Search in sources :

Example 36 with TezTaskAttemptID

use of org.apache.tez.dag.records.TezTaskAttemptID in project tez by apache.

the class TestContainerReuse method testReuseLocalResourcesChanged.

@Test(timeout = 30000l)
public void testReuseLocalResourcesChanged() throws IOException, InterruptedException, ExecutionException {
    LOG.info("Test testReuseLocalResourcesChanged");
    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.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_NON_LOCAL_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, -1);
    CapturingEventHandler eventHandler = new CapturingEventHandler();
    TezDAGID dagID1 = TezDAGID.getInstance("0", 1, 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();
    ChangingDAGIDAnswer dagIDAnswer = new ChangingDAGIDAnswer(dagID1);
    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(true).when(appContext).isSession();
    doAnswer(dagIDAnswer).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);
    String[] host1 = { "host1" };
    String[] racks = { "/default-rack" };
    Priority priority1 = Priority.newInstance(1);
    String rsrc1 = "rsrc1";
    String rsrc2 = "rsrc2";
    String rsrc3 = "rsrc3";
    LocalResource lr1 = mock(LocalResource.class);
    LocalResource lr2 = mock(LocalResource.class);
    LocalResource lr3 = mock(LocalResource.class);
    AMContainerEventAssignTA assignEvent = null;
    Map<String, LocalResource> dag1LRs = Maps.newHashMap();
    dag1LRs.put(rsrc1, lr1);
    TezVertexID vertexID11 = TezVertexID.getInstance(dagID1, 1);
    // Vertex 1, Task 1, Attempt 1, host1, lr1
    TezTaskAttemptID taID111 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID11, 1), 1);
    TaskAttempt ta111 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent11 = createLaunchRequestEvent(taID111, ta111, resource1, host1, racks, priority1, dag1LRs);
    // Vertex 1, Task 2, Attempt 1, host1, lr1
    TezTaskAttemptID taID112 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID11, 2), 1);
    TaskAttempt ta112 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent12 = createLaunchRequestEvent(taID112, ta112, resource1, host1, racks, priority1, dag1LRs);
    taskSchedulerManager.handleEvent(lrEvent11);
    taskSchedulerManager.handleEvent(lrEvent12);
    Container container1 = createContainer(1, "host1", resource1, priority1);
    // One container allocated.
    taskScheduler.onContainersAllocated(Collections.singletonList(container1));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta111), any(Object.class), eq(container1));
    assignEvent = (AMContainerEventAssignTA) eventHandler.verifyInvocation(AMContainerEventAssignTA.class);
    assertEquals(1, assignEvent.getRemoteTaskLocalResources().size());
    // Task assigned to container completed successfully. Container should be re-used.
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta111, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta111, true, null, null);
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta112), any(Object.class), eq(container1));
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    assignEvent = (AMContainerEventAssignTA) eventHandler.verifyInvocation(AMContainerEventAssignTA.class);
    assertEquals(1, assignEvent.getRemoteTaskLocalResources().size());
    eventHandler.reset();
    // Task assigned to container completed successfully.
    // Verify reuse across hosts.
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta112, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta112, true, null, null);
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // Setup DAG2 with additional resources. Make sure the container, even without all resources, is reused.
    TezDAGID dagID2 = TezDAGID.getInstance("0", 2, 0);
    dagIDAnswer.setDAGID(dagID2);
    Map<String, LocalResource> dag2LRs = Maps.newHashMap();
    dag2LRs.put(rsrc2, lr2);
    dag2LRs.put(rsrc3, lr3);
    TezVertexID vertexID21 = TezVertexID.getInstance(dagID2, 1);
    // Vertex 2, Task 1, Attempt 1, host1, lr2
    TezTaskAttemptID taID211 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID21, 1), 1);
    TaskAttempt ta211 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent21 = createLaunchRequestEvent(taID211, ta211, resource1, host1, racks, priority1, dag2LRs);
    // Vertex 2, Task 2, Attempt 1, host1, lr2
    TezTaskAttemptID taID212 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID21, 2), 1);
    TaskAttempt ta212 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent22 = createLaunchRequestEvent(taID212, ta212, resource1, host1, racks, priority1, dag2LRs);
    taskSchedulerManager.handleEvent(lrEvent21);
    taskSchedulerManager.handleEvent(lrEvent22);
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta211), any(Object.class), eq(container1));
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    assignEvent = (AMContainerEventAssignTA) eventHandler.verifyInvocation(AMContainerEventAssignTA.class);
    assertEquals(2, assignEvent.getRemoteTaskLocalResources().size());
    eventHandler.reset();
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta211, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta211, true, null, null);
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta212), any(Object.class), eq(container1));
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    assignEvent = (AMContainerEventAssignTA) eventHandler.verifyInvocation(AMContainerEventAssignTA.class);
    assertEquals(2, assignEvent.getRemoteTaskLocalResources().size());
    eventHandler.reset();
    taskScheduler.shutdown();
    taskSchedulerManager.close();
}
Also used : CapturingEventHandler(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.CapturingEventHandler) AMContainerEventAssignTA(org.apache.tez.dag.app.rm.container.AMContainerEventAssignTA) Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) TaskSchedulerWithDrainableContext(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerWithDrainableContext) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) AlwaysMatchesContainerMatcher(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AlwaysMatchesContainerMatcher) Container(org.apache.hadoop.yarn.api.records.Container) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) TezDAGID(org.apache.tez.dag.records.TezDAGID) AMRMClientAsyncForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientAsyncForTest) TaskAttempt(org.apache.tez.dag.app.dag.TaskAttempt) ContainerHeartbeatHandler(org.apache.tez.dag.app.ContainerHeartbeatHandler) TaskSchedulerContextDrainable(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerContextDrainable) TezVertexID(org.apache.tez.dag.records.TezVertexID) Priority(org.apache.hadoop.yarn.api.records.Priority) AppContext(org.apache.tez.dag.app.AppContext) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) TaskCommunicatorManagerInterface(org.apache.tez.dag.app.TaskCommunicatorManagerInterface) ContainerContextMatcher(org.apache.tez.dag.app.rm.container.ContainerContextMatcher) AMNodeTracker(org.apache.tez.dag.app.rm.node.AMNodeTracker) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) AtomicBoolean(java.util.concurrent.atomic.AtomicBoolean) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) TaskSchedulerManagerForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerManagerForTest) CookieContainerRequest(org.apache.tez.dag.app.rm.YarnTaskSchedulerService.CookieContainerRequest) AMRMClientForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientForTest) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) TaskSchedulerManagerForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerManagerForTest) AMRMClientForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientForTest) AMRMClientAsyncForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientAsyncForTest) Test(org.junit.Test)

