Examples with StandaloneLeaderRetrievalService org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService used on opensource projects

Example 6 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class JobManagerTest method testSavepointRestoreSettings.

/**
	 * Tests that configured {@link SavepointRestoreSettings} are respected.
	 */
@Test
public void testSavepointRestoreSettings() throws Exception {
    FiniteDuration timeout = new FiniteDuration(30, TimeUnit.SECONDS);
    ActorSystem actorSystem = null;
    ActorGateway jobManager = null;
    ActorGateway archiver = null;
    ActorGateway taskManager = null;
    try {
        actorSystem = AkkaUtils.createLocalActorSystem(new Configuration());
        Tuple2<ActorRef, ActorRef> master = JobManager.startJobManagerActors(new Configuration(), actorSystem, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), Option.apply("jm"), Option.apply("arch"), TestingJobManager.class, TestingMemoryArchivist.class);
        jobManager = new AkkaActorGateway(master._1(), null);
        archiver = new AkkaActorGateway(master._2(), null);
        Configuration tmConfig = new Configuration();
        tmConfig.setInteger(ConfigConstants.TASK_MANAGER_NUM_TASK_SLOTS, 4);
        ActorRef taskManagerRef = TaskManager.startTaskManagerComponentsAndActor(tmConfig, ResourceID.generate(), actorSystem, "localhost", Option.apply("tm"), Option.<LeaderRetrievalService>apply(new StandaloneLeaderRetrievalService(jobManager.path())), true, TestingTaskManager.class);
        taskManager = new AkkaActorGateway(taskManagerRef, null);
        // Wait until connected
        Object msg = new TestingTaskManagerMessages.NotifyWhenRegisteredAtJobManager(jobManager.actor());
        Await.ready(taskManager.ask(msg, timeout), timeout);
        // Create job graph
        JobVertex sourceVertex = new JobVertex("Source");
        sourceVertex.setInvokableClass(BlockingStatefulInvokable.class);
        sourceVertex.setParallelism(1);
        JobGraph jobGraph = new JobGraph("TestingJob", sourceVertex);
        JobSnapshottingSettings snapshottingSettings = new JobSnapshottingSettings(Collections.singletonList(sourceVertex.getID()), Collections.singletonList(sourceVertex.getID()), Collections.singletonList(sourceVertex.getID()), // deactivated checkpointing
        Long.MAX_VALUE, 360000, 0, Integer.MAX_VALUE, ExternalizedCheckpointSettings.none(), null, true);
        jobGraph.setSnapshotSettings(snapshottingSettings);
        // Submit job graph
        msg = new JobManagerMessages.SubmitJob(jobGraph, ListeningBehaviour.DETACHED);
        Await.result(jobManager.ask(msg, timeout), timeout);
        // Wait for all tasks to be running
        msg = new TestingJobManagerMessages.WaitForAllVerticesToBeRunning(jobGraph.getJobID());
        Await.result(jobManager.ask(msg, timeout), timeout);
        // Trigger savepoint
        File targetDirectory = tmpFolder.newFolder();
        msg = new TriggerSavepoint(jobGraph.getJobID(), Option.apply(targetDirectory.getAbsolutePath()));
        Future<Object> future = jobManager.ask(msg, timeout);
        Object result = Await.result(future, timeout);
