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

Example 16 with HistoryEvent

use of org.apache.tez.dag.history.HistoryEvent in project tez by apache.

the class RecoveryParser method parseRecoveryData.

/**
 * 1. Read Summary Recovery file and build DAGSummaryData
 *    Check whether it is recoverable based on the summary file (whether dag is
 *    in the middle of committing)
 * 2. Read the non-Summary Recovery file and build DAGRecoveryData
 *    Check whether it is recoverable based on both the summary file and non-summary file
 *    (whether vertex has completed its committing, but its full non-summary recovery events are not seen)
 * @return DAGRecoveryData
 * @throws IOException
 */
public DAGRecoveryData parseRecoveryData() throws IOException {
    int dagCounter = 0;
    Map<TezDAGID, DAGSummaryData> dagSummaryDataMap = new HashMap<TezDAGID, DAGSummaryData>();
    List<Path> summaryFiles = getSummaryFiles();
    LOG.debug("SummaryFile size:" + summaryFiles.size());
    for (Path summaryFile : summaryFiles) {
        FileStatus summaryFileStatus = recoveryFS.getFileStatus(summaryFile);
        LOG.info("Parsing summary file" + ", path=" + summaryFile.toString() + ", len=" + summaryFileStatus.getLen() + ", lastModTime=" + summaryFileStatus.getModificationTime());
        FSDataInputStream summaryStream = getSummaryStream(summaryFile);
        while (true) {
            RecoveryProtos.SummaryEventProto proto;
            try {
                proto = RecoveryProtos.SummaryEventProto.parseDelimitedFrom(summaryStream);
                if (proto == null) {
                    LOG.info("Reached end of summary stream");
                    break;
                }
            } catch (EOFException eof) {
                LOG.info("Reached end of summary stream");
                break;
            }
            HistoryEventType eventType = HistoryEventType.values()[proto.getEventType()];
            if (LOG.isDebugEnabled()) {
                LOG.debug("[RECOVERY SUMMARY]" + " dagId=" + proto.getDagId() + ", timestamp=" + proto.getTimestamp() + ", event=" + eventType);
            }
            TezDAGID dagId;
            try {
                dagId = TezDAGID.fromString(proto.getDagId());
            } catch (IllegalArgumentException e) {
                throw new IOException("Invalid dagId, summary records may be corrupted", e);
            }
            if (dagCounter < dagId.getId()) {
                dagCounter = dagId.getId();
            }
            if (!dagSummaryDataMap.containsKey(dagId)) {
                dagSummaryDataMap.put(dagId, new DAGSummaryData(dagId));
            }
            try {
                dagSummaryDataMap.get(dagId).handleSummaryEvent(proto);
            } catch (Exception e) {
                // any exception when parsing protobuf
                throw new IOException("Error when parsing summary event proto", e);
            }
        }
        summaryStream.close();
    }
    // Set counter for next set of DAGs & update dagNames Set in DAGAppMaster
    dagAppMaster.setDAGCounter(dagCounter);
    for (DAGSummaryData dagSummaryData : dagSummaryDataMap.values()) {
        dagAppMaster.dagIDs.add(dagSummaryData.dagId.toString());
    }
    DAGSummaryData lastInProgressDAGData = getLastCompletedOrInProgressDAG(dagSummaryDataMap);
    if (lastInProgressDAGData == null) {
        LOG.info("Nothing to recover as no uncompleted/completed DAGs found");
        return null;
    }
    TezDAGID lastInProgressDAG = lastInProgressDAGData.dagId;
    if (lastInProgressDAG == null) {
        LOG.info("Nothing to recover as no uncompleted/completed DAGs found");
        return null;
    }
    LOG.info("Checking if DAG is in recoverable state" + ", dagId=" + lastInProgressDAGData.dagId);
    final DAGRecoveryData recoveredDAGData = new DAGRecoveryData(lastInProgressDAGData);
    List<Path> dagRecoveryFiles = getDAGRecoveryFiles(lastInProgressDAG);
    boolean skipAllOtherEvents = false;
    Path lastRecoveryFile = null;
    // to create the DAGImpl)
    for (Path dagRecoveryFile : dagRecoveryFiles) {
        if (skipAllOtherEvents) {
            LOG.warn("Other recovery files will be skipped due to error in the previous recovery file" + lastRecoveryFile);
            break;
        }
        FileStatus fileStatus = recoveryFS.getFileStatus(dagRecoveryFile);
        lastRecoveryFile = dagRecoveryFile;
        LOG.info("Trying to recover dag from recovery file" + ", dagId=" + lastInProgressDAG.toString() + ", dagRecoveryFile=" + dagRecoveryFile + ", len=" + fileStatus.getLen());
        FSDataInputStream dagRecoveryStream = recoveryFS.open(dagRecoveryFile, recoveryBufferSize);
        while (true) {
            HistoryEvent event;
            try {
                event = getNextEvent(dagRecoveryStream);
                if (event == null) {
                    LOG.info("Reached end of dag recovery stream");
                    break;
                }
            } catch (EOFException eof) {
