use of org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed in project tez by apache.
the class TaskCommunicatorManager method heartbeat.
public TaskHeartbeatResponse heartbeat(TaskHeartbeatRequest request) throws IOException, TezException {
ContainerId containerId = ConverterUtils.toContainerId(request.getContainerIdentifier());
if (LOG.isDebugEnabled()) {
LOG.debug("Received heartbeat from container" + ", request=" + request);
}
if (!registeredContainers.containsKey(containerId)) {
LOG.warn("Received task heartbeat from unknown container with id: " + containerId + ", asking it to die");
return RESPONSE_SHOULD_DIE;
}
// A heartbeat can come in anytime. The AM may have made a decision to kill a running task/container
// meanwhile. If the decision is processed through the pipeline before the heartbeat is processed,
// the heartbeat will be dropped. Otherwise the heartbeat will be processed - and the system
// know how to handle this - via FailedInputEvents for example (relevant only if the heartbeat has events).
// So - avoiding synchronization.
pingContainerHeartbeatHandler(containerId);
TaskAttemptEventInfo eventInfo = new TaskAttemptEventInfo(0, null, 0);
TezTaskAttemptID taskAttemptID = request.getTaskAttemptId();
if (taskAttemptID != null) {
ContainerId containerIdFromMap = registeredAttempts.get(taskAttemptID);
if (containerIdFromMap == null || !containerIdFromMap.equals(containerId)) {
// This can happen when a task heartbeats. Meanwhile the container is unregistered.
// The information will eventually make it through to the plugin via a corresponding unregister.
// There's a race in that case between the unregister making it through, and this method returning.
// TODO TEZ-2003 (post) TEZ-2666. An exception back is likely a better approach than sending a shouldDie = true,
// so that the plugin can handle the scenario. Alternately augment the response with error codes.
// Error codes would be better than exceptions.
LOG.info("Attempt: " + taskAttemptID + " is not recognized for heartbeats");
return RESPONSE_SHOULD_DIE;
}
List<TezEvent> inEvents = request.getEvents();
if (LOG.isDebugEnabled()) {
LOG.debug("Ping from " + taskAttemptID.toString() + " events: " + (inEvents != null ? inEvents.size() : -1));
}
long currTime = context.getClock().getTime();
// taFinishedEvents - means the TaskAttemptFinishedEvent
// taGeneratedEvents - for recovery, means the events generated by this task attempt and is needed by its downstream vertices
// eventsForVertex - including all the taGeneratedEvents and other events such as INPUT_READ_ERROR_EVENT/INPUT_FAILED_EVENT
// taGeneratedEvents is routed both to TaskAttempt & Vertex. Route to Vertex is for performance consideration
// taFinishedEvents must be routed before taGeneratedEvents
List<TezEvent> taFinishedEvents = new ArrayList<TezEvent>();
List<TezEvent> taGeneratedEvents = new ArrayList<TezEvent>();
List<TezEvent> eventsForVertex = new ArrayList<TezEvent>();
TaskAttemptEventStatusUpdate taskAttemptEvent = null;
boolean readErrorReported = false;
for (TezEvent tezEvent : ListUtils.emptyIfNull(inEvents)) {
// for now, set the event time on the AM when it is received.
// this avoids any time disparity between machines.
