Examples with TaskSchedulerDescriptor org.apache.tez.serviceplugins.api.TaskSchedulerDescriptor used on opensource projects

Example 1 with TaskSchedulerDescriptor

use of org.apache.tez.serviceplugins.api.TaskSchedulerDescriptor in project tez by apache.

the class TestExtServicesWithLocalMode method test1.

@Test(timeout = 30000)
public void test1() throws Exception {
    UserPayload userPayload = TezUtils.createUserPayloadFromConf(confForJobs);
    TaskSchedulerDescriptor[] taskSchedulerDescriptors = new TaskSchedulerDescriptor[] { TaskSchedulerDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceTaskSchedulerService.class.getName()).setUserPayload(userPayload) };
    ContainerLauncherDescriptor[] containerLauncherDescriptors = new ContainerLauncherDescriptor[] { ContainerLauncherDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceNoOpContainerLauncher.class.getName()).setUserPayload(userPayload) };
    TaskCommunicatorDescriptor[] taskCommunicatorDescriptors = new TaskCommunicatorDescriptor[] { TaskCommunicatorDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceTaskCommunicatorImpl.class.getName()).setUserPayload(userPayload) };
    ServicePluginsDescriptor servicePluginsDescriptor = ServicePluginsDescriptor.create(true, false, taskSchedulerDescriptors, containerLauncherDescriptors, taskCommunicatorDescriptors);
    TezConfiguration tezConf = new TezConfiguration(confForJobs);
    TezClient tezClient = TezClient.newBuilder("test1", tezConf).setIsSession(true).setServicePluginDescriptor(servicePluginsDescriptor).build();
    try {
        tezClient.start();
        Path dataPath1 = new Path(SRC_DATA_DIR, "inPath1");
        Path dataPath2 = new Path(SRC_DATA_DIR, "inPath2");
        Path expectedResultPath = new Path(SRC_DATA_DIR, "expectedOutputPath");
        JoinDataGen dataGen = new JoinDataGen();
        String[] dataGenArgs = new String[] { dataPath1.toString(), "1048576", dataPath2.toString(), "524288", expectedResultPath.toString(), "2" };
        assertEquals(0, dataGen.run(tezConf, dataGenArgs, tezClient));
        Path outputPath = new Path(SRC_DATA_DIR, "outPath");
        HashJoinExample joinExample = new HashJoinExample();
        String[] args = new String[] { dataPath1.toString(), dataPath2.toString(), "2", outputPath.toString() };
        assertEquals(0, joinExample.run(tezConf, args, tezClient));
        LOG.info("Completed generating Data - Expected Hash Result and Actual Join Result");
        assertEquals(0, tezTestServiceCluster.getNumSubmissions());
        // ext can consume from ext.
        runJoinValidate(tezClient, "allInExt", 7, EXECUTION_CONTEXT_EXT_SERVICE_PUSH, EXECUTION_CONTEXT_EXT_SERVICE_PUSH, EXECUTION_CONTEXT_EXT_SERVICE_PUSH);
        LOG.info("Completed allInExt");
        // uber can consume from uber.
        runJoinValidate(tezClient, "noneInExt", 0, null, null, null);
        LOG.info("Completed noneInExt");
        // uber can consume from ext
        runJoinValidate(tezClient, "lhsInExt", 2, EXECUTION_CONTEXT_EXT_SERVICE_PUSH, null, null);
        LOG.info("Completed lhsInExt");
    // ext cannot consume from uber in this mode since there's no shuffle handler working,
    // and the local data transfer semantics may not match.
    } finally {
        tezClient.stop();
    }
}

Example 2 with TaskSchedulerDescriptor

use of org.apache.tez.serviceplugins.api.TaskSchedulerDescriptor in project tez by apache.

the class TestExternalTezServicesErrors method setup.

