Examples with TaskAttemptRecoveryData org.apache.tez.dag.app.RecoveryParser.TaskAttemptRecoveryData used on opensource projects

Example 1 with TaskAttemptRecoveryData

use of org.apache.tez.dag.app.RecoveryParser.TaskAttemptRecoveryData in project tez by apache.

the class TestDAGRecovery method testTARecoverFromNewToFailed.

/**
 * RecoveryEvents: TaskAttemptFinishedEvent (FAILED)
 * Recover it to FAILED
 */
@Test(timeout = 5000)
public void testTARecoverFromNewToFailed() {
    initMockDAGRecoveryDataForTaskAttempt();
    TaskAttemptFinishedEvent taFinishedEvent = new TaskAttemptFinishedEvent(ta1t1v1Id, "v1", ta1LaunchTime, ta1FinishedTime, TaskAttemptState.FAILED, TaskFailureType.NON_FATAL, TaskAttemptTerminationCause.CONTAINER_LAUNCH_FAILED, "", null, null, null, 0L, null, 0L, null, null, null, null, null);
    TaskAttemptRecoveryData taRecoveryData = new TaskAttemptRecoveryData(null, taFinishedEvent);
    doReturn(taRecoveryData).when(dagRecoveryData).getTaskAttemptRecoveryData(ta1t1v1Id);
    dag.handle(new DAGEventRecoverEvent(dagId, dagRecoveryData));
    dispatcher.await();
    TaskImpl task = (TaskImpl) dag.getVertex(v1Id).getTask(t1v1Id);
    TaskAttemptImpl taskAttempt = (TaskAttemptImpl) task.getAttempt(ta1t1v1Id);
    assertEquals(TaskAttemptStateInternal.FAILED, taskAttempt.getInternalState());
    assertEquals(TaskAttemptTerminationCause.CONTAINER_LAUNCH_FAILED, taskAttempt.getTerminationCause());
    historyEventHandler.verifyHistoryEvent(0, HistoryEventType.TASK_ATTEMPT_STARTED);
    historyEventHandler.verifyHistoryEvent(0, HistoryEventType.TASK_ATTEMPT_FINISHED);
    assertEquals(1, task.failedAttempts);
    // new task attempt is scheduled
    assertEquals(2, task.getAttempts().size());
    assertEquals(ta1FinishedTime, taskAttempt.getFinishTime());
}

Example 2 with TaskAttemptRecoveryData

use of org.apache.tez.dag.app.RecoveryParser.TaskAttemptRecoveryData in project tez by apache.

the class TestDAGRecovery method testTARecoverFromSucceeded.

/**
 * RecoveryEvents: TaskAttemptStartedEvent -> TaskAttemptFinishedEvent (SUCCEEDED)
 * Recover it to SUCCEEDED
 */
@Test(timeout = 5000)
public void testTARecoverFromSucceeded() {
    initMockDAGRecoveryDataForTaskAttempt();
    TaskAttemptStartedEvent taStartedEvent = new TaskAttemptStartedEvent(ta1t1v1Id, "v1", ta1LaunchTime, mock(ContainerId.class), mock(NodeId.class), "", "", "");
    List<TezEvent> taGeneratedEvents = new ArrayList<TezEvent>();
    EventMetaData sourceInfo = new EventMetaData(EventProducerConsumerType.INPUT, "vertex1", "vertex3", ta1t1v1Id);
    taGeneratedEvents.add(new TezEvent(DataMovementEvent.create(ByteBuffer.wrap(new byte[0])), sourceInfo));
    TaskAttemptFinishedEvent taFinishedEvent = new TaskAttemptFinishedEvent(ta1t1v1Id, "v1", ta1LaunchTime, ta1FinishedTime, TaskAttemptState.SUCCEEDED, null, null, "", null, null, taGeneratedEvents, 0L, null, 0L, null, null, null, null, null);
    TaskAttemptRecoveryData taRecoveryData = new TaskAttemptRecoveryData(taStartedEvent, taFinishedEvent);
    doReturn(taRecoveryData).when(dagRecoveryData).getTaskAttemptRecoveryData(ta1t1v1Id);
    dag.handle(new DAGEventRecoverEvent(dagId, dagRecoveryData));
    dispatcher.await();
    TaskImpl task = (TaskImpl) dag.getVertex(v1Id).getTask(t1v1Id);
    TaskAttemptImpl taskAttempt = (TaskAttemptImpl) task.getAttempt(ta1t1v1Id);
    assertEquals(TaskAttemptStateInternal.SUCCEEDED, taskAttempt.getInternalState());
    historyEventHandler.verifyHistoryEvent(0, HistoryEventType.TASK_ATTEMPT_FINISHED);
    assertEquals(TaskStateInternal.SUCCEEDED, task.getInternalState());
    assertEquals(ta1LaunchTime, taskAttempt.getLaunchTime());
    assertEquals(ta1FinishedTime, taskAttempt.getFinishTime());
}

Example 3 with TaskAttemptRecoveryData

use of org.apache.tez.dag.app.RecoveryParser.TaskAttemptRecoveryData in project tez by apache.

the class TestDAGRecovery method testTARecoverFromRunning.

