Examples with SubmitJob org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob used on opensource projects

Example 6 with SubmitJob

use of org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob in project flink by apache.

the class JobManagerHACheckpointRecoveryITCase method testCheckpointedStreamingSumProgram.

/**
	 * Simple checkpointed streaming sum.
	 *
	 * <p>The sources (Parallelism) count until sequenceEnd. The sink (1) sums up all counts and
	 * returns it to the main thread via a static variable. We wait until some checkpoints are
	 * completed and sanity check that the sources recover with an updated state to make sure that
	 * this test actually tests something.
	 */
@Test
@RetryOnFailure(times = 1)
public void testCheckpointedStreamingSumProgram() throws Exception {
    // Config
    final int checkpointingInterval = 200;
    final int sequenceEnd = 5000;
    final long expectedSum = Parallelism * sequenceEnd * (sequenceEnd + 1) / 2;
    final StreamExecutionEnvironment env = StreamExecutionEnvironment.createLocalEnvironment();
    env.setParallelism(Parallelism);
    env.enableCheckpointing(checkpointingInterval);
    env.addSource(new CheckpointedSequenceSource(sequenceEnd)).addSink(new CountingSink()).setParallelism(1);
    JobGraph jobGraph = env.getStreamGraph().getJobGraph();
    Configuration config = ZooKeeperTestUtils.createZooKeeperHAConfig(ZooKeeper.getConnectString(), FileStateBackendBasePath.getAbsoluteFile().toURI().toString());
    config.setInteger(ConfigConstants.TASK_MANAGER_NUM_TASK_SLOTS, Parallelism);
    ActorSystem testSystem = null;
    final JobManagerProcess[] jobManagerProcess = new JobManagerProcess[2];
    LeaderRetrievalService leaderRetrievalService = null;
    ActorSystem taskManagerSystem = null;
    try {
        final Deadline deadline = TestTimeOut.fromNow();
        // Test actor system
        testSystem = AkkaUtils.createActorSystem(new Configuration(), new Some<>(new Tuple2<String, Object>("localhost", 0)));
        // The job managers
        jobManagerProcess[0] = new JobManagerProcess(0, config);
        jobManagerProcess[1] = new JobManagerProcess(1, config);
        jobManagerProcess[0].startProcess();
        jobManagerProcess[1].startProcess();
        // Leader listener
        TestingListener leaderListener = new TestingListener();
        leaderRetrievalService = ZooKeeperUtils.createLeaderRetrievalService(config);
        leaderRetrievalService.start(leaderListener);
        // The task manager
        taskManagerSystem = AkkaUtils.createActorSystem(config, Option.apply(new Tuple2<String, Object>("localhost", 0)));
        TaskManager.startTaskManagerComponentsAndActor(config, ResourceID.generate(), taskManagerSystem, "localhost", Option.<String>empty(), Option.<LeaderRetrievalService>empty(), false, TaskManager.class);
        {
            // Initial submission
            leaderListener.waitForNewLeader(deadline.timeLeft().toMillis());
            String leaderAddress = leaderListener.getAddress();
            UUID leaderId = leaderListener.getLeaderSessionID();
            // Get the leader ref
            ActorRef leaderRef = AkkaUtils.getActorRef(leaderAddress, testSystem, deadline.timeLeft());
            ActorGateway leader = new AkkaActorGateway(leaderRef, leaderId);
            // Submit the job in detached mode
            leader.tell(new SubmitJob(jobGraph, ListeningBehaviour.DETACHED));
            JobManagerActorTestUtils.waitForJobStatus(jobGraph.getJobID(), JobStatus.RUNNING, leader, deadline.timeLeft());
        }
        // Who's the boss?
        JobManagerProcess leadingJobManagerProcess;
        if (jobManagerProcess[0].getJobManagerAkkaURL(deadline.timeLeft()).equals(leaderListener.getAddress())) {
            leadingJobManagerProcess = jobManagerProcess[0];
        } else {
            leadingJobManagerProcess = jobManagerProcess[1];
        }
        CompletedCheckpointsLatch.await();
        // Kill the leading job manager process
        leadingJobManagerProcess.destroy();
        {
            // Recovery by the standby JobManager
            leaderListener.waitForNewLeader(deadline.timeLeft().toMillis());
            String leaderAddress = leaderListener.getAddress();
            UUID leaderId = leaderListener.getLeaderSessionID();
            ActorRef leaderRef = AkkaUtils.getActorRef(leaderAddress, testSystem, deadline.timeLeft());
            ActorGateway leader = new AkkaActorGateway(leaderRef, leaderId);
            JobManagerActorTestUtils.waitForJobStatus(jobGraph.getJobID(), JobStatus.RUNNING, leader, deadline.timeLeft());
        }
        // Wait to finish
        FinalCountLatch.await();
        assertEquals(expectedSum, (long) FinalCount.get());
        for (int i = 0; i < Parallelism; i++) {
            assertNotEquals(0, RecoveredStates.get(i));
        }
    } catch (Throwable t) {
        // Reset all static state for test retries
        CompletedCheckpointsLatch = new CountDownLatch(2);
        RecoveredStates = new AtomicLongArray(Parallelism);
        FinalCountLatch = new CountDownLatch(1);
        FinalCount = new AtomicReference<>();
        LastElement = -1;
        // Print early (in some situations the process logs get too big
        // for Travis and the root problem is not shown)
        t.printStackTrace();
        // In case of an error, print the job manager process logs.
        if (jobManagerProcess[0] != null) {
            jobManagerProcess[0].printProcessLog();
        }
        if (jobManagerProcess[1] != null) {
            jobManagerProcess[1].printProcessLog();
        }
        throw t;
    } finally {
        if (jobManagerProcess[0] != null) {
            jobManagerProcess[0].destroy();
        }
        if (jobManagerProcess[1] != null) {
            jobManagerProcess[1].destroy();
        }
        if (leaderRetrievalService != null) {
            leaderRetrievalService.stop();
        }
        if (taskManagerSystem != null) {
            taskManagerSystem.shutdown();
        }
        if (testSystem != null) {
            testSystem.shutdown();
        }
    }
}