@BeforeClass
public static void setup() throws Exception {
    extServiceTestHelper = new ExternalTezServiceTestHelper(TEST_ROOT_DIR);
    UserPayload userPayload = TezUtils.createUserPayloadFromConf(extServiceTestHelper.getConfForJobs());
    UserPayload userPayloadThrowError = ErrorPluginConfiguration.toUserPayload(ErrorPluginConfiguration.createThrowErrorConf());
    UserPayload userPayloadReportFatalErrorLauncher = ErrorPluginConfiguration.toUserPayload(ErrorPluginConfiguration.createReportFatalErrorConf(SUFFIX_LAUNCHER));
    UserPayload userPayloadReportFatalErrorTaskComm = ErrorPluginConfiguration.toUserPayload(ErrorPluginConfiguration.createReportFatalErrorConf(SUFFIX_TASKCOMM));
    UserPayload userPayloadReportFatalErrorScheduler = ErrorPluginConfiguration.toUserPayload(ErrorPluginConfiguration.createReportFatalErrorConf(SUFFIX_SCHEDULER));
    UserPayload userPayloadReportNonFatalErrorLauncher = ErrorPluginConfiguration.toUserPayload(ErrorPluginConfiguration.createReportNonFatalErrorConf(SUFFIX_LAUNCHER));
    UserPayload userPayloadReportNonFatalErrorTaskComm = ErrorPluginConfiguration.toUserPayload(ErrorPluginConfiguration.createReportNonFatalErrorConf(SUFFIX_TASKCOMM));
    UserPayload userPayloadReportNonFatalErrorScheduler = ErrorPluginConfiguration.toUserPayload(ErrorPluginConfiguration.createReportNonFatalErrorConf(SUFFIX_SCHEDULER));
    TaskSchedulerDescriptor[] taskSchedulerDescriptors = new TaskSchedulerDescriptor[] { TaskSchedulerDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceTaskSchedulerService.class.getName()).setUserPayload(userPayload), TaskSchedulerDescriptor.create(EXT_THROW_ERROR_ENTITY_NAME, TezTestServiceTaskSchedulerServiceWithErrors.class.getName()).setUserPayload(userPayloadThrowError), TaskSchedulerDescriptor.create(EXT_REPORT_FATAL_ERROR_ENTITY_NAME, TezTestServiceTaskSchedulerServiceWithErrors.class.getName()).setUserPayload(userPayloadReportFatalErrorScheduler), TaskSchedulerDescriptor.create(EXT_REPORT_NON_FATAL_ERROR_ENTITY_NAME, TezTestServiceTaskSchedulerServiceWithErrors.class.getName()).setUserPayload(userPayloadReportNonFatalErrorScheduler) };
    ContainerLauncherDescriptor[] containerLauncherDescriptors = new ContainerLauncherDescriptor[] { ContainerLauncherDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceNoOpContainerLauncher.class.getName()).setUserPayload(userPayload), ContainerLauncherDescriptor.create(EXT_THROW_ERROR_ENTITY_NAME, TezTestServiceContainerLauncherWithErrors.class.getName()).setUserPayload(userPayloadThrowError), ContainerLauncherDescriptor.create(EXT_REPORT_FATAL_ERROR_ENTITY_NAME, TezTestServiceContainerLauncherWithErrors.class.getName()).setUserPayload(userPayloadReportFatalErrorLauncher), ContainerLauncherDescriptor.create(EXT_REPORT_NON_FATAL_ERROR_ENTITY_NAME, TezTestServiceContainerLauncherWithErrors.class.getName()).setUserPayload(userPayloadReportNonFatalErrorLauncher) };
    TaskCommunicatorDescriptor[] taskCommunicatorDescriptors = new TaskCommunicatorDescriptor[] { TaskCommunicatorDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceTaskCommunicatorImpl.class.getName()).setUserPayload(userPayload), TaskCommunicatorDescriptor.create(EXT_THROW_ERROR_ENTITY_NAME, TezTestServiceTaskCommunicatorWithErrors.class.getName()).setUserPayload(userPayloadThrowError), TaskCommunicatorDescriptor.create(EXT_REPORT_FATAL_ERROR_ENTITY_NAME, TezTestServiceTaskCommunicatorWithErrors.class.getName()).setUserPayload(userPayloadReportFatalErrorTaskComm), TaskCommunicatorDescriptor.create(EXT_REPORT_NON_FATAL_ERROR_ENTITY_NAME, TezTestServiceTaskCommunicatorWithErrors.class.getName()).setUserPayload(userPayloadReportNonFatalErrorTaskComm) };
    servicePluginsDescriptor = ServicePluginsDescriptor.create(true, true, taskSchedulerDescriptors, containerLauncherDescriptors, taskCommunicatorDescriptors);
    extServiceTestHelper.setupSharedTezClient(servicePluginsDescriptor);
    // Generate the join data set used for each run.
    // Can a timeout be enforced here ?
    Path dataPath1 = new Path(SRC_DATA_DIR, "inPath1");
    Path dataPath2 = new Path(SRC_DATA_DIR, "inPath2");
    extServiceTestHelper.setupHashJoinData(SRC_DATA_DIR, dataPath1, dataPath2, HASH_JOIN_EXPECTED_RESULT_PATH, HASH_JOIN_OUTPUT_PATH);
    extServiceTestHelper.shutdownSharedTezClient();
}

