Examples with DAGAppMasterEventUserServiceFatalError org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError used on opensource projects

Example 1 with DAGAppMasterEventUserServiceFatalError

use of org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError in project tez by apache.

the class TaskSchedulerManager method setShouldUnregisterFlag.

public void setShouldUnregisterFlag() {
    LOG.info("TaskScheduler notified that it should unregister from RM");
    this.shouldUnregisterFlag.set(true);
    for (int i = 0; i < taskSchedulers.length; i++) {
        if (this.taskSchedulers[i] != null) {
            try {
                this.taskSchedulers[i].setShouldUnregister();
            } catch (Exception e) {
                String msg = "Error in TaskScheduler when setting Unregister Flag" + ", scheduler=" + Utils.getTaskSchedulerIdentifierString(i, appContext);
                LOG.error(msg, e);
                sendEvent(new DAGAppMasterEventUserServiceFatalError(DAGAppMasterEventType.TASK_SCHEDULER_SERVICE_FATAL_ERROR, msg, e));
            }
        }
    }
}

Example 2 with DAGAppMasterEventUserServiceFatalError

use of org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError in project tez by apache.

the class TaskSchedulerManager method handleTASucceeded.

private void handleTASucceeded(AMSchedulerEventTAEnded event) {
    TaskAttempt attempt = event.getAttempt();
    ContainerId usedContainerId = event.getUsedContainerId();
    // assigned to it.
    if (event.getUsedContainerId() != null) {
        sendEvent(new AMContainerEventTASucceeded(usedContainerId, event.getAttemptID()));
        sendEvent(new AMNodeEventTaskAttemptSucceeded(appContext.getAllContainers().get(usedContainerId).getContainer().getNodeId(), event.getSchedulerId(), usedContainerId, event.getAttemptID()));
    }
    boolean wasContainerAllocated = false;
    try {
        wasContainerAllocated = taskSchedulers[event.getSchedulerId()].deallocateTask(attempt, true, null, event.getDiagnostics());
    } catch (Exception e) {
        String msg = "Error in TaskScheduler for handling Task De-allocation" + ", eventType=" + event.getType() + ", scheduler=" + Utils.getTaskSchedulerIdentifierString(event.getSchedulerId(), appContext) + ", taskAttemptId=" + attempt.getID();
        LOG.error(msg, e);
        sendEvent(new DAGAppMasterEventUserServiceFatalError(DAGAppMasterEventType.TASK_SCHEDULER_SERVICE_FATAL_ERROR, msg, e));
        return;
    }
    if (!wasContainerAllocated) {
        LOG.error("De-allocated successful task: " + attempt.getID() + ", but TaskScheduler reported no container assigned to task");
    }
}

Example 3 with DAGAppMasterEventUserServiceFatalError

use of org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError in project tez by apache.

the class TaskSchedulerManager method handleContainerDeallocate.

private void handleContainerDeallocate(AMSchedulerEventDeallocateContainer event) {
    ContainerId containerId = event.getContainerId();
    // TaskAttempt taskAttempt = (TaskAttempt)
    try {
        taskSchedulers[event.getSchedulerId()].deallocateContainer(containerId);
    } catch (Exception e) {
        String msg = "Error in TaskScheduler for handling Container De-allocation" + ", eventType=" + event.getType() + ", scheduler=" + Utils.getTaskSchedulerIdentifierString(event.getSchedulerId(), appContext) + ", containerId=" + containerId;
        LOG.error(msg, e);
        sendEvent(new DAGAppMasterEventUserServiceFatalError(DAGAppMasterEventType.TASK_SCHEDULER_SERVICE_FATAL_ERROR, msg, e));
        return;
    }
    // TODO does this container need to be stopped via C_STOP_REQUEST
    sendEvent(new AMContainerEventStopRequest(containerId));
}

Example 4 with DAGAppMasterEventUserServiceFatalError

use of org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError in project tez by apache.

the class TestContainerLauncherManager method testReportFailureFromContainerLauncher.

