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

Example 11 with TaskAttempt

use of org.apache.tez.dag.app.dag.TaskAttempt in project tez by apache.

the class TestContainerReuse method testDifferentResourceContainerReuse.

@Test(timeout = 5000)
public void testDifferentResourceContainerReuse() throws Exception {
    Configuration tezConf = new Configuration(new YarnConfiguration());
    tezConf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_ENABLED, true);
    tezConf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_RACK_FALLBACK_ENABLED, true);
    tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_REUSE_LOCALITY_DELAY_ALLOCATION_MILLIS, 0);
    tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MIN_MILLIS, 0);
    tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MAX_MILLIS, 0);
    CapturingEventHandler eventHandler = new CapturingEventHandler();
    TezDAGID dagID = TezDAGID.getInstance("0", 0, 0);
    AMRMClient<CookieContainerRequest> rmClientCore = new AMRMClientForTest();
    TezAMRMClientAsync<CookieContainerRequest> rmClient = spy(new AMRMClientAsyncForTest(rmClientCore, 100));
    AppContext appContext = mock(AppContext.class);
    doReturn(new Configuration(false)).when(appContext).getAMConf();
    AMContainerMap amContainerMap = new AMContainerMap(mock(ContainerHeartbeatHandler.class), mock(TaskCommunicatorManagerInterface.class), new ContainerContextMatcher(), appContext);
    AMNodeTracker amNodeTracker = new AMNodeTracker(eventHandler, appContext);
    doReturn(amContainerMap).when(appContext).getAllContainers();
    doReturn(amNodeTracker).when(appContext).getNodeTracker();
    doReturn(DAGAppMasterState.RUNNING).when(appContext).getAMState();
    doReturn(dagID).when(appContext).getCurrentDAGID();
    doReturn(mock(ClusterInfo.class)).when(appContext).getClusterInfo();
    TaskSchedulerManager taskSchedulerManagerReal = new TaskSchedulerManagerForTest(appContext, eventHandler, rmClient, new AlwaysMatchesContainerMatcher(), TezUtils.createUserPayloadFromConf(tezConf));
    TaskSchedulerManager taskSchedulerManager = spy(taskSchedulerManagerReal);
    taskSchedulerManager.init(tezConf);
    taskSchedulerManager.start();
    TaskSchedulerWithDrainableContext taskScheduler = (TaskSchedulerWithDrainableContext) ((TaskSchedulerManagerForTest) taskSchedulerManager).getSpyTaskScheduler();
    TaskSchedulerContextDrainable drainableAppCallback = taskScheduler.getDrainableAppCallback();
    AtomicBoolean drainNotifier = new AtomicBoolean(false);
    taskScheduler.delayedContainerManager.drainedDelayedContainersForTest = drainNotifier;
    Resource resource1 = Resource.newInstance(1024, 1);
    Resource resource2 = Resource.newInstance(2048, 1);
    String[] host1 = { "host1" };
    String[] host2 = { "host2" };
    String[] racks = { "/default-rack" };
    Priority priority1 = Priority.newInstance(1);
    TezVertexID vertexID1 = TezVertexID.getInstance(dagID, 1);
    // Vertex 1, Task 1, Attempt 1, host1
    TezTaskAttemptID taID11 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 1), 1);
    TaskAttempt ta11 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent1 = createLaunchRequestEvent(taID11, ta11, resource1, host1, racks, priority1);
    // Vertex 1, Task 2, Attempt 1, host1
    TezTaskAttemptID taID12 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 2), 1);
    TaskAttempt ta12 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent2 = createLaunchRequestEvent(taID12, ta12, resource1, host1, racks, priority1);
    // Vertex 1, Task 3, Attempt 1, host2
    TezTaskAttemptID taID13 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 3), 1);
    TaskAttempt ta13 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent3 = createLaunchRequestEvent(taID13, ta13, resource2, host2, racks, priority1);
    // Vertex 1, Task 4, Attempt 1, host2
    TezTaskAttemptID taID14 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 4), 1);
    TaskAttempt ta14 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent4 = createLaunchRequestEvent(taID14, ta14, resource2, host2, racks, priority1);
    taskSchedulerManager.handleEvent(lrEvent1);
    taskSchedulerManager.handleEvent(lrEvent2);
    taskSchedulerManager.handleEvent(lrEvent3);
    taskSchedulerManager.handleEvent(lrEvent4);
    Container container1 = createContainer(1, "host1", resource1, priority1);
    Container container2 = createContainer(2, "host2", resource2, priority1);
    // One container allocated, should start ta11
    taskScheduler.onContainersAllocated(Collections.singletonList(container1));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta11), any(Object.class), eq(container1));
    // Second container allocated, should start ta13
    taskScheduler.onContainersAllocated(Collections.singletonList(container2));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta13), any(Object.class), eq(container2));
    // ta11 finished, should start ta12
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta11, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta11, true, null, null);
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta12), any(Object.class), eq(container1));
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // ta13 finished, should start ta14
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta13, container2.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta13, true, null, null);
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta14), any(Object.class), eq(container2));
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container2.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // ta12 finished no pending requests, should release container1
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta12, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta12, true, null, null);
    verify(rmClient).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // ta14 finished no pending requests, should release container2.
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta14, container2.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta14, true, null, null);
    verify(rmClient).releaseAssignedContainer(eq(container2.getId()));
    eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
    eventHandler.reset();
    taskScheduler.shutdown();
    taskSchedulerManager.close();
}

