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

Example 1 with VertexRecoveryData

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

the class DAGImpl method vertexSucceeded.

private boolean vertexSucceeded(Vertex vertex) {
    numSuccessfulVertices++;
    boolean recoveryFailed = false;
    if (!commitAllOutputsOnSuccess && isCommittable()) {
        // committing successful outputs immediately. check for shared outputs
        List<VertexGroupInfo> groupsList = vertexGroupInfo.get(vertex.getName());
        if (groupsList != null) {
            List<VertexGroupInfo> commitList = Lists.newArrayListWithCapacity(groupsList.size());
            for (VertexGroupInfo groupInfo : groupsList) {
                groupInfo.successfulMembers++;
                if (groupInfo.groupMembers.size() == groupInfo.successfulMembers && !groupInfo.outputs.isEmpty()) {
                    // group has outputs and all vertex members are done
                    LOG.info("All members of group: " + groupInfo.groupName + " are succeeded. Commiting outputs");
                    commitList.add(groupInfo);
                }
            }
            for (VertexGroupInfo groupInfo : commitList) {
                if (recoveryData != null && recoveryData.isVertexGroupCommitted(groupInfo.groupName)) {
                    LOG.info("VertexGroup was already committed as per recovery" + " data, groupName=" + groupInfo.groupName);
                    for (String vertexName : groupInfo.groupMembers) {
                        VertexRecoveryData vertexRecoveryData = recoveryData.getVertexRecoveryData(getVertex(vertexName).getVertexId());
                        Preconditions.checkArgument(vertexRecoveryData != null, "Vertex Group has been committed" + ", but no VertexRecoveryData found for its vertex " + vertexName);
                        VertexFinishedEvent vertexFinishedEvent = vertexRecoveryData.getVertexFinishedEvent();
                        Preconditions.checkArgument(vertexFinishedEvent != null, "Vertex Group has been committed" + ", but no VertexFinishedEvent found in its vertex " + vertexName);
                        Preconditions.checkArgument(vertexFinishedEvent.getState() == VertexState.SUCCEEDED, "Vertex Group has been committed, but unexpected vertex state of its vertex " + vertexName + ", vertexstate=" + vertexFinishedEvent.getState());
                    }
                    continue;
                }
                groupInfo.commitStarted = true;
                final Vertex v = getVertex(groupInfo.groupMembers.iterator().next());
                try {
                    Collection<TezVertexID> vertexIds = getVertexIds(groupInfo.groupMembers);
                    appContext.getHistoryHandler().handleCriticalEvent(new DAGHistoryEvent(getID(), new VertexGroupCommitStartedEvent(dagId, groupInfo.groupName, vertexIds, clock.getTime())));
                } catch (IOException e) {
                    LOG.error("Failed to send commit recovery event to handler", e);
                    recoveryFailed = true;
                }
                if (!recoveryFailed) {
                    for (final String outputName : groupInfo.outputs) {
                        OutputKey outputKey = new OutputKey(outputName, groupInfo.groupName, true);
                        CommitCallback groupCommitCallback = new CommitCallback(outputKey);
                        CallableEvent groupCommitCallableEvent = new CallableEvent(groupCommitCallback) {

                            public Void call() throws Exception {
                                OutputCommitter committer = v.getOutputCommitters().get(outputName);
                                LOG.info("Committing output: " + outputName);
                                commitOutput(committer);
                                return null;
                            }
                        };
                        ListenableFuture<Void> groupCommitFuture = appContext.getExecService().submit(groupCommitCallableEvent);
                        Futures.addCallback(groupCommitFuture, groupCommitCallableEvent.getCallback());
                        commitFutures.put(outputKey, groupCommitFuture);
                    }
                }
            }
        }
    }
    if (recoveryFailed) {
        LOG.info("Recovery failure occurred during commit");
        enactKill(DAGTerminationCause.RECOVERY_FAILURE, VertexTerminationCause.COMMIT_FAILURE);
    }
    return !recoveryFailed;
}

Example 2 with VertexRecoveryData

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

the class TestDAGRecovery method testVertexRecoverFromStart.

/**
 * RecoveryEvents:
 *  DAG:  DAGInitedEvent -> DAGStartedEvent
 *  V1:   VertexReconfigrationDoneEvent -> VertexInitializedEvent -> VertexStartedEvent
 *
 * V1 skip initialization.
 */
@Test(timeout = 5000)
public void testVertexRecoverFromStart() {
    initMockDAGRecoveryDataForVertex();
    List<TezEvent> inputGeneratedTezEvents = new ArrayList<TezEvent>();
    VertexInitializedEvent v1InitedEvent = new VertexInitializedEvent(v1Id, "vertex1", 0L, v1InitedTime, v1NumTask, "", null, inputGeneratedTezEvents, null);
    VertexConfigurationDoneEvent v1ReconfigureDoneEvent = new VertexConfigurationDoneEvent(v1Id, 0L, v1NumTask, null, null, null, true);
    VertexStartedEvent v1StartedEvent = new VertexStartedEvent(v1Id, 0L, v1StartedTime);
    VertexRecoveryData vertexRecoveryData = new VertexRecoveryData(v1InitedEvent, v1ReconfigureDoneEvent, v1StartedEvent, null, new HashMap<TezTaskID, TaskRecoveryData>(), false);
    doReturn(vertexRecoveryData).when(dagRecoveryData).getVertexRecoveryData(v1Id);
    DAGEventRecoverEvent recoveryEvent = new DAGEventRecoverEvent(dagId, dagRecoveryData);
    dag.handle(recoveryEvent);
    dispatcher.await();
    VertexImpl v1 = (VertexImpl) dag.getVertex("vertex1");
    VertexImpl v2 = (VertexImpl) dag.getVertex("vertex2");
    VertexImpl v3 = (VertexImpl) dag.getVertex("vertex3");
    assertEquals(DAGState.RUNNING, dag.getState());
    // v1 skip initialization
    assertEquals(VertexState.RUNNING, v1.getState());
    assertEquals(v1InitedTime, v1.initedTime);
    assertEquals(v1StartedTime, v1.startedTime);
    assertEquals(v1NumTask, v1.getTotalTasks());
    assertEquals(VertexState.RUNNING, v2.getState());
    assertEquals(VertexState.RUNNING, v3.getState());
}

