Examples with ContainerContextMatcher org.apache.tez.dag.app.rm.container.ContainerContextMatcher used on opensource projects

Example 11 with ContainerContextMatcher

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

the class TestTaskAttempt method testCompletedAtSubmitted.

@Test(timeout = 5000)
public void testCompletedAtSubmitted() throws ServicePluginException {
    ApplicationId appId = ApplicationId.newInstance(1, 2);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, 0);
    TezDAGID dagID = TezDAGID.getInstance(appId, 1);
    TezVertexID vertexID = TezVertexID.getInstance(dagID, 1);
    TezTaskID taskID = TezTaskID.getInstance(vertexID, 1);
    MockEventHandler eventHandler = spy(new MockEventHandler());
    TaskCommunicatorManagerInterface taListener = createMockTaskAttemptListener();
    Configuration taskConf = new Configuration();
    taskConf.setClass("fs.file.impl", StubbedFS.class, FileSystem.class);
    taskConf.setBoolean("fs.file.impl.disable.cache", true);
    locationHint = TaskLocationHint.createTaskLocationHint(new HashSet<String>(Arrays.asList(new String[] { "127.0.0.1" })), null);
    Resource resource = Resource.newInstance(1024, 1);
    NodeId nid = NodeId.newInstance("127.0.0.1", 0);
    @SuppressWarnings("deprecation") ContainerId contId = ContainerId.newInstance(appAttemptId, 3);
    Container container = mock(Container.class);
    when(container.getId()).thenReturn(contId);
    when(container.getNodeId()).thenReturn(nid);
    when(container.getNodeHttpAddress()).thenReturn("localhost:0");
    AMContainerMap containers = new AMContainerMap(mock(ContainerHeartbeatHandler.class), mock(TaskCommunicatorManagerInterface.class), new ContainerContextMatcher(), appCtx);
    containers.addContainerIfNew(container, 0, 0, 0);
    doReturn(new ClusterInfo()).when(appCtx).getClusterInfo();
    doReturn(containers).when(appCtx).getAllContainers();
    TaskHeartbeatHandler mockHeartbeatHandler = mock(TaskHeartbeatHandler.class);
    TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, new SystemClock(), mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
    TezTaskAttemptID taskAttemptID = taImpl.getID();
    ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
    taImpl.handle(new TaskAttemptEventSchedule(taskAttemptID, 0, 0));
    taImpl.handle(new TaskAttemptEventSubmitted(taskAttemptID, contId));
    assertEquals("Task attempt is not in the RUNNING state", taImpl.getState(), TaskAttemptState.STARTING);
    verify(mockHeartbeatHandler).register(taskAttemptID);
    int expectedEventsAtStarting = 4;
    verify(eventHandler, times(expectedEventsAtStarting)).handle(arg.capture());
    // Ensure status_updates are handled in the submitted state.
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
    taImpl.handle(new TaskAttemptEvent(taskAttemptID, TaskAttemptEventType.TA_DONE));
    assertEquals("Task attempt is not in the  SUCCEEDED state", taImpl.getState(), TaskAttemptState.SUCCEEDED);
    verify(mockHeartbeatHandler).unregister(taskAttemptID);
    assertEquals(0, taImpl.getDiagnostics().size());
    int expectedEvenstAfterTerminating = expectedEventsAtStarting + 3;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEvenstAfterTerminating)).handle(arg.capture());
    Event e = verifyEventType(arg.getAllValues().subList(expectedEventsAtStarting, expectedEvenstAfterTerminating), TaskEventTASucceeded.class, 1);
    assertEquals(TaskEventType.T_ATTEMPT_SUCCEEDED, e.getType());
    verifyEventType(arg.getAllValues().subList(expectedEventsAtStarting, expectedEvenstAfterTerminating), AMSchedulerEventTAEnded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtStarting, expectedEvenstAfterTerminating), DAGEventCounterUpdate.class, 1);
}

Example 12 with ContainerContextMatcher

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

the class TestContainerReuse method testReuseWithTaskSpecificLaunchCmdOption.

