Examples with JobVertexID org.apache.flink.runtime.jobgraph.JobVertexID used on opensource projects

Example 76 with JobVertexID

use of org.apache.flink.runtime.jobgraph.JobVertexID in project flink by apache.

the class AbstractJobVertexRequestHandler method handleRequest.

@Override
public final String handleRequest(AccessExecutionGraph graph, Map<String, String> params) throws Exception {
    final JobVertexID vid = parseJobVertexId(params);
    final AccessExecutionJobVertex jobVertex = graph.getJobVertex(vid);
    if (jobVertex == null) {
        throw new IllegalArgumentException("No vertex with ID '" + vid + "' exists.");
    }
    return handleRequest(jobVertex, params);
}

Example 77 with JobVertexID

use of org.apache.flink.runtime.jobgraph.JobVertexID in project flink by apache.

the class JsonGeneratorTest method checkVertexExists.

private void checkVertexExists(String vertexId, JobGraph graph) {
    // validate that the vertex has a valid 
    JobVertexID id = JobVertexID.fromHexString(vertexId);
    for (JobVertex vertex : graph.getVertices()) {
        if (vertex.getID().equals(id)) {
            return;
        }
    }
    fail("could not find vertex with id " + vertexId + " in JobGraph");
}

Example 78 with JobVertexID

use of org.apache.flink.runtime.jobgraph.JobVertexID in project flink by apache.

the class JobManagerHARecoveryTest method testJobRecoveryWhenLosingLeadership.