Example 7 with SubmitJob

use of org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob in project flink by apache.

the class JobManagerHAJobGraphRecoveryITCase method testJobPersistencyWhenJobManagerShutdown.

// ---------------------------------------------------------------------------------------------
/**
	 * Tests that the HA job is not cleaned up when the jobmanager is stopped.
	 */
@Test
public void testJobPersistencyWhenJobManagerShutdown() throws Exception {
    Configuration config = ZooKeeperTestUtils.createZooKeeperHAConfig(ZooKeeper.getConnectString(), FileStateBackendBasePath.getPath());
    // Configure the cluster
    config.setInteger(ConfigConstants.LOCAL_NUMBER_JOB_MANAGER, 1);
    config.setInteger(ConfigConstants.LOCAL_NUMBER_TASK_MANAGER, 1);
    TestingCluster flink = new TestingCluster(config, false, false);
    try {
        final Deadline deadline = TestTimeOut.fromNow();
        // Start the JobManager and TaskManager
        flink.start(true);
        JobGraph jobGraph = createBlockingJobGraph();
        // Set restart strategy to guard against shut down races.
        // If the TM fails before the JM, it might happen that the
        // Job is failed, leading to state removal.
        ExecutionConfig ec = new ExecutionConfig();
        ec.setRestartStrategy(RestartStrategies.fixedDelayRestart(Integer.MAX_VALUE, 100));
        jobGraph.setExecutionConfig(ec);
        ActorGateway jobManager = flink.getLeaderGateway(deadline.timeLeft());
        // Submit the job
        jobManager.tell(new SubmitJob(jobGraph, ListeningBehaviour.DETACHED));
        // Wait for the job to start
        JobManagerActorTestUtils.waitForJobStatus(jobGraph.getJobID(), JobStatus.RUNNING, jobManager, deadline.timeLeft());
    } finally {
        flink.shutdown();
    }
    // verify that the persisted job data has not been removed from ZooKeeper when the JM has
    // been shutdown
    verifyRecoveryState(config);
}

Example 8 with SubmitJob

use of org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob in project flink by apache.

the class JobManagerHAJobGraphRecoveryITCase method testSubmitJobToNonLeader.

