use of org.apache.tez.dag.history.DAGHistoryEvent in project tez by apache.
the class TestTaskAttempt method testMultipleOutputFailed.
@Test(timeout = 5000)
public // TaskAttempt.
void testMultipleOutputFailed() 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 mockEh = new MockEventHandler();
MockEventHandler eventHandler = spy(mockEh);
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();
HistoryEventHandler mockHistHandler = mock(HistoryEventHandler.class);
doReturn(mockHistHandler).when(appCtx).getHistoryHandler();
TaskHeartbeatHandler mockHeartbeatHandler = mock(TaskHeartbeatHandler.class);
MockTaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, new SystemClock(), mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
TezTaskAttemptID taskAttemptID = taImpl.getID();
taImpl.handle(new TaskAttemptEventSchedule(taskAttemptID, 0, 0));
taImpl.handle(new TaskAttemptEventSubmitted(taskAttemptID, contId));
taImpl.handle(new TaskAttemptEventStartedRemotely(taskAttemptID));
verify(mockHeartbeatHandler).register(taskAttemptID);
taImpl.handle(new TaskAttemptEvent(taskAttemptID, TaskAttemptEventType.TA_DONE));
assertEquals("Task attempt is not in succeeded state", taImpl.getState(), TaskAttemptState.SUCCEEDED);
verify(mockHeartbeatHandler).unregister(taskAttemptID);
int expectedEventsTillSucceeded = 8;
ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
ArgumentCaptor<DAGHistoryEvent> histArg = ArgumentCaptor.forClass(DAGHistoryEvent.class);
verify(eventHandler, times(expectedEventsTillSucceeded)).handle(arg.capture());
// start and finish
verify(mockHistHandler, times(2)).handle(histArg.capture());
DAGHistoryEvent histEvent = histArg.getValue();
TaskAttemptFinishedEvent finishEvent = (TaskAttemptFinishedEvent) histEvent.getHistoryEvent();
long finishTime = finishEvent.getFinishTime();
verifyEventType(arg.getAllValues(), TaskEventTAUpdate.class, 2);
InputReadErrorEvent mockReEvent = InputReadErrorEvent.create("", 0, 1);
EventMetaData mockMeta = mock(EventMetaData.class);
TezTaskAttemptID mockDestId1 = mock(TezTaskAttemptID.class);
when(mockMeta.getTaskAttemptID()).thenReturn(mockDestId1);
TezEvent tzEvent = new TezEvent(mockReEvent, mockMeta);
taImpl.handle(new TaskAttemptEventOutputFailed(taskAttemptID, tzEvent, 11));
// failure threshold not met. state is SUCCEEDED
assertEquals("Task attempt is not in succeeded state", taImpl.getState(), TaskAttemptState.SUCCEEDED);
// sending same error again doesnt change anything
taImpl.handle(new TaskAttemptEventOutputFailed(taskAttemptID, tzEvent, 11));
assertEquals("Task attempt is not in succeeded state", taImpl.getState(), TaskAttemptState.SUCCEEDED);
// default value of error cause
assertEquals(TaskAttemptTerminationCause.UNKNOWN_ERROR, taImpl.getTerminationCause());
// different destination attempt reports error. now threshold crossed
TezTaskAttemptID mockDestId2 = mock(TezTaskAttemptID.class);
when(mockMeta.getTaskAttemptID()).thenReturn(mockDestId2);
taImpl.handle(new TaskAttemptEventOutputFailed(taskAttemptID, tzEvent, 11));
assertEquals("Task attempt is not in FAILED state", taImpl.getState(), TaskAttemptState.FAILED);
assertEquals(TaskAttemptTerminationCause.OUTPUT_LOST, taImpl.getTerminationCause());
// verify unregister is not invoked again
verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
verify(mockHistHandler, times(3)).handle(histArg.capture());
histEvent = histArg.getValue();
finishEvent = (TaskAttemptFinishedEvent) histEvent.getHistoryEvent();
assertEquals(TaskFailureType.NON_FATAL, finishEvent.getTaskFailureType());
long newFinishTime = finishEvent.getFinishTime();
Assert.assertEquals(finishTime, newFinishTime);
assertEquals(true, taImpl.inputFailedReported);
int expectedEventsAfterFetchFailure = expectedEventsTillSucceeded + 2;
arg.getAllValues().clear();
verify(eventHandler, times(expectedEventsAfterFetchFailure)).handle(arg.capture());
Event e = verifyEventType(arg.getAllValues().subList(expectedEventsTillSucceeded, expectedEventsAfterFetchFailure), TaskEventTAFailed.class, 1);
TaskEventTAFailed failedEvent = (TaskEventTAFailed) e;
assertEquals(TaskFailureType.NON_FATAL, failedEvent.getTaskFailureType());
taImpl.handle(new TaskAttemptEventOutputFailed(taskAttemptID, tzEvent, 1));
assertEquals("Task attempt is not in FAILED state, still", taImpl.getState(), TaskAttemptState.FAILED);
assertFalse("InternalError occurred trying to handle TA_TOO_MANY_FETCH_FAILURES", eventHandler.internalError);
// No new events.
