use of org.apache.tez.dag.app.ClusterInfo 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.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, 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.getID(), 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.ClusterInfo in project tez by apache.
the class TestTaskAttempt method testLastDataEventRecording.
@Test(timeout = 5000)
public void testLastDataEventRecording() 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();
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);
long ts1 = 1024;
long ts2 = 2048;
TezTaskAttemptID mockId1 = mock(TezTaskAttemptID.class);
TezTaskAttemptID mockId2 = mock(TezTaskAttemptID.class);
TezEvent mockTezEvent1 = mock(TezEvent.class, RETURNS_DEEP_STUBS);
when(mockTezEvent1.getEventReceivedTime()).thenReturn(ts1);
when(mockTezEvent1.getSourceInfo().getTaskAttemptID()).thenReturn(mockId1);
TezEvent mockTezEvent2 = mock(TezEvent.class, RETURNS_DEEP_STUBS);
when(mockTezEvent2.getEventReceivedTime()).thenReturn(ts2);
when(mockTezEvent2.getSourceInfo().getTaskAttemptID()).thenReturn(mockId2);
TaskAttemptEventStatusUpdate statusEvent = new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false));
assertEquals(0, taImpl.lastDataEvents.size());
taImpl.setLastEventSent(mockTezEvent1);
assertEquals(1, taImpl.lastDataEvents.size());
assertEquals(ts1, taImpl.lastDataEvents.get(0).getTimestamp());
assertEquals(mockId1, taImpl.lastDataEvents.get(0).getTaskAttemptId());
taImpl.handle(statusEvent);
taImpl.setLastEventSent(mockTezEvent2);
assertEquals(1, taImpl.lastDataEvents.size());
assertEquals(ts2, taImpl.lastDataEvents.get(0).getTimestamp());
// over-write earlier value
assertEquals(mockId2, taImpl.lastDataEvents.get(0).getTaskAttemptId());
statusEvent.setReadErrorReported(true);
taImpl.handle(statusEvent);
taImpl.setLastEventSent(mockTezEvent1);
assertEquals(2, taImpl.lastDataEvents.size());
assertEquals(ts1, taImpl.lastDataEvents.get(1).getTimestamp());
// add new event
assertEquals(mockId1, taImpl.lastDataEvents.get(1).getTaskAttemptId());
taImpl.setLastEventSent(mockTezEvent2);
assertEquals(2, taImpl.lastDataEvents.size());
assertEquals(ts2, taImpl.lastDataEvents.get(1).getTimestamp());
// over-write earlier value
assertEquals(mockId2, taImpl.lastDataEvents.get(1).getTaskAttemptId());
}
use of org.apache.tez.dag.app.ClusterInfo 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.getID(), "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.ClusterInfo 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.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", 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());
}
Aggregations