/**
	 * Tests that the persisted job is not removed from the SubmittedJobGraphStore if the JobManager
	 * loses its leadership. Furthermore, it tests that the job manager can recover the job from
	 * the SubmittedJobGraphStore and checkpoint state is recovered as well.
	 */
@Test
public void testJobRecoveryWhenLosingLeadership() throws Exception {
    FiniteDuration timeout = new FiniteDuration(30, TimeUnit.SECONDS);
    FiniteDuration jobRecoveryTimeout = new FiniteDuration(3, TimeUnit.SECONDS);
    Deadline deadline = new FiniteDuration(2, TimeUnit.MINUTES).fromNow();
    Configuration flinkConfiguration = new Configuration();
    UUID leaderSessionID = UUID.randomUUID();
    UUID newLeaderSessionID = UUID.randomUUID();
    int slots = 2;
    ActorRef archive = null;
    ActorRef jobManager = null;
    ActorRef taskManager = null;
    flinkConfiguration.setString(HighAvailabilityOptions.HA_MODE, "zookeeper");
    flinkConfiguration.setString(HighAvailabilityOptions.HA_STORAGE_PATH, temporaryFolder.newFolder().toString());
    flinkConfiguration.setInteger(ConfigConstants.TASK_MANAGER_NUM_TASK_SLOTS, slots);
    try {
        Scheduler scheduler = new Scheduler(TestingUtils.defaultExecutionContext());
        MySubmittedJobGraphStore mySubmittedJobGraphStore = new MySubmittedJobGraphStore();
        MyCheckpointStore checkpointStore = new MyCheckpointStore();
        CheckpointIDCounter checkpointCounter = new StandaloneCheckpointIDCounter();
        CheckpointRecoveryFactory checkpointStateFactory = new MyCheckpointRecoveryFactory(checkpointStore, checkpointCounter);
        TestingLeaderElectionService myLeaderElectionService = new TestingLeaderElectionService();
        TestingLeaderRetrievalService myLeaderRetrievalService = new TestingLeaderRetrievalService();
        InstanceManager instanceManager = new InstanceManager();
        instanceManager.addInstanceListener(scheduler);
        archive = system.actorOf(Props.create(MemoryArchivist.class, 10));
        Props jobManagerProps = Props.create(TestingJobManager.class, flinkConfiguration, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), instanceManager, scheduler, new BlobLibraryCacheManager(new BlobServer(flinkConfiguration), 3600000), archive, new FixedDelayRestartStrategy.FixedDelayRestartStrategyFactory(Int.MaxValue(), 100), timeout, myLeaderElectionService, mySubmittedJobGraphStore, checkpointStateFactory, jobRecoveryTimeout, Option.apply(null));
        jobManager = system.actorOf(jobManagerProps);
        ActorGateway gateway = new AkkaActorGateway(jobManager, leaderSessionID);
        taskManager = TaskManager.startTaskManagerComponentsAndActor(flinkConfiguration, ResourceID.generate(), system, "localhost", Option.apply("taskmanager"), Option.apply((LeaderRetrievalService) myLeaderRetrievalService), true, TestingTaskManager.class);
        ActorGateway tmGateway = new AkkaActorGateway(taskManager, leaderSessionID);
        Future<Object> tmAlive = tmGateway.ask(TestingMessages.getAlive(), deadline.timeLeft());
        Await.ready(tmAlive, deadline.timeLeft());
        JobVertex sourceJobVertex = new JobVertex("Source");
        sourceJobVertex.setInvokableClass(BlockingStatefulInvokable.class);
        sourceJobVertex.setParallelism(slots);
        JobGraph jobGraph = new JobGraph("TestingJob", sourceJobVertex);
        List<JobVertexID> vertexId = Collections.singletonList(sourceJobVertex.getID());
        jobGraph.setSnapshotSettings(new JobSnapshottingSettings(vertexId, vertexId, vertexId, 100, 10 * 60 * 1000, 0, 1, ExternalizedCheckpointSettings.none(), null, true));
        BlockingStatefulInvokable.initializeStaticHelpers(slots);
        Future<Object> isLeader = gateway.ask(TestingJobManagerMessages.getNotifyWhenLeader(), deadline.timeLeft());
        Future<Object> isConnectedToJobManager = tmGateway.ask(new TestingTaskManagerMessages.NotifyWhenRegisteredAtJobManager(jobManager), deadline.timeLeft());
        // tell jobManager that he's the leader
        myLeaderElectionService.isLeader(leaderSessionID);
        // tell taskManager who's the leader
        myLeaderRetrievalService.notifyListener(gateway.path(), leaderSessionID);
        Await.ready(isLeader, deadline.timeLeft());
        Await.ready(isConnectedToJobManager, deadline.timeLeft());
        // submit blocking job
        Future<Object> jobSubmitted = gateway.ask(new JobManagerMessages.SubmitJob(jobGraph, ListeningBehaviour.DETACHED), deadline.timeLeft());
        Await.ready(jobSubmitted, deadline.timeLeft());
        // Wait for some checkpoints to complete
        BlockingStatefulInvokable.awaitCompletedCheckpoints();
        Future<Object> jobRemoved = gateway.ask(new TestingJobManagerMessages.NotifyWhenJobRemoved(jobGraph.getJobID()), deadline.timeLeft());
        // Revoke leadership
        myLeaderElectionService.notLeader();
        // check that the job gets removed from the JobManager
        Await.ready(jobRemoved, deadline.timeLeft());
        // but stays in the submitted job graph store
        assertTrue(mySubmittedJobGraphStore.contains(jobGraph.getJobID()));
        Future<Object> jobRunning = gateway.ask(new TestingJobManagerMessages.NotifyWhenJobStatus(jobGraph.getJobID(), JobStatus.RUNNING), deadline.timeLeft());
        // Make JobManager again a leader
        myLeaderElectionService.isLeader(newLeaderSessionID);
        // tell the TaskManager about it
        myLeaderRetrievalService.notifyListener(gateway.path(), newLeaderSessionID);
        // wait that the job is recovered and reaches state RUNNING
        Await.ready(jobRunning, deadline.timeLeft());
        Future<Object> jobFinished = gateway.ask(new TestingJobManagerMessages.NotifyWhenJobRemoved(jobGraph.getJobID()), deadline.timeLeft());
        BlockingInvokable.unblock();
        // wait til the job has finished
        Await.ready(jobFinished, deadline.timeLeft());
        // check that the job has been removed from the submitted job graph store
        assertFalse(mySubmittedJobGraphStore.contains(jobGraph.getJobID()));
        // Check that state has been recovered
        long[] recoveredStates = BlockingStatefulInvokable.getRecoveredStates();
        for (long state : recoveredStates) {
            boolean isExpected = state >= BlockingStatefulInvokable.NUM_CHECKPOINTS_TO_COMPLETE;
            assertTrue("Did not recover checkpoint state correctly, expecting >= " + BlockingStatefulInvokable.NUM_CHECKPOINTS_TO_COMPLETE + ", but state was " + state, isExpected);
        }
    } finally {
        if (archive != null) {
            archive.tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (jobManager != null) {
            jobManager.tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (taskManager != null) {
            taskManager.tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
    }
}