Example 12 with TaskAttempt

use of org.apache.tez.dag.app.dag.TaskAttempt in project tez by apache.

the class TestSpeculation method testSingleTaskSpeculation.

@Test(timeout = 10000)
public void testSingleTaskSpeculation() throws Exception {
    // Map<Timeout conf value, expected number of tasks>
    Map<Long, Integer> confToExpected = new HashMap<Long, Integer>();
    // Really long time to speculate
    confToExpected.put(Long.MAX_VALUE >> 1, 1);
    confToExpected.put(100L, 2);
    // Don't speculate
    confToExpected.put(-1L, 1);
    for (Map.Entry<Long, Integer> entry : confToExpected.entrySet()) {
        defaultConf.setLong(TezConfiguration.TEZ_AM_LEGACY_SPECULATIVE_SINGLE_TASK_VERTEX_TIMEOUT, entry.getKey());
        DAG dag = DAG.create("test");
        Vertex vA = Vertex.create("A", ProcessorDescriptor.create("Proc.class"), 1);
        dag.addVertex(vA);
        MockTezClient tezClient = createTezSession();
        DAGClient dagClient = tezClient.submitDAG(dag);
        DAGImpl dagImpl = (DAGImpl) mockApp.getContext().getCurrentDAG();
        TezVertexID vertexId = TezVertexID.getInstance(dagImpl.getID(), 0);
        // original attempt is killed and speculative one is successful
        TezTaskAttemptID killedTaId = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexId, 0), 0);
        TezTaskAttemptID successTaId = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexId, 0), 1);
        Thread.sleep(200);
        // cause speculation trigger
        mockLauncher.setStatusUpdatesForTask(killedTaId, 100);
        mockLauncher.startScheduling(true);
        dagClient.waitForCompletion();
        Assert.assertEquals(DAGStatus.State.SUCCEEDED, dagClient.getDAGStatus(null).getState());
        Task task = dagImpl.getTask(killedTaId.getTaskID());
        Assert.assertEquals(entry.getValue().intValue(), task.getAttempts().size());
        if (entry.getValue() > 1) {
            Assert.assertEquals(successTaId, task.getSuccessfulAttempt().getID());
            TaskAttempt killedAttempt = task.getAttempt(killedTaId);
            Joiner.on(",").join(killedAttempt.getDiagnostics()).contains("Killed as speculative attempt");
            Assert.assertEquals(TaskAttemptTerminationCause.TERMINATED_EFFECTIVE_SPECULATION, killedAttempt.getTerminationCause());
        }
        tezClient.stop();
    }
}