verify(eventHandler, times(expectedEventsAfterFetchFailure)).handle(arg.capture());
taskConf.setInt(TezConfiguration.TEZ_TASK_MAX_ALLOWED_OUTPUT_FAILURES, 1);
TezTaskID taskID2 = TezTaskID.getInstance(vertexID, 2);
MockTaskAttemptImpl taImpl2 = new MockTaskAttemptImpl(taskID2, 1, eventHandler, taListener, taskConf, new SystemClock(), mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
TezTaskAttemptID taskAttemptID2 = taImpl2.getID();
taImpl2.handle(new TaskAttemptEventSchedule(taskAttemptID2, 0, 0));
taImpl2.handle(new TaskAttemptEventSubmitted(taskAttemptID2, contId));
taImpl2.handle(new TaskAttemptEventStartedRemotely(taskAttemptID2));
verify(mockHeartbeatHandler).register(taskAttemptID2);
taImpl2.handle(new TaskAttemptEvent(taskAttemptID2, TaskAttemptEventType.TA_DONE));
assertEquals("Task attempt is not in succeeded state", taImpl2.getState(), TaskAttemptState.SUCCEEDED);
verify(mockHeartbeatHandler).unregister(taskAttemptID2);
mockReEvent = InputReadErrorEvent.create("", 1, 1);
mockMeta = mock(EventMetaData.class);
mockDestId1 = mock(TezTaskAttemptID.class);
when(mockMeta.getTaskAttemptID()).thenReturn(mockDestId1);
tzEvent = new TezEvent(mockReEvent, mockMeta);
// This should fail even when MAX_ALLOWED_OUTPUT_FAILURES_FRACTION is within limits, as
// MAX_ALLOWED_OUTPUT_FAILURES has crossed the limit.
taImpl2.handle(new TaskAttemptEventOutputFailed(taskAttemptID2, tzEvent, 8));
assertEquals("Task attempt is not in failed state", taImpl2.getState(), TaskAttemptState.FAILED);
assertEquals(TaskAttemptTerminationCause.OUTPUT_LOST, taImpl2.getTerminationCause());
// verify unregister is not invoked again
verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID2);
Clock mockClock = mock(Clock.class);
int readErrorTimespanSec = 1;
taskConf.setInt(TezConfiguration.TEZ_TASK_MAX_ALLOWED_OUTPUT_FAILURES, 10);
taskConf.setInt(TezConfiguration.TEZ_AM_MAX_ALLOWED_TIME_FOR_TASK_READ_ERROR_SEC, readErrorTimespanSec);
TezTaskID taskID3 = TezTaskID.getInstance(vertexID, 3);
MockTaskAttemptImpl taImpl3 = new MockTaskAttemptImpl(taskID3, 1, eventHandler, taListener, taskConf, mockClock, mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
TezTaskAttemptID taskAttemptID3 = taImpl3.getID();
taImpl3.handle(new TaskAttemptEventSchedule(taskAttemptID3, 0, 0));
taImpl3.handle(new TaskAttemptEventSubmitted(taskAttemptID3, contId));
taImpl3.handle(new TaskAttemptEventStartedRemotely(taskAttemptID3));
verify(mockHeartbeatHandler).register(taskAttemptID3);
taImpl3.handle(new TaskAttemptEvent(taskAttemptID3, TaskAttemptEventType.TA_DONE));
assertEquals("Task attempt is not in succeeded state", taImpl3.getState(), TaskAttemptState.SUCCEEDED);
verify(mockHeartbeatHandler).unregister(taskAttemptID3);
mockReEvent = InputReadErrorEvent.create("", 1, 1);
mockMeta = mock(EventMetaData.class);
mockDestId1 = mock(TezTaskAttemptID.class);
when(mockMeta.getTaskAttemptID()).thenReturn(mockDestId1);
tzEvent = new TezEvent(mockReEvent, mockMeta);
when(mockClock.getTime()).thenReturn(1000L);
// time deadline not exceeded for a couple of read error events
taImpl3.handle(new TaskAttemptEventOutputFailed(taskAttemptID3, tzEvent, 1000));
assertEquals("Task attempt is not in succeeded state", taImpl3.getState(), TaskAttemptState.SUCCEEDED);
when(mockClock.getTime()).thenReturn(1500L);
taImpl3.handle(new TaskAttemptEventOutputFailed(taskAttemptID3, tzEvent, 1000));
assertEquals("Task attempt is not in succeeded state", taImpl3.getState(), TaskAttemptState.SUCCEEDED);
// exceed the time threshold
when(mockClock.getTime()).thenReturn(2001L);
taImpl3.handle(new TaskAttemptEventOutputFailed(taskAttemptID3, tzEvent, 1000));
assertEquals("Task attempt is not in FAILED state", taImpl3.getState(), TaskAttemptState.FAILED);
assertEquals(TaskAttemptTerminationCause.OUTPUT_LOST, taImpl3.getTerminationCause());
// verify unregister is not invoked again
verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID3);
}
use of org.apache.tez.dag.history.DAGHistoryEvent in project tez by apache.
