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

Example 1 with JobStatusResponse

use of org.apache.flink.runtime.messages.JobManagerMessages.JobStatusResponse 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 2 with JobStatusResponse

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

the class JobManagerActorTestUtils method requestJobStatus.

/**
	 * Request a {@link JobStatusResponse}.
	 *
	 * @param jobId      Job ID of the job to request the status of
	 * @param jobManager Job manager actor to ask
	 * @param timeout    Timeout after which the operation fails
	 * @return The {@link JobStatusResponse} from the job manager
	 * @throws Exception If there is no answer within the timeout.
	 */
public static JobStatusResponse requestJobStatus(JobID jobId, ActorGateway jobManager, FiniteDuration timeout) throws Exception {
    checkNotNull(jobId, "Job ID");
    checkNotNull(jobManager, "Job manager");
    checkNotNull(timeout, "Timeout");
    // Ask the JobManager
    RequestJobStatus request = (RequestJobStatus) getRequestJobStatus(jobId);
    Future<Object> ask = jobManager.ask(request, timeout);
    Object response = Await.result(ask, timeout);
    if (response instanceof JobStatusResponse) {
        return (JobStatusResponse) response;
    }
    throw new IllegalStateException("Unexpected response.");
}

Example 3 with JobStatusResponse

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

the class JobManagerActorTestUtils method waitForJobStatus.

/**
	 * Waits for the expected {@link JobStatus}.
	 *
	 * <p>Repeatedly queries the JobManager via {@link RequestJobStatus} messages.
	 *
	 * @param jobId             Job ID of the job to wait for
	 * @param expectedJobStatus Expected job status
	 * @param jobManager        Job manager actor to ask
	 * @param timeout           Timeout after which the operation fails
	 * @throws Exception If the job is not found within the timeout or the job is in another state.
	 */
public static void waitForJobStatus(JobID jobId, JobStatus expectedJobStatus, ActorGateway jobManager, FiniteDuration timeout) throws Exception {
    checkNotNull(jobId, "Job ID");
    checkNotNull(expectedJobStatus, "Expected job status");
    checkNotNull(jobManager, "Job manager");
    checkNotNull(timeout, "Timeout");
    final Deadline deadline = timeout.fromNow();
    while (deadline.hasTimeLeft()) {
        // Request the job status
        JobStatusResponse response = requestJobStatus(jobId, jobManager, deadline.timeLeft());
        // Found the job
        if (response instanceof CurrentJobStatus) {
            JobStatus jobStatus = ((CurrentJobStatus) response).status();
            // OK, that's what we were waiting for
            if (jobStatus == expectedJobStatus) {
                return;
            } else if (jobStatus.isGloballyTerminalState()) {
                throw new IllegalStateException("Job is in terminal state " + jobStatus + ", " + "but was waiting for " + expectedJobStatus + ".");
            }
        } else // Did not find the job... retry
        if (response instanceof JobNotFound) {
            Thread.sleep(Math.min(100, deadline.timeLeft().toMillis()));
        } else {
            throw new IllegalStateException("Unexpected response.");
        }
    }
    throw new IllegalStateException("Job not found within deadline.");
}