Examples with ActorRef akka.actor.ActorRef used on opensource projects

Example 66 with ActorRef

use of akka.actor.ActorRef 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 67 with ActorRef

use of akka.actor.ActorRef in project flink by apache.

the class JobManagerHARecoveryTest method testFailingJobRecovery.

/**
	 * Tests that a failing job recovery won't cause other job recoveries to fail.
	 */
@Test
public void testFailingJobRecovery() throws Exception {
    final FiniteDuration timeout = new FiniteDuration(10, TimeUnit.SECONDS);
    final FiniteDuration jobRecoveryTimeout = new FiniteDuration(0, TimeUnit.SECONDS);
    Deadline deadline = new FiniteDuration(1, TimeUnit.MINUTES).fromNow();
    final Configuration flinkConfiguration = new Configuration();
    UUID leaderSessionID = UUID.randomUUID();
    ActorRef jobManager = null;
    JobID jobId1 = new JobID();
    JobID jobId2 = new JobID();
    // set HA mode to zookeeper so that we try to recover jobs
    flinkConfiguration.setString(HighAvailabilityOptions.HA_MODE, "zookeeper");
    try {
        final SubmittedJobGraphStore submittedJobGraphStore = mock(SubmittedJobGraphStore.class);
        SubmittedJobGraph submittedJobGraph = mock(SubmittedJobGraph.class);
        when(submittedJobGraph.getJobId()).thenReturn(jobId2);
        when(submittedJobGraphStore.getJobIds()).thenReturn(Arrays.asList(jobId1, jobId2));
        // fail the first job recovery
        when(submittedJobGraphStore.recoverJobGraph(eq(jobId1))).thenThrow(new Exception("Test exception"));
        // succeed the second job recovery
        when(submittedJobGraphStore.recoverJobGraph(eq(jobId2))).thenReturn(submittedJobGraph);
        final TestingLeaderElectionService myLeaderElectionService = new TestingLeaderElectionService();
        final Collection<JobID> recoveredJobs = new ArrayList<>(2);
        Props jobManagerProps = Props.create(TestingFailingHAJobManager.class, flinkConfiguration, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), mock(InstanceManager.class), mock(Scheduler.class), new BlobLibraryCacheManager(mock(BlobService.class), 1 << 20), ActorRef.noSender(), new FixedDelayRestartStrategy.FixedDelayRestartStrategyFactory(Int.MaxValue(), 100), timeout, myLeaderElectionService, submittedJobGraphStore, mock(CheckpointRecoveryFactory.class), jobRecoveryTimeout, Option.<MetricRegistry>apply(null), recoveredJobs).withDispatcher(CallingThreadDispatcher.Id());
        jobManager = system.actorOf(jobManagerProps);
        Future<Object> started = Patterns.ask(jobManager, new Identify(42), deadline.timeLeft().toMillis());
        Await.ready(started, deadline.timeLeft());
        // make the job manager the leader --> this triggers the recovery of all jobs
        myLeaderElectionService.isLeader(leaderSessionID);
        // check that we have successfully recovered the second job
        assertThat(recoveredJobs, containsInAnyOrder(jobId2));
    } finally {
        TestingUtils.stopActor(jobManager);
    }
}

Example 68 with ActorRef

use of akka.actor.ActorRef in project flink by apache.

the class JobManagerProcessReapingTest method testReapProcessOnFailure.