@Test(timeout = 10000l)
public void testReuseWithTaskSpecificLaunchCmdOption() throws IOException, InterruptedException, ExecutionException {
    LOG.info("Test testReuseWithTaskSpecificLaunchCmdOption");
    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);
    // Profile 3 tasks
    tezConf.set(TezConfiguration.TEZ_TASK_SPECIFIC_LAUNCH_CMD_OPTS_LIST, "v1[1,3,4]");
    tezConf.set(TezConfiguration.TEZ_TASK_SPECIFIC_LAUNCH_CMD_OPTS, "dir=/tmp/__VERTEX_NAME__/__TASK_INDEX__");
    TaskSpecificLaunchCmdOption taskSpecificLaunchCmdOption = new TaskSpecificLaunchCmdOption(tezConf);
    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();
    // Use ContainerContextMatcher here.  Otherwise it would not match the JVM options
    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[] host2 = { "host2" };
    String[] host3 = { "host3" };
    String[] racks = { "/default-rack" };
    Priority priority1 = Priority.newInstance(1);
    TezVertexID vertexID1 = TezVertexID.getInstance(dagID, 1);
    Map<String, LocalResource> localResources = new HashMap<String, LocalResource>();
    String tsLaunchCmdOpts = taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 1);
    /**
     * Schedule 2 tasks (1 with additional launch-cmd option and another in normal mode).
     * Container should not be reused in this case.
     */
    // 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, localResources, tsLaunchCmdOpts);
    // 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);
    taskSchedulerManager.handleEvent(lrEvent1);
    taskSchedulerManager.handleEvent(lrEvent2);
    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(ta11), any(Object.class), eq(container1));
    // First task had profiling on. This container can not be reused further.
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta11, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta11, true, null, null);
    verify(taskSchedulerManager, times(0)).taskAllocated(eq(0), eq(ta12), any(Object.class), eq(container1));
    verify(rmClient, times(1)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
    eventHandler.reset();
    /**
     * Schedule 2 tasks (both having different task specific JVM option).
     * Container should not be reused.
     */
    // Vertex 1, Task 3, Attempt 1, host2
    tsLaunchCmdOpts = taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 3);
    TezTaskAttemptID taID13 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 3), 1);
    TaskAttempt ta13 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent3 = createLaunchRequestEvent(taID13, ta13, resource1, host2, racks, priority1, localResources, tsLaunchCmdOpts);
    // Vertex 1, Task 4, Attempt 1, host2
    tsLaunchCmdOpts = taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 4);
    TezTaskAttemptID taID14 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 4), 1);
    TaskAttempt ta14 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent4 = createLaunchRequestEvent(taID14, ta14, resource1, host2, racks, priority1, localResources, tsLaunchCmdOpts);
    taskSchedulerManager.handleEvent(lrEvent3);
    taskSchedulerManager.handleEvent(lrEvent4);
    // Container started
    Container container2 = createContainer(2, "host2", resource1, priority1);
    taskScheduler.onContainersAllocated(Collections.singletonList(container2));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta13), any(Object.class), eq(container2));
    // Verify that the container can not be reused when profiling option is turned on
    // Even for 2 tasks having same profiling option can have container reusability.
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta13, container2.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta13, true, null, null);
    verify(taskSchedulerManager, times(0)).taskAllocated(eq(0), eq(ta14), any(Object.class), eq(container2));
    verify(rmClient, times(1)).releaseAssignedContainer(eq(container2.getId()));
    eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
    eventHandler.reset();
    /**
     * Schedule 2 tasks with same jvm profiling option.
     * Container should be reused.
     */
    tezConf.set(TezConfiguration.TEZ_TASK_SPECIFIC_LAUNCH_CMD_OPTS_LIST, "v1[1,2,3,5,6]");
    tezConf.set(TezConfiguration.TEZ_TASK_SPECIFIC_LAUNCH_CMD_OPTS, "dummyOpts");
    taskSpecificLaunchCmdOption = new TaskSpecificLaunchCmdOption(tezConf);
    // Vertex 1, Task 5, Attempt 1, host3
    TezTaskAttemptID taID15 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 3), 1);
    TaskAttempt ta15 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent5 = createLaunchRequestEvent(taID15, ta15, resource1, host3, racks, priority1, localResources, taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 5));
    // Vertex 1, Task 6, Attempt 1, host3
    tsLaunchCmdOpts = taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 4);
    TezTaskAttemptID taID16 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 4), 1);
    TaskAttempt ta16 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent6 = createLaunchRequestEvent(taID16, ta16, resource1, host3, racks, priority1, localResources, taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 6));
    // Container started
    Container container3 = createContainer(3, "host3", resource1, priority1);
    taskSchedulerManager.handleEvent(lrEvent5);
    taskSchedulerManager.handleEvent(lrEvent6);
    taskScheduler.onContainersAllocated(Collections.singletonList(container3));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta15), any(Object.class), eq(container3));
    // Ensure task 6 (of vertex 1) is allocated to same container
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta15, container3.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta15, true, null, null);
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta16), any(Object.class), eq(container3));
    eventHandler.reset();
    taskScheduler.shutdown();
    taskSchedulerManager.close();
}