        String savepointPath = ((TriggerSavepointSuccess) result).savepointPath();
        // Cancel because of restarts
        msg = new TestingJobManagerMessages.NotifyWhenJobRemoved(jobGraph.getJobID());
        Future<?> removedFuture = jobManager.ask(msg, timeout);
        Future<?> cancelFuture = jobManager.ask(new CancelJob(jobGraph.getJobID()), timeout);
        Object response = Await.result(cancelFuture, timeout);
        assertTrue("Unexpected response: " + response, response instanceof CancellationSuccess);
        Await.ready(removedFuture, timeout);
        // Adjust the job (we need a new operator ID)
        JobVertex newSourceVertex = new JobVertex("NewSource");
        newSourceVertex.setInvokableClass(BlockingStatefulInvokable.class);
        newSourceVertex.setParallelism(1);
        JobGraph newJobGraph = new JobGraph("NewTestingJob", newSourceVertex);
        JobSnapshottingSettings newSnapshottingSettings = new JobSnapshottingSettings(Collections.singletonList(newSourceVertex.getID()), Collections.singletonList(newSourceVertex.getID()), Collections.singletonList(newSourceVertex.getID()), // deactivated checkpointing
        Long.MAX_VALUE, 360000, 0, Integer.MAX_VALUE, ExternalizedCheckpointSettings.none(), null, true);
        newJobGraph.setSnapshotSettings(newSnapshottingSettings);
        SavepointRestoreSettings restoreSettings = SavepointRestoreSettings.forPath(savepointPath, false);
        newJobGraph.setSavepointRestoreSettings(restoreSettings);
        msg = new JobManagerMessages.SubmitJob(newJobGraph, ListeningBehaviour.DETACHED);
        response = Await.result(jobManager.ask(msg, timeout), timeout);
        assertTrue("Unexpected response: " + response, response instanceof JobManagerMessages.JobResultFailure);
        JobManagerMessages.JobResultFailure failure = (JobManagerMessages.JobResultFailure) response;
        Throwable cause = failure.cause().deserializeError(ClassLoader.getSystemClassLoader());
        assertTrue(cause instanceof IllegalStateException);
        assertTrue(cause.getMessage().contains("allowNonRestoredState"));
        // Wait until removed
        msg = new TestingJobManagerMessages.NotifyWhenJobRemoved(newJobGraph.getJobID());
        Await.ready(jobManager.ask(msg, timeout), timeout);
        // Resubmit, but allow non restored state now
        restoreSettings = SavepointRestoreSettings.forPath(savepointPath, true);
        newJobGraph.setSavepointRestoreSettings(restoreSettings);
        msg = new JobManagerMessages.SubmitJob(newJobGraph, ListeningBehaviour.DETACHED);
        response = Await.result(jobManager.ask(msg, timeout), timeout);
        assertTrue("Unexpected response: " + response, response instanceof JobManagerMessages.JobSubmitSuccess);
    } finally {
        if (actorSystem != null) {
            actorSystem.shutdown();
        }
        if (archiver != null) {
            archiver.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (jobManager != null) {
            jobManager.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (taskManager != null) {
            taskManager.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
    }
}