Example 79 with JobVertexID

use of org.apache.flink.runtime.jobgraph.JobVertexID in project flink by apache.

the class JobManagerTest method testKvStateMessages.

/**
	 * Tests that the JobManager handles {@link org.apache.flink.runtime.query.KvStateMessage}
	 * instances as expected.
	 */
@Test
public void testKvStateMessages() throws Exception {
    Deadline deadline = new FiniteDuration(100, TimeUnit.SECONDS).fromNow();
    Configuration config = new Configuration();
    config.setString(ConfigConstants.AKKA_ASK_TIMEOUT, "100ms");
    UUID leaderSessionId = null;
    ActorGateway jobManager = new AkkaActorGateway(JobManager.startJobManagerActors(config, system, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), TestingJobManager.class, MemoryArchivist.class)._1(), leaderSessionId);
    LeaderRetrievalService leaderRetrievalService = new StandaloneLeaderRetrievalService(AkkaUtils.getAkkaURL(system, jobManager.actor()));
    Configuration tmConfig = new Configuration();
    tmConfig.setInteger(ConfigConstants.TASK_MANAGER_MEMORY_SIZE_KEY, 4);
    tmConfig.setInteger(ConfigConstants.TASK_MANAGER_NUM_TASK_SLOTS, 8);
    ActorRef taskManager = TaskManager.startTaskManagerComponentsAndActor(tmConfig, ResourceID.generate(), system, "localhost", scala.Option.<String>empty(), scala.Option.apply(leaderRetrievalService), true, TestingTaskManager.class);
    Future<Object> registrationFuture = jobManager.ask(new NotifyWhenAtLeastNumTaskManagerAreRegistered(1), deadline.timeLeft());
    Await.ready(registrationFuture, deadline.timeLeft());
    //
    // Location lookup
    //
    LookupKvStateLocation lookupNonExistingJob = new LookupKvStateLocation(new JobID(), "any-name");
    Future<KvStateLocation> lookupFuture = jobManager.ask(lookupNonExistingJob, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    try {
        Await.result(lookupFuture, deadline.timeLeft());
        fail("Did not throw expected Exception");
    } catch (IllegalStateException ignored) {
    // Expected
    }
    JobGraph jobGraph = new JobGraph("croissant");
    JobVertex jobVertex1 = new JobVertex("cappuccino");
    jobVertex1.setParallelism(4);
    jobVertex1.setMaxParallelism(16);
    jobVertex1.setInvokableClass(BlockingNoOpInvokable.class);
    JobVertex jobVertex2 = new JobVertex("americano");
    jobVertex2.setParallelism(4);
    jobVertex2.setMaxParallelism(16);
    jobVertex2.setInvokableClass(BlockingNoOpInvokable.class);
    jobGraph.addVertex(jobVertex1);
    jobGraph.addVertex(jobVertex2);
    Future<JobSubmitSuccess> submitFuture = jobManager.ask(new SubmitJob(jobGraph, ListeningBehaviour.DETACHED), deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<JobSubmitSuccess>apply(JobSubmitSuccess.class));
    Await.result(submitFuture, deadline.timeLeft());
    Object lookupUnknownRegistrationName = new LookupKvStateLocation(jobGraph.getJobID(), "unknown");
    lookupFuture = jobManager.ask(lookupUnknownRegistrationName, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    try {
        Await.result(lookupFuture, deadline.timeLeft());
        fail("Did not throw expected Exception");
    } catch (UnknownKvStateLocation ignored) {
    // Expected
    }
    //
    // Registration
    //
    NotifyKvStateRegistered registerNonExistingJob = new NotifyKvStateRegistered(new JobID(), new JobVertexID(), new KeyGroupRange(0, 0), "any-name", new KvStateID(), new KvStateServerAddress(InetAddress.getLocalHost(), 1233));
