Examples with RecoveryService org.apache.tez.dag.history.recovery.RecoveryService used on opensource projects

Example 6 with RecoveryService

use of org.apache.tez.dag.history.recovery.RecoveryService in project tez by apache.

the class TestRecoveryParser method testLastCorruptedRecoveryRecord.

@Test(timeout = 5000)
public void testLastCorruptedRecoveryRecord() throws IOException {
    ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 1);
    TezDAGID dagID = TezDAGID.getInstance(appId, 1);
    AppContext appContext = mock(AppContext.class);
    when(appContext.getCurrentRecoveryDir()).thenReturn(new Path(recoveryPath + "/1"));
    when(appContext.getClock()).thenReturn(new SystemClock());
    when(appContext.getHadoopShim()).thenReturn(new DefaultHadoopShim());
    when(appContext.getApplicationID()).thenReturn(appId);
    DAGPlan dagPlan = TestDAGImpl.createTestDAGPlan();
    // write data in attempt_1
    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();
    rService.handle(new DAGHistoryEvent(dagID, new DAGSubmittedEvent(dagID, 1L, dagPlan, ApplicationAttemptId.newInstance(appId, 1), null, "user", new Configuration(), null, null)));
    // wait until DAGSubmittedEvent is handled in the RecoveryEventHandling thread
    rService.await();
    rService.outputStreamMap.get(dagID).writeUTF("INVALID_DATA");
    rService.stop();
    // write data in attempt_2
    when(appContext.getCurrentRecoveryDir()).thenReturn(new Path(recoveryPath + "/2"));
    rService = new RecoveryService(appContext);
    rService.init(conf);
    rService.start();
    rService.handle(new DAGHistoryEvent(dagID, new DAGInitializedEvent(dagID, 1L, "user", dagPlan.getName(), null)));
    rService.await();
    rService.outputStreamMap.get(dagID).writeUTF("INVALID_DATA");
    rService.stop();
    // corrupted last records will be skipped but the whole recovery logs will be read
    DAGRecoveryData dagData = parser.parseRecoveryData();
    assertEquals(false, dagData.isCompleted);
    assertEquals(null, dagData.reason);
    assertEquals(false, dagData.nonRecoverable);
    // verify DAGSubmitedEvent & DAGInititlizedEvent is handled.
    verify(mockAppMaster).createDAG(any(DAGPlan.class), any(TezDAGID.class));
    assertNotNull(dagData.getDAGInitializedEvent());
}

Example 7 with RecoveryService

use of org.apache.tez.dag.history.recovery.RecoveryService 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 8 with RecoveryService

use of org.apache.tez.dag.history.recovery.RecoveryService in project tez by apache.

the class TestRecoveryParser method testRecoverableSummary_VertexGroupFinishCommitting.

@Test(timeout = 5000)
public void testRecoverableSummary_VertexGroupFinishCommitting() throws IOException {
    ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 1);
    TezDAGID dagID = TezDAGID.getInstance(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();
    // write a DAGSubmittedEvent first to initialize summaryStream
    rService.handle(new DAGHistoryEvent(dagID, new DAGSubmittedEvent(dagID, 1L, dagPlan, ApplicationAttemptId.newInstance(appId, 1), null, "user", new Configuration(), null, null)));
    // It should be fine to skip other events, just for testing.
    TezVertexID v0 = TezVertexID.getInstance(dagID, 0);
    TezVertexID v1 = TezVertexID.getInstance(dagID, 1);
    rService.handle(new DAGHistoryEvent(dagID, new VertexGroupCommitStartedEvent(dagID, "group_1", Lists.newArrayList(v0, v1), 0L)));
    rService.handle(new DAGHistoryEvent(dagID, new VertexGroupCommitFinishedEvent(dagID, "group_1", Lists.newArrayList(v0, v1), 0L)));
    // also write VertexFinishedEvent, otherwise it is still non-recoverable
    // when checking with non-summary event
    rService.handle(new DAGHistoryEvent(dagID, new VertexFinishedEvent(v0, "v1", 10, 0L, 0L, 0L, 0L, 0L, VertexState.SUCCEEDED, "", null, null, null, null)));
    rService.handle(new DAGHistoryEvent(dagID, new VertexFinishedEvent(v1, "v1", 10, 0L, 0L, 0L, 0L, 0L, VertexState.SUCCEEDED, "", null, null, null, null)));
    rService.stop();
    DAGRecoveryData dagData = parser.parseRecoveryData();
    assertEquals(dagID, dagData.recoveredDagID);
    assertFalse(dagData.nonRecoverable);
}

Example 9 with RecoveryService

use of org.apache.tez.dag.history.recovery.RecoveryService in project tez by apache.

the class TestRecoveryParser method testLastCorruptedSummaryRecord.

@Test(timeout = 5000)
public void testLastCorruptedSummaryRecord() throws IOException {
    ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 1);
    TezDAGID dagID = TezDAGID.getInstance(appId, 1);
    AppContext appContext = mock(AppContext.class);
    when(appContext.getCurrentRecoveryDir()).thenReturn(new Path(recoveryPath + "/1"));
    when(appContext.getClock()).thenReturn(new SystemClock());
    // write data in attempt_1
    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();
    // write a DAGSubmittedEvent first to initialize summaryStream
    rService.handle(new DAGHistoryEvent(dagID, new DAGSubmittedEvent(dagID, 1L, dagPlan, ApplicationAttemptId.newInstance(appId, 1), null, "user", new Configuration(), null, null)));
    // write an corrupted SummaryEvent
    rService.summaryStream.writeChars("INVALID_DATA");
    rService.stop();
    try {
        // Corrupted SummaryEvent will cause recovery fail (throw exception here)
        parser.parseRecoveryData();
        fail();
    } catch (IOException e) {
        // exception when parsing protobuf object
        e.printStackTrace();
    }
}

Example 10 with RecoveryService

use of org.apache.tez.dag.history.recovery.RecoveryService in project tez by apache.

the class TestRecoveryParser method testRecoverableSummary_DAGFinishCommitting.

@Test(timeout = 5000)
public void testRecoverableSummary_DAGFinishCommitting() throws IOException {
    ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 1);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, 1);
    TezDAGID dagID = TezDAGID.getInstance(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);
    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();
    // write a DAGSubmittedEvent first to initialize summaryStream
    rService.handle(new DAGHistoryEvent(dagID, new DAGSubmittedEvent(dagID, 1L, dagPlan, ApplicationAttemptId.newInstance(appId, 1), null, "user", new Configuration(), null, null)));
    // It should be fine to skip other events, just for testing.
    rService.handle(new DAGHistoryEvent(dagID, new DAGCommitStartedEvent(dagID, 0L)));
    rService.handle(new DAGHistoryEvent(dagID, new DAGFinishedEvent(dagID, 1L, 2L, DAGState.FAILED, "diag", null, "user", "dag1", null, appAttemptId, dagPlan)));
    rService.stop();
    DAGRecoveryData dagData = parser.parseRecoveryData();
    assertEquals(dagID, dagData.recoveredDagID);
    assertEquals(DAGState.FAILED, dagData.dagState);
    assertFalse(dagData.nonRecoverable);
    assertNull(dagData.reason);
    assertTrue(dagData.isCompleted);
}