Search in sources :

Example 1 with DAGClientServer

use of org.apache.tez.dag.api.client.DAGClientServer in project tez by apache.

the class DAGAppMaster method serviceInit.

@Override
public synchronized void serviceInit(final Configuration conf) throws Exception {
    this.amConf = conf;
    initResourceCalculatorPlugins();
    this.hadoopShim = new HadoopShimsLoader(this.amConf).getHadoopShim();
    long sleepTimeBeforeSecs = this.amConf.getLong(TezConfiguration.TEZ_AM_SLEEP_TIME_BEFORE_EXIT_MILLIS, TezConstants.TEZ_DAG_SLEEP_TIME_BEFORE_EXIT);
    if (sleepTimeBeforeSecs >= 0) {
        this.shutdownHandler.setSleepTimeBeforeExit(sleepTimeBeforeSecs);
    }
    this.isLocal = conf.getBoolean(TezConfiguration.TEZ_LOCAL_MODE, TezConfiguration.TEZ_LOCAL_MODE_DEFAULT);
    UserPayload defaultPayload = TezUtils.createUserPayloadFromConf(amConf);
    List<NamedEntityDescriptor> taskSchedulerDescriptors = Lists.newLinkedList();
    List<NamedEntityDescriptor> containerLauncherDescriptors = Lists.newLinkedList();
    List<NamedEntityDescriptor> taskCommunicatorDescriptors = Lists.newLinkedList();
    parseAllPlugins(taskSchedulerDescriptors, taskSchedulers, containerLauncherDescriptors, containerLaunchers, taskCommunicatorDescriptors, taskCommunicators, amPluginDescriptorProto, isLocal, defaultPayload);
    LOG.info(buildPluginComponentLog(taskSchedulerDescriptors, taskSchedulers, "TaskSchedulers"));
    LOG.info(buildPluginComponentLog(containerLauncherDescriptors, containerLaunchers, "ContainerLaunchers"));
    LOG.info(buildPluginComponentLog(taskCommunicatorDescriptors, taskCommunicators, "TaskCommunicators"));
    boolean disableVersionCheck = conf.getBoolean(TezConfiguration.TEZ_AM_DISABLE_CLIENT_VERSION_CHECK, TezConfiguration.TEZ_AM_DISABLE_CLIENT_VERSION_CHECK_DEFAULT);
    // Check client - AM version compatibility
    LOG.info("Comparing client version with AM version" + ", clientVersion=" + clientVersion + ", AMVersion=" + dagVersionInfo.getVersion());
    Simple2LevelVersionComparator versionComparator = new Simple2LevelVersionComparator();
    if (versionComparator.compare(clientVersion, dagVersionInfo.getVersion()) != 0) {
        versionMismatchDiagnostics = "Incompatible versions found" + ", clientVersion=" + clientVersion + ", AMVersion=" + dagVersionInfo.getVersion();
        addDiagnostic(versionMismatchDiagnostics);
        if (disableVersionCheck) {
            LOG.warn("Ignoring client-AM version mismatch as check disabled. " + versionMismatchDiagnostics);
        } else {
            LOG.error(versionMismatchDiagnostics);
            versionMismatch = true;
        }
    }
    dispatcher = createDispatcher();
    if (isLocal) {
        conf.setBoolean(TezConfiguration.TEZ_AM_NODE_BLACKLISTING_ENABLED, false);
        conf.set(TezConfiguration.TEZ_HISTORY_LOGGING_SERVICE_CLASS, TezConfiguration.TEZ_HISTORY_LOGGING_SERVICE_CLASS_DEFAULT);
    } else {
        dispatcher.enableExitOnDispatchException();
    }
    String strAppId = this.appAttemptID.getApplicationId().toString();
    this.tezSystemStagingDir = TezCommonUtils.getTezSystemStagingPath(conf, strAppId);
    context = new RunningAppContext(conf);
    this.aclManager = new ACLManager(appMasterUgi.getShortUserName(), this.amConf);
    clientHandler = new DAGClientHandler(this);
    addIfService(dispatcher, false);
    recoveryDataDir = TezCommonUtils.getRecoveryPath(tezSystemStagingDir, conf);
    recoveryFS = recoveryDataDir.getFileSystem(conf);
    currentRecoveryDataDir = TezCommonUtils.getAttemptRecoveryPath(recoveryDataDir, appAttemptID.getAttemptId());
    if (LOG.isDebugEnabled()) {