/**
	 * Tests that submissions to non-leaders are handled.
	 */
@Test
public void testSubmitJobToNonLeader() throws Exception {
    Configuration config = ZooKeeperTestUtils.createZooKeeperHAConfig(ZooKeeper.getConnectString(), FileStateBackendBasePath.getPath());
    // Configure the cluster
    config.setInteger(ConfigConstants.LOCAL_NUMBER_JOB_MANAGER, 2);
    config.setInteger(ConfigConstants.LOCAL_NUMBER_TASK_MANAGER, 1);
    TestingCluster flink = new TestingCluster(config, false, false);
    try {
        final Deadline deadline = TestTimeOut.fromNow();
        // Start the JobManager and TaskManager
        flink.start(true);
        JobGraph jobGraph = createBlockingJobGraph();
        List<ActorRef> bothJobManagers = flink.getJobManagersAsJava();
        ActorGateway leadingJobManager = flink.getLeaderGateway(deadline.timeLeft());
        ActorGateway nonLeadingJobManager;
        if (bothJobManagers.get(0).equals(leadingJobManager.actor())) {
            nonLeadingJobManager = new AkkaActorGateway(bothJobManagers.get(1), null);
        } else {
            nonLeadingJobManager = new AkkaActorGateway(bothJobManagers.get(0), null);
        }
        log.info("Leading job manager: " + leadingJobManager);
        log.info("Non-leading job manager: " + nonLeadingJobManager);
        // Submit the job
        nonLeadingJobManager.tell(new SubmitJob(jobGraph, ListeningBehaviour.DETACHED));
        log.info("Submitted job graph to " + nonLeadingJobManager);
        // Wait for the job to start. We are asking the *leading** JM here although we've
        // submitted the job to the non-leading JM. This is the behaviour under test.
        JobManagerActorTestUtils.waitForJobStatus(jobGraph.getJobID(), JobStatus.RUNNING, leadingJobManager, deadline.timeLeft());
        log.info("Wait that the non-leader removes the submitted job.");
        // Make sure that the **non-leading** JM has actually removed the job graph from its
        // local state.
        boolean success = false;
        while (!success && deadline.hasTimeLeft()) {
            JobStatusResponse jobStatusResponse = JobManagerActorTestUtils.requestJobStatus(jobGraph.getJobID(), nonLeadingJobManager, deadline.timeLeft());
            if (jobStatusResponse instanceof JobManagerMessages.JobNotFound) {
                success = true;
            } else {
                log.info(((JobManagerMessages.CurrentJobStatus) jobStatusResponse).status().toString());
                Thread.sleep(100);
            }
        }
        if (!success) {
            fail("Non-leading JM was still holding reference to the job graph.");
        }
        Future<Object> jobRemoved = leadingJobManager.ask(new TestingJobManagerMessages.NotifyWhenJobRemoved(jobGraph.getJobID()), deadline.timeLeft());
        leadingJobManager.tell(new JobManagerMessages.CancelJob(jobGraph.getJobID()));
        Await.ready(jobRemoved, deadline.timeLeft());
    } finally {
        flink.shutdown();
    }
    // Verify that everything is clean
    verifyCleanRecoveryState(config);
}

Example 9 with SubmitJob

use of org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob in project flink by apache.

the class JobManagerTest method testStopSignalFail.

@Test
public void testStopSignalFail() 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());
                        // Create a task
                        final JobVertex sender = new JobVertex("Sender");
                        sender.setParallelism(1);
                        // just block
                        sender.setInvokableClass(BlockingNoOpInvokable.class);
                        final JobGraph jobGraph = new JobGraph("Non-Stoppable batching 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 WaitForAllVerticesToBeRunning(jid), testActorGateway);
                        expectMsgClass(AllVerticesRunning.class);
                        jobManagerGateway.tell(new StopJob(jid), testActorGateway);
                        // - The test ----------------------------------------------------------------------
                        expectMsgClass(StoppingFailure.class);
                        jobManagerGateway.tell(new RequestExecutionGraph(jid), testActorGateway);
                        expectMsgClass(ExecutionGraphFound.class);
                    } finally {
                        if (cluster != null) {
                            cluster.shutdown();
                        }
                    }
                }
            };
        }
    };
}

Example 10 with SubmitJob

use of org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob in project flink by apache.

the class JobManagerHACheckpointRecoveryITCase method testCheckpointRecoveryFailure.

