use of org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule in project tez by apache.
the class TestTaskAttempt method testRetriesAtSamePriorityConfig.
@Test(timeout = 5000)
public void testRetriesAtSamePriorityConfig() {
// Override the test defaults to setup the config change
TezConfiguration vertexConf = new TezConfiguration();
vertexConf.setBoolean(TezConfiguration.TEZ_AM_TASK_RESCHEDULE_HIGHER_PRIORITY, false);
vertexConf.setBoolean(TezConfiguration.TEZ_AM_TASK_RESCHEDULE_RELAXED_LOCALITY, true);
when(mockVertex.getVertexConfig()).thenReturn(new VertexImpl.VertexConfigImpl(vertexConf));
// set locality
Set<String> hosts = new TreeSet<String>();
hosts.add("host1");
locationHint = TaskLocationHint.createTaskLocationHint(hosts, null);
TaskAttemptImpl.ScheduleTaskattemptTransition sta = new TaskAttemptImpl.ScheduleTaskattemptTransition();
EventHandler eventHandler = mock(EventHandler.class);
TezTaskID taskID = TezTaskID.getInstance(TezVertexID.getInstance(TezDAGID.getInstance("1", 1, 1), 1), 1);
TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, mock(TaskCommunicatorManagerInterface.class), new Configuration(), new SystemClock(), mock(TaskHeartbeatHandler.class), appCtx, false, Resource.newInstance(1024, 1), createFakeContainerContext(), false);
TaskAttemptImpl taImplReScheduled = new MockTaskAttemptImpl(taskID, 1, eventHandler, mock(TaskCommunicatorManagerInterface.class), new Configuration(), new SystemClock(), mock(TaskHeartbeatHandler.class), appCtx, true, Resource.newInstance(1024, 1), createFakeContainerContext(), false);
ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
TaskAttemptEventSchedule sEvent = mock(TaskAttemptEventSchedule.class);
when(sEvent.getPriorityLowLimit()).thenReturn(3);
when(sEvent.getPriorityHighLimit()).thenReturn(1);
// Verify priority for a non-retried attempt
sta.transition(taImpl, sEvent);
verify(eventHandler, times(1)).handle(arg.capture());
AMSchedulerEventTALaunchRequest launchEvent = (AMSchedulerEventTALaunchRequest) arg.getValue();
Assert.assertEquals(2, launchEvent.getPriority());
Assert.assertEquals(1, launchEvent.getLocationHint().getHosts().size());
Assert.assertTrue(launchEvent.getLocationHint().getHosts().contains("host1"));
// Verify priority for a retried attempt is the same
sta.transition(taImplReScheduled, sEvent);
verify(eventHandler, times(2)).handle(arg.capture());
launchEvent = (AMSchedulerEventTALaunchRequest) arg.getValue();
Assert.assertEquals(2, launchEvent.getPriority());
Assert.assertNull(launchEvent.getLocationHint());
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule in project tez by apache.
the class TestTaskAttempt method testFailureFatalError.
@Test(timeout = 5000)
public void testFailureFatalError() throws Exception {
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);
taskConf.setBoolean(TezConfiguration.TEZ_AM_SPECULATION_ENABLED, 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));
taImpl.handle(new TaskAttemptEventStartedRemotely(taskAttemptID));
assertEquals("Task attempt is not in the RUNNING state", taImpl.getState(), TaskAttemptState.RUNNING);
verify(mockHeartbeatHandler).register(taskAttemptID);
int expectedEventsAtRunning = 6;
verify(eventHandler, times(expectedEventsAtRunning)).handle(arg.capture());
verifyEventType(arg.getAllValues().subList(0, expectedEventsAtRunning), SpeculatorEventTaskAttemptStatusUpdate.class, 1);
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
taImpl.handle(new TaskAttemptEventAttemptFailed(taskAttemptID, TaskAttemptEventType.TA_FAILED, TaskFailureType.FATAL, "0", TaskAttemptTerminationCause.APPLICATION_ERROR));
assertEquals("Task attempt is not in the FAIL_IN_PROGRESS state", taImpl.getInternalState(), TaskAttemptStateInternal.FAIL_IN_PROGRESS);
verify(mockHeartbeatHandler).unregister(taskAttemptID);
assertEquals(1, taImpl.getDiagnostics().size());
assertEquals("0", taImpl.getDiagnostics().get(0));
assertEquals(TaskAttemptTerminationCause.APPLICATION_ERROR, taImpl.getTerminationCause());
assertEquals(TaskAttemptStateInternal.FAIL_IN_PROGRESS, taImpl.getInternalState());
taImpl.handle(new TaskAttemptEventTezEventUpdate(taImpl.getID(), Collections.EMPTY_LIST));
assertFalse("InternalError occurred trying to handle TA_TEZ_EVENT_UPDATE in FAIL_IN_PROGRESS state", eventHandler.internalError);
taImpl.handle(new TaskAttemptEventContainerTerminated(contId, taskAttemptID, "1", TaskAttemptTerminationCause.CONTAINER_EXITED));
// verify unregister is not invoked again
verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
assertEquals(2, taImpl.getDiagnostics().size());
assertEquals("1", taImpl.getDiagnostics().get(1));
// err cause does not change
assertEquals(TaskAttemptTerminationCause.APPLICATION_ERROR, taImpl.getTerminationCause());
int expectedEvenstAfterTerminating = expectedEventsAtRunning + 5;
arg = ArgumentCaptor.forClass(Event.class);
verify(eventHandler, times(expectedEvenstAfterTerminating)).handle(arg.capture());
Event e = verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), TaskEventTAFailed.class, 1);
TaskEventTAFailed failedEvent = (TaskEventTAFailed) e;
assertEquals(TaskFailureType.FATAL, failedEvent.getTaskFailureType());
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), AMSchedulerEventTAEnded.class, 1);
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), DAGEventCounterUpdate.class, 1);
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), SpeculatorEventTaskAttemptStatusUpdate.class, 2);
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule in project tez by apache.
