Examples with StandaloneSessionClusterEntrypoint org.apache.flink.runtime.entrypoint.StandaloneSessionClusterEntrypoint used on opensource projects

Example 1 with StandaloneSessionClusterEntrypoint

use of org.apache.flink.runtime.entrypoint.StandaloneSessionClusterEntrypoint in project flink by apache.

the class AbstractTaskManagerProcessFailureRecoveryTest method testTaskManagerProcessFailure.

@Test
public void testTaskManagerProcessFailure() throws Exception {
    TestProcess taskManagerProcess1 = null;
    TestProcess taskManagerProcess2 = null;
    TestProcess taskManagerProcess3 = null;
    File coordinateTempDir = null;
    Configuration config = new Configuration();
    config.setString(JobManagerOptions.ADDRESS, "localhost");
    config.setString(RestOptions.BIND_PORT, "0");
    config.setLong(HeartbeatManagerOptions.HEARTBEAT_INTERVAL, 200L);
    config.setLong(HeartbeatManagerOptions.HEARTBEAT_TIMEOUT, 10000L);
    config.set(HeartbeatManagerOptions.HEARTBEAT_RPC_FAILURE_THRESHOLD, 1);
    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(NettyShuffleEnvironmentOptions.NETWORK_SORT_SHUFFLE_MIN_BUFFERS, 16);
    config.set(TaskManagerOptions.TASK_HEAP_MEMORY, MemorySize.parse("128m"));
    config.set(TaskManagerOptions.CPU_CORES, 1.0);
    config.setString(JobManagerOptions.EXECUTION_FAILOVER_STRATEGY, "full");
    config.set(JobManagerOptions.RESOURCE_WAIT_TIMEOUT, Duration.ofSeconds(30L));
    try (final StandaloneSessionClusterEntrypoint clusterEntrypoint = new StandaloneSessionClusterEntrypoint(config)) {
        // check that we run this test only if the java command
        // is available on this machine
        String javaCommand = getJavaCommandPath();
        if (javaCommand == null) {
            System.out.println("---- Skipping Process Failure test : Could not find java executable ----");
            return;
        }
        clusterEntrypoint.startCluster();
        // coordination between the processes goes through a directory
        coordinateTempDir = temporaryFolder.newFolder();
        TestProcessBuilder taskManagerProcessBuilder = new TestProcessBuilder(TaskExecutorProcessEntryPoint.class.getName());
        taskManagerProcessBuilder.addConfigAsMainClassArgs(config);
        // start the first two TaskManager processes
        taskManagerProcess1 = taskManagerProcessBuilder.start();
        taskManagerProcess2 = taskManagerProcessBuilder.start();
        // the program will set a marker file in each of its parallel tasks once they are ready,
        // so that
        // this coordinating code is aware of this.
        // the program will very slowly consume elements until the marker file (later created by
        // the
        // test driver code) is present
        final File coordinateDirClosure = coordinateTempDir;
        final AtomicReference<Throwable> errorRef = new AtomicReference<>();
        // we trigger program execution in a separate thread
        Thread programTrigger = new Thread("Program Trigger") {

            @Override
            public void run() {
                try {
                    testTaskManagerFailure(config, coordinateDirClosure);
                } catch (Throwable t) {
                    t.printStackTrace();
                    errorRef.set(t);
                }
            }
        };
        // start the test program
        programTrigger.start();
        // max 20 seconds
        if (!waitForMarkerFiles(coordinateTempDir, READY_MARKER_FILE_PREFIX, PARALLELISM, 120000)) {
            // check if the program failed for some reason
            if (errorRef.get() != null) {
                Throwable error = errorRef.get();
                error.printStackTrace();
                fail("The program encountered a " + error.getClass().getSimpleName() + " : " + error.getMessage());
            } else {
                // no error occurred, simply a timeout
                fail("The tasks were not started within time (" + 120000 + "msecs)");
            }
        }
        // start the third TaskManager
        taskManagerProcess3 = taskManagerProcessBuilder.start();
        // kill one of the previous TaskManagers, triggering a failure and recovery
        taskManagerProcess1.destroy();
        waitForShutdown("TaskManager 1", taskManagerProcess1);
        // we create the marker file which signals the program functions tasks that they can
        // complete
        touchFile(new File(coordinateTempDir, PROCEED_MARKER_FILE));
        // wait for at most 5 minutes for the program to complete
        programTrigger.join(300000);
        // check that the program really finished
        assertFalse("The program did not finish in time", programTrigger.isAlive());
        // check whether the program encountered an error
        if (errorRef.get() != null) {
            Throwable error = errorRef.get();
            error.printStackTrace();
            fail("The program encountered a " + error.getClass().getSimpleName() + " : " + error.getMessage());
        }
    // all seems well :-)
    } catch (Exception e) {
        e.printStackTrace();
        printProcessLog("TaskManager 1", taskManagerProcess1);
        printProcessLog("TaskManager 2", taskManagerProcess2);
        printProcessLog("TaskManager 3", taskManagerProcess3);
        fail(e.getMessage());
    } catch (Error e) {
        e.printStackTrace();
        printProcessLog("TaskManager 1", taskManagerProcess1);
        printProcessLog("TaskManager 2", taskManagerProcess2);
        printProcessLog("TaskManager 3", taskManagerProcess3);
        throw e;
    } finally {
        if (taskManagerProcess1 != null) {
            taskManagerProcess1.destroy();
        }
        if (taskManagerProcess2 != null) {
            taskManagerProcess2.destroy();
        }
        if (taskManagerProcess3 != null) {
            taskManagerProcess3.destroy();
        }
        waitForShutdown("TaskManager 1", taskManagerProcess1);
        waitForShutdown("TaskManager 2", taskManagerProcess2);
        waitForShutdown("TaskManager 3", taskManagerProcess3);
    }
}