tezEvent.setEventReceivedTime(currTime);
final EventType eventType = tezEvent.getEventType();
if (eventType == EventType.TASK_STATUS_UPDATE_EVENT) {
// send TA_STATUS_UPDATE before TA_DONE/TA_FAILED/TA_KILLED otherwise Status may be missed
taskAttemptEvent = new TaskAttemptEventStatusUpdate(taskAttemptID, (TaskStatusUpdateEvent) tezEvent.getEvent());
} else if (eventType == EventType.TASK_ATTEMPT_COMPLETED_EVENT || eventType == EventType.TASK_ATTEMPT_FAILED_EVENT || eventType == EventType.TASK_ATTEMPT_KILLED_EVENT) {
taFinishedEvents.add(tezEvent);
} else {
if (eventType == EventType.INPUT_READ_ERROR_EVENT) {
readErrorReported = true;
}
if (eventType == EventType.DATA_MOVEMENT_EVENT || eventType == EventType.COMPOSITE_DATA_MOVEMENT_EVENT || eventType == EventType.ROOT_INPUT_INITIALIZER_EVENT || eventType == EventType.VERTEX_MANAGER_EVENT) {
taGeneratedEvents.add(tezEvent);
}
eventsForVertex.add(tezEvent);
}
}
if (taskAttemptEvent != null) {
taskAttemptEvent.setReadErrorReported(readErrorReported);
sendEvent(taskAttemptEvent);
}
// route taGeneratedEvents to TaskAttempt
if (!taGeneratedEvents.isEmpty()) {
sendEvent(new TaskAttemptEventTezEventUpdate(taskAttemptID, taGeneratedEvents));
}
// route events to TaskAttempt
Preconditions.checkArgument(taFinishedEvents.size() <= 1, "Multiple TaskAttemptFinishedEvent");
for (TezEvent e : taFinishedEvents) {
EventMetaData sourceMeta = e.getSourceInfo();
switch(e.getEventType()) {
case TASK_ATTEMPT_FAILED_EVENT:
case TASK_ATTEMPT_KILLED_EVENT:
TaskAttemptTerminationCause errCause = null;
switch(sourceMeta.getEventGenerator()) {
case INPUT:
errCause = TaskAttemptTerminationCause.INPUT_READ_ERROR;
break;
case PROCESSOR:
errCause = TaskAttemptTerminationCause.APPLICATION_ERROR;
break;
case OUTPUT:
errCause = TaskAttemptTerminationCause.OUTPUT_WRITE_ERROR;
break;
case SYSTEM:
errCause = TaskAttemptTerminationCause.FRAMEWORK_ERROR;
break;
default:
throw new TezUncheckedException("Unknown EventProducerConsumerType: " + sourceMeta.getEventGenerator());
}
if (e.getEventType() == EventType.TASK_ATTEMPT_FAILED_EVENT) {
TaskAttemptFailedEvent taskFailedEvent = (TaskAttemptFailedEvent) e.getEvent();
sendEvent(new TaskAttemptEventAttemptFailed(sourceMeta.getTaskAttemptID(), TaskAttemptEventType.TA_FAILED, taskFailedEvent.getTaskFailureType(), "Error: " + taskFailedEvent.getDiagnostics(), errCause));
} else {
// Killed
TaskAttemptKilledEvent taskKilledEvent = (TaskAttemptKilledEvent) e.getEvent();
sendEvent(new TaskAttemptEventAttemptKilled(sourceMeta.getTaskAttemptID(), "Error: " + taskKilledEvent.getDiagnostics(), errCause));
}
break;
case TASK_ATTEMPT_COMPLETED_EVENT:
sendEvent(new TaskAttemptEvent(sourceMeta.getTaskAttemptID(), TaskAttemptEventType.TA_DONE));
break;
default:
throw new TezUncheckedException("Unhandled tez event type: " + e.getEventType());
}
}
if (!eventsForVertex.isEmpty()) {
TezVertexID vertexId = taskAttemptID.getTaskID().getVertexID();
sendEvent(new VertexEventRouteEvent(vertexId, Collections.unmodifiableList(eventsForVertex)));
}
taskHeartbeatHandler.pinged(taskAttemptID);
eventInfo = context.getCurrentDAG().getVertex(taskAttemptID.getTaskID().getVertexID()).getTaskAttemptTezEvents(taskAttemptID, request.getStartIndex(), request.getPreRoutedStartIndex(), request.getMaxEvents());
}
return new TaskHeartbeatResponse(false, eventInfo.getEvents(), eventInfo.getNextFromEventId(), eventInfo.getNextPreRoutedFromEventId());
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed in project tez by apache.
the class TestVertexImpl method testVertexTaskAttemptOutputFailure.
@SuppressWarnings("deprecation")
@Test(timeout = 5000)
public void testVertexTaskAttemptOutputFailure() 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.getID(), 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.getID(), contId));
ta.handle(new TaskAttemptEventStartedRemotely(ta.getID()));
Assert.assertEquals(TaskAttemptStateInternal.RUNNING, ta.getInternalState());
ta.handle(new TaskAttemptEventAttemptFailed(ta.getID(), TaskAttemptEventType.TA_FAILED, TaskFailureType.NON_FATAL, "diag", TaskAttemptTerminationCause.OUTPUT_WRITE_ERROR));
dispatcher.await();
Assert.assertEquals(VertexState.RUNNING, v.getState());
Assert.assertEquals(TaskAttemptTerminationCause.OUTPUT_WRITE_ERROR, ta.getTerminationCause());
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed in project tez by apache.
the class TestTaskAttempt method testProgressTimeStampUpdate.
