org.apache.flink.runtime.client.JobCancellationException Java Examples

@Test
public void testCancelledJobThrowsJobCancellationException() throws Exception {
	final FlinkException cause = new FlinkException("Test exception");
	final JobResult jobResult = JobResult.createFrom(
		new ArchivedExecutionGraphBuilder()
			.setJobID(new JobID())
			.setState(JobStatus.CANCELED)
			.setFailureCause(new ErrorInfo(cause, 42L))
			.build());

	try {
		jobResult.toJobExecutionResult(getClass().getClassLoader());
		fail("Job should fail with an JobCancellationException.");
	} catch (JobCancellationException expected) {
		assertThat(expected.getCause(), is(equalTo(cause)));
	}
}
 

@Test
public void testCancelledJobThrowsJobCancellationException() throws Exception {
	final FlinkException cause = new FlinkException("Test exception");
	final JobResult jobResult = JobResult.createFrom(
		new ArchivedExecutionGraphBuilder()
			.setJobID(new JobID())
			.setState(JobStatus.CANCELED)
			.setFailureCause(new ErrorInfo(cause, 42L))
			.build());

	try {
		jobResult.toJobExecutionResult(getClass().getClassLoader());
		fail("Job should fail with an JobCancellationException.");
	} catch (JobCancellationException expected) {
		assertThat(expected.getCause(), is(equalTo(cause)));
	}
}
 

@Test
public void testCancelledJobThrowsJobCancellationException() throws Exception {
	final FlinkException cause = new FlinkException("Test exception");
	final JobResult jobResult = JobResult.createFrom(
		new ArchivedExecutionGraphBuilder()
			.setJobID(new JobID())
			.setState(JobStatus.CANCELED)
			.setFailureCause(new ErrorInfo(cause, 42L))
			.build());

	try {
		jobResult.toJobExecutionResult(getClass().getClassLoader());
		fail("Job should fail with an JobCancellationException.");
	} catch (JobCancellationException expected) {
		// the failure cause in the execution graph should not be the cause of the canceled job result
		assertThat(expected.getCause(), is(nullValue()));
	}
}
 

/**
 * Runs the user program entrypoint using {@link #runApplicationAsync(DispatcherGateway,
 * ScheduledExecutor, boolean)} and shuts down the given dispatcher when the application
 * completes (either successfully or in case of failure).
 */
@VisibleForTesting
CompletableFuture<Acknowledge> runApplicationAndShutdownClusterAsync(
		final DispatcherGateway dispatcher,
		final ScheduledExecutor scheduledExecutor) {

	applicationCompletionFuture = fixJobIdAndRunApplicationAsync(dispatcher, scheduledExecutor);

	return applicationCompletionFuture
			.handle((r, t) -> {
				final ApplicationStatus applicationStatus;
				if (t != null) {

					final Optional<JobCancellationException> cancellationException =
							ExceptionUtils.findThrowable(t, JobCancellationException.class);

					if (cancellationException.isPresent()) {
						// this means the Flink Job was cancelled
						applicationStatus = ApplicationStatus.CANCELED;
					} else if (t instanceof CancellationException) {
						// this means that the future was cancelled
						applicationStatus = ApplicationStatus.UNKNOWN;
					} else {
						applicationStatus = ApplicationStatus.FAILED;
					}

					LOG.warn("Application {}: ", applicationStatus, t);
				} else {
					applicationStatus = ApplicationStatus.SUCCEEDED;
					LOG.info("Application completed SUCCESSFULLY");
				}
				return dispatcher.shutDownCluster(applicationStatus);
			})
			.thenCompose(Function.identity());
}
 

/**
 * If the given {@link JobResult} indicates success, this passes through the {@link JobResult}.
 * Otherwise, this returns a future that is finished exceptionally (potentially with an
 * exception from the {@link JobResult}.
 */
private CompletableFuture<JobResult> unwrapJobResultException(
		CompletableFuture<JobResult> jobResult) {
	return jobResult.thenApply(result -> {
		if (result.isSuccess()) {
			return result;
		}