Example 13 with ContainerContextMatcher

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

the class TestContainerReuse method testEnvironmentVarsContainerReuse.

@Test(timeout = 5000)
public void testEnvironmentVarsContainerReuse() 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 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);
    TezVertexID vertexID1 = TezVertexID.getInstance(dagID, 1);
    // Create containers with same resources and then see how tasks are allocated
    // when the environments vary
    Map<String, String> env1 = ImmutableMap.of("env1", "val1", "env2", "val2");
    Map<String, String> env2 = ImmutableMap.of("env3", "val3", "env4", "val4");
    Map<String, String> env3 = ImmutableMap.of("env1", "val1");
    Map<String, String> env4 = ImmutableMap.of("env1", "val1", "env4", "val4");
    // Vertex 1, Task 1, Attempt 1, host1, r1=lr1
    TezTaskAttemptID taID11 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 1), 1);
    TaskAttempt ta11 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent1 = createLaunchRequestEvent(taID11, ta11, resource1, host1, racks, priority1, "", env1);
    // Vertex 1, Task 2, Attempt 1, host1, r1=lr2
    TezTaskAttemptID taID12 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 2), 1);
    TaskAttempt ta12 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent2 = createLaunchRequestEvent(taID12, ta12, resource1, host1, racks, priority1, "", env2);
    // Vertex 1, Task 3, Attempt 1, host1, r1=lr1 & r2=lr2
    TezTaskAttemptID taID13 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 3), 1);
    TaskAttempt ta13 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent3 = createLaunchRequestEvent(taID13, ta13, resource1, host1, racks, priority1, "", env3);
    // Vertex 1, Task 4, Attempt 1, host1, r2=lr2
    TezTaskAttemptID taID14 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 4), 1);
    TaskAttempt ta14 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrEvent4 = createLaunchRequestEvent(taID14, ta14, resource1, host1, racks, priority1, "", env4);
    taskSchedulerManager.handleEvent(lrEvent1);
    taskSchedulerManager.handleEvent(lrEvent2);
    taskSchedulerManager.handleEvent(lrEvent3);
    taskSchedulerManager.handleEvent(lrEvent4);
    Container container1 = createContainer(1, "host1", resource1, priority1);
    Container container2 = createContainer(2, "host1", resource1, priority1);
    Container container3 = createContainer(3, "host1", resource1, priority1);
    // One container allocated, should start ta11 on container1
    taskScheduler.onContainersAllocated(Collections.singletonList(container1));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta11), any(Object.class), eq(container1));
    // finish ta11, should start ta13
    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(ta13), any(Object.class), eq(container1));
    verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // ta13 finished, cannot reuse container1, should release container1
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta13, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta13, true, null, null);
    verify(rmClient).releaseAssignedContainer(eq(container1.getId()));
    eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // Second container allocated, should start ta12.
    taskScheduler.onContainersAllocated(Collections.singletonList(container2));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta12), any(Object.class), eq(container2));
    // ta12 finished, cannot reuse container, should release container2
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta12, container2.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta12, true, null, null);
    verify(rmClient).releaseAssignedContainer(eq(container2.getId()));
    eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
    eventHandler.reset();
    // Third container allocated, should start ta14.
    taskScheduler.onContainersAllocated(Collections.singletonList(container3));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta14), any(Object.class), eq(container3));
    // ta14 finished, should release container3
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta14, container3.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta14, true, null, null);
    verify(rmClient).releaseAssignedContainer(eq(container3.getId()));
    eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
    eventHandler.reset();
    taskScheduler.shutdown();
    taskSchedulerManager.close();
}

Example 14 with ContainerContextMatcher

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

the class TestContainerReuse method testDelayedReuseContainerNotAvailable.