@Test
public void testReapProcessOnFailure() {
    Process jmProcess = null;
    ActorSystem localSystem = null;
    final StringWriter processOutput = new StringWriter();
    try {
        String javaCommand = getJavaCommandPath();
        // is available on this machine
        if (javaCommand == null) {
            System.out.println("---- Skipping JobManagerProcessReapingTest : Could not find java executable ----");
            return;
        }
        // create a logging file for the process
        File tempLogFile = File.createTempFile("testlogconfig", "properties");
        tempLogFile.deleteOnExit();
        CommonTestUtils.printLog4jDebugConfig(tempLogFile);
        // start a JobManger process
        // the log level must be at least INFO, otherwise the bound port cannot be retrieved
        String[] command = new String[] { javaCommand, "-Dlog.level=DEBUG", "-Dlog4j.configuration=file:" + tempLogFile.getAbsolutePath(), "-Xms256m", "-Xmx256m", "-classpath", getCurrentClasspath(), JobManagerTestEntryPoint.class.getName() };
        // spawn the process and collect its output
        ProcessBuilder bld = new ProcessBuilder(command);
        jmProcess = bld.start();
        new PipeForwarder(jmProcess.getErrorStream(), processOutput);
        // start another actor system so we can send something to the JobManager
        Tuple2<String, Object> localAddress = new Tuple2<String, Object>("localhost", 0);
        localSystem = AkkaUtils.createActorSystem(new Configuration(), new Some<Tuple2<String, Object>>(localAddress));
        // grab the reference to the JobManager. try multiple times, until the process
        // is started and the JobManager is up
        ActorRef jobManagerRef = null;
        Throwable lastError = null;
        // Log message on JobManager must be: Starting JobManager at ...://flink@...:port/..."
        // otherwise, the pattern does not match and, thus, cannot retrieve the bound port
        String pattern = "Starting JobManager at [^:]*://flink@[^:]*:(\\d*)/";
        Pattern r = Pattern.compile(pattern);
        int jobManagerPort = -1;
        for (int i = 0; i < 40; i++) {
            Matcher m = r.matcher(processOutput.toString());
            if (m.find()) {
                jobManagerPort = Integer.parseInt(m.group(1));
                break;
            }
            Thread.sleep(500);
        }
        if (jobManagerPort != -1) {
            try {
                jobManagerRef = JobManager.getJobManagerActorRef("akka.tcp", NetUtils.unresolvedHostAndPortToNormalizedString("localhost", jobManagerPort), localSystem, new FiniteDuration(25, TimeUnit.SECONDS));
            } catch (Throwable t) {
                // job manager probably not ready yet
                lastError = t;
            }
        } else {
            fail("Could not determine port of started JobManager.");
        }
        assertTrue("JobManager process died", isProcessAlive(jmProcess));
        if (jobManagerRef == null) {
            if (lastError != null) {
                lastError.printStackTrace();
            }
            fail("JobManager process did not launch the JobManager properly. Failed to look up JobManager actor at" + " localhost:" + jobManagerPort);
        }
        // kill the JobManager actor
        jobManagerRef.tell(PoisonPill.getInstance(), ActorRef.noSender());
        // wait for max 5 seconds for the process to terminate
        {
            long now = System.currentTimeMillis();
            long deadline = now + 5000;
            while (now < deadline && isProcessAlive(jmProcess)) {
                Thread.sleep(100);
                now = System.currentTimeMillis();
            }
        }
        assertFalse("JobManager process did not terminate upon actor death", isProcessAlive(jmProcess));
        int returnCode = jmProcess.exitValue();
        assertEquals("JobManager died, but not because of the process reaper", JobManager.RUNTIME_FAILURE_RETURN_CODE(), returnCode);
    } catch (Exception e) {
        e.printStackTrace();
        printProcessLog(processOutput.toString());
        fail(e.getMessage());
    } catch (Error e) {
        e.printStackTrace();
        printProcessLog(processOutput.toString());
        throw e;
    } finally {
        if (jmProcess != null) {
            jmProcess.destroy();
        }
        if (localSystem != null) {
            localSystem.shutdown();
        }
    }
}

Example 69 with ActorRef

use of akka.actor.ActorRef in project flink by apache.

the class AkkaKvStateLocationLookupServiceTest method testRetryOnUnknownJobManager.

/**
	 * Tests that lookups are retried when no leader notification is available.
	 */
@Test
public void testRetryOnUnknownJobManager() throws Exception {
    final Queue<LookupRetryStrategy> retryStrategies = new LinkedBlockingQueue<>();
    LookupRetryStrategyFactory retryStrategy = new LookupRetryStrategyFactory() {

        @Override
        public LookupRetryStrategy createRetryStrategy() {
            return retryStrategies.poll();
        }
    };
    final TestingLeaderRetrievalService leaderRetrievalService = new TestingLeaderRetrievalService();
    AkkaKvStateLocationLookupService lookupService = new AkkaKvStateLocationLookupService(leaderRetrievalService, testActorSystem, TIMEOUT, retryStrategy);
    lookupService.start();
    //
    // Test call to retry
    //
    final AtomicBoolean hasRetried = new AtomicBoolean();
    retryStrategies.add(new LookupRetryStrategy() {

        @Override
        public FiniteDuration getRetryDelay() {
            return FiniteDuration.Zero();
        }

        @Override
        public boolean tryRetry() {
            if (hasRetried.compareAndSet(false, true)) {
                return true;
            }
            return false;
        }
    });
    Future<KvStateLocation> locationFuture = lookupService.getKvStateLookupInfo(new JobID(), "yessir");
    Await.ready(locationFuture, TIMEOUT);
    assertTrue("Did not retry ", hasRetried.get());
    //
    // Test leader notification after retry
    //
    Queue<LookupKvStateLocation> received = new LinkedBlockingQueue<>();
    KvStateLocation expected = new KvStateLocation(new JobID(), new JobVertexID(), 12122, "garlic");
    ActorRef testActor = LookupResponseActor.create(received, null, expected);
    final String testActorAddress = AkkaUtils.getAkkaURL(testActorSystem, testActor);
    retryStrategies.add(new LookupRetryStrategy() {

        @Override
        public FiniteDuration getRetryDelay() {
            return FiniteDuration.apply(100, TimeUnit.MILLISECONDS);
        }

        @Override
        public boolean tryRetry() {
            leaderRetrievalService.notifyListener(testActorAddress, null);
            return true;
        }
    });
    KvStateLocation location = Await.result(lookupService.getKvStateLookupInfo(new JobID(), "yessir"), TIMEOUT);
    assertEquals(expected, location);
}

Example 70 with ActorRef

use of akka.actor.ActorRef 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();
        }
    }
}