Example 2 with StandaloneSessionClusterEntrypoint

use of org.apache.flink.runtime.entrypoint.StandaloneSessionClusterEntrypoint in project flink by apache.

the class LocalRecoveryITCase method submitJob.

private JobClient submitJob(int parallelism, StandaloneSessionClusterEntrypoint clusterEntrypoint) throws Exception {
    final StreamExecutionEnvironment env = StreamExecutionEnvironment.createRemoteEnvironment("localhost", clusterEntrypoint.getRestPort(), new Configuration());
    env.setParallelism(parallelism);
    env.enableCheckpointing(100, CheckpointingMode.EXACTLY_ONCE);
    env.addSource(new LocalRecoverySource()).keyBy(x -> x).addSink(new DiscardingSink<>());
    final JobClient jobClient = env.executeAsync();
    return jobClient;
}

Example 3 with StandaloneSessionClusterEntrypoint

use of org.apache.flink.runtime.entrypoint.StandaloneSessionClusterEntrypoint in project flink by apache.

the class LocalRecoveryITCase method testRecoverLocallyFromProcessCrashWithWorkingDirectory.

@Test
public void testRecoverLocallyFromProcessCrashWithWorkingDirectory() throws Exception {
    final Configuration configuration = new Configuration();
    configuration.set(JobManagerOptions.ADDRESS, "localhost");
    configuration.set(JobManagerOptions.PORT, 0);
    configuration.set(RestOptions.BIND_PORT, "0");
    configuration.set(HeartbeatManagerOptions.HEARTBEAT_TIMEOUT, 10000L);
    configuration.set(HeartbeatManagerOptions.HEARTBEAT_INTERVAL, 1000L);
    configuration.set(HeartbeatManagerOptions.HEARTBEAT_RPC_FAILURE_THRESHOLD, 1);
    configuration.set(ClusterOptions.PROCESS_WORKING_DIR_BASE, tmpDirectory.getAbsolutePath());
    configuration.set(CheckpointingOptions.LOCAL_RECOVERY, true);
    configuration.set(TaskManagerOptions.SLOT_TIMEOUT, Duration.ofSeconds(30L));
    final int parallelism = 3;
    boolean success = false;
    Collection<TaskManagerProcess> taskManagerProcesses = Collections.emptyList();
    try (final StandaloneSessionClusterEntrypoint clusterEntrypoint = new StandaloneSessionClusterEntrypoint(configuration)) {
        clusterEntrypoint.startCluster();
        final Configuration configurationTemplate = new Configuration(configuration);
        configurationTemplate.set(JobManagerOptions.PORT, clusterEntrypoint.getRpcPort());
        taskManagerProcesses = startTaskManagerProcesses(parallelism, configurationTemplate);
        final JobClient jobClient = submitJob(parallelism, clusterEntrypoint);
        final long waitingTimeInSeconds = 45L;
        waitUntilCheckpointCompleted(configuration, clusterEntrypoint.getRestPort(), jobClient.getJobID(), Deadline.fromNow(Duration.ofSeconds(waitingTimeInSeconds)));
        restartTaskManagerProcesses(taskManagerProcesses, parallelism - 1);
        jobClient.getJobExecutionResult().get(waitingTimeInSeconds, TimeUnit.SECONDS);
        success = true;
    } finally {
        if (!success) {
            for (TaskManagerProcess taskManagerProcess : taskManagerProcesses) {
                printLogOutput(taskManagerProcess);
            }
        }
        for (TaskManagerProcess taskManagerProcess : taskManagerProcesses) {
            taskManagerProcess.terminate();
        }
    }
}