Example 3 with TaskSchedulerDescriptor

use of org.apache.tez.serviceplugins.api.TaskSchedulerDescriptor in project tez by apache.

the class TestExternalTezServices method setup.

@BeforeClass
public static void setup() throws Exception {
    extServiceTestHelper = new ExternalTezServiceTestHelper(TEST_ROOT_DIR);
    UserPayload userPayload = TezUtils.createUserPayloadFromConf(extServiceTestHelper.getConfForJobs());
    TaskSchedulerDescriptor[] taskSchedulerDescriptors = new TaskSchedulerDescriptor[] { TaskSchedulerDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceTaskSchedulerService.class.getName()).setUserPayload(userPayload) };
    ContainerLauncherDescriptor[] containerLauncherDescriptors = new ContainerLauncherDescriptor[] { ContainerLauncherDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceNoOpContainerLauncher.class.getName()).setUserPayload(userPayload) };
    TaskCommunicatorDescriptor[] taskCommunicatorDescriptors = new TaskCommunicatorDescriptor[] { TaskCommunicatorDescriptor.create(EXT_PUSH_ENTITY_NAME, TezTestServiceTaskCommunicatorImpl.class.getName()).setUserPayload(userPayload) };
    ServicePluginsDescriptor servicePluginsDescriptor = ServicePluginsDescriptor.create(true, true, taskSchedulerDescriptors, containerLauncherDescriptors, taskCommunicatorDescriptors);
    extServiceTestHelper.setupSharedTezClient(servicePluginsDescriptor);
    // Generate the join data set used for each run.
    // Can a timeout be enforced here ?
    Path dataPath1 = new Path(SRC_DATA_DIR, "inPath1");
    Path dataPath2 = new Path(SRC_DATA_DIR, "inPath2");
    extServiceTestHelper.setupHashJoinData(SRC_DATA_DIR, dataPath1, dataPath2, HASH_JOIN_EXPECTED_RESULT_PATH, HASH_JOIN_OUTPUT_PATH);
}

Example 4 with TaskSchedulerDescriptor

use of org.apache.tez.serviceplugins.api.TaskSchedulerDescriptor in project hive by apache.

the class TezSessionState method openInternal.