    jobManager.tell(registerNonExistingJob);
    LookupKvStateLocation lookupAfterRegistration = new LookupKvStateLocation(registerNonExistingJob.getJobId(), registerNonExistingJob.getRegistrationName());
    lookupFuture = jobManager.ask(lookupAfterRegistration, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    try {
        Await.result(lookupFuture, deadline.timeLeft());
        fail("Did not throw expected Exception");
    } catch (IllegalStateException ignored) {
    // Expected
    }
    NotifyKvStateRegistered registerForExistingJob = new NotifyKvStateRegistered(jobGraph.getJobID(), jobVertex1.getID(), new KeyGroupRange(0, 0), "register-me", new KvStateID(), new KvStateServerAddress(InetAddress.getLocalHost(), 1293));
    jobManager.tell(registerForExistingJob);
    lookupAfterRegistration = new LookupKvStateLocation(registerForExistingJob.getJobId(), registerForExistingJob.getRegistrationName());
    lookupFuture = jobManager.ask(lookupAfterRegistration, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    KvStateLocation location = Await.result(lookupFuture, deadline.timeLeft());
    assertNotNull(location);
    assertEquals(jobGraph.getJobID(), location.getJobId());
    assertEquals(jobVertex1.getID(), location.getJobVertexId());
    assertEquals(jobVertex1.getMaxParallelism(), location.getNumKeyGroups());
    assertEquals(1, location.getNumRegisteredKeyGroups());
    KeyGroupRange keyGroupRange = registerForExistingJob.getKeyGroupRange();
    assertEquals(1, keyGroupRange.getNumberOfKeyGroups());
    assertEquals(registerForExistingJob.getKvStateId(), location.getKvStateID(keyGroupRange.getStartKeyGroup()));
    assertEquals(registerForExistingJob.getKvStateServerAddress(), location.getKvStateServerAddress(keyGroupRange.getStartKeyGroup()));
    //
    // Unregistration
    //
    NotifyKvStateUnregistered unregister = new NotifyKvStateUnregistered(registerForExistingJob.getJobId(), registerForExistingJob.getJobVertexId(), registerForExistingJob.getKeyGroupRange(), registerForExistingJob.getRegistrationName());
    jobManager.tell(unregister);
    lookupFuture = jobManager.ask(lookupAfterRegistration, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    try {
        Await.result(lookupFuture, deadline.timeLeft());
        fail("Did not throw expected Exception");
    } catch (UnknownKvStateLocation ignored) {
    // Expected
    }
    //
    // Duplicate registration fails task
    //
    NotifyKvStateRegistered register = new NotifyKvStateRegistered(jobGraph.getJobID(), jobVertex1.getID(), new KeyGroupRange(0, 0), "duplicate-me", new KvStateID(), new KvStateServerAddress(InetAddress.getLocalHost(), 1293));
    NotifyKvStateRegistered duplicate = new NotifyKvStateRegistered(jobGraph.getJobID(), // <--- different operator, but...
    jobVertex2.getID(), new KeyGroupRange(0, 0), // ...same name
    "duplicate-me", new KvStateID(), new KvStateServerAddress(InetAddress.getLocalHost(), 1293));
    Future<TestingJobManagerMessages.JobStatusIs> failedFuture = jobManager.ask(new NotifyWhenJobStatus(jobGraph.getJobID(), JobStatus.FAILED), deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<JobStatusIs>apply(JobStatusIs.class));
    jobManager.tell(register);
    jobManager.tell(duplicate);
    // Wait for failure
    JobStatusIs jobStatus = Await.result(failedFuture, deadline.timeLeft());
    assertEquals(JobStatus.FAILED, jobStatus.state());
}