Example 3 with VertexRecoveryData

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

the class TestDAGRecovery method initMockDAGRecoveryDataForTask.

// ///////////////////////////// Task ////////////////////////////////////////////////////////////
private void initMockDAGRecoveryDataForTask() {
    List<TezEvent> inputGeneratedTezEvents = new ArrayList<TezEvent>();
    VertexInitializedEvent v1InitedEvent = new VertexInitializedEvent(v1Id, "vertex1", 0L, v1InitedTime, v1NumTask, "", null, inputGeneratedTezEvents, null);
    Map<String, InputSpecUpdate> rootInputSpecs = new HashMap<String, InputSpecUpdate>();
    VertexConfigurationDoneEvent v1ReconfigureDoneEvent = new VertexConfigurationDoneEvent(v1Id, 0L, v1NumTask, null, null, rootInputSpecs, true);
    VertexStartedEvent v1StartedEvent = new VertexStartedEvent(v1Id, 0L, v1StartedTime);
    VertexRecoveryData v1RecoveryData = new VertexRecoveryData(v1InitedEvent, v1ReconfigureDoneEvent, v1StartedEvent, null, new HashMap<TezTaskID, TaskRecoveryData>(), false);
    DAGInitializedEvent dagInitedEvent = new DAGInitializedEvent(dagId, dagInitedTime, "user", "dagName", null);
    DAGStartedEvent dagStartedEvent = new DAGStartedEvent(dagId, dagStartedTime, "user", "dagName");
    doReturn(v1RecoveryData).when(dagRecoveryData).getVertexRecoveryData(v1Id);
    doReturn(dagInitedEvent).when(dagRecoveryData).getDAGInitializedEvent();
    doReturn(dagStartedEvent).when(dagRecoveryData).getDAGStartedEvent();
}

Example 4 with VertexRecoveryData

use of org.apache.tez.dag.app.RecoveryParser.VertexRecoveryData 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 VertexRecoveryData

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

the class TestDAGRecovery method initMockDAGRecoveryDataForTaskAttempt.

// ///////////////////////////// TaskAttempt Recovery /////////////////////////////////////////////////////
private void initMockDAGRecoveryDataForTaskAttempt() {
    TaskStartedEvent t1StartedEvent = new TaskStartedEvent(t1v1Id, "vertex1", 0L, t1StartedTime);
    TaskRecoveryData taskRecoveryData = new TaskRecoveryData(t1StartedEvent, null, null);
    Map<TezTaskID, TaskRecoveryData> taskRecoveryDataMap = new HashMap<TezTaskID, TaskRecoveryData>();
    taskRecoveryDataMap.put(t1v1Id, taskRecoveryData);
    List<TezEvent> inputGeneratedTezEvents = new ArrayList<TezEvent>();
    VertexInitializedEvent v1InitedEvent = new VertexInitializedEvent(v1Id, "vertex1", 0L, v1InitedTime, v1NumTask, "", null, inputGeneratedTezEvents, null);
    Map<String, InputSpecUpdate> rootInputSpecs = new HashMap<String, InputSpecUpdate>();
    VertexConfigurationDoneEvent v1ReconfigureDoneEvent = new VertexConfigurationDoneEvent(v1Id, 0L, v1NumTask, null, null, rootInputSpecs, true);
    VertexStartedEvent v1StartedEvent = new VertexStartedEvent(v1Id, 0L, v1StartedTime);
    VertexRecoveryData v1RecoveryData = new VertexRecoveryData(v1InitedEvent, v1ReconfigureDoneEvent, v1StartedEvent, null, taskRecoveryDataMap, false);
    DAGInitializedEvent dagInitedEvent = new DAGInitializedEvent(dagId, dagInitedTime, "user", "dagName", null);
    DAGStartedEvent dagStartedEvent = new DAGStartedEvent(dagId, dagStartedTime, "user", "dagName");
    doReturn(v1RecoveryData).when(dagRecoveryData).getVertexRecoveryData(v1Id);
    doReturn(dagInitedEvent).when(dagRecoveryData).getDAGInitializedEvent();
    doReturn(dagStartedEvent).when(dagRecoveryData).getDAGStartedEvent();
}