Example 37 with TezTaskAttemptID

use of org.apache.tez.dag.records.TezTaskAttemptID in project tez by apache.

the class TestContainerReuse method testAssignmentOnShutdown.

@Test(timeout = 10000l)
public void testAssignmentOnShutdown() throws IOException, InterruptedException, ExecutionException {
    LOG.info("Test testAssignmentOnShutdown");
    Configuration tezConf = new Configuration(new YarnConfiguration());
    tezConf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_ENABLED, false);
    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.SUCCEEDED).when(appContext).getAMState();
    doReturn(true).when(appContext).isAMInCompletionState();
    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);
    String[] host1 = { "host1" };
    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);
    taskSchedulerManager.handleEvent(lrEvent1);
    Container container1 = createContainer(1, "host1", resource1, priority1);
    // One container allocated.
    taskScheduler.onContainersAllocated(Collections.singletonList(container1));
    drainableAppCallback.drain();
    verify(taskSchedulerManager, times(0)).taskAllocated(eq(0), eq(ta11), any(Object.class), eq(container1));
    taskScheduler.shutdown();
    taskSchedulerManager.close();
}
Also used : CapturingEventHandler(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.CapturingEventHandler) Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) TaskSchedulerWithDrainableContext(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerWithDrainableContext) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) AlwaysMatchesContainerMatcher(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AlwaysMatchesContainerMatcher) Container(org.apache.hadoop.yarn.api.records.Container) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) TezDAGID(org.apache.tez.dag.records.TezDAGID) AMRMClientAsyncForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientAsyncForTest) TaskAttempt(org.apache.tez.dag.app.dag.TaskAttempt) ContainerHeartbeatHandler(org.apache.tez.dag.app.ContainerHeartbeatHandler) TaskSchedulerContextDrainable(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerContextDrainable) TezVertexID(org.apache.tez.dag.records.TezVertexID) Priority(org.apache.hadoop.yarn.api.records.Priority) AppContext(org.apache.tez.dag.app.AppContext) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) TaskCommunicatorManagerInterface(org.apache.tez.dag.app.TaskCommunicatorManagerInterface) ContainerContextMatcher(org.apache.tez.dag.app.rm.container.ContainerContextMatcher) AMNodeTracker(org.apache.tez.dag.app.rm.node.AMNodeTracker) AtomicBoolean(java.util.concurrent.atomic.AtomicBoolean) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) TaskSchedulerManagerForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerManagerForTest) CookieContainerRequest(org.apache.tez.dag.app.rm.YarnTaskSchedulerService.CookieContainerRequest) AMRMClientForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientForTest) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) TaskSchedulerManagerForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerManagerForTest) AMRMClientForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientForTest) AMRMClientAsyncForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientAsyncForTest) Test(org.junit.Test)