        LOG.debug("Stage directory information for AppAttemptId :" + this.appAttemptID + " tezSystemStagingDir :" + tezSystemStagingDir + " recoveryDataDir :" + recoveryDataDir + " recoveryAttemptDir :" + currentRecoveryDataDir);
    }
    recoveryEnabled = conf.getBoolean(TezConfiguration.DAG_RECOVERY_ENABLED, TezConfiguration.DAG_RECOVERY_ENABLED_DEFAULT);
    clientRpcServer = new DAGClientServer(clientHandler, appAttemptID, recoveryFS);
    addIfService(clientRpcServer, true);
    taskHeartbeatHandler = createTaskHeartbeatHandler(context, conf);
    addIfService(taskHeartbeatHandler, true);
    containerHeartbeatHandler = createContainerHeartbeatHandler(context, conf);
    addIfService(containerHeartbeatHandler, true);
    sessionToken = TokenCache.getSessionToken(amCredentials);
    if (sessionToken == null) {
        throw new RuntimeException("Could not find session token in AM Credentials");
    }
    // Prepare the TaskAttemptListener server for authentication of Containers
    // TaskAttemptListener gets the information via jobTokenSecretManager.
    jobTokenSecretManager.addTokenForJob(appAttemptID.getApplicationId().toString(), sessionToken);
    // service to handle requests to TaskUmbilicalProtocol
    taskCommunicatorManager = createTaskCommunicatorManager(context, taskHeartbeatHandler, containerHeartbeatHandler, taskCommunicatorDescriptors);
    addIfService(taskCommunicatorManager, true);
    containerSignatureMatcher = createContainerSignatureMatcher();
    containers = new AMContainerMap(containerHeartbeatHandler, taskCommunicatorManager, containerSignatureMatcher, context);
    addIfService(containers, true);
    dispatcher.register(AMContainerEventType.class, containers);
    nodes = new AMNodeTracker(dispatcher.getEventHandler(), context);
    addIfService(nodes, true);
    dispatcher.register(AMNodeEventType.class, nodes);
    this.dagEventDispatcher = new DagEventDispatcher();
    this.vertexEventDispatcher = new VertexEventDispatcher();
    // register the event dispatchers
    dispatcher.register(DAGAppMasterEventType.class, new DAGAppMasterEventHandler());
    dispatcher.register(DAGEventType.class, dagEventDispatcher);
    dispatcher.register(VertexEventType.class, vertexEventDispatcher);
    boolean useConcurrentDispatcher = conf.getBoolean(TezConfiguration.TEZ_AM_USE_CONCURRENT_DISPATCHER, TezConfiguration.TEZ_AM_USE_CONCURRENT_DISPATCHER_DEFAULT);
    LOG.info("Using concurrent dispatcher: " + useConcurrentDispatcher);
    if (!useConcurrentDispatcher) {
        dispatcher.register(TaskEventType.class, new TaskEventDispatcher());
        dispatcher.register(TaskAttemptEventType.class, new TaskAttemptEventDispatcher());
    } else {
        int concurrency = conf.getInt(TezConfiguration.TEZ_AM_CONCURRENT_DISPATCHER_CONCURRENCY, TezConfiguration.TEZ_AM_CONCURRENT_DISPATCHER_CONCURRENCY_DEFAULT);
        AsyncDispatcherConcurrent sharedDispatcher = dispatcher.registerAndCreateDispatcher(TaskEventType.class, new TaskEventDispatcher(), "TaskAndAttemptEventThread", concurrency);
        dispatcher.registerWithExistingDispatcher(TaskAttemptEventType.class, new TaskAttemptEventDispatcher(), sharedDispatcher);
    }
    // register other delegating dispatchers
    dispatcher.registerAndCreateDispatcher(SpeculatorEventType.class, new SpeculatorEventHandler(), "Speculator");
    if (enableWebUIService()) {
        this.webUIService = new WebUIService(context);
        addIfService(webUIService, false);
    } else {
        if (LOG.isDebugEnabled()) {
            LOG.debug("Web UI Service is not enabled.");
        }
    }