the class TestTaskImpl method testTaskSucceedAndRetroActiveFailure.
@SuppressWarnings("rawtypes")
@Test(timeout = 5000)
public void testTaskSucceedAndRetroActiveFailure() {
TezTaskID taskId = getNewTaskID();
scheduleTaskAttempt(taskId);
launchTaskAttempt(mockTask.getLastAttempt().getID());
updateAttemptState(mockTask.getLastAttempt(), TaskAttemptState.RUNNING);
mockTask.handle(createTaskTASucceededEvent(mockTask.getLastAttempt().getID()));
// The task should now have succeeded
assertTaskSucceededState();
verify(mockTask.stateChangeNotifier).taskSucceeded(any(String.class), eq(taskId), eq(mockTask.getLastAttempt().getID().getId()));
ArgumentCaptor<DAGHistoryEvent> argumentCaptor = ArgumentCaptor.forClass(DAGHistoryEvent.class);
verify(mockHistoryHandler).handle(argumentCaptor.capture());
DAGHistoryEvent dagHistoryEvent = argumentCaptor.getValue();
HistoryEvent historyEvent = dagHistoryEvent.getHistoryEvent();
assertTrue(historyEvent instanceof TaskFinishedEvent);
TaskFinishedEvent taskFinishedEvent = (TaskFinishedEvent) historyEvent;
assertEquals(taskFinishedEvent.getFinishTime(), mockTask.getFinishTime());
eventHandler.events.clear();
// Now fail the attempt after it has succeeded
TezTaskAttemptID mockDestId = mock(TezTaskAttemptID.class);
TezEvent mockTezEvent = mock(TezEvent.class);
EventMetaData meta = new EventMetaData(EventProducerConsumerType.INPUT, "Vertex", "Edge", mockDestId);
when(mockTezEvent.getSourceInfo()).thenReturn(meta);
TaskAttemptEventOutputFailed outputFailedEvent = new TaskAttemptEventOutputFailed(mockDestId, mockTezEvent, 1);
mockTask.handle(createTaskTAFailedEvent(mockTask.getLastAttempt().getID(), TaskFailureType.NON_FATAL, outputFailedEvent));
// The task should still be in the scheduled state
assertTaskScheduledState();
Event event = eventHandler.events.get(0);
Assert.assertEquals(AMNodeEventType.N_TA_ENDED, event.getType());
event = eventHandler.events.get(eventHandler.events.size() - 1);
Assert.assertEquals(VertexEventType.V_TASK_RESCHEDULED, event.getType());
// report of output read error should be the causal TA
List<MockTaskAttemptImpl> attempts = mockTask.getAttemptList();
Assert.assertEquals(2, attempts.size());
MockTaskAttemptImpl newAttempt = attempts.get(1);
Assert.assertEquals(mockDestId, newAttempt.getSchedulingCausalTA());
}
use of org.apache.tez.dag.history.DAGHistoryEvent in project tez by apache.
the class TestRecoveryParser method testRecoverableSummary_VertexGroupInCommitting.
@Test(timeout = 5000)
public void testRecoverableSummary_VertexGroupInCommitting() throws IOException {
ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 1);
TezDAGID dagID = TezDAGID.getInstance(appId, 1);
AppContext appContext = mock(AppContext.class);
when(appContext.getCurrentRecoveryDir()).thenReturn(new Path(recoveryPath + "/1"));
when(appContext.getClock()).thenReturn(new SystemClock());
when(mockDAGImpl.getID()).thenReturn(dagID);
when(appContext.getHadoopShim()).thenReturn(new DefaultHadoopShim());
when(appContext.getApplicationID()).thenReturn(appId);
RecoveryService rService = new RecoveryService(appContext);
Configuration conf = new Configuration();
conf.setBoolean(RecoveryService.TEZ_TEST_RECOVERY_DRAIN_EVENTS_WHEN_STOPPED, true);
rService.init(conf);
rService.start();
DAGPlan dagPlan = TestDAGImpl.createTestDAGPlan();
// write a DAGSubmittedEvent first to initialize summaryStream
rService.handle(new DAGHistoryEvent(dagID, new DAGSubmittedEvent(dagID, 1L, dagPlan, ApplicationAttemptId.newInstance(appId, 1), null, "user", new Configuration(), null, null)));
// It should be fine to skip other events, just for testing.
