Examples with ActorGateway org.apache.flink.runtime.instance.ActorGateway used on opensource projects

Example 71 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class LeaderChangeStateCleanupTest method testReelectionOfSameJobManager.

/**
	 * Tests that the same JobManager can be reelected as the leader. Even though, the same JM
	 * is elected as the next leader, all currently running jobs should be canceled properly and
	 * all TMs should disconnect from the leader and then reconnect to it.
	 */
@Test
public void testReelectionOfSameJobManager() throws Exception {
    UUID leaderSessionID = UUID.randomUUID();
    UUID newLeaderSessionID = UUID.randomUUID();
    FiniteDuration shortTimeout = new FiniteDuration(10, TimeUnit.SECONDS);
    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);
    LOG.info("Make JM(0) again the leader. This should first revoke the leadership.");
    // make JM(0) again the leader --> this implies first a leadership revocation
    cluster.grantLeadership(0, newLeaderSessionID);
    Await.ready(jobRemoval, timeout);
    LOG.info("Job removed.");
    // session ID
    try {
        cluster.waitForTaskManagersToBeRegistered(shortTimeout);
        fail("TaskManager should not be able to register at JobManager.");
    } catch (TimeoutException e) {
    // expected exception since the TMs have still the old leader session ID
    }
    LOG.info("Notify TMs about the new (old) leader.");
    // notify the TMs about the new (old) leader
    cluster.notifyRetrievalListeners(0, newLeaderSessionID);
    cluster.waitForTaskManagersToBeRegistered(timeout);
    ActorGateway leaderGateway = cluster.getLeaderGateway(timeout);
    // try to resubmit now the non-blocking job, it should complete successfully
    Tasks.BlockingOnceReceiver$.MODULE$.blocking_$eq(false);
    cluster.submitJobAndWait(job, false, timeout, new TestingLeaderRetrievalService(leaderGateway.path(), leaderGateway.leaderSessionID()));
}

Example 72 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class LeaderChangeStateCleanupTest method testStateCleanupAfterNewLeaderElectionAndListenerNotification.

/**
	 * Tests that a job is properly canceled in the case of a leader change. In such an event all
	 * TaskManagers have to disconnect from the previous leader and connect to the newly elected
	 * leader.
	 */
@Test
public void testStateCleanupAfterNewLeaderElectionAndListenerNotification() throws Exception {
    UUID leaderSessionID1 = UUID.randomUUID();
    UUID leaderSessionID2 = UUID.randomUUID();
    // first make JM(0) the leader
    cluster.grantLeadership(0, leaderSessionID1);
    // notify all listeners
    cluster.notifyRetrievalListeners(0, leaderSessionID1);
    cluster.waitForTaskManagersToBeRegistered(timeout);
    // submit blocking job so that it is not finished when we cancel it
    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);
    // make the JM(1) the new leader
    cluster.grantLeadership(1, leaderSessionID2);
    // notify all listeners about the event
    cluster.notifyRetrievalListeners(1, leaderSessionID2);
    Await.ready(jobRemoval, timeout);
    cluster.waitForTaskManagersToBeRegistered(timeout);
    ActorGateway jm2 = cluster.getLeaderGateway(timeout);
    Future<Object> futureNumberSlots = jm2.ask(JobManagerMessages.getRequestTotalNumberOfSlots(), timeout);
    // check that all TMs have registered at the new leader
    int numberSlots = (Integer) Await.result(futureNumberSlots, timeout);
    assertEquals(parallelism, numberSlots);
    // try to resubmit now the non-blocking job, it should complete successfully
    Tasks.BlockingOnceReceiver$.MODULE$.blocking_$eq(false);
    cluster.submitJobAndWait(job, false, timeout, new TestingLeaderRetrievalService(jm2.path(), jm2.leaderSessionID()));
}

Example 73 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class TaskManagerTest method testUpdateTaskInputPartitionsFailure.

/**
	 * Tests that the TaskManager sends a proper exception back to the sender if the trigger stack
	 * trace message fails.
	 */
@Test
public void testUpdateTaskInputPartitionsFailure() throws Exception {
    ActorGateway jobManager = null;
    ActorGateway taskManager = null;
    try {
        final ExecutionAttemptID executionAttemptId = new ExecutionAttemptID();
        ActorRef jm = system.actorOf(Props.create(SimpleJobManager.class, leaderSessionID));
        jobManager = new AkkaActorGateway(jm, leaderSessionID);
        taskManager = TestingUtils.createTaskManager(system, jobManager, new Configuration(), true, true);
        TaskDeploymentDescriptor tdd = createTaskDeploymentDescriptor(new JobID(), "test job", new JobVertexID(), executionAttemptId, new SerializedValue<>(new ExecutionConfig()), "test task", 1, 0, 1, 0, new Configuration(), new Configuration(), BlockingNoOpInvokable.class.getName(), Collections.<ResultPartitionDeploymentDescriptor>emptyList(), Collections.<InputGateDeploymentDescriptor>emptyList(), Collections.<BlobKey>emptyList(), Collections.<URL>emptyList(), 0);
        Future<Object> submitResponse = taskManager.ask(new SubmitTask(tdd), timeout);
        Await.result(submitResponse, timeout);
        Future<Object> partitionUpdateResponse = taskManager.ask(new TaskMessages.UpdateTaskSinglePartitionInfo(executionAttemptId, new IntermediateDataSetID(), new InputChannelDeploymentDescriptor(new ResultPartitionID(), ResultPartitionLocation.createLocal())), timeout);
        try {
            Await.result(partitionUpdateResponse, timeout);
            fail("The update task input partitions message should have failed.");
        } catch (Exception e) {
        // expected
        }
    } finally {
        TestingUtils.stopActor(jobManager);
        TestingUtils.stopActor(taskManager);
    }
}

Example 74 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class TaskManagerTest method testGateChannelEdgeMismatch.