    this.taskSchedulerManager = createTaskSchedulerManager(taskSchedulerDescriptors);
    addIfService(taskSchedulerManager, true);
    if (enableWebUIService()) {
        addIfServiceDependency(taskSchedulerManager, webUIService);
    }
    dispatcher.register(AMSchedulerEventType.class, taskSchedulerManager);
    addIfServiceDependency(taskSchedulerManager, clientRpcServer);
    this.containerLauncherManager = createContainerLauncherManager(containerLauncherDescriptors, isLocal);
    addIfService(containerLauncherManager, true);
    dispatcher.register(ContainerLauncherEventType.class, containerLauncherManager);
    historyEventHandler = createHistoryEventHandler(context);
    addIfService(historyEventHandler, true);
    this.sessionTimeoutInterval = TezCommonUtils.getDAGSessionTimeout(amConf);
    this.clientAMHeartbeatTimeoutIntervalMillis = TezCommonUtils.getAMClientHeartBeatTimeoutMillis(amConf);
    if (!versionMismatch) {
        if (isSession) {
            FileInputStream sessionResourcesStream = null;
            try {
                sessionResourcesStream = new FileInputStream(new File(workingDirectory, TezConstants.TEZ_AM_LOCAL_RESOURCES_PB_FILE_NAME));
                PlanLocalResourcesProto amLocalResourceProto = PlanLocalResourcesProto.parseDelimitedFrom(sessionResourcesStream);
                amResources.putAll(DagTypeConverters.convertFromPlanLocalResources(amLocalResourceProto));
            } finally {
                if (sessionResourcesStream != null) {
                    sessionResourcesStream.close();
                }
            }
        }
    }
    rawExecutor = Executors.newCachedThreadPool(new ThreadFactoryBuilder().setDaemon(true).setNameFormat("App Shared Pool - " + "#%d").build());
    execService = MoreExecutors.listeningDecorator(rawExecutor);
    initServices(conf);
    super.serviceInit(conf);
    if (!versionMismatch) {
        if (this.appAttemptID.getAttemptId() == 1) {
            AppLaunchedEvent appLaunchedEvent = new AppLaunchedEvent(appAttemptID.getApplicationId(), startTime, appSubmitTime, appMasterUgi.getShortUserName(), this.amConf, dagVersionInfo);
            historyEventHandler.handle(new DAGHistoryEvent(appLaunchedEvent));
        }
        AMLaunchedEvent launchedEvent = new AMLaunchedEvent(appAttemptID, startTime, appSubmitTime, appMasterUgi.getShortUserName());
        historyEventHandler.handle(new DAGHistoryEvent(launchedEvent));
        this.state = DAGAppMasterState.INITED;
    } else {
        this.state = DAGAppMasterState.ERROR;
    }
}
Also used : PlanLocalResourcesProto(org.apache.tez.dag.api.records.DAGProtos.PlanLocalResourcesProto) ACLManager(org.apache.tez.common.security.ACLManager) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) HadoopShimsLoader(org.apache.tez.hadoop.shim.HadoopShimsLoader) AsyncDispatcherConcurrent(org.apache.tez.common.AsyncDispatcherConcurrent) ThreadFactoryBuilder(com.google.common.util.concurrent.ThreadFactoryBuilder) DAGClientHandler(org.apache.tez.dag.api.client.DAGClientHandler) Simple2LevelVersionComparator(org.apache.tez.dag.utils.Simple2LevelVersionComparator) UserPayload(org.apache.tez.dag.api.UserPayload) DAGHistoryEvent(org.apache.tez.dag.history.DAGHistoryEvent) NamedEntityDescriptor(org.apache.tez.dag.api.NamedEntityDescriptor) FileInputStream(java.io.FileInputStream) AMNodeTracker(org.apache.tez.dag.app.rm.node.AMNodeTracker) WebUIService(org.apache.tez.dag.app.web.WebUIService) AppLaunchedEvent(org.apache.tez.dag.history.events.AppLaunchedEvent) AMLaunchedEvent(org.apache.tez.dag.history.events.AMLaunchedEvent) DAGClientServer(org.apache.tez.dag.api.client.DAGClientServer) File(java.io.File)