Example 7 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class JobManagerTest method testCancelWithSavepoint.

@Test
public void testCancelWithSavepoint() throws Exception {
    File defaultSavepointDir = tmpFolder.newFolder();
    FiniteDuration timeout = new FiniteDuration(30, TimeUnit.SECONDS);
    Configuration config = new Configuration();
    config.setString(ConfigConstants.SAVEPOINT_DIRECTORY_KEY, defaultSavepointDir.getAbsolutePath());
    ActorSystem actorSystem = null;
    ActorGateway jobManager = null;
    ActorGateway archiver = null;
    ActorGateway taskManager = null;
    try {
        actorSystem = AkkaUtils.createLocalActorSystem(new Configuration());
        Tuple2<ActorRef, ActorRef> master = JobManager.startJobManagerActors(config, actorSystem, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), Option.apply("jm"), Option.apply("arch"), TestingJobManager.class, TestingMemoryArchivist.class);
        jobManager = new AkkaActorGateway(master._1(), null);
        archiver = new AkkaActorGateway(master._2(), null);
        ActorRef taskManagerRef = TaskManager.startTaskManagerComponentsAndActor(config, ResourceID.generate(), actorSystem, "localhost", Option.apply("tm"), Option.<LeaderRetrievalService>apply(new StandaloneLeaderRetrievalService(jobManager.path())), true, TestingTaskManager.class);
        taskManager = new AkkaActorGateway(taskManagerRef, null);
        // Wait until connected
        Object msg = new TestingTaskManagerMessages.NotifyWhenRegisteredAtJobManager(jobManager.actor());
        Await.ready(taskManager.ask(msg, timeout), timeout);
        // Create job graph
        JobVertex sourceVertex = new JobVertex("Source");
        sourceVertex.setInvokableClass(BlockingStatefulInvokable.class);
        sourceVertex.setParallelism(1);
        JobGraph jobGraph = new JobGraph("TestingJob", sourceVertex);
        JobSnapshottingSettings snapshottingSettings = new JobSnapshottingSettings(Collections.singletonList(sourceVertex.getID()), Collections.singletonList(sourceVertex.getID()), Collections.singletonList(sourceVertex.getID()), 3600000, 3600000, 0, Integer.MAX_VALUE, ExternalizedCheckpointSettings.none(), null, true);
        jobGraph.setSnapshotSettings(snapshottingSettings);
        // Submit job graph
        msg = new JobManagerMessages.SubmitJob(jobGraph, ListeningBehaviour.DETACHED);
        Await.result(jobManager.ask(msg, timeout), timeout);
        // Wait for all tasks to be running
        msg = new TestingJobManagerMessages.WaitForAllVerticesToBeRunning(jobGraph.getJobID());
        Await.result(jobManager.ask(msg, timeout), timeout);
        // Notify when canelled
        msg = new NotifyWhenJobStatus(jobGraph.getJobID(), JobStatus.CANCELED);
        Future<Object> cancelled = jobManager.ask(msg, timeout);
        // Cancel with savepoint
        String savepointPath = null;
        for (int i = 0; i < 10; i++) {
            msg = new JobManagerMessages.CancelJobWithSavepoint(jobGraph.getJobID(), null);
            CancellationResponse cancelResp = (CancellationResponse) Await.result(jobManager.ask(msg, timeout), timeout);
            if (cancelResp instanceof CancellationFailure) {
                CancellationFailure failure = (CancellationFailure) cancelResp;
                if (failure.cause().getMessage().contains(CheckpointDeclineReason.NOT_ALL_REQUIRED_TASKS_RUNNING.message())) {
                    // wait and retry
                    Thread.sleep(200);
                } else {
                    failure.cause().printStackTrace();
                    fail("Failed to cancel job: " + failure.cause().getMessage());
                }
            } else {
                savepointPath = ((CancellationSuccess) cancelResp).savepointPath();
                break;
            }
        }
        // Verify savepoint path
        assertNotEquals("Savepoint not triggered", null, savepointPath);
        // Wait for job status change
        Await.ready(cancelled, timeout);
        File savepointFile = new File(savepointPath);
        assertEquals(true, savepointFile.exists());
    } finally {
        if (actorSystem != null) {
            actorSystem.shutdown();
        }
        if (archiver != null) {
            archiver.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (jobManager != null) {
            jobManager.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (taskManager != null) {
            taskManager.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
    }
}

Example 8 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class JobManagerTest method testSavepointWithDeactivatedPeriodicCheckpointing.

