use of org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely in project tez by apache.
the class TestTaskAttempt method testStatusUpdateWithNullCounters.
@Test
public void testStatusUpdateWithNullCounters() 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);
ContainerId contId = ContainerId.newContainerId(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();
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);
TezCounters counters = new TezCounters();
counters.findCounter("group", "counter").increment(1);
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(counters, 0.1f, null, false)));
assertEquals(1, taImpl.getCounters().findCounter("group", "counter").getValue());
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
assertEquals(1, taImpl.getCounters().findCounter("group", "counter").getValue());
counters.findCounter("group", "counter").increment(1);
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(counters, 0.1f, null, false)));
assertEquals(2, taImpl.getCounters().findCounter("group", "counter").getValue());
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
assertEquals(2, taImpl.getCounters().findCounter("group", "counter").getValue());
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely in project tez by apache.
the class TestTaskAttempt method testContainerPreemptedAfterSuccess.
@Test(timeout = 5000)
public // the TaskAttempt
void testContainerPreemptedAfterSuccess() 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 expectedEventsAfterTerminating = expectedEventsAtRunning + 3;
arg = ArgumentCaptor.forClass(Event.class);
verify(eventHandler, times(expectedEventsAfterTerminating)).handle(arg.capture());
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEventsAfterTerminating), TaskEventTASucceeded.class, 1);
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEventsAfterTerminating), AMSchedulerEventTAEnded.class, 1);
verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEventsAfterTerminating), DAGEventCounterUpdate.class, 1);
taImpl.handle(new TaskAttemptEvent(taskAttemptID, TaskAttemptEventType.TA_CONTAINER_TERMINATED_BY_SYSTEM));
// verify unregister is not invoked again
verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
int expectedEventAfterTerminated = expectedEventsAfterTerminating + 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 should be the default value
assertEquals(0, taImpl.getDiagnostics().size());
assertEquals(TaskAttemptTerminationCause.UNKNOWN_ERROR, taImpl.getTerminationCause());
}
use of org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely 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.TaskAttemptEventStartedRemotely 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.TaskAttemptEventStartedRemotely 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());
}
Aggregations