Example 2 with DAGClientServer

use of org.apache.tez.dag.api.client.DAGClientServer in project tez by apache.

the class TestTaskSchedulerManager method testTaskSchedulerUserError.

@SuppressWarnings("unchecked")
@Test(timeout = 5000)
public void testTaskSchedulerUserError() {
    TaskScheduler taskScheduler = mock(TaskScheduler.class, new ExceptionAnswer());
    EventHandler eventHandler = mock(EventHandler.class);
    AppContext appContext = mock(AppContext.class, RETURNS_DEEP_STUBS);
    when(appContext.getEventHandler()).thenReturn(eventHandler);
    doReturn("testTaskScheduler").when(appContext).getTaskSchedulerName(0);
    String expectedId = "[0:testTaskScheduler]";
    Configuration conf = new Configuration(false);
    InetSocketAddress address = new InetSocketAddress(15222);
    DAGClientServer mockClientService = mock(DAGClientServer.class);
    doReturn(address).when(mockClientService).getBindAddress();
    TaskSchedulerManager taskSchedulerManager = new TaskSchedulerManager(taskScheduler, appContext, mock(ContainerSignatureMatcher.class), mockClientService, Executors.newFixedThreadPool(1)) {

        @Override
        protected void instantiateSchedulers(String host, int port, String trackingUrl, AppContext appContext) throws TezException {
        // Stubbed out since these are setup up front in the constructor used for testing
        }
    };
    try {
        taskSchedulerManager.init(conf);
        taskSchedulerManager.start();
        // Invoking a couple of random methods
        AMSchedulerEventTALaunchRequest launchRequest = new AMSchedulerEventTALaunchRequest(mock(TezTaskAttemptID.class), mock(Resource.class), mock(TaskSpec.class), mock(TaskAttempt.class), mock(TaskLocationHint.class), 0, mock(ContainerContext.class), 0, 0, 0);
        taskSchedulerManager.handleEvent(launchRequest);
        ArgumentCaptor<Event> argumentCaptor = ArgumentCaptor.forClass(Event.class);
        verify(eventHandler, times(1)).handle(argumentCaptor.capture());
        Event rawEvent = argumentCaptor.getValue();
        assertTrue(rawEvent instanceof DAGAppMasterEventUserServiceFatalError);
        DAGAppMasterEventUserServiceFatalError event = (DAGAppMasterEventUserServiceFatalError) rawEvent;
        assertEquals(DAGAppMasterEventType.TASK_SCHEDULER_SERVICE_FATAL_ERROR, event.getType());
        assertTrue(event.getError().getMessage().contains("TestException_" + "allocateTask"));
        assertTrue(event.getDiagnosticInfo().contains("Task Allocation"));
        assertTrue(event.getDiagnosticInfo().contains(expectedId));
        taskSchedulerManager.dagCompleted();
        argumentCaptor = ArgumentCaptor.forClass(Event.class);
        verify(eventHandler, times(2)).handle(argumentCaptor.capture());
        rawEvent = argumentCaptor.getAllValues().get(1);
        assertTrue(rawEvent instanceof DAGAppMasterEventUserServiceFatalError);
        event = (DAGAppMasterEventUserServiceFatalError) rawEvent;
        assertEquals(DAGAppMasterEventType.TASK_SCHEDULER_SERVICE_FATAL_ERROR, event.getType());
        assertTrue(event.getError().getMessage().contains("TestException_" + "dagComplete"));
        assertTrue(event.getDiagnosticInfo().contains("Dag Completion"));
        assertTrue(event.getDiagnosticInfo().contains(expectedId));
    } finally {
        taskSchedulerManager.stop();
    }
}
Also used : DAGAppMasterEventUserServiceFatalError(org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError) Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) InetSocketAddress(java.net.InetSocketAddress) AppContext(org.apache.tez.dag.app.AppContext) TaskSpec(org.apache.tez.runtime.api.impl.TaskSpec) ContainerSignatureMatcher(org.apache.tez.common.ContainerSignatureMatcher) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) EventHandler(org.apache.hadoop.yarn.event.EventHandler) TaskScheduler(org.apache.tez.serviceplugins.api.TaskScheduler) TaskLocationHint(org.apache.tez.dag.api.TaskLocationHint) ContainerContext(org.apache.tez.dag.app.ContainerContext) Event(org.apache.hadoop.yarn.event.Event) DAGClientServer(org.apache.tez.dag.api.client.DAGClientServer) TaskAttempt(org.apache.tez.dag.app.dag.TaskAttempt) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) DagInfoImplForTest(org.apache.tez.dag.helpers.DagInfoImplForTest) Test(org.junit.Test)