@Test(timeout = 15000l)
public void testDelayedReuseContainerNotAvailable() throws IOException, InterruptedException, ExecutionException {
    LOG.info("Test testDelayedReuseContainerNotAvailable");
    Configuration conf = new Configuration(new YarnConfiguration());
    conf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_ENABLED, true);
    conf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_RACK_FALLBACK_ENABLED, false);
    conf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_NON_LOCAL_FALLBACK_ENABLED, false);
    conf.setLong(TezConfiguration.TEZ_AM_CONTAINER_REUSE_LOCALITY_DELAY_ALLOCATION_MILLIS, 1000l);
    conf.setLong(TezConfiguration.TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MIN_MILLIS, 0);
    conf.setLong(TezConfiguration.TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MAX_MILLIS, 0);
    CapturingEventHandler eventHandler = new CapturingEventHandler();
    TezDAGID dagID = TezDAGID.getInstance("0", 0, 0);
    TezVertexID vertexID = TezVertexID.getInstance(dagID, 1);
    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(conf));
    TaskSchedulerManager taskSchedulerManager = spy(taskSchedulerManagerReal);
    taskSchedulerManager.init(conf);
    taskSchedulerManager.start();
    TaskSchedulerWithDrainableContext taskScheduler = (TaskSchedulerWithDrainableContext) ((TaskSchedulerManagerForTest) taskSchedulerManager).getSpyTaskScheduler();
    TaskSchedulerContextDrainable drainableAppCallback = taskScheduler.getDrainableAppCallback();
    AtomicBoolean drainNotifier = new AtomicBoolean(false);
    taskScheduler.delayedContainerManager.drainedDelayedContainersForTest = drainNotifier;
    Resource resource = Resource.newInstance(1024, 1);
    Priority priority = Priority.newInstance(5);
    String[] host1 = { "host1" };
    String[] host2 = { "host2" };
    String[] defaultRack = { "/default-rack" };
    TezTaskAttemptID taID11 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID, 1), 1);
    TezTaskAttemptID taID21 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID, 2), 1);
    TezTaskAttemptID taID31 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID, 3), 1);
    TaskAttempt ta11 = mock(TaskAttempt.class);
    TaskAttempt ta21 = mock(TaskAttempt.class);
    TaskAttempt ta31 = mock(TaskAttempt.class);
    AMSchedulerEventTALaunchRequest lrTa11 = createLaunchRequestEvent(taID11, ta11, resource, host1, defaultRack, priority);
    AMSchedulerEventTALaunchRequest lrTa21 = createLaunchRequestEvent(taID21, ta21, resource, host2, defaultRack, priority);
    AMSchedulerEventTALaunchRequest lrTa31 = createLaunchRequestEvent(taID31, ta31, resource, host1, defaultRack, priority);
    taskSchedulerManager.handleEvent(lrTa11);
    taskSchedulerManager.handleEvent(lrTa21);
    Container containerHost1 = createContainer(1, host1[0], resource, priority);
    Container containerHost2 = createContainer(2, host2[0], resource, priority);
    taskScheduler.onContainersAllocated(Lists.newArrayList(containerHost1, containerHost2));
    TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
    drainableAppCallback.drain();
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta11), any(Object.class), eq(containerHost1));
    verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta21), any(Object.class), eq(containerHost2));
    // Adding the event later so that task1 assigned to containerHost1 is deterministic.
    taskSchedulerManager.handleEvent(lrTa31);
    taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta21, containerHost2.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
    drainableAppCallback.drain();
    verifyDeAllocateTask(taskScheduler, ta21, true, null, null);
    verify(taskSchedulerManager, times(0)).taskAllocated(eq(0), eq(ta31), any(Object.class), eq(containerHost2));
    verify(rmClient, times(1)).releaseAssignedContainer(eq(containerHost2.getId()));
    eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
    taskScheduler.shutdown();
    taskSchedulerManager.close();
}

Example 15 with ContainerContextMatcher

use of org.apache.tez.dag.app.rm.container.ContainerContextMatcher 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();
}