List of usage examples for org.apache.hadoop.yarn.api.records Container getNodeId
@Public @Stable public abstract NodeId getNodeId();
From source file:org.apache.tez.dag.app.dag.impl.TestTaskAttempt.java
License:Apache License
@Test(timeout = 5000) // Ensure node failure on Successful Leaf tasks do not cause them to be marked as KILLED public 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()); TaskAttemptListener taListener = mock(TaskAttemptListener.class); when(taListener.getAddress()).thenReturn(new InetSocketAddress("localhost", 0)); Configuration taskConf = new Configuration(); taskConf.setClass("fs.file.impl", StubbedFS.class, FileSystem.class); taskConf.setBoolean("fs.file.impl.disable.cache", true); TaskLocationHint 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"); AppContext appCtx = mock(AppContext.class); AMContainerMap containers = new AMContainerMap(mock(ContainerHeartbeatHandler.class), mock(TaskAttemptListener.class), new ContainerContextMatcher(), appCtx); containers.addContainerIfNew(container); 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, locationHint, false, resource, createFakeContainerContext(), true); TezTaskAttemptID taskAttemptID = taImpl.getID(); ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class); taImpl.handle(new TaskAttemptEventSchedule(taskAttemptID, 0, 0)); // At state STARTING. taImpl.handle(new TaskAttemptEventStartedRemotely(taskAttemptID, contId, null)); assertEquals("Task attempt is not in the RUNNING state", TaskAttemptState.RUNNING, taImpl.getState()); verify(mockHeartbeatHandler).register(taskAttemptID); int expectedEventsAtRunning = 3; 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), TaskEventTAUpdate.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()); }
From source file:org.apache.tez.dag.app.dag.impl.TestTaskAttempt.java
License:Apache License
@Test(timeout = 5000) // Verifies that multiple TooManyFetchFailures are handled correctly by the // TaskAttempt./*w w w .j a v a 2 s . c om*/ public 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); TaskAttemptListener taListener = mock(TaskAttemptListener.class); when(taListener.getAddress()).thenReturn(new InetSocketAddress("localhost", 0)); Configuration taskConf = new Configuration(); taskConf.setClass("fs.file.impl", StubbedFS.class, FileSystem.class); taskConf.setBoolean("fs.file.impl.disable.cache", true); TaskLocationHint 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"); AppContext appCtx = mock(AppContext.class); AMContainerMap containers = new AMContainerMap(mock(ContainerHeartbeatHandler.class), mock(TaskAttemptListener.class), new ContainerContextMatcher(), appCtx); containers.addContainerIfNew(container); 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, locationHint, false, resource, createFakeContainerContext(), false); TezTaskAttemptID taskAttemptID = taImpl.getID(); taImpl.handle(new TaskAttemptEventSchedule(taskAttemptID, 0, 0)); // At state STARTING. taImpl.handle(new TaskAttemptEventStartedRemotely(taskAttemptID, contId, null)); 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 = 6; ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class); verify(eventHandler, times(expectedEventsTillSucceeded)).handle(arg.capture()); 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, 4)); // 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, 4)); 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, 4)); 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); assertEquals(true, taImpl.inputFailedReported); int expectedEventsAfterFetchFailure = expectedEventsTillSucceeded + 2; arg.getAllValues().clear(); verify(eventHandler, times(expectedEventsAfterFetchFailure)).handle(arg.capture()); verifyEventType(arg.getAllValues().subList(expectedEventsTillSucceeded, expectedEventsAfterFetchFailure), TaskEventTAUpdate.class, 1); 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()); }
From source file:org.apache.tez.dag.app.dag.impl.TestVertexImpl.java
License:Apache License