@Test
public void testProgressTimeStampUpdate() 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.setLong(TezConfiguration.TEZ_TASK_PROGRESS_STUCK_INTERVAL_MS, 75);
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);
Clock mockClock = mock(Clock.class);
when(mockClock.getTime()).thenReturn(50l);
TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, mockClock, 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);
when(mockClock.getTime()).thenReturn(100l);
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
verify(eventHandler, atLeast(1)).handle(arg.capture());
if (arg.getValue() instanceof TaskAttemptEventAttemptFailed) {
TaskAttemptEventAttemptFailed fEvent = (TaskAttemptEventAttemptFailed) arg.getValue();
assertEquals(taImpl.getID(), fEvent.getTaskAttemptID());
assertEquals(TaskAttemptTerminationCause.NO_PROGRESS, fEvent.getTerminationCause());
taImpl.handle(fEvent);
fail("Should not fail since the timestamps do not differ by progress interval config");
} else {
Assert.assertEquals("Task Attempt's internal state should be RUNNING!", taImpl.getInternalState(), TaskAttemptStateInternal.RUNNING);
}
when(mockClock.getTime()).thenReturn(200l);
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
verify(eventHandler, atLeast(1)).handle(arg.capture());
Assert.assertTrue("This should have been an attempt failed event!", arg.getValue() instanceof TaskAttemptEventAttemptFailed);
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed 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.TaskAttemptEventAttemptFailed in project tez by apache.
the class TestTaskImpl method testSucceededLeafTaskWithRetroFailures.
@SuppressWarnings("rawtypes")
@Test(timeout = 10000L)
public void testSucceededLeafTaskWithRetroFailures() throws InterruptedException {
Configuration newConf = new Configuration(conf);
newConf.setInt(TezConfiguration.TEZ_AM_TASK_MAX_FAILED_ATTEMPTS, 1);
Vertex vertex = mock(Vertex.class);
doReturn(new VertexImpl.VertexConfigImpl(newConf)).when(vertex).getVertexConfig();
mockTask = new MockTaskImpl(vertexId, partition, eventHandler, conf, taskCommunicatorManagerInterface, clock, taskHeartbeatHandler, appContext, true, taskResource, containerContext, vertex);
TezTaskID taskId = getNewTaskID();
scheduleTaskAttempt(taskId);
MockTaskAttemptImpl firstMockTaskAttempt = mockTask.getAttemptList().get(0);
launchTaskAttempt(firstMockTaskAttempt.getID());
mockTask.handle(createTaskTAAddSpecAttempt(mockTask.getLastAttempt().getID()));
MockTaskAttemptImpl secondMockTaskAttempt = mockTask.getAttemptList().get(1);
launchTaskAttempt(secondMockTaskAttempt.getID());
firstMockTaskAttempt.handle(new TaskAttemptEventSchedule(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString()), 10, 10));
secondMockTaskAttempt.handle(new TaskAttemptEventSchedule(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString()), 10, 10));
firstMockTaskAttempt.handle(new TaskAttemptEventSubmitted(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString()), mockContainer.getId()));
secondMockTaskAttempt.handle(new TaskAttemptEventSubmitted(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString()), mockContainer.getId()));
secondMockTaskAttempt.handle(new TaskAttemptEventStartedRemotely(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString())));
firstMockTaskAttempt.handle(new TaskAttemptEventStartedRemotely(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString())));
secondMockTaskAttempt.handle(new TaskAttemptEvent(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString()), TaskAttemptEventType.TA_DONE));
firstMockTaskAttempt.handle(new TaskAttemptEventAttemptFailed(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString()), TaskAttemptEventType.TA_FAILED, TaskFailureType.NON_FATAL, "test", TaskAttemptTerminationCause.CONTAINER_EXITED));
mockTask.handle(new TaskEventTASucceeded(secondMockTaskAttempt.getID()));
firstMockTaskAttempt.handle(new TaskAttemptEventContainerTerminated(mockContainerId, firstMockTaskAttempt.getID(), "test", TaskAttemptTerminationCause.NO_PROGRESS));
InputReadErrorEvent mockReEvent = InputReadErrorEvent.create("", 0, 0);
TezTaskAttemptID mockDestId = firstMockTaskAttempt.getID();
EventMetaData meta = new EventMetaData(EventProducerConsumerType.INPUT, "Vertex", "Edge", mockDestId);
TezEvent tzEvent = new TezEvent(mockReEvent, meta);
TaskAttemptEventOutputFailed outputFailedEvent = new TaskAttemptEventOutputFailed(mockDestId, tzEvent, 1);
firstMockTaskAttempt.handle(outputFailedEvent);
mockTask.handle(new TaskEventTAFailed(firstMockTaskAttempt.getID(), TaskFailureType.NON_FATAL, mock(TaskAttemptEvent.class)));
Assert.assertEquals(mockTask.getInternalState(), TaskStateInternal.SUCCEEDED);
}
Aggregations