Example 3 with DAGClientServer

use of org.apache.tez.dag.api.client.DAGClientServer 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();
    }
}
Also used : DAGAppMasterEventUserServiceFatalError(org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError) Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) TaskSchedulerDescriptor(org.apache.tez.serviceplugins.api.TaskSchedulerDescriptor) TaskSchedulerContext(org.apache.tez.serviceplugins.api.TaskSchedulerContext) InetSocketAddress(java.net.InetSocketAddress) AppContext(org.apache.tez.dag.app.AppContext) ContainerSignatureMatcher(org.apache.tez.common.ContainerSignatureMatcher) EventHandler(org.apache.hadoop.yarn.event.EventHandler) DAG(org.apache.tez.dag.app.dag.DAG) DAGEventTerminateDag(org.apache.tez.dag.app.dag.event.DAGEventTerminateDag) NamedEntityDescriptor(org.apache.tez.dag.api.NamedEntityDescriptor) LinkedList(java.util.LinkedList) WebUIService(org.apache.tez.dag.app.web.WebUIService) HadoopShimsLoader(org.apache.tez.hadoop.shim.HadoopShimsLoader) TezDAGID(org.apache.tez.dag.records.TezDAGID) Event(org.apache.hadoop.yarn.event.Event) DAGClientServer(org.apache.tez.dag.api.client.DAGClientServer) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) DagInfoImplForTest(org.apache.tez.dag.helpers.DagInfoImplForTest) Test(org.junit.Test)

Aggregations

DAGClientServer (org.apache.tez.dag.api.client.DAGClientServer)3 InetSocketAddress (java.net.InetSocketAddress)2 Configuration (org.apache.hadoop.conf.Configuration)2 Event (org.apache.hadoop.yarn.event.Event)2 EventHandler (org.apache.hadoop.yarn.event.EventHandler)2 ContainerSignatureMatcher (org.apache.tez.common.ContainerSignatureMatcher)2 NamedEntityDescriptor (org.apache.tez.dag.api.NamedEntityDescriptor)2 TezConfiguration (org.apache.tez.dag.api.TezConfiguration)2 AppContext (org.apache.tez.dag.app.AppContext)2 DAGAppMasterEventUserServiceFatalError (org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError)2 WebUIService (org.apache.tez.dag.app.web.WebUIService)2 DagInfoImplForTest (org.apache.tez.dag.helpers.DagInfoImplForTest)2 HadoopShimsLoader (org.apache.tez.hadoop.shim.HadoopShimsLoader)2 Test (org.junit.Test)2 ThreadFactoryBuilder (com.google.common.util.concurrent.ThreadFactoryBuilder)1 File (java.io.File)1 FileInputStream (java.io.FileInputStream)1 LinkedList (java.util.LinkedList)1 LocalResource (org.apache.hadoop.yarn.api.records.LocalResource)1 Resource (org.apache.hadoop.yarn.api.records.Resource)1