@SuppressWarnings("unchecked") @Test(timeout = 5000)/*from www .j a va 2 s .c o m*/ public void testVertexTaskAttemptProcessorFailure() 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(TaskAttemptListener.class), new ContainerContextMatcher(), appContext); containers.addContainerIfNew(container); doReturn(containers).when(appContext).getAllContainers(); ta.handle(new TaskAttemptEventStartedRemotely(ta.getID(), contId, null)); Assert.assertEquals(TaskAttemptStateInternal.RUNNING, ta.getInternalState()); dispatcher.getEventHandler() .handle(new VertexEventRouteEvent(v.getVertexId(), Collections.singletonList(new TezEvent( new TaskAttemptFailedEvent("Failed"), new EventMetaData(EventProducerConsumerType.PROCESSOR, v.getName(), null, ta.getID()))))); dispatcher.await(); Assert.assertEquals(VertexState.RUNNING, v.getState()); Assert.assertEquals(TaskAttemptTerminationCause.APPLICATION_ERROR, ta.getTerminationCause()); }
From source file:org.apache.tez.dag.app.dag.impl.TestVertexImpl.java
License:Apache License
@SuppressWarnings("unchecked") @Test(timeout = 5000)//from w w w. j av a 2 s.co m public void testVertexTaskAttemptInputFailure() 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(TaskAttemptListener.class), new ContainerContextMatcher(), appContext); containers.addContainerIfNew(container); doReturn(containers).when(appContext).getAllContainers(); ta.handle(new TaskAttemptEventStartedRemotely(ta.getID(), contId, null)); Assert.assertEquals(TaskAttemptStateInternal.RUNNING, ta.getInternalState()); dispatcher.getEventHandler() .handle(new VertexEventRouteEvent(v.getVertexId(), Collections.singletonList(new TezEvent( new TaskAttemptFailedEvent("Failed"), new EventMetaData(EventProducerConsumerType.INPUT, v.getName(), null, ta.getID()))))); dispatcher.await(); Assert.assertEquals(VertexState.RUNNING, v.getState()); Assert.assertEquals(TaskAttemptTerminationCause.INPUT_READ_ERROR, ta.getTerminationCause()); }
From source file:org.apache.tez.dag.app.dag.impl.TestVertexImpl.java
License:Apache License
@SuppressWarnings("unchecked") @Test(timeout = 5000)/*from ww w . j a v a2s. c o m*/ 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(TaskAttemptListener.class), new ContainerContextMatcher(), appContext); containers.addContainerIfNew(container); doReturn(containers).when(appContext).getAllContainers(); ta.handle(new TaskAttemptEventStartedRemotely(ta.getID(), contId, null)); Assert.assertEquals(TaskAttemptStateInternal.RUNNING, ta.getInternalState()); dispatcher.getEventHandler() .handle(new VertexEventRouteEvent(v.getVertexId(), Collections.singletonList(new TezEvent( new TaskAttemptFailedEvent("Failed"), new EventMetaData(EventProducerConsumerType.OUTPUT, v.getName(), null, ta.getID()))))); dispatcher.await(); Assert.assertEquals(VertexState.RUNNING, v.getState()); Assert.assertEquals(TaskAttemptTerminationCause.OUTPUT_WRITE_ERROR, ta.getTerminationCause()); }
From source file:org.apache.tez.dag.app.rm.ContainerLauncherLaunchRequestEvent.java
License:Apache License
public ContainerLauncherLaunchRequestEvent(ContainerLaunchContext clc, Container container, int launcherId, int schedulerId, int taskCommId) { super(container.getId(), container.getNodeId(), container.getContainerToken(), ContainerLauncherEventType.CONTAINER_LAUNCH_REQUEST, launcherId, schedulerId, taskCommId); this.clc = clc; this.container = container; }
From source file:org.apache.tez.dag.app.rm.DagAwareYarnTaskScheduler.java
License:Apache License
