Examples with ActorGateway org.apache.flink.runtime.instance.ActorGateway used on opensource projects

Example 26 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class JobCancellationWithSavepointHandlersTest method testFailedCancellation.

/**
	 * Tests response when a request fails.
	 */
@Test
public void testFailedCancellation() throws Exception {
    JobID jobId = new JobID();
    ExecutionGraphHolder holder = mock(ExecutionGraphHolder.class);
    ExecutionGraph graph = mock(ExecutionGraph.class);
    CheckpointCoordinator coord = mock(CheckpointCoordinator.class);
    when(holder.getExecutionGraph(eq(jobId), any(ActorGateway.class))).thenReturn(graph);
    when(graph.getCheckpointCoordinator()).thenReturn(coord);
    JobCancellationWithSavepointHandlers handlers = new JobCancellationWithSavepointHandlers(holder, EC);
    JobCancellationWithSavepointHandlers.TriggerHandler trigger = handlers.getTriggerHandler();
    JobCancellationWithSavepointHandlers.InProgressHandler progress = handlers.getInProgressHandler();
    Map<String, String> params = new HashMap<>();
    params.put("jobid", jobId.toString());
    params.put("targetDirectory", "custom-directory");
    ActorGateway jobManager = mock(ActorGateway.class);
    // Successful
    Future<Object> future = Futures.failed(new Exception("Test Exception"));
    when(jobManager.ask(any(Object.class), any(FiniteDuration.class))).thenReturn(future);
    // Trigger
    trigger.handleRequest(params, Collections.<String, String>emptyMap(), jobManager);
    verify(jobManager).ask(eq(new CancelJobWithSavepoint(jobId, "custom-directory")), any(FiniteDuration.class));
    // Query progress
    params.put("requestId", "1");
    FullHttpResponse response = progress.handleRequest(params, Collections.<String, String>emptyMap(), jobManager);
    assertEquals(HttpResponseStatus.INTERNAL_SERVER_ERROR, response.getStatus());
    assertEquals("application/json", response.headers().get(HttpHeaders.Names.CONTENT_TYPE));
    assertEquals(Integer.toString(response.content().readableBytes()), response.headers().get(HttpHeaders.Names.CONTENT_LENGTH));
    String json = response.content().toString(Charset.forName("UTF-8"));
    JsonNode root = new ObjectMapper().readTree(json);
    assertEquals("failed", root.get("status").getValueAsText());
    assertEquals("1", root.get("request-id").getValueAsText());
    assertEquals("Test Exception", root.get("cause").getValueAsText());
}

Example 27 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class RescalingITCase method testSavepointRescalingPartitionedOperatorState.