Example 80 with JobVertexID

use of org.apache.flink.runtime.jobgraph.JobVertexID in project flink by apache.

the class SchedulerSlotSharingTest method scheduleSingleVertexType.

@Test
public void scheduleSingleVertexType() {
    try {
        JobVertexID jid1 = new JobVertexID();
        SlotSharingGroup sharingGroup = new SlotSharingGroup(jid1);
        Scheduler scheduler = new Scheduler(TestingUtils.directExecutionContext());
        Instance i1 = getRandomInstance(2);
        Instance i2 = getRandomInstance(2);
        scheduler.newInstanceAvailable(i1);
        scheduler.newInstanceAvailable(i2);
        // schedule 4 tasks from the first vertex group
        SimpleSlot s1 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 0, 8), sharingGroup), false).get();
        SimpleSlot s2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 1, 8), sharingGroup), false).get();
        SimpleSlot s3 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 2, 8), sharingGroup), false).get();
        SimpleSlot s4 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 3, 8), sharingGroup), false).get();
        assertNotNull(s1);
        assertNotNull(s2);
        assertNotNull(s3);
        assertNotNull(s4);
        assertTrue(areAllDistinct(s1, s2, s3, s4));
        // we cannot schedule another task from the first vertex group
        try {
            scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 4, 8), sharingGroup), false).get();
            fail("Scheduler accepted too many tasks at the same time");
        } catch (ExecutionException e) {
            assertTrue(e.getCause() instanceof NoResourceAvailableException);
        } catch (Exception e) {
            fail("Wrong exception.");
        }
        // release something
        s3.releaseSlot();
        // allocate another slot from that group
        SimpleSlot s5 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 4, 8), sharingGroup), false).get();
        assertNotNull(s5);
        // release all old slots
        s1.releaseSlot();
        s2.releaseSlot();
        s4.releaseSlot();
        SimpleSlot s6 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 5, 8), sharingGroup), false).get();
        SimpleSlot s7 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 6, 8), sharingGroup), false).get();
        SimpleSlot s8 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 7, 8), sharingGroup), false).get();
        assertNotNull(s6);
        assertNotNull(s7);
        assertNotNull(s8);
        // make sure we have two slots on the first instance, and two on the second
        int c = 0;
        c += (s5.getTaskManagerID().equals(i1.getTaskManagerID())) ? 1 : -1;
        c += (s6.getTaskManagerID().equals(i1.getTaskManagerID())) ? 1 : -1;
        c += (s7.getTaskManagerID().equals(i1.getTaskManagerID())) ? 1 : -1;
        c += (s8.getTaskManagerID().equals(i1.getTaskManagerID())) ? 1 : -1;
        assertEquals(0, c);
        // release all
        s5.releaseSlot();
        s6.releaseSlot();
        s7.releaseSlot();
        s8.releaseSlot();
        // test that everything is released
        assertEquals(4, scheduler.getNumberOfAvailableSlots());
        // check the scheduler's bookkeeping
        assertEquals(0, scheduler.getNumberOfLocalizedAssignments());
        assertEquals(0, scheduler.getNumberOfNonLocalizedAssignments());
        assertEquals(8, scheduler.getNumberOfUnconstrainedAssignments());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    }
}