protected void openInternal(final HiveConf conf, Collection<String> additionalFiles, boolean isAsync, LogHelper console, Path scratchDir) throws IOException, LoginException, IllegalArgumentException, URISyntaxException, TezException {
    this.conf = conf;
    // TODO Why is the queue name set again. It has already been setup via setQueueName. Do only one of the two.
    String confQueueName = conf.get(TezConfiguration.TEZ_QUEUE_NAME);
    if (queueName != null && !queueName.equals(confQueueName)) {
        LOG.warn("Resetting a queue name that was already set: was " + queueName + ", now " + confQueueName);
    }
    this.queueName = confQueueName;
    this.doAsEnabled = conf.getBoolVar(HiveConf.ConfVars.HIVE_SERVER2_ENABLE_DOAS);
    final boolean llapMode = "llap".equalsIgnoreCase(HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_EXECUTION_MODE));
    // TODO This - at least for the session pool - will always be the hive user. How does doAs above this affect things ?
    UserGroupInformation ugi = Utils.getUGI();
    user = ugi.getShortUserName();
    LOG.info("User of session id " + sessionId + " is " + user);
    // create the tez tmp dir
    tezScratchDir = scratchDir == null ? createTezDir(sessionId) : scratchDir;
    additionalFilesNotFromConf.clear();
    if (additionalFiles != null) {
        additionalFilesNotFromConf.addAll(additionalFiles);
    }
    refreshLocalResourcesFromConf(conf);
    // unless already installed on all the cluster nodes, we'll have to
    // localize hive-exec.jar as well.
    appJarLr = createJarLocalResource(utils.getExecJarPathLocal());
    // configuration for the application master
    final Map<String, LocalResource> commonLocalResources = new HashMap<String, LocalResource>();
    commonLocalResources.put(utils.getBaseName(appJarLr), appJarLr);
    for (LocalResource lr : localizedResources) {
        commonLocalResources.put(utils.getBaseName(lr), lr);
    }
    if (llapMode) {
        // localize llap client jars
        addJarLRByClass(LlapTaskSchedulerService.class, commonLocalResources);
        addJarLRByClass(LlapProtocolClientImpl.class, commonLocalResources);
        addJarLRByClass(LlapProtocolClientProxy.class, commonLocalResources);
        addJarLRByClassName("org.apache.hadoop.registry.client.api.RegistryOperations", commonLocalResources);
    }
    // Create environment for AM.
    Map<String, String> amEnv = new HashMap<String, String>();
    MRHelpers.updateEnvBasedOnMRAMEnv(conf, amEnv);
    // and finally we're ready to create and start the session
    // generate basic tez config
    final TezConfiguration tezConfig = new TezConfiguration(conf);
    // set up the staging directory to use
    tezConfig.set(TezConfiguration.TEZ_AM_STAGING_DIR, tezScratchDir.toUri().toString());
    conf.stripHiddenConfigurations(tezConfig);
    ServicePluginsDescriptor servicePluginsDescriptor;
    Credentials llapCredentials = null;
    if (llapMode) {
        if (UserGroupInformation.isSecurityEnabled()) {
            llapCredentials = new Credentials();
            llapCredentials.addToken(LlapTokenIdentifier.KIND_NAME, getLlapToken(user, tezConfig));
        }
        // TODO Change this to not serialize the entire Configuration - minor.
        UserPayload servicePluginPayload = TezUtils.createUserPayloadFromConf(tezConfig);
        // we need plugins to handle llap and uber mode
        servicePluginsDescriptor = ServicePluginsDescriptor.create(true, new TaskSchedulerDescriptor[] { TaskSchedulerDescriptor.create(LLAP_SERVICE, LLAP_SCHEDULER).setUserPayload(servicePluginPayload) }, new ContainerLauncherDescriptor[] { ContainerLauncherDescriptor.create(LLAP_SERVICE, LLAP_LAUNCHER) }, new TaskCommunicatorDescriptor[] { TaskCommunicatorDescriptor.create(LLAP_SERVICE, LLAP_TASK_COMMUNICATOR).setUserPayload(servicePluginPayload) });
    } else {
        servicePluginsDescriptor = ServicePluginsDescriptor.create(true);
    }
    // container prewarming. tell the am how many containers we need
    if (HiveConf.getBoolVar(conf, ConfVars.HIVE_PREWARM_ENABLED)) {
        int n = HiveConf.getIntVar(conf, ConfVars.HIVE_PREWARM_NUM_CONTAINERS);
        n = Math.max(tezConfig.getInt(TezConfiguration.TEZ_AM_SESSION_MIN_HELD_CONTAINERS, TezConfiguration.TEZ_AM_SESSION_MIN_HELD_CONTAINERS_DEFAULT), n);
        tezConfig.setInt(TezConfiguration.TEZ_AM_SESSION_MIN_HELD_CONTAINERS, n);
    }
    setupSessionAcls(tezConfig, conf);
    final TezClient session = TezClient.newBuilder("HIVE-" + sessionId, tezConfig).setIsSession(true).setLocalResources(commonLocalResources).setCredentials(llapCredentials).setServicePluginDescriptor(servicePluginsDescriptor).build();
    LOG.info("Opening new Tez Session (id: " + sessionId + ", scratch dir: " + tezScratchDir + ")");
    TezJobMonitor.initShutdownHook();
    if (!isAsync) {
        startSessionAndContainers(session, conf, commonLocalResources, tezConfig, false);
        this.session = session;
    } else {
        FutureTask<TezClient> sessionFuture = new FutureTask<>(new Callable<TezClient>() {

            @Override
            public TezClient call() throws Exception {
                try {
                    return startSessionAndContainers(session, conf, commonLocalResources, tezConfig, true);
                } catch (Throwable t) {
                    LOG.error("Failed to start Tez session", t);
                    throw (t instanceof Exception) ? (Exception) t : new Exception(t);
                }
            }
        });
        new Thread(sessionFuture, "Tez session start thread").start();
        // We assume here nobody will try to get session before open() returns.
        this.console = console;
        this.sessionFuture = sessionFuture;
    }
}

Example 5 with TaskSchedulerDescriptor

use of org.apache.tez.serviceplugins.api.TaskSchedulerDescriptor in project tez by apache.

the class TestTaskSchedulerManager method testReportFailureFromTaskScheduler.