/**
	 * Tests that the JobManager logs failures during recovery properly.
	 *
	 * @see <a href="https://issues.apache.org/jira/browse/FLINK-3185">FLINK-3185</a>
	 */
@Test
@RetryOnFailure(times = 1)
public void testCheckpointRecoveryFailure() throws Exception {
    final Deadline testDeadline = TestTimeOut.fromNow();
    final String zooKeeperQuorum = ZooKeeper.getConnectString();
    final String fileStateBackendPath = FileStateBackendBasePath.getAbsoluteFile().toString();
    Configuration config = ZooKeeperTestUtils.createZooKeeperHAConfig(zooKeeperQuorum, fileStateBackendPath);
    config.setInteger(ConfigConstants.LOCAL_NUMBER_JOB_MANAGER, 2);
    JobManagerProcess[] jobManagerProcess = new JobManagerProcess[2];
    LeaderRetrievalService leaderRetrievalService = null;
    ActorSystem taskManagerSystem = null;
    ActorSystem testActorSystem = null;
    try {
        // Test actor system
        testActorSystem = AkkaUtils.createActorSystem(new Configuration(), new Some<>(new Tuple2<String, Object>("localhost", 0)));
        // The job managers
        jobManagerProcess[0] = new JobManagerProcess(0, config);
        jobManagerProcess[1] = new JobManagerProcess(1, config);
        jobManagerProcess[0].startProcess();
        jobManagerProcess[1].startProcess();
        // Leader listener
        TestingListener leaderListener = new TestingListener();
        leaderRetrievalService = ZooKeeperUtils.createLeaderRetrievalService(config);
        leaderRetrievalService.start(leaderListener);
        // The task manager
        taskManagerSystem = AkkaUtils.createActorSystem(config, Option.apply(new Tuple2<String, Object>("localhost", 0)));
        TaskManager.startTaskManagerComponentsAndActor(config, ResourceID.generate(), taskManagerSystem, "localhost", Option.<String>empty(), Option.<LeaderRetrievalService>empty(), false, TaskManager.class);
        // Get the leader
        leaderListener.waitForNewLeader(testDeadline.timeLeft().toMillis());
        String leaderAddress = leaderListener.getAddress();
        UUID leaderId = leaderListener.getLeaderSessionID();
        // Get the leader ref
        ActorRef leaderRef = AkkaUtils.getActorRef(leaderAddress, testActorSystem, testDeadline.timeLeft());
        ActorGateway leader = new AkkaActorGateway(leaderRef, leaderId);
        // Who's the boss?
        JobManagerProcess leadingJobManagerProcess;
        JobManagerProcess nonLeadingJobManagerProcess;
        if (jobManagerProcess[0].getJobManagerAkkaURL(testDeadline.timeLeft()).equals(leaderListener.getAddress())) {
            leadingJobManagerProcess = jobManagerProcess[0];
            nonLeadingJobManagerProcess = jobManagerProcess[1];
        } else {
            leadingJobManagerProcess = jobManagerProcess[1];
            nonLeadingJobManagerProcess = jobManagerProcess[0];
        }
        // Blocking JobGraph
        JobVertex blockingVertex = new JobVertex("Blocking vertex");
        blockingVertex.setInvokableClass(BlockingNoOpInvokable.class);
        JobGraph jobGraph = new JobGraph(blockingVertex);
        // Submit the job in detached mode
        leader.tell(new SubmitJob(jobGraph, ListeningBehaviour.DETACHED));
        // Wait for the job to be running
        JobManagerActorTestUtils.waitForJobStatus(jobGraph.getJobID(), JobStatus.RUNNING, leader, testDeadline.timeLeft());
        // Remove all files
        FileUtils.deleteDirectory(FileStateBackendBasePath);
        // Kill the leader
        leadingJobManagerProcess.destroy();
        // Verify that the job manager logs the failed recovery. We can not
        // do more at this point. :(
        boolean success = false;
        while (testDeadline.hasTimeLeft()) {
            String output = nonLeadingJobManagerProcess.getProcessOutput();
            if (output != null) {
                if (output.contains("Failed to recover job") && output.contains("java.io.FileNotFoundException")) {
                    success = true;
                    break;
                }
            } else {
                log.warn("No process output available.");
            }
            Thread.sleep(500);
        }
        assertTrue("Did not find expected output in logs.", success);
    } catch (Throwable t) {
        // Print early (in some situations the process logs get too big
        // for Travis and the root problem is not shown)
        t.printStackTrace();
        // In case of an error, print the job manager process logs.
        if (jobManagerProcess[0] != null) {
            jobManagerProcess[0].printProcessLog();
        }
        if (jobManagerProcess[1] != null) {
            jobManagerProcess[1].printProcessLog();
        }
        throw t;
    } finally {
        if (jobManagerProcess[0] != null) {
            jobManagerProcess[0].destroy();
        }
        if (jobManagerProcess[1] != null) {
            jobManagerProcess[1].destroy();
        }
        if (leaderRetrievalService != null) {
            leaderRetrievalService.stop();
        }
        if (taskManagerSystem != null) {
            taskManagerSystem.shutdown();
        }
        if (testActorSystem != null) {
            testActorSystem.shutdown();
        }
    }
}