Examples with TestingFatalErrorHandler org.apache.flink.runtime.util.TestingFatalErrorHandler used on opensource projects

Example 21 with TestingFatalErrorHandler

use of org.apache.flink.runtime.util.TestingFatalErrorHandler in project flink by apache.

the class JobMasterTest method testTriggerSavepointTimeout.

/**
 * Tests that the timeout in {@link JobMasterGateway#triggerSavepoint(String, boolean,
 * SavepointFormatType, Time)} is respected.
 */
@Test
public void testTriggerSavepointTimeout() throws Exception {
    final TestingSchedulerNG testingSchedulerNG = TestingSchedulerNG.newBuilder().setTriggerSavepointFunction((ignoredA, ignoredB, formatType) -> new CompletableFuture<>()).build();
    final JobMaster jobMaster = new JobMasterBuilder(jobGraph, rpcService).withFatalErrorHandler(testingFatalErrorHandler).withSlotPoolServiceSchedulerFactory(DefaultSlotPoolServiceSchedulerFactory.create(TestingSlotPoolServiceBuilder.newBuilder(), new TestingSchedulerNGFactory(testingSchedulerNG))).createJobMaster();
    try {
        jobMaster.start();
        final JobMasterGateway jobMasterGateway = jobMaster.getSelfGateway(JobMasterGateway.class);
        final CompletableFuture<String> savepointFutureLowTimeout = jobMasterGateway.triggerSavepoint("/tmp", false, SavepointFormatType.CANONICAL, Time.milliseconds(1));
        final CompletableFuture<String> savepointFutureHighTimeout = jobMasterGateway.triggerSavepoint("/tmp", false, SavepointFormatType.CANONICAL, RpcUtils.INF_TIMEOUT);
        try {
            savepointFutureLowTimeout.get(testingTimeout.getSize(), testingTimeout.getUnit());
            fail();
        } catch (final ExecutionException e) {
            final Throwable cause = ExceptionUtils.stripExecutionException(e);
            assertThat(cause, instanceOf(TimeoutException.class));
        }
        assertThat(savepointFutureHighTimeout.isDone(), is(equalTo(false)));
    } finally {
        RpcUtils.terminateRpcEndpoint(jobMaster, testingTimeout);
    }
}

Example 22 with TestingFatalErrorHandler

use of org.apache.flink.runtime.util.TestingFatalErrorHandler in project flink by apache.

the class JobMasterTest method setup.

@Before
public void setup() throws IOException {
    configuration = new Configuration();
    haServices = new TestingHighAvailabilityServices();
    jobMasterId = JobMasterId.generate();
    jmResourceId = ResourceID.generate();
    testingFatalErrorHandler = new TestingFatalErrorHandler();
    haServices.setCheckpointRecoveryFactory(new StandaloneCheckpointRecoveryFactory());
    rmLeaderRetrievalService = new SettableLeaderRetrievalService(null, null);
    haServices.setResourceManagerLeaderRetriever(rmLeaderRetrievalService);
    configuration.setString(BlobServerOptions.STORAGE_DIRECTORY, temporaryFolder.newFolder().getAbsolutePath());
}

Example 23 with TestingFatalErrorHandler

use of org.apache.flink.runtime.util.TestingFatalErrorHandler in project flink by apache.

the class AdaptiveSchedulerTest method testFatalErrorsForwardedToFatalErrorHandler.

@Test
public void testFatalErrorsForwardedToFatalErrorHandler() throws Exception {
    final TestingFatalErrorHandler fatalErrorHandler = new TestingFatalErrorHandler();
    final AdaptiveScheduler scheduler = new AdaptiveSchedulerBuilder(createJobGraph(), mainThreadExecutor).setFatalErrorHandler(fatalErrorHandler).build();
    final RuntimeException exception = new RuntimeException();
    scheduler.runIfState(scheduler.getState(), () -> {
        throw exception;
    });
    assertThat(fatalErrorHandler.getException()).isEqualTo(exception);
}

Example 24 with TestingFatalErrorHandler

use of org.apache.flink.runtime.util.TestingFatalErrorHandler in project flink by apache.

the class ProcessFailureCancelingITCase method testCancelingOnProcessFailure.