the class TestTaskAttempt method testContainerTerminatedAfterSuccess.
@Test(timeout = 5000)
public // the TaskAttempt
void testContainerTerminatedAfterSuccess() throws Exception {
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));
taImpl.handle(new TaskAttemptEventStartedRemotely(taskAttemptID));
assertEquals("Task attempt is not in the RUNNING state", taImpl.getState(), TaskAttemptState.RUNNING);
verify(mockHeartbeatHandler).register(taskAttemptID);
int expectedEventsAtRunning = 5;
verify(eventHandler, times(expectedEventsAtRunning)).handle(arg.capture());
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 = expectedEventsAtRunning + 3;
arg = ArgumentCaptor.forClass(Event.class);
verify(eventHandler, times(expectedEvenstAfterTerminating)).handle(arg.capture());
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), TaskEventTASucceeded.class, 1);
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), AMSchedulerEventTAEnded.class, 1);
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), DAGEventCounterUpdate.class, 1);
taImpl.handle(new TaskAttemptEventContainerTerminated(contId, taskAttemptID, "Terminated", TaskAttemptTerminationCause.CONTAINER_EXITED));
// verify unregister is not invoked again
verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
int expectedEventAfterTerminated = expectedEvenstAfterTerminating + 0;
arg = ArgumentCaptor.forClass(Event.class);
verify(eventHandler, times(expectedEventAfterTerminated)).handle(arg.capture());
// Verify that the diagnostic message included in the Terminated event is not
// captured - TA already succeeded. Error cause is the default value.
assertEquals(0, taImpl.getDiagnostics().size());
assertEquals(TaskAttemptTerminationCause.UNKNOWN_ERROR, taImpl.getTerminationCause());
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule in project tez by apache.
the class TestTaskAttempt method testLocalityRequest.
@Test(timeout = 5000)
public void testLocalityRequest() {
TaskAttemptImpl.ScheduleTaskattemptTransition sta = new TaskAttemptImpl.ScheduleTaskattemptTransition();
EventHandler eventHandler = mock(EventHandler.class);
Set<String> hosts = new TreeSet<String>();
hosts.add("host1");
hosts.add("host2");
hosts.add("host3");
locationHint = TaskLocationHint.createTaskLocationHint(hosts, null);
TezTaskID taskID = TezTaskID.getInstance(TezVertexID.getInstance(TezDAGID.getInstance("1", 1, 1), 1), 1);
TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, mock(TaskCommunicatorManagerInterface.class), new Configuration(), new SystemClock(), mock(TaskHeartbeatHandler.class), appCtx, false, Resource.newInstance(1024, 1), createFakeContainerContext(), false);
TaskAttemptEventSchedule sEvent = mock(TaskAttemptEventSchedule.class);
sta.transition(taImpl, sEvent);
ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
verify(eventHandler, times(1)).handle(arg.capture());
if (!(arg.getAllValues().get(0) instanceof AMSchedulerEventTALaunchRequest)) {
fail("Second event not of type " + AMSchedulerEventTALaunchRequest.class.getName());
}
// TODO Move the Rack request check to the client after TEZ-125 is fixed.
Set<String> requestedRacks = taImpl.taskRacks;
assertEquals(1, requestedRacks.size());
assertEquals(3, taImpl.taskHosts.size());
for (int i = 0; i < 3; i++) {
String host = ("host" + (i + 1));
assertEquals(host, true, taImpl.taskHosts.contains(host));
}
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule 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);
}
Aggregations