Example 13 with TaskAttempt

use of org.apache.tez.dag.app.dag.TaskAttempt in project tez by apache.

the class TestSpeculation method testBasicSpeculation.

public void testBasicSpeculation(boolean withProgress) throws Exception {
    DAG dag = DAG.create("test");
    Vertex vA = Vertex.create("A", ProcessorDescriptor.create("Proc.class"), 5);
    dag.addVertex(vA);
    MockTezClient tezClient = createTezSession();
    DAGClient dagClient = tezClient.submitDAG(dag);
    DAGImpl dagImpl = (DAGImpl) mockApp.getContext().getCurrentDAG();
    TezVertexID vertexId = TezVertexID.getInstance(dagImpl.getID(), 0);
    // original attempt is killed and speculative one is successful
    TezTaskAttemptID killedTaId = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexId, 0), 0);
    TezTaskAttemptID successTaId = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexId, 0), 1);
    mockLauncher.updateProgress(withProgress);
    // cause speculation trigger
    mockLauncher.setStatusUpdatesForTask(killedTaId, 100);
    mockLauncher.startScheduling(true);
    dagClient.waitForCompletion();
    Assert.assertEquals(DAGStatus.State.SUCCEEDED, dagClient.getDAGStatus(null).getState());
    Task task = dagImpl.getTask(killedTaId.getTaskID());
    Assert.assertEquals(2, task.getAttempts().size());
    Assert.assertEquals(successTaId, task.getSuccessfulAttempt().getID());
    TaskAttempt killedAttempt = task.getAttempt(killedTaId);
    Joiner.on(",").join(killedAttempt.getDiagnostics()).contains("Killed as speculative attempt");
    Assert.assertEquals(TaskAttemptTerminationCause.TERMINATED_EFFECTIVE_SPECULATION, killedAttempt.getTerminationCause());
    if (withProgress) {
        // without progress updates occasionally more than 1 task speculates
        Assert.assertEquals(1, task.getCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
        Assert.assertEquals(1, dagImpl.getAllCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
        org.apache.tez.dag.app.dag.Vertex v = dagImpl.getVertex(killedTaId.getTaskID().getVertexID());
        Assert.assertEquals(1, v.getAllCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
    }
    tezClient.stop();
}

Example 14 with TaskAttempt

use of org.apache.tez.dag.app.dag.TaskAttempt in project tez by apache.

the class TestSpeculation method testBasicSpeculationNotUseful.

@Test(timeout = 10000)
public void testBasicSpeculationNotUseful() throws Exception {
    DAG dag = DAG.create("test");
    Vertex vA = Vertex.create("A", ProcessorDescriptor.create("Proc.class"), 5);
    dag.addVertex(vA);
    MockTezClient tezClient = createTezSession();
    DAGClient dagClient = tezClient.submitDAG(dag);
    DAGImpl dagImpl = (DAGImpl) mockApp.getContext().getCurrentDAG();
    TezVertexID vertexId = TezVertexID.getInstance(dagImpl.getID(), 0);
    // original attempt is successful and speculative one is killed
    TezTaskAttemptID successTaId = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexId, 0), 0);
    TezTaskAttemptID killedTaId = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexId, 0), 1);
    mockLauncher.setStatusUpdatesForTask(successTaId, 100);
    mockLauncher.setStatusUpdatesForTask(killedTaId, 100);
    mockLauncher.startScheduling(true);
    dagClient.waitForCompletion();
    Assert.assertEquals(DAGStatus.State.SUCCEEDED, dagClient.getDAGStatus(null).getState());
    Task task = dagImpl.getTask(killedTaId.getTaskID());
    Assert.assertEquals(2, task.getAttempts().size());
    Assert.assertEquals(successTaId, task.getSuccessfulAttempt().getID());
    TaskAttempt killedAttempt = task.getAttempt(killedTaId);
    Joiner.on(",").join(killedAttempt.getDiagnostics()).contains("Killed speculative attempt as");
    Assert.assertEquals(TaskAttemptTerminationCause.TERMINATED_INEFFECTIVE_SPECULATION, killedAttempt.getTerminationCause());
    Assert.assertEquals(1, task.getCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
    Assert.assertEquals(1, dagImpl.getAllCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
    org.apache.tez.dag.app.dag.Vertex v = dagImpl.getVertex(killedTaId.getTaskID().getVertexID());
    Assert.assertEquals(1, v.getAllCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
    tezClient.stop();
}

Example 15 with TaskAttempt

use of org.apache.tez.dag.app.dag.TaskAttempt in project tez by apache.

the class TestTaskSchedulerManager method testTaskSchedulerRouting.

