use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestVertexImpl method testVertexTaskAttemptInputFailure.
@SuppressWarnings("deprecation")
@Test(timeout = 5000)
public void testVertexTaskAttemptInputFailure() throws Exception {
initAllVertices(VertexState.INITED);
VertexImpl v = vertices.get("vertex1");
startVertex(v);
dispatcher.await();
TaskAttemptImpl ta = (TaskAttemptImpl) v.getTask(0).getAttempts().values().iterator().next();
ta.handle(new TaskAttemptEventSchedule(ta.getTaskAttemptID(), 2, 2));
NodeId nid = NodeId.newInstance("127.0.0.1", 0);
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(), appContext);
containers.addContainerIfNew(container, 0, 0, 0);
doReturn(containers).when(appContext).getAllContainers();
ta.handle(new TaskAttemptEventSubmitted(ta.getTaskAttemptID(), contId));
ta.handle(new TaskAttemptEventStartedRemotely(ta.getTaskAttemptID()));
Assert.assertEquals(TaskAttemptStateInternal.RUNNING, ta.getInternalState());
ta.handle(new TaskAttemptEventAttemptFailed(ta.getTaskAttemptID(), TaskAttemptEventType.TA_FAILED, TaskFailureType.NON_FATAL, "diag", TaskAttemptTerminationCause.INPUT_READ_ERROR));
dispatcher.await();
Assert.assertEquals(VertexState.RUNNING, v.getState());
Assert.assertEquals(TaskAttemptTerminationCause.INPUT_READ_ERROR, ta.getTerminationCause());
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestContainerReuse method testDelayedReuseContainerBecomesAvailable.
@Test(timeout = 15000l)
public void testDelayedReuseContainerBecomesAvailable() throws IOException, InterruptedException, ExecutionException {
LOG.info("Test testDelayedReuseContainerBecomesAvailable");
Configuration conf = new Configuration();
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, 3000l);
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(conf).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(DAGAppMasterState.RUNNING).when(appContext).getAMState();
doReturn(amNodeTracker).when(appContext).getNodeTracker();
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(ta11, containerHost1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
drainableAppCallback.drain();
verifyDeAllocateTask(taskScheduler, ta11, true, null, null);
verify(taskSchedulerManager, times(1)).taskAllocated(eq(0), eq(ta31), any(Object.class), eq(containerHost1));
verify(rmClient, times(0)).releaseAssignedContainer(eq(containerHost1.getId()));
eventHandler.verifyNoInvocations(AMContainerEventStopRequest.class);
eventHandler.reset();
taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta21, containerHost2.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
long currentTs = System.currentTimeMillis();
Throwable exception = null;
while (System.currentTimeMillis() < currentTs + 5000l) {
try {
verify(taskSchedulerManager, times(1)).containerBeingReleased(eq(0), eq(containerHost2.getId()));
exception = null;
break;
} catch (Throwable e) {
exception = e;
}
}
assertTrue("containerHost2 was not released", exception == null);
taskScheduler.shutdown();
taskSchedulerManager.close();
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestContainerReuse method testReuseAcrossVertices.
@Test(timeout = 30000l)
public void testReuseAcrossVertices() throws IOException, InterruptedException, ExecutionException {
LOG.info("Test testReuseAcrossVertices");
Configuration tezConf = new Configuration();
tezConf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_ENABLED, true);
tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_REUSE_LOCALITY_DELAY_ALLOCATION_MILLIS, 1l);
tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MIN_MILLIS, 20l);
tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MAX_MILLIS, 30l);
tezConf.setInt(TezConfiguration.TEZ_AM_SESSION_MIN_HELD_CONTAINERS, 1);
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(true).when(appContext).isSession();
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(3);
Priority priority2 = Priority.newInstance(4);
TezVertexID vertexID1 = TezVertexID.getInstance(dagID, 1);
TezVertexID vertexID2 = TezVertexID.getInstance(dagID, 2);
// Vertex 1, Task 1, Attempt 1, host1
TezTaskAttemptID taID11 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 1), 1);
TaskAttempt ta11 = mock(TaskAttempt.class);
doReturn(vertexID1).when(ta11).getVertexID();
AMSchedulerEventTALaunchRequest lrEvent11 = createLaunchRequestEvent(taID11, ta11, resource1, host1, racks, priority1);
// Vertex2, Task1, Attempt 1, host1
TezTaskAttemptID taID21 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID2, 1), 1);
TaskAttempt ta21 = mock(TaskAttempt.class);
doReturn(vertexID2).when(ta21).getVertexID();
AMSchedulerEventTALaunchRequest lrEvent21 = createLaunchRequestEvent(taID21, ta21, resource1, host1, racks, priority2);
// Send launch request for task 1 onle, deterministic assignment to this task.
taskSchedulerManager.handleEvent(lrEvent11);
Container container1 = createContainer(1, host1[0], 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));
// Send launch request for task2 (vertex2)
taskSchedulerManager.handleEvent(lrEvent21);
// Task assigned to container completed successfully.
// Container should be assigned to task21.
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(ta21), any(Object.class), eq(container1));
verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
// Task 2 completes.
taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta21, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
LOG.info("Sleeping to ensure that the container has been idled longer " + "than TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MAX_MILLIS ");
Thread.sleep(50l);
// container should not get released due to min held containers
verify(rmClient, times(0)).releaseAssignedContainer(eq(container1.getId()));
taskScheduler.shutdown();
taskSchedulerManager.close();
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestContainerReuse method testSimpleReuse.
@Test(timeout = 10000l)
public void testSimpleReuse() throws IOException, InterruptedException, ExecutionException {
LOG.info("Test testSimpleReuse");
Configuration tezConf = new Configuration();
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);
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, resource1, 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, resource1, host2, racks, priority1);
taskSchedulerManager.handleEvent(lrEvent1);
taskSchedulerManager.handleEvent(lrEvent2);
taskSchedulerManager.handleEvent(lrEvent3);
taskSchedulerManager.handleEvent(lrEvent4);
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));
// Task assigned to container completed successfully. Container should be re-used.
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();
// Task assigned to container completed successfully.
// Verify reuse across hosts.
taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta12, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
drainableAppCallback.drain();
verifyDeAllocateTask(taskScheduler, ta12, 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();
// Verify no re-use if a previous task fails.
taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta13, container1.getId(), TaskAttemptState.FAILED, null, "TIMEOUT", 0));
drainableAppCallback.drain();
verify(taskSchedulerManager, times(0)).taskAllocated(eq(0), eq(ta14), any(Object.class), eq(container1));
verifyDeAllocateTask(taskScheduler, ta13, false, null, "TIMEOUT");
verify(rmClient).releaseAssignedContainer(eq(container1.getId()));
eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
eventHandler.reset();
Container container2 = createContainer(2, "host2", resource1, priority1);
// Second container allocated. Should be allocated to the last task.
taskScheduler.onContainersAllocated(Collections.singletonList(container2));
TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
drainableAppCallback.drain();
verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta14), any(Object.class), eq(container2));
// Task assigned to container completed successfully. No pending requests. Container should be released.
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();
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap 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();
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).getTaskAttemptID();
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).getTaskAttemptID();
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).getTaskAttemptID();
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).getTaskAttemptID();
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).getTaskAttemptID();
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).getTaskAttemptID();
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();
}
Aggregations