Search in sources :

Example 16 with TaskHeartbeatHandler

use of org.apache.tez.dag.app.TaskHeartbeatHandler 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);
}
Also used : Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) SystemClock(org.apache.hadoop.yarn.util.SystemClock) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) TaskCommunicatorManagerInterface(org.apache.tez.dag.app.TaskCommunicatorManagerInterface) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) ContainerContextMatcher(org.apache.tez.dag.app.rm.container.ContainerContextMatcher) TezTaskID(org.apache.tez.dag.records.TezTaskID) Container(org.apache.hadoop.yarn.api.records.Container) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) TezDAGID(org.apache.tez.dag.records.TezDAGID) NodeId(org.apache.hadoop.yarn.api.records.NodeId) TaskHeartbeatHandler(org.apache.tez.dag.app.TaskHeartbeatHandler) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) TezVertexID(org.apache.tez.dag.records.TezVertexID) ContainerHeartbeatHandler(org.apache.tez.dag.app.ContainerHeartbeatHandler) HashSet(java.util.HashSet) TaskAttemptEventKillRequest(org.apache.tez.dag.app.dag.event.TaskAttemptEventKillRequest) Test(org.junit.Test)

Example 17 with TaskHeartbeatHandler

use of org.apache.tez.dag.app.TaskHeartbeatHandler 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());
}
Also used : Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) Container(org.apache.hadoop.yarn.api.records.Container) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) TezDAGID(org.apache.tez.dag.records.TezDAGID) TaskHeartbeatHandler(org.apache.tez.dag.app.TaskHeartbeatHandler) TaskAttemptEventSchedule(org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule) TezVertexID(org.apache.tez.dag.records.TezVertexID) ContainerHeartbeatHandler(org.apache.tez.dag.app.ContainerHeartbeatHandler) HashSet(java.util.HashSet) TaskAttemptEventStartedRemotely(org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely) SystemClock(org.apache.hadoop.yarn.util.SystemClock) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) TaskCommunicatorManagerInterface(org.apache.tez.dag.app.TaskCommunicatorManagerInterface) ContainerContextMatcher(org.apache.tez.dag.app.rm.container.ContainerContextMatcher) TaskAttemptEventSubmitted(org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted) TaskAttemptEventNodeFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventNodeFailed) TezTaskID(org.apache.tez.dag.records.TezTaskID) TaskLocationHint(org.apache.tez.dag.api.TaskLocationHint) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) NodeId(org.apache.hadoop.yarn.api.records.NodeId) TaskStatusUpdateEvent(org.apache.tez.runtime.api.events.TaskStatusUpdateEvent) TaskAttemptFinishedEvent(org.apache.tez.dag.history.events.TaskAttemptFinishedEvent) TaskEvent(org.apache.tez.dag.app.dag.event.TaskEvent) DAGEvent(org.apache.tez.dag.app.dag.event.DAGEvent) InputReadErrorEvent(org.apache.tez.runtime.api.events.InputReadErrorEvent) TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) Event(org.apache.hadoop.yarn.event.Event) DAGHistoryEvent(org.apache.tez.dag.history.DAGHistoryEvent) TezEvent(org.apache.tez.runtime.api.impl.TezEvent) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) Test(org.junit.Test)

Aggregations

Configuration (org.apache.hadoop.conf.Configuration)17 Container (org.apache.hadoop.yarn.api.records.Container)17 ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)17 LocalResource (org.apache.hadoop.yarn.api.records.LocalResource)17 NodeId (org.apache.hadoop.yarn.api.records.NodeId)17 SystemClock (org.apache.hadoop.yarn.util.SystemClock)17 TezConfiguration (org.apache.tez.dag.api.TezConfiguration)17 TaskCommunicatorManagerInterface (org.apache.tez.dag.app.TaskCommunicatorManagerInterface)17 TaskHeartbeatHandler (org.apache.tez.dag.app.TaskHeartbeatHandler)17 HashSet (java.util.HashSet)16 ApplicationAttemptId (org.apache.hadoop.yarn.api.records.ApplicationAttemptId)16 ApplicationId (org.apache.hadoop.yarn.api.records.ApplicationId)16 Resource (org.apache.hadoop.yarn.api.records.Resource)16 ClusterInfo (org.apache.tez.dag.app.ClusterInfo)16 ContainerHeartbeatHandler (org.apache.tez.dag.app.ContainerHeartbeatHandler)16 AMContainerMap (org.apache.tez.dag.app.rm.container.AMContainerMap)16 ContainerContextMatcher (org.apache.tez.dag.app.rm.container.ContainerContextMatcher)16 TezDAGID (org.apache.tez.dag.records.TezDAGID)16 TezTaskID (org.apache.tez.dag.records.TezTaskID)16 TezVertexID (org.apache.tez.dag.records.TezVertexID)16