@SuppressWarnings("unchecked")
@Test(timeout = 5000)
public void testReportFailureFromContainerLauncher() throws ServicePluginException, TezException {
    final String dagName = DAG_NAME;
    final int dagIndex = DAG_INDEX;
    TezDAGID dagId = TezDAGID.getInstance(ApplicationId.newInstance(0, 0), dagIndex);
    DAG dag = mock(DAG.class);
    doReturn(dagName).when(dag).getName();
    doReturn(dagId).when(dag).getID();
    EventHandler eventHandler = mock(EventHandler.class);
    AppContext appContext = mock(AppContext.class);
    doReturn(eventHandler).when(appContext).getEventHandler();
    doReturn(dag).when(appContext).getCurrentDAG();
    doReturn("testlauncher").when(appContext).getContainerLauncherName(0);
    NamedEntityDescriptor<TaskCommunicatorDescriptor> taskCommDescriptor = new NamedEntityDescriptor<>("testlauncher", ContainerLauncherForTest.class.getName());
    List<NamedEntityDescriptor> list = new LinkedList<>();
    list.add(taskCommDescriptor);
    ContainerLauncherManager containerLauncherManager = new ContainerLauncherManager(appContext, mock(TaskCommunicatorManagerInterface.class), "", list, false);
    try {
        ContainerLaunchContext clc1 = mock(ContainerLaunchContext.class);
        Container container1 = mock(Container.class);
        ContainerLauncherLaunchRequestEvent launchRequestEvent = new ContainerLauncherLaunchRequestEvent(clc1, container1, 0, 0, 0);
        containerLauncherManager.handle(launchRequestEvent);
        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.CONTAINER_LAUNCHER_SERVICE_FATAL_ERROR, event.getType());
        assertTrue(event.getDiagnosticInfo().contains("ReportedFatalError"));
        assertTrue(event.getDiagnosticInfo().contains(ServicePluginErrorDefaults.INCONSISTENT_STATE.name()));
        assertTrue(event.getDiagnosticInfo().contains("[0:testlauncher]"));
        reset(eventHandler);
        // stop container
        ContainerId containerId2 = mock(ContainerId.class);
        NodeId nodeId2 = mock(NodeId.class);
        ContainerLauncherStopRequestEvent stopRequestEvent = new ContainerLauncherStopRequestEvent(containerId2, nodeId2, null, 0, 0, 0);
        argumentCaptor = ArgumentCaptor.forClass(Event.class);
        containerLauncherManager.handle(stopRequestEvent);
        verify(eventHandler, times(1)).handle(argumentCaptor.capture());
        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("[0:testlauncher]"));
    } finally {
        containerLauncherManager.stop();
    }
}

Example 5 with DAGAppMasterEventUserServiceFatalError

use of org.apache.tez.dag.app.dag.event.DAGAppMasterEventUserServiceFatalError in project tez by apache.

the class TestTaskCommunicatorManager method testReportFailureFromTaskCommunicator.

@SuppressWarnings("unchecked")
@Test(timeout = 5000)
public void testReportFailureFromTaskCommunicator() throws TezException {
    String dagName = DAG_NAME;
    EventHandler eventHandler = mock(EventHandler.class);
    AppContext appContext = mock(AppContext.class, RETURNS_DEEP_STUBS);
    doReturn("testTaskCommunicator").when(appContext).getTaskCommunicatorName(0);
    doReturn(eventHandler).when(appContext).getEventHandler();
    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<TaskCommunicatorDescriptor> namedEntityDescriptor = new NamedEntityDescriptor<>("testTaskCommunicator", TaskCommForFailureTest.class.getName());
    List<NamedEntityDescriptor> list = new LinkedList<>();
    list.add(namedEntityDescriptor);
    TaskCommunicatorManager taskCommManager = new TaskCommunicatorManager(appContext, mock(TaskHeartbeatHandler.class), mock(ContainerHeartbeatHandler.class), list);
    try {
        taskCommManager.init(new Configuration());
        taskCommManager.start();
        taskCommManager.registerRunningContainer(mock(ContainerId.class), 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("[0:testTaskCommunicator]"));
        reset(eventHandler);
        taskCommManager.dagComplete(dag);
        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_COMMUNICATOR_SERVICE_FATAL_ERROR, event.getType());
        assertTrue(event.getDiagnosticInfo().contains("ReportedFatalError"));
        assertTrue(event.getDiagnosticInfo().contains(ServicePluginErrorDefaults.INCONSISTENT_STATE.name()));
        assertTrue(event.getDiagnosticInfo().contains("[0:testTaskCommunicator]"));
    } finally {
        taskCommManager.stop();
    }
}