use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestTaskAttempt method testNoProgressFail.
@Test(timeout = 5000)
public void testNoProgressFail() 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);
TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, mockClock, mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
TezTaskAttemptID taskAttemptID = taImpl.getTaskAttemptID();
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, true)));
// invocations and time updated
assertEquals(100l, taImpl.lastNotifyProgressTimestamp);
when(mockClock.getTime()).thenReturn(150l);
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, true)));
// invocations and time updated
assertEquals(150l, taImpl.lastNotifyProgressTimestamp);
when(mockClock.getTime()).thenReturn(200l);
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
// invocations and time not updated
assertEquals(150l, taImpl.lastNotifyProgressTimestamp);
when(mockClock.getTime()).thenReturn(250l);
taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
// invocations and time not updated
assertEquals(150l, taImpl.lastNotifyProgressTimestamp);
// failed event sent to self
verify(eventHandler, atLeast(1)).handle(arg.capture());
TaskAttemptEventAttemptFailed fEvent = (TaskAttemptEventAttemptFailed) arg.getValue();
assertEquals(taImpl.getTaskAttemptID(), fEvent.getTaskAttemptID());
assertEquals(TaskAttemptTerminationCause.NO_PROGRESS, fEvent.getTerminationCause());
assertEquals(TaskFailureType.NON_FATAL, fEvent.getTaskFailureType());
taImpl.handle(fEvent);
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(TaskAttemptTerminationCause.NO_PROGRESS, taImpl.getTerminationCause());
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestTaskAttempt method testNodeFailedNonLeafVertex.
@Test(timeout = 5000)
public // Ensure node failure on Successful Non-Leaf tasks cause them to be marked as KILLED
void testNodeFailedNonLeafVertex() 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);
MockTaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, new SystemClock(), mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
TezTaskAttemptID taskAttemptID = taImpl.getTaskAttemptID();
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", TaskAttemptState.RUNNING, taImpl.getState());
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", TaskAttemptState.SUCCEEDED, taImpl.getState());
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);
// Send out a Node Failure.
taImpl.handle(new TaskAttemptEventNodeFailed(taskAttemptID, "NodeDecomissioned", TaskAttemptTerminationCause.NODE_FAILED));
// Verify in KILLED state
assertEquals("Task attempt is not in the KILLED state", TaskAttemptState.KILLED, taImpl.getState());
// verify unregister is not invoked again
verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
assertEquals(true, taImpl.inputFailedReported);
// Verify one event to the Task informing it about FAILURE. No events to scheduler. Counter event.
int expectedEventsNodeFailure = expectedEvenstAfterTerminating + 2;
arg = ArgumentCaptor.forClass(Event.class);
verify(eventHandler, times(expectedEventsNodeFailure)).handle(arg.capture());
verifyEventType(arg.getAllValues().subList(expectedEvenstAfterTerminating, expectedEventsNodeFailure), TaskEventTAKilled.class, 1);
// Verify still in KILLED state
assertEquals("Task attempt is not in the KILLED state", TaskAttemptState.KILLED, taImpl.getState());
assertEquals(TaskAttemptTerminationCause.NODE_FAILED, taImpl.getTerminationCause());
assertEquals(TaskAttemptStateInternal.KILLED, taImpl.getInternalState());
taImpl.handle(new TaskAttemptEventTezEventUpdate(taImpl.getTaskAttemptID(), Collections.EMPTY_LIST));
assertFalse("InternalError occurred trying to handle TA_TEZ_EVENT_UPDATE in KILLED state", eventHandler.internalError);
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestTaskAttempt method testNodeFailedLeafVertex.
@Test(timeout = 5000)
public // Ensure node failure on Successful Leaf tasks do not cause them to be marked as KILLED
void testNodeFailedLeafVertex() 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(), true);
TezTaskAttemptID taskAttemptID = taImpl.getTaskAttemptID();
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", TaskAttemptState.RUNNING, taImpl.getState());
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", TaskAttemptState.SUCCEEDED, taImpl.getState());
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);
// Send out a Node Failure.
taImpl.handle(new TaskAttemptEventNodeFailed(taskAttemptID, "NodeDecomissioned", TaskAttemptTerminationCause.NODE_FAILED));
// Verify no additional events
int expectedEventsNodeFailure = expectedEvenstAfterTerminating + 0;
arg = ArgumentCaptor.forClass(Event.class);
verify(eventHandler, times(expectedEventsNodeFailure)).handle(arg.capture());
// Verify still in SUCCEEDED state
assertEquals("Task attempt is not in the SUCCEEDED state", TaskAttemptState.SUCCEEDED, taImpl.getState());
// verify unregister is not invoked again
verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
// error cause remains as default value
assertEquals(TaskAttemptTerminationCause.UNKNOWN_ERROR, taImpl.getTerminationCause());
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestTaskAttempt method testKilledInNew.
@SuppressWarnings("deprecation")
@Test(timeout = 5000)
public void testKilledInNew() 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);
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);
MockTaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, new SystemClock(), mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), true);
Assert.assertEquals(TaskAttemptStateInternal.NEW, taImpl.getInternalState());
taImpl.handle(new TaskAttemptEventKillRequest(taImpl.getTaskAttemptID(), "kill it", TaskAttemptTerminationCause.TERMINATED_BY_CLIENT));
Assert.assertEquals(TaskAttemptStateInternal.KILLED, taImpl.getInternalState());
Assert.assertEquals(0, taImpl.taskAttemptStartedEventLogged);
Assert.assertEquals(1, taImpl.taskAttemptFinishedEventLogged);
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestTaskAttempt method testFailure.
@Test(timeout = 5000)
public void testFailure() 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.getTaskAttemptID();
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.NON_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.getTaskAttemptID(), 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.NON_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);
}
Aggregations