Example 38 with TezTaskAttemptID

use of org.apache.tez.dag.records.TezTaskAttemptID 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();
}
Also used : CapturingEventHandler(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.CapturingEventHandler) Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) TaskSchedulerWithDrainableContext(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerWithDrainableContext) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) AlwaysMatchesContainerMatcher(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AlwaysMatchesContainerMatcher) Container(org.apache.hadoop.yarn.api.records.Container) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) TezDAGID(org.apache.tez.dag.records.TezDAGID) AMRMClientAsyncForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientAsyncForTest) TaskAttempt(org.apache.tez.dag.app.dag.TaskAttempt) ContainerHeartbeatHandler(org.apache.tez.dag.app.ContainerHeartbeatHandler) TaskSchedulerContextDrainable(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerContextDrainable) TezVertexID(org.apache.tez.dag.records.TezVertexID) Priority(org.apache.hadoop.yarn.api.records.Priority) AppContext(org.apache.tez.dag.app.AppContext) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) TaskCommunicatorManagerInterface(org.apache.tez.dag.app.TaskCommunicatorManagerInterface) ContainerContextMatcher(org.apache.tez.dag.app.rm.container.ContainerContextMatcher) AMNodeTracker(org.apache.tez.dag.app.rm.node.AMNodeTracker) AtomicBoolean(java.util.concurrent.atomic.AtomicBoolean) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) TaskSchedulerManagerForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerManagerForTest) CookieContainerRequest(org.apache.tez.dag.app.rm.YarnTaskSchedulerService.CookieContainerRequest) AMRMClientForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientForTest) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) TaskSchedulerManagerForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.TaskSchedulerManagerForTest) AMRMClientForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientForTest) AMRMClientAsyncForTest(org.apache.tez.dag.app.rm.TestTaskSchedulerHelpers.AMRMClientAsyncForTest) Test(org.junit.Test)

Example 39 with TezTaskAttemptID

use of org.apache.tez.dag.records.TezTaskAttemptID 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();
    }
}
Also used : Vertex(org.apache.tez.dag.api.Vertex) Task(org.apache.tez.dag.app.dag.Task) HashMap(java.util.HashMap) DAG(org.apache.tez.dag.api.DAG) DAGImpl(org.apache.tez.dag.app.dag.impl.DAGImpl) DAGClient(org.apache.tez.dag.api.client.DAGClient) TaskAttempt(org.apache.tez.dag.app.dag.TaskAttempt) HashMap(java.util.HashMap) Map(java.util.Map) TezVertexID(org.apache.tez.dag.records.TezVertexID) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) Test(org.junit.Test)

Example 40 with TezTaskAttemptID

use of org.apache.tez.dag.records.TezTaskAttemptID 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();
}
Also used : Vertex(org.apache.tez.dag.api.Vertex) Task(org.apache.tez.dag.app.dag.Task) DAGImpl(org.apache.tez.dag.app.dag.impl.DAGImpl) DAGClient(org.apache.tez.dag.api.client.DAGClient) DAG(org.apache.tez.dag.api.DAG) TaskAttempt(org.apache.tez.dag.app.dag.TaskAttempt) TezVertexID(org.apache.tez.dag.records.TezVertexID) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID)

Aggregations

TezTaskAttemptID (org.apache.tez.dag.records.TezTaskAttemptID)167 Test (org.junit.Test)124 TezTaskID (org.apache.tez.dag.records.TezTaskID)61 TezVertexID (org.apache.tez.dag.records.TezVertexID)54 Container (org.apache.hadoop.yarn.api.records.Container)48 TezEvent (org.apache.tez.runtime.api.impl.TezEvent)46 TezDAGID (org.apache.tez.dag.records.TezDAGID)43 Configuration (org.apache.hadoop.conf.Configuration)42 ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)41 Priority (org.apache.hadoop.yarn.api.records.Priority)41 TezConfiguration (org.apache.tez.dag.api.TezConfiguration)41 LocalResource (org.apache.hadoop.yarn.api.records.LocalResource)33 Resource (org.apache.hadoop.yarn.api.records.Resource)30 TaskCommunicatorManagerInterface (org.apache.tez.dag.app.TaskCommunicatorManagerInterface)28 EventMetaData (org.apache.tez.runtime.api.impl.EventMetaData)28 ClusterInfo (org.apache.tez.dag.app.ClusterInfo)27 ContainerHeartbeatHandler (org.apache.tez.dag.app.ContainerHeartbeatHandler)27 AMContainerMap (org.apache.tez.dag.app.rm.container.AMContainerMap)27 ContainerContextMatcher (org.apache.tez.dag.app.rm.container.ContainerContextMatcher)27 ApplicationId (org.apache.hadoop.yarn.api.records.ApplicationId)25