/**
	 * Tests that we can trigger a savepoint when periodic checkpoints are disabled.
	 */
@Test
public void testSavepointWithDeactivatedPeriodicCheckpointing() throws Exception {
    File defaultSavepointDir = tmpFolder.newFolder();
    FiniteDuration timeout = new FiniteDuration(30, TimeUnit.SECONDS);
    Configuration config = new Configuration();
    config.setString(ConfigConstants.SAVEPOINT_DIRECTORY_KEY, defaultSavepointDir.getAbsolutePath());
    ActorSystem actorSystem = null;
    ActorGateway jobManager = null;
    ActorGateway archiver = null;
    ActorGateway taskManager = null;
    try {
        actorSystem = AkkaUtils.createLocalActorSystem(new Configuration());
        Tuple2<ActorRef, ActorRef> master = JobManager.startJobManagerActors(config, actorSystem, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), Option.apply("jm"), Option.apply("arch"), TestingJobManager.class, TestingMemoryArchivist.class);
        jobManager = new AkkaActorGateway(master._1(), null);
        archiver = new AkkaActorGateway(master._2(), null);
        ActorRef taskManagerRef = TaskManager.startTaskManagerComponentsAndActor(config, ResourceID.generate(), actorSystem, "localhost", Option.apply("tm"), Option.<LeaderRetrievalService>apply(new StandaloneLeaderRetrievalService(jobManager.path())), true, TestingTaskManager.class);
        taskManager = new AkkaActorGateway(taskManagerRef, null);
        // Wait until connected
        Object msg = new TestingTaskManagerMessages.NotifyWhenRegisteredAtJobManager(jobManager.actor());
        Await.ready(taskManager.ask(msg, timeout), timeout);
        // Create job graph
        JobVertex sourceVertex = new JobVertex("Source");
        sourceVertex.setInvokableClass(BlockingStatefulInvokable.class);
        sourceVertex.setParallelism(1);
        JobGraph jobGraph = new JobGraph("TestingJob", sourceVertex);
        JobSnapshottingSettings snapshottingSettings = new JobSnapshottingSettings(Collections.singletonList(sourceVertex.getID()), Collections.singletonList(sourceVertex.getID()), Collections.singletonList(sourceVertex.getID()), // deactivated checkpointing
        Long.MAX_VALUE, 360000, 0, Integer.MAX_VALUE, ExternalizedCheckpointSettings.none(), null, true);
        jobGraph.setSnapshotSettings(snapshottingSettings);
        // Submit job graph
        msg = new JobManagerMessages.SubmitJob(jobGraph, ListeningBehaviour.DETACHED);
        Await.result(jobManager.ask(msg, timeout), timeout);
        // Wait for all tasks to be running
        msg = new TestingJobManagerMessages.WaitForAllVerticesToBeRunning(jobGraph.getJobID());
        Await.result(jobManager.ask(msg, timeout), timeout);
        // Cancel with savepoint
        File targetDirectory = tmpFolder.newFolder();
        msg = new TriggerSavepoint(jobGraph.getJobID(), Option.apply(targetDirectory.getAbsolutePath()));
        Future<Object> future = jobManager.ask(msg, timeout);
        Object result = Await.result(future, timeout);
        assertTrue("Did not trigger savepoint", result instanceof TriggerSavepointSuccess);
        assertEquals(1, targetDirectory.listFiles().length);
    } finally {
        if (actorSystem != null) {
            actorSystem.shutdown();
        }
        if (archiver != null) {
            archiver.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (jobManager != null) {
            jobManager.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (taskManager != null) {
            taskManager.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
    }
}

Example 9 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class JobManagerTest method testCancelWithSavepointNoDirectoriesConfigured.

