Search in sources :

Example 31 with ContainerContextMatcher

use of org.apache.tez.dag.app.rm.container.ContainerContextMatcher in project tez by apache.

the class TestContainerReuse method testReuseConflictLocalResources.

@Test(timeout = 30000l)
public void testReuseConflictLocalResources() 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 ContainerContextMatcher(), 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";
    LocalResource lr1 = mock(LocalResource.class);
    doReturn(LocalResourceType.FILE).when(lr1).getType();
    LocalResource lr2 = mock(LocalResource.class);
    doReturn(LocalResourceType.FILE).when(lr2).getType();
    LocalResource lr3 = mock(LocalResource.class);
    doReturn(LocalResourceType.FILE).when(lr3).getType();
    AMContainerEventAssignTA assignEvent = null;
    Map<String, LocalResource> v11LR = Maps.newHashMap();
    v11LR.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);
    doReturn(taID111).when(ta111).getID();
    doReturn("Mock for TA " + taID111.toString()).when(ta111).toString();
    AMSchedulerEventTALaunchRequest lrEvent11 = createLaunchRequestEvent(taID111, ta111, resource1, host1, racks, priority1, v11LR);
    Map<String, LocalResource> v12LR = Maps.newHashMap();
    v12LR.put(rsrc1, lr1);
    v12LR.put(rsrc2, lr2);
    // Vertex 1, Task 2, Attempt 1, host1, lr1
    TezTaskAttemptID taID112 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID11, 2), 1);
    TaskAttempt ta112 = mock(TaskAttempt.class);
    doReturn(taID112).when(ta112).getID();
    doReturn("Mock for TA " + taID112.toString()).when(ta112).toString();
    AMSchedulerEventTALaunchRequest lrEvent12 = createLaunchRequestEvent(taID112, ta112, resource1, host1, racks, priority1, v12LR);
    // Vertex 1, Task 3, Attempt 1, host1
    TezTaskAttemptID taID113 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID11, 3), 1);
    TaskAttempt ta113 = mock(TaskAttempt.class);
    doReturn(taID113).when(ta113).getID();
    doReturn("Mock for TA " + taID113.toString()).when(ta113).toString();
    AMSchedulerEventTALaunchRequest lrEvent13 = createLaunchRequestEvent(taID113, ta113, resource1, host1, racks, priority1, new HashMap<String, LocalResource>());
    // Vertex 1, Task 4, Attempt 1, host1
    TezTaskAttemptID taID114 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID11, 4), 1);
    TaskAttempt ta114 = mock(TaskAttempt.class);
    doReturn(taID114).when(ta114).getID();
    doReturn("Mock for TA " + taID114.toString()).when(ta114).toString();
    AMSchedulerEventTALaunchRequest lrEvent14 = createLaunchRequestEvent(taID114, ta114, resource1, host1, racks, priority1, new HashMap<String, LocalResource>());
    taskSchedulerManager.handleEvent(lrEvent11);
    taskSchedulerManager.handleEvent(lrEvent12);
    Container container1 = createContainer(1, "host1", resource1, priority1);
    Container container2 = createContainer(2, "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();
    // Task 3
    taskSchedulerManager.handleEvent(lrEvent13);
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta113), any(Object.class), eq(container1));
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    eventHandler.reset();
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta113, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta113, true, null, null);
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // Task 4
    taskSchedulerManager.handleEvent(lrEvent14);
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta114), any(Object.class), eq(container1));
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    eventHandler.reset();
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta114, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta114, true, null, null);
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // Setup DAG2 with different resources.
    TezDAGID dagID2 = TezDAGID.getInstance("0", 2, 0);
    dagIDAnswer.setDAGID(dagID2);
    Map<String, LocalResource> v21LR = Maps.newHashMap();
    v21LR.put(rsrc1, lr1);
    v21LR.put(rsrc2, 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);
    doReturn(taID211).when(ta211).getID();
    doReturn("Mock for TA " + taID211.toString()).when(ta211).toString();
    AMSchedulerEventTALaunchRequest lrEvent21 = createLaunchRequestEvent(taID211, ta211, resource1, host1, racks, priority1, v21LR);
    taskSchedulerManager.handleEvent(lrEvent21);
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(rmClient, times(1)).releaseAssignedContainer(eq(container1.getId()));
    taskScheduler.onContainersAllocated(Collections.singletonList(container2));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta211), any(Object.class), eq(container2));
    eventHandler.reset();
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta211, container2.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta211, true, null, null);
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container2.getId()));
    eventHandler.reset();
    // Setup DAG3 with a subset of resources from DAG2 (container should be reused).
    TezDAGID dagID3 = TezDAGID.getInstance("0", 3, 0);
    dagIDAnswer.setDAGID(dagID3);
    Map<String, LocalResource> v31LR = Collections.singletonMap(rsrc1, lr1);
    // dag3, vertex1, task1, ta1
    TezTaskAttemptID taID311 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(TezVertexID.getInstance(dagID3, 1), 1), 1);
    TaskAttempt ta311 = mock(TaskAttempt.class);
    doReturn(taID311).when(ta311).getID();
    doReturn("Mock for TA " + taID311).when(ta311).toString();
    AMSchedulerEventTALaunchRequest lrEvent31 = createLaunchRequestEvent(taID311, ta311, resource1, host1, racks, priority1, v31LR);
    taskSchedulerManager.handleEvent(lrEvent31);
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta311), any(Object.class), eq(container2));
    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) 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)

Aggregations

Container (org.apache.hadoop.yarn.api.records.Container)31 ContainerHeartbeatHandler (org.apache.tez.dag.app.ContainerHeartbeatHandler)31 TaskCommunicatorManagerInterface (org.apache.tez.dag.app.TaskCommunicatorManagerInterface)31 AMContainerMap (org.apache.tez.dag.app.rm.container.AMContainerMap)31 ContainerContextMatcher (org.apache.tez.dag.app.rm.container.ContainerContextMatcher)31 Test (org.junit.Test)31 Configuration (org.apache.hadoop.conf.Configuration)27 LocalResource (org.apache.hadoop.yarn.api.records.LocalResource)27 Resource (org.apache.hadoop.yarn.api.records.Resource)27 TezConfiguration (org.apache.tez.dag.api.TezConfiguration)27 ClusterInfo (org.apache.tez.dag.app.ClusterInfo)27 TezDAGID (org.apache.tez.dag.records.TezDAGID)27 TezTaskAttemptID (org.apache.tez.dag.records.TezTaskAttemptID)27 TezVertexID (org.apache.tez.dag.records.TezVertexID)27 ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)20 NodeId (org.apache.hadoop.yarn.api.records.NodeId)20 TaskAttemptEventSchedule (org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule)19 TaskAttemptEventSubmitted (org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted)19 TaskAttemptEventStartedRemotely (org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely)18 TezTaskID (org.apache.tez.dag.records.TezTaskID)17