		Optional<SerializedThrowable> serializedThrowable = result.getSerializedThrowable();

		if (result.getApplicationStatus() == ApplicationStatus.CANCELED) {
			throw new CompletionException(
					new JobCancellationException(
							result.getJobId(),
							"Job was cancelled.",
							serializedThrowable.orElse(null)));
		}

		if (serializedThrowable.isPresent()) {
			Throwable throwable =
					serializedThrowable
							.get()
							.deserializeError(application.getUserCodeClassLoader());
			throw new CompletionException(throwable);
		}
		throw new RuntimeException("Job execution failed for unknown reason.");
	});
}
 

private static JobResult createCancelledJobResult(final JobID jobId) {
	return new JobResult.Builder()
			.jobId(jobId)
			.netRuntime(2L)
			.serializedThrowable(
					new SerializedThrowable(
							new JobCancellationException(jobId, "Hello", null)))
			.applicationStatus(ApplicationStatus.CANCELED)
			.build();
}
 

/**
 * Tests disposal of a savepoint, which contains custom user code KvState.
 */
@Test
public void testDisposeSavepointWithCustomKvState() throws Exception {
	ClusterClient<?> clusterClient = new MiniClusterClient(new Configuration(), miniClusterResource.getMiniCluster());

	Deadline deadline = new FiniteDuration(100, TimeUnit.SECONDS).fromNow();

	File checkpointDir = FOLDER.newFolder();
	File outputDir = FOLDER.newFolder();

	final PackagedProgram program = new PackagedProgram(
			new File(CUSTOM_KV_STATE_JAR_PATH),
			new String[] {
					String.valueOf(parallelism),
					checkpointDir.toURI().toString(),
					"5000",
					outputDir.toURI().toString()
			});

	TestStreamEnvironment.setAsContext(
		miniClusterResource.getMiniCluster(),
		parallelism,
		Collections.singleton(new Path(CUSTOM_KV_STATE_JAR_PATH)),
		Collections.<URL>emptyList()
	);

	// Execute detached
	Thread invokeThread = new Thread(new Runnable() {
		@Override
		public void run() {
			try {
				program.invokeInteractiveModeForExecution();
			} catch (ProgramInvocationException ignored) {
				if (ignored.getCause() == null ||
					!(ignored.getCause() instanceof JobCancellationException)) {
					ignored.printStackTrace();
				}
			}
		}
	});

	LOG.info("Starting program invoke thread");
	invokeThread.start();

	// The job ID
	JobID jobId = null;

	LOG.info("Waiting for job status running.");

	// Wait for running job
	while (jobId == null && deadline.hasTimeLeft()) {

		Collection<JobStatusMessage> jobs = clusterClient.listJobs().get(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
		for (JobStatusMessage job : jobs) {
			if (job.getJobState() == JobStatus.RUNNING) {
				jobId = job.getJobId();
				LOG.info("Job running. ID: " + jobId);
				break;
			}
		}

		// Retry if job is not available yet
		if (jobId == null) {
			Thread.sleep(100L);
		}
	}

	// Trigger savepoint
	String savepointPath = null;
	for (int i = 0; i < 20; i++) {
		LOG.info("Triggering savepoint (" + (i + 1) + "/20).");
		try {
			savepointPath = clusterClient.triggerSavepoint(jobId, null)
				.get(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
		} catch (Exception cause) {
			LOG.info("Failed to trigger savepoint. Retrying...", cause);
			// This can fail if the operators are not opened yet
			Thread.sleep(500);
		}
	}

	assertNotNull("Failed to trigger savepoint", savepointPath);

	clusterClient.disposeSavepoint(savepointPath).get();

	clusterClient.cancel(jobId);

	// make sure, the execution is finished to not influence other test methods
	invokeThread.join(deadline.timeLeft().toMillis());
	assertFalse("Program invoke thread still running", invokeThread.isAlive());
}
 

/**
 * Tests disposal of a savepoint, which contains custom user code KvState.
 */
@Test
public void testDisposeSavepointWithCustomKvState() throws Exception {
	ClusterClient<?> clusterClient = new MiniClusterClient(new Configuration(), miniClusterResource.getMiniCluster());

	Deadline deadline = new FiniteDuration(100, TimeUnit.SECONDS).fromNow();