/**
	 * Tests rescaling of partitioned operator state. More specific, we test the mechanism with {@link ListCheckpointed}
	 * as it subsumes {@link org.apache.flink.streaming.api.checkpoint.CheckpointedFunction}.
	 */
public void testSavepointRescalingPartitionedOperatorState(boolean scaleOut, OperatorCheckpointMethod checkpointMethod) throws Exception {
    final int parallelism = scaleOut ? numSlots : numSlots / 2;
    final int parallelism2 = scaleOut ? numSlots / 2 : numSlots;
    final int maxParallelism = 13;
    FiniteDuration timeout = new FiniteDuration(3, TimeUnit.MINUTES);
    Deadline deadline = timeout.fromNow();
    JobID jobID = null;
    ActorGateway jobManager = null;
    int counterSize = Math.max(parallelism, parallelism2);
    if (checkpointMethod == OperatorCheckpointMethod.CHECKPOINTED_FUNCTION || checkpointMethod == OperatorCheckpointMethod.CHECKPOINTED_FUNCTION_BROADCAST) {
        PartitionedStateSource.CHECK_CORRECT_SNAPSHOT = new int[counterSize];
        PartitionedStateSource.CHECK_CORRECT_RESTORE = new int[counterSize];
    } else {
        PartitionedStateSourceListCheckpointed.CHECK_CORRECT_SNAPSHOT = new int[counterSize];
        PartitionedStateSourceListCheckpointed.CHECK_CORRECT_RESTORE = new int[counterSize];
    }
    try {
        jobManager = cluster.getLeaderGateway(deadline.timeLeft());
        JobGraph jobGraph = createJobGraphWithOperatorState(parallelism, maxParallelism, checkpointMethod);
        jobID = jobGraph.getJobID();
        cluster.submitJobDetached(jobGraph);
        Object savepointResponse = null;
        // wait until the operator is started
        StateSourceBase.workStartedLatch.await();
        while (deadline.hasTimeLeft()) {
            Future<Object> savepointPathFuture = jobManager.ask(new JobManagerMessages.TriggerSavepoint(jobID, Option.<String>empty()), deadline.timeLeft());
            FiniteDuration waitingTime = new FiniteDuration(10, TimeUnit.SECONDS);
            savepointResponse = Await.result(savepointPathFuture, waitingTime);
            if (savepointResponse instanceof JobManagerMessages.TriggerSavepointSuccess) {
                break;
            }
            System.out.println(savepointResponse);
        }
        assertTrue(savepointResponse instanceof JobManagerMessages.TriggerSavepointSuccess);
        final String savepointPath = ((JobManagerMessages.TriggerSavepointSuccess) savepointResponse).savepointPath();
        Future<Object> jobRemovedFuture = jobManager.ask(new TestingJobManagerMessages.NotifyWhenJobRemoved(jobID), deadline.timeLeft());
        Future<Object> cancellationResponseFuture = jobManager.ask(new JobManagerMessages.CancelJob(jobID), deadline.timeLeft());
        Object cancellationResponse = Await.result(cancellationResponseFuture, deadline.timeLeft());
        assertTrue(cancellationResponse instanceof JobManagerMessages.CancellationSuccess);
        Await.ready(jobRemovedFuture, deadline.timeLeft());
        // job successfully removed
        jobID = null;
        JobGraph scaledJobGraph = createJobGraphWithOperatorState(parallelism2, maxParallelism, checkpointMethod);
        scaledJobGraph.setSavepointRestoreSettings(SavepointRestoreSettings.forPath(savepointPath));
        jobID = scaledJobGraph.getJobID();
        cluster.submitJobAndWait(scaledJobGraph, false);
        int sumExp = 0;
        int sumAct = 0;
        if (checkpointMethod == OperatorCheckpointMethod.CHECKPOINTED_FUNCTION) {
            for (int c : PartitionedStateSource.CHECK_CORRECT_SNAPSHOT) {
                sumExp += c;
            }
            for (int c : PartitionedStateSource.CHECK_CORRECT_RESTORE) {
                sumAct += c;
            }
        } else if (checkpointMethod == OperatorCheckpointMethod.CHECKPOINTED_FUNCTION_BROADCAST) {
            for (int c : PartitionedStateSource.CHECK_CORRECT_SNAPSHOT) {
                sumExp += c;
            }
            for (int c : PartitionedStateSource.CHECK_CORRECT_RESTORE) {
                sumAct += c;
            }
            sumExp *= parallelism2;
        } else {
            for (int c : PartitionedStateSourceListCheckpointed.CHECK_CORRECT_SNAPSHOT) {
                sumExp += c;
            }
            for (int c : PartitionedStateSourceListCheckpointed.CHECK_CORRECT_RESTORE) {
                sumAct += c;
            }
        }
        assertEquals(sumExp, sumAct);
        jobID = null;
    } finally {
        // clear any left overs from a possibly failed job
        if (jobID != null && jobManager != null) {
            Future<Object> jobRemovedFuture = jobManager.ask(new TestingJobManagerMessages.NotifyWhenJobRemoved(jobID), timeout);
            try {
                Await.ready(jobRemovedFuture, timeout);
            } catch (TimeoutException | InterruptedException ie) {
                fail("Failed while cleaning up the cluster.");
            }
        }
    }
}

Example 28 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class SavepointITCase method testCanRestoreWithModifiedStatelessOperators.

/**
	 * FLINK-5985
	 *
	 * This test ensures we can restore from a savepoint under modifications to the job graph that only concern
	 * stateless operators.
	 */
@Test
public void testCanRestoreWithModifiedStatelessOperators() throws Exception {
    // Config
    int numTaskManagers = 2;
    int numSlotsPerTaskManager = 2;
    int parallelism = 2;
    // Test deadline
    final Deadline deadline = new FiniteDuration(5, TimeUnit.MINUTES).fromNow();
    final File tmpDir = CommonTestUtils.createTempDirectory();
    final File savepointDir = new File(tmpDir, "savepoints");
    TestingCluster flink = null;
    String savepointPath;
    try {
        // Flink configuration
        final Configuration config = new Configuration();
        config.setInteger(ConfigConstants.LOCAL_NUMBER_TASK_MANAGER, numTaskManagers);
        config.setInteger(ConfigConstants.TASK_MANAGER_NUM_TASK_SLOTS, numSlotsPerTaskManager);
        config.setString(ConfigConstants.SAVEPOINT_DIRECTORY_KEY, savepointDir.toURI().toString());
        LOG.info("Flink configuration: " + config + ".");
        // Start Flink
        flink = new TestingCluster(config);
        LOG.info("Starting Flink cluster.");
        flink.start(true);
        // Retrieve the job manager
        LOG.info("Retrieving JobManager.");
        ActorGateway jobManager = Await.result(flink.leaderGateway().future(), deadline.timeLeft());
        LOG.info("JobManager: " + jobManager + ".");
        final StatefulCounter statefulCounter = new StatefulCounter();
        StatefulCounter.resetForTest(parallelism);
        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(parallelism);
        env.addSource(new InfiniteTestSource()).shuffle().map(new MapFunction<Integer, Integer>() {