@Test(timeout = 5000)
public void testTaskSchedulerRouting() throws Exception {
    Configuration conf = new Configuration(false);
    UserPayload defaultPayload = TezUtils.createUserPayloadFromConf(conf);
    String customSchedulerName = "fakeScheduler";
    List<NamedEntityDescriptor> taskSchedulers = new LinkedList<>();
    ByteBuffer bb = ByteBuffer.allocate(4);
    bb.putInt(0, 3);
    UserPayload userPayload = UserPayload.create(bb);
    taskSchedulers.add(new NamedEntityDescriptor(customSchedulerName, FakeTaskScheduler.class.getName()).setUserPayload(userPayload));
    taskSchedulers.add(new NamedEntityDescriptor(TezConstants.getTezYarnServicePluginName(), null).setUserPayload(defaultPayload));
    TSEHForMultipleSchedulersTest tseh = new TSEHForMultipleSchedulersTest(mockAppContext, mockClientService, mockEventHandler, mockSigMatcher, mockWebUIService, taskSchedulers, false);
    tseh.init(conf);
    tseh.start();
    // Verify that the YARN task scheduler is installed by default
    assertTrue(tseh.getYarnSchedulerCreated());
    assertFalse(tseh.getUberSchedulerCreated());
    assertEquals(2, tseh.getNumCreateInvocations());
    // Verify the order of the schedulers
    assertEquals(customSchedulerName, tseh.getTaskSchedulerName(0));
    assertEquals(TezConstants.getTezYarnServicePluginName(), tseh.getTaskSchedulerName(1));
    verify(tseh.getTestTaskScheduler(0)).initialize();
    verify(tseh.getTestTaskScheduler(0)).start();
    ApplicationId appId = ApplicationId.newInstance(1000, 1);
    TezDAGID dagId = TezDAGID.getInstance(appId, 1);
    TezVertexID vertexID = TezVertexID.getInstance(dagId, 1);
    TezTaskID taskId1 = TezTaskID.getInstance(vertexID, 1);
    TezTaskAttemptID attemptId11 = TezTaskAttemptID.getInstance(taskId1, 1);
    TezTaskID taskId2 = TezTaskID.getInstance(vertexID, 2);
    TezTaskAttemptID attemptId21 = TezTaskAttemptID.getInstance(taskId2, 1);
    Resource resource = Resource.newInstance(1024, 1);
    TaskAttempt mockTaskAttempt1 = mock(TaskAttempt.class);
    TaskAttempt mockTaskAttempt2 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest launchRequest1 = new AMSchedulerEventTALaunchRequest(attemptId11, resource, mock(TaskSpec.class), mockTaskAttempt1, mock(TaskLocationHint.class), 1, mock(ContainerContext.class), 0, 0, 0);
    tseh.handle(launchRequest1);
    verify(tseh.getTestTaskScheduler(0)).allocateTask(eq(mockTaskAttempt1), eq(resource), any(String[].class), any(String[].class), any(Priority.class), any(Object.class), eq(launchRequest1));
    AMSchedulerEventTALaunchRequest launchRequest2 = new AMSchedulerEventTALaunchRequest(attemptId21, resource, mock(TaskSpec.class), mockTaskAttempt2, mock(TaskLocationHint.class), 1, mock(ContainerContext.class), 1, 0, 0);
    tseh.handle(launchRequest2);
    verify(tseh.getTestTaskScheduler(1)).allocateTask(eq(mockTaskAttempt2), eq(resource), any(String[].class), any(String[].class), any(Priority.class), any(Object.class), eq(launchRequest2));
}