	File checkpointDir = FOLDER.newFolder();
	File outputDir = FOLDER.newFolder();

	final PackagedProgram program = new PackagedProgram(
			new File(CUSTOM_KV_STATE_JAR_PATH),
			new String[] {
					String.valueOf(parallelism),
					checkpointDir.toURI().toString(),
					"5000",
					outputDir.toURI().toString()
			});

	TestStreamEnvironment.setAsContext(
		miniClusterResource.getMiniCluster(),
		parallelism,
		Collections.singleton(new Path(CUSTOM_KV_STATE_JAR_PATH)),
		Collections.<URL>emptyList()
	);

	// Execute detached
	Thread invokeThread = new Thread(new Runnable() {
		@Override
		public void run() {
			try {
				program.invokeInteractiveModeForExecution();
			} catch (ProgramInvocationException ignored) {
				if (ignored.getCause() == null ||
					!(ignored.getCause() instanceof JobCancellationException)) {
					ignored.printStackTrace();
				}
			}
		}
	});

	LOG.info("Starting program invoke thread");
	invokeThread.start();

	// The job ID
	JobID jobId = null;

	LOG.info("Waiting for job status running.");

	// Wait for running job
	while (jobId == null && deadline.hasTimeLeft()) {

		Collection<JobStatusMessage> jobs = clusterClient.listJobs().get(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
		for (JobStatusMessage job : jobs) {
			if (job.getJobState() == JobStatus.RUNNING) {
				jobId = job.getJobId();
				LOG.info("Job running. ID: " + jobId);
				break;
			}
		}

		// Retry if job is not available yet
		if (jobId == null) {
			Thread.sleep(100L);
		}
	}

	// Trigger savepoint
	String savepointPath = null;
	for (int i = 0; i < 20; i++) {
		LOG.info("Triggering savepoint (" + (i + 1) + "/20).");
		try {
			savepointPath = clusterClient.triggerSavepoint(jobId, null)
				.get(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
		} catch (Exception cause) {
			LOG.info("Failed to trigger savepoint. Retrying...", cause);
			// This can fail if the operators are not opened yet
			Thread.sleep(500);
		}
	}

	assertNotNull("Failed to trigger savepoint", savepointPath);

	clusterClient.disposeSavepoint(savepointPath).get();

	clusterClient.cancel(jobId);

	// make sure, the execution is finished to not influence other test methods
	invokeThread.join(deadline.timeLeft().toMillis());
	assertFalse("Program invoke thread still running", invokeThread.isAlive());
}
 

@Test
public void testPravegaWithSavepoint() throws Exception {
    final int sourceParallelism = 4;
    final int numPravegaSegments = 4;
    final int numElements = NUM_STREAM_ELEMENTS;

    // set up the stream
    final String streamName = RandomStringUtils.randomAlphabetic(20);
    SETUP_UTILS.createTestStream(streamName, numPravegaSegments);

    // we create two independent Flink jobs (that come from the same program)
    final JobGraph program1 = getFlinkJob(sourceParallelism, streamName, numElements);

    try (
            final EventStreamWriter<Integer> eventWriter = SETUP_UTILS.getIntegerWriter(streamName);

            // create the producer that writes to the stream
            final ThrottledIntegerWriter producer = new ThrottledIntegerWriter(
                    eventWriter,
                    numElements,
                    numElements / 2,  // the latest when the thread must be un-throttled
                    1,                 // the initial sleep time per element
                    false
            )

    ) {
        // the object on which we block while waiting for the checkpoint completion
        final OneShotLatch sync = new OneShotLatch();
        NotifyingMapper.TO_CALL_ON_COMPLETION.set( sync::trigger );

        // launch the Flink program from a separate thread
        final CheckedThread flinkRunner = new CheckedThread() {
            @Override
            public void go() throws Exception {
                MINI_CLUSTER.submitJob(program1);
            }
        };

        producer.start();
        flinkRunner.start();

        // wait until at least one checkpoint is complete before triggering the safepoints
        sync.await();

        // now that we are comfortably into the program, trigger a savepoint
        String savepointPath = null;