/**
	 * Tests that a meaningful exception is returned if no savepoint directory is
	 * configured.
	 */
@Test
public void testCancelWithSavepointNoDirectoriesConfigured() throws Exception {
    FiniteDuration timeout = new FiniteDuration(30, TimeUnit.SECONDS);
    Configuration config = new Configuration();
    ActorSystem actorSystem = null;
    ActorGateway jobManager = null;
    ActorGateway archiver = null;
    ActorGateway taskManager = null;
    try {
        actorSystem = AkkaUtils.createLocalActorSystem(new Configuration());
        Tuple2<ActorRef, ActorRef> master = JobManager.startJobManagerActors(config, actorSystem, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), Option.apply("jm"), Option.apply("arch"), TestingJobManager.class, TestingMemoryArchivist.class);
        jobManager = new AkkaActorGateway(master._1(), null);
        archiver = new AkkaActorGateway(master._2(), null);
        ActorRef taskManagerRef = TaskManager.startTaskManagerComponentsAndActor(config, ResourceID.generate(), actorSystem, "localhost", Option.apply("tm"), Option.<LeaderRetrievalService>apply(new StandaloneLeaderRetrievalService(jobManager.path())), true, TestingTaskManager.class);
        taskManager = new AkkaActorGateway(taskManagerRef, null);
        // Wait until connected
        Object msg = new TestingTaskManagerMessages.NotifyWhenRegisteredAtJobManager(jobManager.actor());
        Await.ready(taskManager.ask(msg, timeout), timeout);
        // Create job graph
        JobVertex sourceVertex = new JobVertex("Source");
        sourceVertex.setInvokableClass(BlockingStatefulInvokable.class);
        sourceVertex.setParallelism(1);
        JobGraph jobGraph = new JobGraph("TestingJob", sourceVertex);
        JobSnapshottingSettings snapshottingSettings = new JobSnapshottingSettings(Collections.singletonList(sourceVertex.getID()), Collections.singletonList(sourceVertex.getID()), Collections.singletonList(sourceVertex.getID()), 3600000, 3600000, 0, Integer.MAX_VALUE, ExternalizedCheckpointSettings.none(), null, true);
        jobGraph.setSnapshotSettings(snapshottingSettings);
        // Submit job graph
        msg = new JobManagerMessages.SubmitJob(jobGraph, ListeningBehaviour.DETACHED);
        Await.result(jobManager.ask(msg, timeout), timeout);
        // Wait for all tasks to be running
        msg = new TestingJobManagerMessages.WaitForAllVerticesToBeRunning(jobGraph.getJobID());
        Await.result(jobManager.ask(msg, timeout), timeout);
        // Cancel with savepoint
        msg = new JobManagerMessages.CancelJobWithSavepoint(jobGraph.getJobID(), null);
        CancellationResponse cancelResp = (CancellationResponse) Await.result(jobManager.ask(msg, timeout), timeout);
        if (cancelResp instanceof CancellationFailure) {
            CancellationFailure failure = (CancellationFailure) cancelResp;
            assertTrue(failure.cause() instanceof IllegalStateException);
            assertTrue(failure.cause().getMessage().contains("savepoint directory"));
        } else {
            fail("Unexpected cancellation response from JobManager: " + cancelResp);
        }
    } finally {
        if (actorSystem != null) {
            actorSystem.shutdown();
        }
        if (archiver != null) {
            archiver.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (jobManager != null) {
            jobManager.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
        if (taskManager != null) {
            taskManager.actor().tell(PoisonPill.getInstance(), ActorRef.noSender());
        }
    }
}

Example 10 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class StandaloneLeaderElectionTest method testStandaloneLeaderElectionRetrieval.

/**
 * Tests that the standalone leader election and retrieval service return the same leader URL.
 */
@Test
public void testStandaloneLeaderElectionRetrieval() throws Exception {
    StandaloneLeaderElectionService leaderElectionService = new StandaloneLeaderElectionService();
    StandaloneLeaderRetrievalService leaderRetrievalService = new StandaloneLeaderRetrievalService(TEST_URL);
    TestingContender contender = new TestingContender(TEST_URL, leaderElectionService);
    TestingListener testingListener = new TestingListener();
    try {
        leaderElectionService.start(contender);
        leaderRetrievalService.start(testingListener);
        contender.waitForLeader(1000l);
        assertTrue(contender.isLeader());
        assertEquals(HighAvailabilityServices.DEFAULT_LEADER_ID, contender.getLeaderSessionID());
        testingListener.waitForNewLeader(1000l);
        assertEquals(TEST_URL, testingListener.getAddress());
        assertEquals(HighAvailabilityServices.DEFAULT_LEADER_ID, testingListener.getLeaderSessionID());
    } finally {
        leaderElectionService.stop();
        leaderRetrievalService.stop();
    }
}