            @Override
            public Integer map(Integer value) throws Exception {
                return 4 * value;
            }
        }).shuffle().map(statefulCounter).uid("statefulCounter").shuffle().map(new MapFunction<Integer, Integer>() {

            @Override
            public Integer map(Integer value) throws Exception {
                return 2 * value;
            }
        }).addSink(new DiscardingSink<Integer>());
        JobGraph originalJobGraph = env.getStreamGraph().getJobGraph();
        JobSubmissionResult submissionResult = flink.submitJobDetached(originalJobGraph);
        JobID jobID = submissionResult.getJobID();
        // wait for the Tasks to be ready
        StatefulCounter.getProgressLatch().await(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
        Future<Object> savepointPathFuture = jobManager.ask(new TriggerSavepoint(jobID, Option.<String>empty()), deadline.timeLeft());
        savepointPath = ((TriggerSavepointSuccess) Await.result(savepointPathFuture, deadline.timeLeft())).savepointPath();
        Future<Object> savepointFuture = jobManager.ask(new RequestSavepoint(savepointPath), deadline.timeLeft());
        ((ResponseSavepoint) Await.result(savepointFuture, deadline.timeLeft())).savepoint();
        LOG.info("Retrieved savepoint: " + savepointPath + ".");
        // Shut down the Flink cluster (thereby canceling the job)
        LOG.info("Shutting down Flink cluster.");
        flink.shutdown();
        flink.awaitTermination();
    } finally {
        flink.shutdown();
        flink.awaitTermination();
    }
    try {
        LOG.info("Restarting Flink cluster.");
        flink.start(true);
        // Retrieve the job manager
        LOG.info("Retrieving JobManager.");
        ActorGateway jobManager = Await.result(flink.leaderGateway().future(), deadline.timeLeft());
        LOG.info("JobManager: " + jobManager + ".");
        // Reset static test helpers
        StatefulCounter.resetForTest(parallelism);
        // Gather all task deployment descriptors
        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(parallelism);
        // generate a modified job graph that adds a stateless op
        env.addSource(new InfiniteTestSource()).shuffle().map(new StatefulCounter()).uid("statefulCounter").shuffle().map(new MapFunction<Integer, Integer>() {

            @Override
            public Integer map(Integer value) throws Exception {
                return value;
            }
        }).addSink(new DiscardingSink<Integer>());
        JobGraph modifiedJobGraph = env.getStreamGraph().getJobGraph();
        // Set the savepoint path
        modifiedJobGraph.setSavepointRestoreSettings(SavepointRestoreSettings.forPath(savepointPath));
        LOG.info("Resubmitting job " + modifiedJobGraph.getJobID() + " with " + "savepoint path " + savepointPath + " in detached mode.");
        // Submit the job
        flink.submitJobDetached(modifiedJobGraph);
        // Await state is restored
        StatefulCounter.getRestoreLatch().await(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
        // Await some progress after restore
        StatefulCounter.getProgressLatch().await(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
    } finally {
        flink.shutdown();
        flink.awaitTermination();
    }
}

Example 29 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class CancelingTestBase method runAndCancelJob.

public void runAndCancelJob(Plan plan, final int msecsTillCanceling, int maxTimeTillCanceled) throws Exception {
    try {
        // submit job
        final JobGraph jobGraph = getJobGraph(plan);
        executor.submitJobDetached(jobGraph);
        // Wait for the job to make some progress and then cancel
        JobManagerActorTestUtils.waitForJobStatus(jobGraph.getJobID(), JobStatus.RUNNING, executor.getLeaderGateway(TestingUtils.TESTING_DURATION()), TestingUtils.TESTING_DURATION());
        Thread.sleep(msecsTillCanceling);
        FiniteDuration timeout = new FiniteDuration(maxTimeTillCanceled, TimeUnit.MILLISECONDS);
        ActorGateway jobManager = executor.getLeaderGateway(TestingUtils.TESTING_DURATION());
        Future<Object> ask = jobManager.ask(new CancelJob(jobGraph.getJobID()), timeout);
        Object result = Await.result(ask, timeout);
        if (result instanceof CancellationSuccess) {
        // all good
        } else if (result instanceof CancellationFailure) {
            // Failure
            CancellationFailure failure = (CancellationFailure) result;
            throw new Exception("Failed to cancel job with ID " + failure.jobID() + ".", failure.cause());
        } else {
            throw new Exception("Unexpected response to cancel request: " + result);
        }
        // Wait for the job to be cancelled
        JobManagerActorTestUtils.waitForJobStatus(jobGraph.getJobID(), JobStatus.CANCELED, executor.getLeaderGateway(TestingUtils.TESTING_DURATION()), TestingUtils.TESTING_DURATION());
    } catch (Exception e) {
        LOG.error("Exception found in runAndCancelJob.", e);
        e.printStackTrace();
        Assert.fail(e.getMessage());
    }
}

Example 30 with ActorGateway

use of org.apache.flink.runtime.instance.ActorGateway in project flink by apache.

the class YARNHighAvailabilityITCase method testMultipleAMKill.