/** * Inform the app about a task assignment. This should not be called with * any locks held.// w ww . j a va2 s.co m * * @param request the corresponding task request * @param container the container assigned to the task */ private void informAppAboutAssignment(TaskRequest request, Container container) { if (blacklistedNodes.contains(container.getNodeId())) { Object task = request.getTask(); LOG.info("Container {} allocated for task {} on blacklisted node {}", container.getId(), container.getNodeId(), task); deallocateContainer(container.getId()); // its ok to submit the same request again because the RM will not give us // the bad/unhealthy nodes again. The nodes may become healthy/unblacklisted // and so its better to give the RM the full information. allocateTask(task, request.getCapability(), (request.getNodes() == null ? null : request.getNodes().toArray(new String[request.getNodes().size()])), (request.getRacks() == null ? null : request.getRacks().toArray(new String[request.getRacks().size()])), request.getPriority(), request.getContainerSignature(), request.getCookie()); } else { getContext().taskAllocated(request.getTask(), request.getCookie(), container); } }
From source file:org.apache.tez.dag.app.rm.NMCommunicatorLaunchRequestEvent.java
License:Apache License
public NMCommunicatorLaunchRequestEvent(ContainerLaunchContext clc, Container container) { super(container.getId(), container.getNodeId(), container.getContainerToken(), NMCommunicatorEventType.CONTAINER_LAUNCH_REQUEST); this.clc = clc; this.container = container; }
From source file:org.apache.tez.dag.app.rm.TaskScheduler.java
License:Apache License
@Override public void onContainersAllocated(List<Container> containers) { if (isStopped) { return;//from ww w. j av a 2s . c o m } Map<CookieContainerRequest, Container> appContainers = new HashMap<CookieContainerRequest, Container>( containers.size()); synchronized (this) { for (Container container : containers) { String location = container.getNodeId().getHost(); CookieContainerRequest assigned = getMatchingRequest(container, location); if (assigned == null) { location = RackResolver.resolve(location).getNetworkLocation(); assigned = getMatchingRequest(container, location); } if (assigned == null) { location = ResourceRequest.ANY; assigned = getMatchingRequest(container, location); } if (assigned == null) { // not matched anything. release container // Probably we cancelled a request and RM allocated that to us // before RM heard of the cancellation releaseContainer(container.getId(), null); LOG.info("No RM requests matching container: " + container); continue; } Object task = getTask(assigned); assert task != null; assignContainer(task, container, assigned); appContainers.put(assigned, container); LOG.info("Assigning container: " + container + " for task: " + task + " at locality: " + location + " resource memory: " + container.getResource().getMemory() + " cpu: " + container.getResource().getVirtualCores()); } } // upcall to app must be outside locks for (Entry<CookieContainerRequest, Container> entry : appContainers.entrySet()) { CookieContainerRequest assigned = entry.getKey(); appClient.taskAllocated(getTask(assigned), assigned.getCookie().appCookie, entry.getValue()); } }
From source file:org.apache.tez.dag.app.rm.TaskSchedulerEventHandler.java
License:Apache License
@Override public synchronized void taskAllocated(Object task, Object appCookie, Container container) { ContainerId containerId = container.getId(); if (appContext.getAllContainers().addContainerIfNew(container)) { appContext.getNodeTracker().nodeSeen(container.getNodeId()); sendEvent(new AMNodeEventContainerAllocated(container.getNodeId(), container.getId())); }// ww w . jav a2s. c o m AMSchedulerEventTALaunchRequest event = (AMSchedulerEventTALaunchRequest) appCookie; TaskAttempt taskAttempt = event.getTaskAttempt(); // TODO - perhaps check if the task still needs this container // because the deallocateTask downcall may have raced with the // taskAllocated() upcall assert task.equals(taskAttempt); if (appContext.getAllContainers().get(containerId).getState() == AMContainerState.ALLOCATED) { sendEvent(new AMContainerEventLaunchRequest(containerId, taskAttempt.getVertexID(), event.getContainerContext())); } sendEvent(new DAGEventSchedulerUpdateTAAssigned(taskAttempt, container)); sendEvent(new AMContainerEventAssignTA(containerId, taskAttempt.getID(), event.getRemoteTaskSpec(), event.getContainerContext().getLocalResources(), event.getContainerContext().getCredentials())); }