        // since with the short timeouts we configure in these tests, Pravega Checkpoints
        // sometimes don't complete in time, we retry a bit here
        for (int attempt = 1; savepointPath == null && attempt <= 5; attempt++) {
            savepointPath = MINI_CLUSTER.triggerSavepoint(program1.getJobID(), tmpFolder.newFolder().getAbsolutePath(), false).get();
        }

        assertNotNull("Failed to trigger a savepoint", savepointPath);

        // now cancel the job and relaunch a new one
        MINI_CLUSTER.cancelJob(program1.getJobID());

        try {
            // this throws an exception that the job was cancelled
            flinkRunner.sync();
        } catch (JobCancellationException ignored) {
        }

        producer.unthrottle();

        // now, resume with a new program
        final JobGraph program2 = getFlinkJob(sourceParallelism, streamName, numElements);
        program2.setSavepointRestoreSettings(SavepointRestoreSettings.forPath(savepointPath, false));

        // if these calls complete without exception, then the test passes
        try {
            MINI_CLUSTER.executeJobBlocking(program2);
        } catch (Exception e) {
            if (!(ExceptionUtils.getRootCause(e) instanceof SuccessException)) {
                throw e;
            }
        }
    }
}
 

/**
 * Tests disposal of a savepoint, which contains custom user code KvState.
 */
@Test
public void testDisposeSavepointWithCustomKvState() throws Exception {
	ClusterClient<?> clusterClient = new MiniClusterClient(new Configuration(), miniClusterResource.getMiniCluster());

	Deadline deadline = new FiniteDuration(100, TimeUnit.SECONDS).fromNow();

	File checkpointDir = FOLDER.newFolder();
	File outputDir = FOLDER.newFolder();

	final PackagedProgram program = PackagedProgram.newBuilder()
		.setJarFile(new File(CUSTOM_KV_STATE_JAR_PATH))
		.setArguments(new String[] {
			String.valueOf(parallelism),
			checkpointDir.toURI().toString(),
			"5000",
			outputDir.toURI().toString(),
			"false" // Disable unaligned checkpoints as this test is triggering concurrent savepoints/checkpoints
		})
		.build();

	TestStreamEnvironment.setAsContext(
		miniClusterResource.getMiniCluster(),
		parallelism,
		Collections.singleton(new Path(CUSTOM_KV_STATE_JAR_PATH)),
		Collections.emptyList()
	);

	// Execute detached
	Thread invokeThread = new Thread(() -> {
		try {
			program.invokeInteractiveModeForExecution();
		} catch (ProgramInvocationException ex) {
			if (ex.getCause() == null ||
				!(ex.getCause() instanceof JobCancellationException)) {
				ex.printStackTrace();
			}
		}
	});

	LOG.info("Starting program invoke thread");
	invokeThread.start();

	// The job ID
	JobID jobId = null;

	LOG.info("Waiting for job status running.");

	// Wait for running job
	while (jobId == null && deadline.hasTimeLeft()) {

		Collection<JobStatusMessage> jobs = clusterClient.listJobs().get(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
		for (JobStatusMessage job : jobs) {
			if (job.getJobState() == JobStatus.RUNNING) {
				jobId = job.getJobId();
				LOG.info("Job running. ID: " + jobId);
				break;
			}
		}

		// Retry if job is not available yet
		if (jobId == null) {
			Thread.sleep(100L);
		}
	}

	// Trigger savepoint
	String savepointPath = null;
	for (int i = 0; i < 20; i++) {
		LOG.info("Triggering savepoint (" + (i + 1) + "/20).");
		try {
			savepointPath = clusterClient.triggerSavepoint(jobId, null)
				.get(deadline.timeLeft().toMillis(), TimeUnit.MILLISECONDS);
		} catch (Exception cause) {
			LOG.info("Failed to trigger savepoint. Retrying...", cause);
			// This can fail if the operators are not opened yet
			Thread.sleep(500);
		}
	}

	assertNotNull("Failed to trigger savepoint", savepointPath);

	clusterClient.disposeSavepoint(savepointPath).get();

	clusterClient.cancel(jobId).get();

	// make sure, the execution is finished to not influence other test methods
	invokeThread.join(deadline.timeLeft().toMillis());
	assertFalse("Program invoke thread still running", invokeThread.isAlive());
}