/**
	 * Tests that the application master can be killed multiple times and that the surviving
	 * TaskManager successfully reconnects to the newly started JobManager.
	 * @throws Exception
	 */
@Test
public void testMultipleAMKill() throws Exception {
    final int numberKillingAttempts = numberApplicationAttempts - 1;
    TestingYarnClusterDescriptor flinkYarnClient = new TestingYarnClusterDescriptor();
    Assert.assertNotNull("unable to get yarn client", flinkYarnClient);
    flinkYarnClient.setTaskManagerCount(1);
    flinkYarnClient.setJobManagerMemory(768);
    flinkYarnClient.setTaskManagerMemory(1024);
    flinkYarnClient.setLocalJarPath(new Path(flinkUberjar.getAbsolutePath()));
    flinkYarnClient.addShipFiles(Arrays.asList(flinkLibFolder.listFiles()));
    String confDirPath = System.getenv(ConfigConstants.ENV_FLINK_CONF_DIR);
    flinkYarnClient.setConfigurationDirectory(confDirPath);
    String fsStateHandlePath = temp.getRoot().getPath();
    // load the configuration
    File configDirectory = new File(confDirPath);
    GlobalConfiguration.loadConfiguration(configDirectory.getAbsolutePath());
    flinkYarnClient.setFlinkConfiguration(GlobalConfiguration.loadConfiguration());
    flinkYarnClient.setDynamicPropertiesEncoded("recovery.mode=zookeeper@@recovery.zookeeper.quorum=" + zkServer.getConnectString() + "@@yarn.application-attempts=" + numberApplicationAttempts + "@@" + CoreOptions.STATE_BACKEND + "=FILESYSTEM" + "@@" + FsStateBackendFactory.CHECKPOINT_DIRECTORY_URI_CONF_KEY + "=" + fsStateHandlePath + "/checkpoints" + "@@" + HighAvailabilityOptions.HA_STORAGE_PATH.key() + "=" + fsStateHandlePath + "/recovery");
    flinkYarnClient.setConfigurationFilePath(new Path(confDirPath + File.separator + "flink-conf.yaml"));
    ClusterClient yarnCluster = null;
    final FiniteDuration timeout = new FiniteDuration(2, TimeUnit.MINUTES);
    try {
        yarnCluster = flinkYarnClient.deploy();
        final Configuration config = yarnCluster.getFlinkConfiguration();
        new JavaTestKit(actorSystem) {

            {
                for (int attempt = 0; attempt < numberKillingAttempts; attempt++) {
                    new Within(timeout) {

                        @Override
                        protected void run() {
                            try {
                                LeaderRetrievalService lrs = LeaderRetrievalUtils.createLeaderRetrievalService(config);
                                ActorGateway gateway = LeaderRetrievalUtils.retrieveLeaderGateway(lrs, actorSystem, timeout);
                                ActorGateway selfGateway = new AkkaActorGateway(getRef(), gateway.leaderSessionID());
                                gateway.tell(new TestingJobManagerMessages.NotifyWhenAtLeastNumTaskManagerAreRegistered(1), selfGateway);
                                expectMsgEquals(Acknowledge.get());
                                gateway.tell(PoisonPill.getInstance());
                            } catch (Exception e) {
                                throw new AssertionError("Could not complete test.", e);
                            }
                        }
                    };
                }
                new Within(timeout) {

                    @Override
                    protected void run() {
                        try {
                            LeaderRetrievalService lrs = LeaderRetrievalUtils.createLeaderRetrievalService(config);
                            ActorGateway gateway2 = LeaderRetrievalUtils.retrieveLeaderGateway(lrs, actorSystem, timeout);
                            ActorGateway selfGateway = new AkkaActorGateway(getRef(), gateway2.leaderSessionID());
                            gateway2.tell(new TestingJobManagerMessages.NotifyWhenAtLeastNumTaskManagerAreRegistered(1), selfGateway);
                            expectMsgEquals(Acknowledge.get());
                        } catch (Exception e) {
                            throw new AssertionError("Could not complete test.", e);
                        }
                    }
                };
            }
        };
    } finally {
        if (yarnCluster != null) {
            yarnCluster.shutdown();
        }
    }
}