraphTestUtils.setVertexState(vertex, state);
                            Future<ExecutionState> futurePartitionState = jobManagerGateway.ask(request, getRemainingTime()).mapTo(ClassTag$.MODULE$.<ExecutionState>apply(ExecutionState.class));
                            ExecutionState resp = Await.result(futurePartitionState, getRemainingTime());
                            assertEquals(state, resp);
                        }
                        // 2. Non-existing execution
                        request = new RequestPartitionProducerState(jid, rid, new ResultPartitionID());
                        Future<?> futurePartitionState = jobManagerGateway.ask(request, getRemainingTime());
                        try {
                            Await.result(futurePartitionState, getRemainingTime());
                            fail("Did not fail with expected RuntimeException");
                        } catch (RuntimeException e) {
                            assertEquals(IllegalArgumentException.class, e.getCause().getClass());
                        }
                        // 3. Non-existing job
                        request = new RequestPartitionProducerState(new JobID(), rid, new ResultPartitionID());
                        futurePartitionState = jobManagerGateway.ask(request, getRemainingTime());
                        try {
                            Await.result(futurePartitionState, getRemainingTime());
                            fail("Did not fail with expected IllegalArgumentException");
                        } catch (IllegalArgumentException ignored) {
                        }
                    } catch (Exception e) {
                        e.printStackTrace();
                        fail(e.getMessage());
                    } finally {
                        if (cluster != null) {
                            cluster.shutdown();
                        }
                    }
                }
            };
        }
    };
}
Also used : AkkaActorGateway(org.apache.flink.runtime.instance.AkkaActorGateway) RequestPartitionProducerState(org.apache.flink.runtime.messages.JobManagerMessages.RequestPartitionProducerState) ExecutionState(org.apache.flink.runtime.execution.ExecutionState) ExecutionAttemptID(org.apache.flink.runtime.executiongraph.ExecutionAttemptID) ExecutionVertex(org.apache.flink.runtime.executiongraph.ExecutionVertex) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) WaitForAllVerticesToBeRunningOrFinished(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished) TestingUtils.startTestingCluster(org.apache.flink.runtime.testingUtils.TestingUtils.startTestingCluster) TestingCluster(org.apache.flink.runtime.testingUtils.TestingCluster) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) IntermediateResultPartition(org.apache.flink.runtime.executiongraph.IntermediateResultPartition) ActorGateway(org.apache.flink.runtime.instance.ActorGateway) AkkaActorGateway(org.apache.flink.runtime.instance.AkkaActorGateway) ExecutionGraph(org.apache.flink.runtime.executiongraph.ExecutionGraph) RequestExecutionGraph(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph) IntermediateDataSetID(org.apache.flink.runtime.jobgraph.IntermediateDataSetID) RequestExecutionGraph(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph) ResultPartitionID(org.apache.flink.runtime.io.network.partition.ResultPartitionID) SubmitJob(org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob) JavaTestKit(akka.testkit.JavaTestKit) JobID(org.apache.flink.api.common.JobID) Test(org.junit.Test)

Example 3 with WaitForAllVerticesToBeRunningOrFinished

use of org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished in project flink by apache.

the class JobManagerTest method testRequestPartitionStateMoreRecentExecutionAttempt.

/**
	 * Tests the JobManager response when the execution is not registered with
	 * the ExecutionGraph anymore and a new execution attempt is available.
	 */
@Test
public void testRequestPartitionStateMoreRecentExecutionAttempt() throws Exception {
    new JavaTestKit(system) {