rService.handle(new DAGHistoryEvent(dagID, new VertexGroupCommitStartedEvent(dagID, "group_1", Lists.newArrayList(TezVertexID.getInstance(dagID, 0), TezVertexID.getInstance(dagID, 1)), 0L)));
rService.stop();
DAGRecoveryData dagData = parser.parseRecoveryData();
assertEquals(dagID, dagData.recoveredDagID);
assertTrue(dagData.nonRecoverable);
assertTrue(dagData.reason.contains("Vertex Group Commit was in progress"));
}
use of org.apache.tez.dag.history.DAGHistoryEvent in project tez by apache.
the class TestRecoveryParser method testRecoverableSummary_DAGInCommitting.
@Test(timeout = 5000)
public void testRecoverableSummary_DAGInCommitting() throws IOException {
ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 1);
TezDAGID dagID = TezDAGID.getInstance(appId, 1);
AppContext appContext = mock(AppContext.class);
when(appContext.getCurrentRecoveryDir()).thenReturn(new Path(recoveryPath + "/1"));
when(appContext.getClock()).thenReturn(new SystemClock());
when(mockDAGImpl.getID()).thenReturn(dagID);
RecoveryService rService = new RecoveryService(appContext);
Configuration conf = new Configuration();
conf.setBoolean(RecoveryService.TEZ_TEST_RECOVERY_DRAIN_EVENTS_WHEN_STOPPED, true);
rService.init(conf);
rService.start();
DAGPlan dagPlan = TestDAGImpl.createTestDAGPlan();
// write a DAGSubmittedEvent first to initialize summaryStream
rService.handle(new DAGHistoryEvent(dagID, new DAGSubmittedEvent(dagID, 1L, dagPlan, ApplicationAttemptId.newInstance(appId, 1), null, "user", new Configuration(), null, null)));
// It should be fine to skip other events, just for testing.
rService.handle(new DAGHistoryEvent(dagID, new DAGCommitStartedEvent(dagID, 0L)));
rService.stop();
DAGRecoveryData dagData = parser.parseRecoveryData();
assertEquals(dagID, dagData.recoveredDagID);
assertTrue(dagData.nonRecoverable);
assertTrue(dagData.reason.contains("DAG Commit was in progress"));
}
use of org.apache.tez.dag.history.DAGHistoryEvent in project tez by apache.
the class TestRecoveryParser method testRecoverableNonSummary1.
@Test(timeout = 5000)
public void testRecoverableNonSummary1() throws IOException {
ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 1);
TezDAGID dagID = TezDAGID.getInstance(appId, 1);
AppContext appContext = mock(AppContext.class);
when(appContext.getCurrentRecoveryDir()).thenReturn(new Path(recoveryPath + "/1"));
when(appContext.getClock()).thenReturn(new SystemClock());
when(mockDAGImpl.getID()).thenReturn(dagID);
when(appContext.getHadoopShim()).thenReturn(new DefaultHadoopShim());
when(appContext.getApplicationID()).thenReturn(appId);
// MockRecoveryService will skip the non-summary event
MockRecoveryService rService = new MockRecoveryService(appContext);
Configuration conf = new Configuration();
conf.setBoolean(RecoveryService.TEZ_TEST_RECOVERY_DRAIN_EVENTS_WHEN_STOPPED, true);
rService.init(conf);
rService.start();
DAGPlan dagPlan = TestDAGImpl.createTestDAGPlan();
// write a DAGSubmittedEvent first to initialize summaryStream
rService.handle(new DAGHistoryEvent(dagID, new DAGSubmittedEvent(dagID, 1L, dagPlan, ApplicationAttemptId.newInstance(appId, 1), null, "user", new Configuration(), null, null)));
// It should be fine to skip other events, just for testing.
TezVertexID vertexId = TezVertexID.getInstance(dagID, 0);
rService.handle(new DAGHistoryEvent(dagID, new VertexCommitStartedEvent(vertexId, 0L)));
rService.handle(new DAGHistoryEvent(dagID, new VertexFinishedEvent(vertexId, "v1", 10, 0L, 0L, 0L, 0L, 0L, VertexState.SUCCEEDED, "", null, null, null, null)));
rService.stop();
DAGRecoveryData dagData = parser.parseRecoveryData();
assertTrue(dagData.nonRecoverable);
assertTrue(dagData.reason.contains("Vertex has been committed, but its full recovery events are not seen"));
}
Aggregations