/**
 * RecoveryEvents: TaskAttemptStartedEvent
 * Recover it to KILLED
 */
@Test(timeout = 5000)
public void testTARecoverFromRunning() {
    initMockDAGRecoveryDataForTaskAttempt();
    TaskAttemptStartedEvent taStartedEvent = new TaskAttemptStartedEvent(ta1t1v1Id, "v1", ta1LaunchTime, mock(ContainerId.class), mock(NodeId.class), "", "", "");
    TaskAttemptRecoveryData taRecoveryData = new TaskAttemptRecoveryData(taStartedEvent, null);
    doReturn(taRecoveryData).when(dagRecoveryData).getTaskAttemptRecoveryData(ta1t1v1Id);
    dag.handle(new DAGEventRecoverEvent(dagId, dagRecoveryData));
    dispatcher.await();
    TaskImpl task = (TaskImpl) dag.getVertex(v1Id).getTask(t1v1Id);
    TaskAttemptImpl taskAttempt = (TaskAttemptImpl) task.getAttempt(ta1t1v1Id);
    assertEquals(TaskAttemptStateInternal.KILLED, taskAttempt.getInternalState());
    assertEquals(TaskAttemptTerminationCause.TERMINATED_AT_RECOVERY, taskAttempt.getTerminationCause());
    historyEventHandler.verifyHistoryEvent(0, HistoryEventType.TASK_ATTEMPT_STARTED);
    historyEventHandler.verifyHistoryEvent(1, HistoryEventType.TASK_ATTEMPT_FINISHED);
    assertEquals(ta1LaunchTime, taskAttempt.getLaunchTime());
}

Example 4 with TaskAttemptRecoveryData

use of org.apache.tez.dag.app.RecoveryParser.TaskAttemptRecoveryData in project tez by apache.

the class TestRecoveryParser method testRecoveryData.