@Test
public void testGateChannelEdgeMismatch() {
    new JavaTestKit(system) {

        {
            ActorGateway jobManager = null;
            ActorGateway taskManager = null;
            final ActorGateway testActorGateway = new AkkaActorGateway(getTestActor(), leaderSessionID);
            try {
                ActorRef jm = system.actorOf(Props.create(SimpleJobManager.class, leaderSessionID));
                jobManager = new AkkaActorGateway(jm, leaderSessionID);
                taskManager = TestingUtils.createTaskManager(system, jobManager, new Configuration(), true, true);
                final ActorGateway tm = taskManager;
                final JobID jid = new JobID();
                JobVertexID vid1 = new JobVertexID();
                JobVertexID vid2 = new JobVertexID();
                final ExecutionAttemptID eid1 = new ExecutionAttemptID();
                final ExecutionAttemptID eid2 = new ExecutionAttemptID();
                final TaskDeploymentDescriptor tdd1 = createTaskDeploymentDescriptor(jid, "TestJob", vid1, eid1, new SerializedValue<>(new ExecutionConfig()), "Sender", 1, 0, 1, 0, new Configuration(), new Configuration(), Tasks.Sender.class.getName(), Collections.<ResultPartitionDeploymentDescriptor>emptyList(), Collections.<InputGateDeploymentDescriptor>emptyList(), new ArrayList<BlobKey>(), Collections.<URL>emptyList(), 0);
                final TaskDeploymentDescriptor tdd2 = createTaskDeploymentDescriptor(jid, "TestJob", vid2, eid2, new SerializedValue<>(new ExecutionConfig()), "Receiver", 7, 2, 7, 0, new Configuration(), new Configuration(), Tasks.Receiver.class.getName(), Collections.<ResultPartitionDeploymentDescriptor>emptyList(), Collections.<InputGateDeploymentDescriptor>emptyList(), new ArrayList<BlobKey>(), Collections.<URL>emptyList(), 0);
                new Within(d) {

                    @Override
                    protected void run() {
                        try {
                            tm.tell(new SubmitTask(tdd1), testActorGateway);
                            tm.tell(new SubmitTask(tdd2), testActorGateway);
                            expectMsgEquals(Acknowledge.get());
                            expectMsgEquals(Acknowledge.get());
                            tm.tell(new TestingTaskManagerMessages.NotifyWhenTaskRemoved(eid1), testActorGateway);
                            tm.tell(new TestingTaskManagerMessages.NotifyWhenTaskRemoved(eid2), testActorGateway);
                            expectMsgEquals(true);
                            expectMsgEquals(true);
                            tm.tell(TestingTaskManagerMessages.getRequestRunningTasksMessage(), testActorGateway);
                            Map<ExecutionAttemptID, Task> tasks = expectMsgClass(TestingTaskManagerMessages.ResponseRunningTasks.class).asJava();
                            assertEquals(0, tasks.size());
                        } catch (Exception e) {
                            e.printStackTrace();
                            fail(e.getMessage());
                        }
                    }
                };
            } catch (Exception e) {
                e.printStackTrace();
                fail(e.getMessage());
            } finally {
                // shut down the actors
                TestingUtils.stopActor(taskManager);
                TestingUtils.stopActor(jobManager);
            }
        }
    };
}

Example 75 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class TaskManagerTest method testStackTraceSampleFailure.

/**
	 * Tests that the TaskManager sends a proper exception back to the sender if the trigger stack
	 * trace message fails.
	 */
@Test
public void testStackTraceSampleFailure() throws Exception {
    ActorGateway jobManager = null;
    ActorGateway taskManager = null;
    try {
        ActorRef jm = system.actorOf(Props.create(SimpleJobManager.class, leaderSessionID));
        jobManager = new AkkaActorGateway(jm, leaderSessionID);
        taskManager = TestingUtils.createTaskManager(system, jobManager, new Configuration(), true, true);
        Future<Object> stackTraceResponse = taskManager.ask(new TriggerStackTraceSample(0, new ExecutionAttemptID(), 0, Time.milliseconds(1L), 0), timeout);
        try {
            Await.result(stackTraceResponse, timeout);
            fail("The trigger stack trace message should have failed.");
        } catch (IllegalStateException e) {
        // expected
        }
    } finally {
        TestingUtils.stopActor(jobManager);
        TestingUtils.stopActor(taskManager);
    }
}