@Test
public void testCancelingOnProcessFailure() throws Throwable {
    Assume.assumeTrue("---- Skipping Process Failure test : Could not find java executable ----", getJavaCommandPath() != null);
    TestProcess taskManagerProcess = null;
    final TestingFatalErrorHandler fatalErrorHandler = new TestingFatalErrorHandler();
    Configuration config = new Configuration();
    config.setString(JobManagerOptions.ADDRESS, "localhost");
    config.set(AkkaOptions.ASK_TIMEOUT_DURATION, Duration.ofSeconds(100));
    config.setString(HighAvailabilityOptions.HA_MODE, "zookeeper");
    config.setString(HighAvailabilityOptions.HA_ZOOKEEPER_QUORUM, zooKeeperResource.getConnectString());
    config.setString(HighAvailabilityOptions.HA_STORAGE_PATH, temporaryFolder.newFolder().getAbsolutePath());
    config.setInteger(TaskManagerOptions.NUM_TASK_SLOTS, 2);
    config.set(TaskManagerOptions.MANAGED_MEMORY_SIZE, MemorySize.parse("4m"));
    config.set(TaskManagerOptions.NETWORK_MEMORY_MIN, MemorySize.parse("3200k"));
    config.set(TaskManagerOptions.NETWORK_MEMORY_MAX, MemorySize.parse("3200k"));
    config.set(TaskManagerOptions.TASK_HEAP_MEMORY, MemorySize.parse("128m"));
    config.set(TaskManagerOptions.CPU_CORES, 1.0);
    config.setInteger(RestOptions.PORT, 0);
    final RpcService rpcService = RpcSystem.load().remoteServiceBuilder(config, "localhost", "0").createAndStart();
    final int jobManagerPort = rpcService.getPort();
    config.setInteger(JobManagerOptions.PORT, jobManagerPort);
    final DispatcherResourceManagerComponentFactory resourceManagerComponentFactory = DefaultDispatcherResourceManagerComponentFactory.createSessionComponentFactory(StandaloneResourceManagerFactory.getInstance());
    DispatcherResourceManagerComponent dispatcherResourceManagerComponent = null;
    final ScheduledExecutorService ioExecutor = TestingUtils.defaultExecutor();
    final HighAvailabilityServices haServices = HighAvailabilityServicesUtils.createHighAvailabilityServices(config, ioExecutor, AddressResolution.NO_ADDRESS_RESOLUTION, RpcSystem.load(), NoOpFatalErrorHandler.INSTANCE);
    final AtomicReference<Throwable> programException = new AtomicReference<>();
    try {
        dispatcherResourceManagerComponent = resourceManagerComponentFactory.create(config, ResourceID.generate(), ioExecutor, rpcService, haServices, blobServerResource.getBlobServer(), new HeartbeatServices(100L, 10000L, 2), NoOpMetricRegistry.INSTANCE, new MemoryExecutionGraphInfoStore(), VoidMetricQueryServiceRetriever.INSTANCE, fatalErrorHandler);
        TestProcessBuilder taskManagerProcessBuilder = new TestProcessBuilder(TaskExecutorProcessEntryPoint.class.getName());
        taskManagerProcessBuilder.addConfigAsMainClassArgs(config);
        taskManagerProcess = taskManagerProcessBuilder.start();
        // start the test program, which infinitely blocks
        Runnable programRunner = new Runnable() {

            @Override
            public void run() {
                try {
                    ExecutionEnvironment env = ExecutionEnvironment.createRemoteEnvironment("localhost", 1337, config);
                    env.setParallelism(2);
                    env.setRestartStrategy(RestartStrategies.noRestart());
                    env.generateSequence(0, Long.MAX_VALUE).map(new MapFunction<Long, Long>() {

                        @Override
                        public Long map(Long value) throws Exception {
                            synchronized (this) {
                                System.out.println(TASK_DEPLOYED_MARKER);
                                wait();
                            }
                            return 0L;
                        }
                    }).output(new DiscardingOutputFormat<>());
                    env.execute();
                } catch (Throwable t) {
                    programException.set(t);
                }
            }
        };
        Thread programThread = new Thread(programRunner);
        programThread.start();
        waitUntilAtLeastOneTaskHasBeenDeployed(taskManagerProcess);
        // kill the TaskManager after the job started to run
        taskManagerProcess.destroy();
        taskManagerProcess = null;
        // the job should fail within a few seconds due to heartbeat timeouts
        // since the CI environment is often slow, we conservatively give it up to 2 minutes
        programThread.join(TIMEOUT.toMillis());
        assertFalse("The program did not cancel in time", programThread.isAlive());
        Throwable error = programException.get();
        assertNotNull("The program did not fail properly", error);
        assertTrue(error instanceof ProgramInvocationException);
    // all seems well :-)
    } catch (Exception | Error e) {
        if (taskManagerProcess != null) {
            printOutput("TaskManager OUT", taskManagerProcess.getProcessOutput().toString());
            printOutput("TaskManager ERR", taskManagerProcess.getErrorOutput().toString());
        }
        throw ExceptionUtils.firstOrSuppressed(e, programException.get());
    } finally {
        if (taskManagerProcess != null) {
            taskManagerProcess.destroy();
        }
        if (dispatcherResourceManagerComponent != null) {
            dispatcherResourceManagerComponent.stopApplication(ApplicationStatus.SUCCEEDED, null);
        }
        fatalErrorHandler.rethrowError();
        RpcUtils.terminateRpcService(rpcService, Time.seconds(100L));
        haServices.closeAndCleanupAllData();
    }
}