@Test(timeout = 5000)
public void testRecoveryData() throws IOException {
    ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 1);
    TezDAGID dagID = TezDAGID.getInstance(appId, 1);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, 1);
    AppContext appContext = mock(AppContext.class);
    when(appContext.getCurrentRecoveryDir()).thenReturn(new Path(recoveryPath + "/1"));
    when(appContext.getClock()).thenReturn(new SystemClock());
    when(mockDAGImpl.getID()).thenReturn(dagID);
    when(appContext.getHadoopShim()).thenReturn(new DefaultHadoopShim());
    when(appContext.getApplicationID()).thenReturn(appId);
    RecoveryService rService = new RecoveryService(appContext);
    Configuration conf = new Configuration();
    conf.setBoolean(RecoveryService.TEZ_TEST_RECOVERY_DRAIN_EVENTS_WHEN_STOPPED, true);
    rService.init(conf);
    rService.start();
    DAGPlan dagPlan = TestDAGImpl.createTestDAGPlan();
    // DAG  DAGSubmittedEvent -> DAGInitializedEvent -> DAGStartedEvent
    rService.handle(new DAGHistoryEvent(dagID, new DAGSubmittedEvent(dagID, 1L, dagPlan, ApplicationAttemptId.newInstance(appId, 1), null, "user", new Configuration(), null, null)));
    DAGInitializedEvent dagInitedEvent = new DAGInitializedEvent(dagID, 100L, "user", "dagName", null);
    DAGStartedEvent dagStartedEvent = new DAGStartedEvent(dagID, 0L, "user", "dagName");
    rService.handle(new DAGHistoryEvent(dagID, dagInitedEvent));
    rService.handle(new DAGHistoryEvent(dagID, dagStartedEvent));
    // 3 vertices of this dag: v0, v1, v2
    TezVertexID v0Id = TezVertexID.getInstance(dagID, 0);
    TezVertexID v1Id = TezVertexID.getInstance(dagID, 1);
    TezVertexID v2Id = TezVertexID.getInstance(dagID, 2);
    // v0 VertexInitializedEvent
    VertexInitializedEvent v0InitedEvent = new VertexInitializedEvent(v0Id, "v0", 200L, 400L, 2, null, null, null, null);
    rService.handle(new DAGHistoryEvent(dagID, v0InitedEvent));
    // v1 VertexFinishedEvent(KILLED)
    VertexFinishedEvent v1FinishedEvent = new VertexFinishedEvent(v1Id, "v1", 2, 300L, 400L, 500L, 600L, 700L, VertexState.KILLED, "", null, null, null, null);
    rService.handle(new DAGHistoryEvent(dagID, v1FinishedEvent));
    // v2 VertexInitializedEvent -> VertexStartedEvent
    List<TezEvent> initGeneratedEvents = Lists.newArrayList(new TezEvent(DataMovementEvent.create(ByteBuffer.wrap(new byte[0])), null));
    VertexInitializedEvent v2InitedEvent = new VertexInitializedEvent(v2Id, "v2", 200L, 300L, 2, null, null, initGeneratedEvents, null);
    VertexStartedEvent v2StartedEvent = new VertexStartedEvent(v2Id, 0L, 0L);
    rService.handle(new DAGHistoryEvent(dagID, v2InitedEvent));
    rService.handle(new DAGHistoryEvent(dagID, v2StartedEvent));
    // 3 tasks of v2
    TezTaskID t0v2Id = TezTaskID.getInstance(v2Id, 0);
    TezTaskID t1v2Id = TezTaskID.getInstance(v2Id, 1);
    TezTaskID t2v2Id = TezTaskID.getInstance(v2Id, 2);
    // t0v2 TaskStartedEvent
    TaskStartedEvent t0v2StartedEvent = new TaskStartedEvent(t0v2Id, "v2", 400L, 5000L);
    rService.handle(new DAGHistoryEvent(dagID, t0v2StartedEvent));
    // t1v2 TaskFinishedEvent
    TaskFinishedEvent t1v2FinishedEvent = new TaskFinishedEvent(t1v2Id, "v1", 0L, 0L, null, TaskState.KILLED, "", null, 4);
    rService.handle(new DAGHistoryEvent(dagID, t1v2FinishedEvent));
    // t2v2 TaskStartedEvent -> TaskFinishedEvent
    TaskStartedEvent t2v2StartedEvent = new TaskStartedEvent(t2v2Id, "v2", 400L, 500L);
    rService.handle(new DAGHistoryEvent(dagID, t2v2StartedEvent));
    TaskFinishedEvent t2v2FinishedEvent = new TaskFinishedEvent(t2v2Id, "v1", 0L, 0L, null, TaskState.SUCCEEDED, "", null, 4);
    rService.handle(new DAGHistoryEvent(dagID, t2v2FinishedEvent));
    // attempts under t0v2
    ContainerId containerId = ContainerId.newInstance(appAttemptId, 1);
    NodeId nodeId = NodeId.newInstance("localhost", 9999);
    TezTaskAttemptID ta0t0v2Id = TezTaskAttemptID.getInstance(t0v2Id, 0);
    TaskAttemptStartedEvent ta0t0v2StartedEvent = new TaskAttemptStartedEvent(ta0t0v2Id, "v1", 0L, containerId, nodeId, "", "", "");
    rService.handle(new DAGHistoryEvent(dagID, ta0t0v2StartedEvent));
    // attempts under t2v2
    TezTaskAttemptID ta0t2v2Id = TezTaskAttemptID.getInstance(t2v2Id, 0);
    TaskAttemptStartedEvent ta0t2v2StartedEvent = new TaskAttemptStartedEvent(ta0t2v2Id, "v1", 500L, containerId, nodeId, "", "", "");
    rService.handle(new DAGHistoryEvent(dagID, ta0t2v2StartedEvent));
    TaskAttemptFinishedEvent ta0t2v2FinishedEvent = new TaskAttemptFinishedEvent(ta0t2v2Id, "v1", 500L, 600L, TaskAttemptState.SUCCEEDED, null, null, "", null, null, null, 0L, null, 0L, null, null, null, null, null);
    rService.handle(new DAGHistoryEvent(dagID, ta0t2v2FinishedEvent));
    rService.stop();
    DAGRecoveryData dagData = parser.parseRecoveryData();
    assertFalse(dagData.nonRecoverable);
    // There's no equals method for the history event, so here only verify the init/start/finish time of each event for simplicity
    assertEquals(dagInitedEvent.getInitTime(), dagData.getDAGInitializedEvent().getInitTime());
    assertEquals(dagStartedEvent.getStartTime(), dagData.getDAGStartedEvent().getStartTime());
    assertNull(dagData.getDAGFinishedEvent());
    VertexRecoveryData v0Data = dagData.getVertexRecoveryData(v0Id);
    VertexRecoveryData v1Data = dagData.getVertexRecoveryData(v1Id);
    VertexRecoveryData v2Data = dagData.getVertexRecoveryData(v2Id);
    assertNotNull(v0Data);
    assertNotNull(v1Data);
    assertNotNull(v2Data);
    assertEquals(v0InitedEvent.getInitedTime(), v0Data.getVertexInitedEvent().getInitedTime());
    assertNull(v0Data.getVertexStartedEvent());
    assertNull(v1Data.getVertexInitedEvent());
    assertEquals(v1FinishedEvent.getFinishTime(), v1Data.getVertexFinishedEvent().getFinishTime());
    assertEquals(v2InitedEvent.getInitedTime(), v2Data.getVertexInitedEvent().getInitedTime());
    assertEquals(v2StartedEvent.getStartTime(), v2Data.getVertexStartedEvent().getStartTime());
    TaskRecoveryData t0v2Data = dagData.getTaskRecoveryData(t0v2Id);
    TaskRecoveryData t1v2Data = dagData.getTaskRecoveryData(t1v2Id);
    TaskRecoveryData t2v2Data = dagData.getTaskRecoveryData(t2v2Id);
    assertNotNull(t0v2Data);
    assertNotNull(t1v2Data);
    assertNotNull(t2v2Data);
    assertEquals(t0v2StartedEvent.getStartTime(), t0v2Data.getTaskStartedEvent().getStartTime());
    assertNull(t0v2Data.getTaskFinishedEvent());
    assertEquals(t1v2FinishedEvent.getFinishTime(), t1v2Data.getTaskFinishedEvent().getFinishTime());
    assertNull(t1v2Data.getTaskStartedEvent());
    assertEquals(t2v2StartedEvent.getStartTime(), t2v2Data.getTaskStartedEvent().getStartTime());
    assertEquals(t2v2FinishedEvent.getFinishTime(), t2v2Data.getTaskFinishedEvent().getFinishTime());
    TaskAttemptRecoveryData ta0t0v2Data = dagData.getTaskAttemptRecoveryData(ta0t0v2Id);
    TaskAttemptRecoveryData ta0t2v2Data = dagData.getTaskAttemptRecoveryData(ta0t2v2Id);
    assertNotNull(ta0t0v2Data);
    assertNotNull(ta0t2v2Data);
    assertEquals(ta0t0v2StartedEvent.getStartTime(), ta0t0v2Data.getTaskAttemptStartedEvent().getStartTime());
    assertNull(ta0t0v2Data.getTaskAttemptFinishedEvent());
    assertEquals(ta0t2v2StartedEvent.getStartTime(), ta0t2v2Data.getTaskAttemptStartedEvent().getStartTime());
    assertEquals(ta0t2v2FinishedEvent.getFinishTime(), ta0t2v2Data.getTaskAttemptFinishedEvent().getFinishTime());
}