        {
            new Within(duration("15 seconds")) {

                @Override
                protected void run() {
                    // Setup
                    TestingCluster cluster = null;
                    try {
                        cluster = startTestingCluster(4, 1, DEFAULT_AKKA_ASK_TIMEOUT());
                        final IntermediateDataSetID rid = new IntermediateDataSetID();
                        // Create a task
                        final JobVertex sender = new JobVertex("Sender");
                        sender.setParallelism(1);
                        // just finish
                        sender.setInvokableClass(NoOpInvokable.class);
                        sender.createAndAddResultDataSet(rid, PIPELINED);
                        final JobVertex sender2 = new JobVertex("Blocking Sender");
                        sender2.setParallelism(1);
                        // just block
                        sender2.setInvokableClass(BlockingNoOpInvokable.class);
                        sender2.createAndAddResultDataSet(new IntermediateDataSetID(), PIPELINED);
                        final JobGraph jobGraph = new JobGraph("Fast finishing producer test job", sender, sender2);
                        final JobID jid = jobGraph.getJobID();
                        final ActorGateway jobManagerGateway = cluster.getLeaderGateway(TestingUtils.TESTING_DURATION());
                        // we can set the leader session ID to None because we don't use this gateway to send messages
                        final ActorGateway testActorGateway = new AkkaActorGateway(getTestActor(), null);
                        // Submit the job and wait for all vertices to be running
                        jobManagerGateway.tell(new SubmitJob(jobGraph, ListeningBehaviour.EXECUTION_RESULT), testActorGateway);
                        expectMsgClass(JobManagerMessages.JobSubmitSuccess.class);
                        jobManagerGateway.tell(new WaitForAllVerticesToBeRunningOrFinished(jid), testActorGateway);
                        expectMsgClass(TestingJobManagerMessages.AllVerticesRunning.class);
                        Future<Object> egFuture = jobManagerGateway.ask(new RequestExecutionGraph(jobGraph.getJobID()), remaining());
                        ExecutionGraphFound egFound = (ExecutionGraphFound) Await.result(egFuture, remaining());
                        ExecutionGraph eg = (ExecutionGraph) egFound.executionGraph();
                        ExecutionVertex vertex = eg.getJobVertex(sender.getID()).getTaskVertices()[0];
                        while (vertex.getExecutionState() != ExecutionState.FINISHED) {
                            Thread.sleep(1);
                        }
                        IntermediateResultPartition partition = vertex.getProducedPartitions().values().iterator().next();
                        ResultPartitionID partitionId = new ResultPartitionID(partition.getPartitionId(), vertex.getCurrentExecutionAttempt().getAttemptId());
                        // Reset execution => new execution attempt
                        vertex.resetForNewExecution();
                        // Producer finished, request state
                        Object request = new JobManagerMessages.RequestPartitionProducerState(jid, rid, partitionId);
                        Future<?> producerStateFuture = jobManagerGateway.ask(request, getRemainingTime());
                        try {
                            Await.result(producerStateFuture, getRemainingTime());
                            fail("Did not fail with expected Exception");
                        } catch (PartitionProducerDisposedException ignored) {
                        }
                    } catch (Exception e) {
                        e.printStackTrace();
                        fail(e.getMessage());
                    } finally {
                        if (cluster != null) {
                            cluster.shutdown();
                        }
                    }
                }
            };
        }
    };
}
Also used : AkkaActorGateway(org.apache.flink.runtime.instance.AkkaActorGateway) RequestPartitionProducerState(org.apache.flink.runtime.messages.JobManagerMessages.RequestPartitionProducerState) ExecutionVertex(org.apache.flink.runtime.executiongraph.ExecutionVertex) TestingUtils.startTestingCluster(org.apache.flink.runtime.testingUtils.TestingUtils.startTestingCluster) TestingCluster(org.apache.flink.runtime.testingUtils.TestingCluster) TestingJobManagerMessages(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages) ActorGateway(org.apache.flink.runtime.instance.ActorGateway) AkkaActorGateway(org.apache.flink.runtime.instance.AkkaActorGateway) RequestExecutionGraph(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph) ExecutionGraphFound(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.ExecutionGraphFound) ResultPartitionID(org.apache.flink.runtime.io.network.partition.ResultPartitionID) SubmitJob(org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob) JobManagerMessages(org.apache.flink.runtime.messages.JobManagerMessages) TestingJobManagerMessages(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) WaitForAllVerticesToBeRunningOrFinished(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) IntermediateResultPartition(org.apache.flink.runtime.executiongraph.IntermediateResultPartition) ExecutionGraph(org.apache.flink.runtime.executiongraph.ExecutionGraph) RequestExecutionGraph(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph) IntermediateDataSetID(org.apache.flink.runtime.jobgraph.IntermediateDataSetID) JavaTestKit(akka.testkit.JavaTestKit) JobID(org.apache.flink.api.common.JobID) Test(org.junit.Test)

Example 4 with WaitForAllVerticesToBeRunningOrFinished

use of org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished in project flink by apache.

the class LeaderChangeStateCleanupTest method testStateCleanupAfterNewLeaderElection.