@SuppressWarnings("unchecked")
@Test(timeout = 5000)
public void testReportFailureFromTaskScheduler() {
    String dagName = DAG_NAME;
    Configuration conf = new TezConfiguration();
    String taskSchedulerName = "testTaskScheduler";
    String expIdentifier = "[0:" + taskSchedulerName + "]";
    EventHandler eventHandler = mock(EventHandler.class);
    AppContext appContext = mock(AppContext.class, RETURNS_DEEP_STUBS);
    doReturn(taskSchedulerName).when(appContext).getTaskSchedulerName(0);
    doReturn(eventHandler).when(appContext).getEventHandler();
    doReturn(conf).when(appContext).getAMConf();
    InetSocketAddress address = new InetSocketAddress("host", 55000);
    DAGClientServer dagClientServer = mock(DAGClientServer.class);
    doReturn(address).when(dagClientServer).getBindAddress();
    DAG dag = mock(DAG.class);
    TezDAGID dagId = TezDAGID.getInstance(ApplicationId.newInstance(1, 0), DAG_INDEX);
    doReturn(dagName).when(dag).getName();
    doReturn(dagId).when(dag).getID();
    doReturn(dag).when(appContext).getCurrentDAG();
    NamedEntityDescriptor<TaskSchedulerDescriptor> namedEntityDescriptor = new NamedEntityDescriptor<>(taskSchedulerName, TaskSchedulerForFailureTest.class.getName());
    List<NamedEntityDescriptor> list = new LinkedList<>();
    list.add(namedEntityDescriptor);
    TaskSchedulerManager taskSchedulerManager = new TaskSchedulerManager(appContext, dagClientServer, eventHandler, mock(ContainerSignatureMatcher.class), mock(WebUIService.class), list, false, new HadoopShimsLoader(appContext.getAMConf()).getHadoopShim()) {

        @Override
        TaskSchedulerContext wrapTaskSchedulerContext(TaskSchedulerContext rawContext) {
            // Avoid wrapping in threads
            return rawContext;
        }
    };
    try {
        taskSchedulerManager.init(new TezConfiguration());
        taskSchedulerManager.start();
        taskSchedulerManager.getTotalResources(0);
        ArgumentCaptor<Event> argumentCaptor = ArgumentCaptor.forClass(Event.class);
        verify(eventHandler, times(1)).handle(argumentCaptor.capture());
        Event rawEvent = argumentCaptor.getValue();
        assertTrue(rawEvent instanceof DAGEventTerminateDag);
        DAGEventTerminateDag killEvent = (DAGEventTerminateDag) rawEvent;
        assertTrue(killEvent.getDiagnosticInfo().contains("ReportError"));
        assertTrue(killEvent.getDiagnosticInfo().contains(ServicePluginErrorDefaults.SERVICE_UNAVAILABLE.name()));
        assertTrue(killEvent.getDiagnosticInfo().contains(expIdentifier));
        reset(eventHandler);
        taskSchedulerManager.getAvailableResources(0);
        argumentCaptor = ArgumentCaptor.forClass(Event.class);
        verify(eventHandler, times(1)).handle(argumentCaptor.capture());
        rawEvent = argumentCaptor.getValue();
        assertTrue(rawEvent instanceof DAGAppMasterEventUserServiceFatalError);
        DAGAppMasterEventUserServiceFatalError event = (DAGAppMasterEventUserServiceFatalError) rawEvent;
        assertEquals(DAGAppMasterEventType.TASK_SCHEDULER_SERVICE_FATAL_ERROR, event.getType());
        assertTrue(event.getDiagnosticInfo().contains("ReportedFatalError"));
        assertTrue(event.getDiagnosticInfo().contains(ServicePluginErrorDefaults.INCONSISTENT_STATE.name()));
        assertTrue(event.getDiagnosticInfo().contains(expIdentifier));
    } finally {
        taskSchedulerManager.stop();
    }
}