Example 5 with TaskAttemptRecoveryData

use of org.apache.tez.dag.app.RecoveryParser.TaskAttemptRecoveryData in project tez by apache.

the class TestDAGRecovery method testTARecoverFromFailed.

/**
 * RecoveryEvents: TaskAttemptStartedEvent -> TaskAttemptFinishedEvent (FAILED)
 * Recover it to FAILED
 */
@Test(timeout = 5000)
public void testTARecoverFromFailed() {
    initMockDAGRecoveryDataForTaskAttempt();
    TaskAttemptStartedEvent taStartedEvent = new TaskAttemptStartedEvent(ta1t1v1Id, "v1", ta1LaunchTime, mock(ContainerId.class), mock(NodeId.class), "", "", "");
    TaskAttemptFinishedEvent taFinishedEvent = new TaskAttemptFinishedEvent(ta1t1v1Id, "v1", ta1LaunchTime, ta1FinishedTime, TaskAttemptState.FAILED, TaskFailureType.NON_FATAL, TaskAttemptTerminationCause.INPUT_READ_ERROR, "", null, null, null, 0L, null, 0L, null, null, null, null, null);
    TaskAttemptRecoveryData taRecoveryData = new TaskAttemptRecoveryData(taStartedEvent, taFinishedEvent);
    doReturn(taRecoveryData).when(dagRecoveryData).getTaskAttemptRecoveryData(ta1t1v1Id);
    dag.handle(new DAGEventRecoverEvent(dagId, dagRecoveryData));
    dispatcher.await();
    TaskImpl task = (TaskImpl) dag.getVertex(v1Id).getTask(t1v1Id);
    TaskAttemptImpl taskAttempt = (TaskAttemptImpl) task.getAttempt(ta1t1v1Id);
    assertEquals(TaskAttemptStateInternal.FAILED, taskAttempt.getInternalState());
    assertEquals(TaskAttemptTerminationCause.INPUT_READ_ERROR, taskAttempt.getTerminationCause());
    assertEquals(TaskStateInternal.SCHEDULED, task.getInternalState());
    assertEquals(2, task.getAttempts().size());
    historyEventHandler.verifyHistoryEvent(0, HistoryEventType.TASK_ATTEMPT_FINISHED);
    assertEquals(ta1LaunchTime, taskAttempt.getLaunchTime());
    assertEquals(ta1FinishedTime, taskAttempt.getFinishTime());
}