/**
	 * Tests that a job is properly canceled in the case of a leader change. However, this time only the
	 * JMs are notified about the leader change and the TMs still believe the old leader to have
	 * leadership.
	 */
@Test
public void testStateCleanupAfterNewLeaderElection() throws Exception {
    UUID leaderSessionID = UUID.randomUUID();
    UUID newLeaderSessionID = UUID.randomUUID();
    cluster.grantLeadership(0, leaderSessionID);
    cluster.notifyRetrievalListeners(0, leaderSessionID);
    cluster.waitForTaskManagersToBeRegistered(timeout);
    // submit blocking job so that we can test job clean up
    cluster.submitJobDetached(job);
    ActorGateway jm = cluster.getLeaderGateway(timeout);
    Future<Object> wait = jm.ask(new WaitForAllVerticesToBeRunningOrFinished(job.getJobID()), timeout);
    Await.ready(wait, timeout);
    Future<Object> jobRemoval = jm.ask(new NotifyWhenJobRemoved(job.getJobID()), timeout);
    // only notify the JMs about the new leader JM(1)
    cluster.grantLeadership(1, newLeaderSessionID);
    // job should be removed anyway
    Await.ready(jobRemoval, timeout);
}
Also used : WaitForAllVerticesToBeRunningOrFinished(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished) NotifyWhenJobRemoved(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.NotifyWhenJobRemoved) ActorGateway(org.apache.flink.runtime.instance.ActorGateway) UUID(java.util.UUID) Test(org.junit.Test)

Example 5 with WaitForAllVerticesToBeRunningOrFinished

use of org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished in project flink by apache.

the class LeaderChangeStateCleanupTest method testStateCleanupAfterListenerNotification.

/**
	 * Tests that a job is properly canceled in the event of a leader change. However, this time
	 * only the TMs are notified about the changing leader. This should be enough to cancel the
	 * currently running job, though.
	 */
@Test
public void testStateCleanupAfterListenerNotification() throws Exception {
    UUID leaderSessionID = UUID.randomUUID();
    UUID newLeaderSessionID = UUID.randomUUID();
    cluster.grantLeadership(0, leaderSessionID);
    cluster.notifyRetrievalListeners(0, leaderSessionID);
    cluster.waitForTaskManagersToBeRegistered(timeout);
    // submit blocking job
    cluster.submitJobDetached(job);
    ActorGateway jm = cluster.getLeaderGateway(timeout);
    Future<Object> wait = jm.ask(new WaitForAllVerticesToBeRunningOrFinished(job.getJobID()), timeout);
    Await.ready(wait, timeout);
    Future<Object> jobRemoval = jm.ask(new NotifyWhenJobRemoved(job.getJobID()), timeout);
    // notify listeners (TMs) about the leader change
    cluster.notifyRetrievalListeners(1, newLeaderSessionID);
    Await.ready(jobRemoval, timeout);
}
Also used : WaitForAllVerticesToBeRunningOrFinished(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished) NotifyWhenJobRemoved(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.NotifyWhenJobRemoved) ActorGateway(org.apache.flink.runtime.instance.ActorGateway) UUID(java.util.UUID) Test(org.junit.Test)

Aggregations

ActorGateway (org.apache.flink.runtime.instance.ActorGateway)8 WaitForAllVerticesToBeRunningOrFinished (org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished)8 Test (org.junit.Test)8 UUID (java.util.UUID)4 JobGraph (org.apache.flink.runtime.jobgraph.JobGraph)4 JobVertex (org.apache.flink.runtime.jobgraph.JobVertex)4 TestingCluster (org.apache.flink.runtime.testingUtils.TestingCluster)4 NotifyWhenJobRemoved (org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.NotifyWhenJobRemoved)4 JavaTestKit (akka.testkit.JavaTestKit)3 JobID (org.apache.flink.api.common.JobID)3 ExecutionGraph (org.apache.flink.runtime.executiongraph.ExecutionGraph)3 ExecutionVertex (org.apache.flink.runtime.executiongraph.ExecutionVertex)3 IntermediateResultPartition (org.apache.flink.runtime.executiongraph.IntermediateResultPartition)3 AkkaActorGateway (org.apache.flink.runtime.instance.AkkaActorGateway)3 ResultPartitionID (org.apache.flink.runtime.io.network.partition.ResultPartitionID)3 IntermediateDataSetID (org.apache.flink.runtime.jobgraph.IntermediateDataSetID)3 RequestPartitionProducerState (org.apache.flink.runtime.messages.JobManagerMessages.RequestPartitionProducerState)3 SubmitJob (org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob)3 RequestExecutionGraph (org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph)3 TestingUtils.startTestingCluster (org.apache.flink.runtime.testingUtils.TestingUtils.startTestingCluster)3