                LOG.info("Reached end of dag recovery stream");
                break;
            } catch (IOException ioe) {
                LOG.warn("Corrupt data found when trying to read next event", ioe);
                break;
            }
            if (skipAllOtherEvents) {
                // hit an error - skip reading other events
                break;
            }
            HistoryEventType eventType = event.getEventType();
            LOG.info("Recovering from event" + ", eventType=" + eventType + ", event=" + event.toString());
            switch(eventType) {
                case DAG_SUBMITTED:
                    DAGSubmittedEvent submittedEvent = (DAGSubmittedEvent) event;
                    recoveredDAGData.recoveredDAG = dagAppMaster.createDAG(submittedEvent.getDAGPlan(), lastInProgressDAG);
                    recoveredDAGData.cumulativeAdditionalResources = submittedEvent.getCumulativeAdditionalLocalResources();
                    recoveredDAGData.recoveredDagID = recoveredDAGData.recoveredDAG.getID();
                    dagAppMaster.setCurrentDAG(recoveredDAGData.recoveredDAG);
                    if (recoveredDAGData.nonRecoverable) {
                        skipAllOtherEvents = true;
                    }
                    break;
                case DAG_INITIALIZED:
                    recoveredDAGData.dagInitedEvent = (DAGInitializedEvent) event;
                    break;
                case DAG_STARTED:
                    recoveredDAGData.dagStartedEvent = (DAGStartedEvent) event;
                    break;
                case DAG_FINISHED:
                    recoveredDAGData.dagFinishedEvent = (DAGFinishedEvent) event;
                    skipAllOtherEvents = true;
                    break;
                case DAG_COMMIT_STARTED:
                case VERTEX_GROUP_COMMIT_STARTED:
                case VERTEX_GROUP_COMMIT_FINISHED:
                case CONTAINER_LAUNCHED:
                    {
                        // Nothing to do for now
                        break;
                    }
                case DAG_KILL_REQUEST:
                    {
                        break;
                    }
                case VERTEX_INITIALIZED:
                    {
                        VertexInitializedEvent vertexInitEvent = (VertexInitializedEvent) event;
                        VertexRecoveryData vertexRecoveryData = recoveredDAGData.maybeCreateVertexRecoveryData(vertexInitEvent.getVertexID());
                        vertexRecoveryData.vertexInitedEvent = vertexInitEvent;
                        break;
                    }
                case VERTEX_CONFIGURE_DONE:
                    {
                        VertexConfigurationDoneEvent reconfigureDoneEvent = (VertexConfigurationDoneEvent) event;
                        VertexRecoveryData vertexRecoveryData = recoveredDAGData.maybeCreateVertexRecoveryData(reconfigureDoneEvent.getVertexID());
                        vertexRecoveryData.vertexConfigurationDoneEvent = reconfigureDoneEvent;
                        break;
                    }
                case VERTEX_STARTED:
                    {
                        VertexStartedEvent vertexStartedEvent = (VertexStartedEvent) event;
                        VertexRecoveryData vertexRecoveryData = recoveredDAGData.vertexRecoveryDataMap.get(vertexStartedEvent.getVertexID());
                        Preconditions.checkArgument(vertexRecoveryData != null, "No VertexInitializedEvent before VertexStartedEvent");
                        vertexRecoveryData.vertexStartedEvent = vertexStartedEvent;
                        break;
                    }
                case VERTEX_COMMIT_STARTED:
                    {
                        break;
                    }
                case VERTEX_FINISHED:
                    {
                        VertexFinishedEvent vertexFinishedEvent = (VertexFinishedEvent) event;
                        VertexRecoveryData vertexRecoveryData = recoveredDAGData.maybeCreateVertexRecoveryData(vertexFinishedEvent.getVertexID());
                        vertexRecoveryData.vertexFinishedEvent = vertexFinishedEvent;
                        break;
                    }
                case TASK_STARTED:
                    {
                        TaskStartedEvent taskStartedEvent = (TaskStartedEvent) event;
                        VertexRecoveryData vertexRecoveryData = recoveredDAGData.vertexRecoveryDataMap.get(taskStartedEvent.getTaskID().getVertexID());
                        Preconditions.checkArgument(vertexRecoveryData != null, "Invalid TaskStartedEvent, its vertex does not exist:" + taskStartedEvent.getTaskID().getVertexID());
                        TaskRecoveryData taskRecoveryData = vertexRecoveryData.maybeCreateTaskRecoveryData(taskStartedEvent.getTaskID());
                        taskRecoveryData.taskStartedEvent = taskStartedEvent;
                        break;
                    }
                case TASK_FINISHED:
                    {
                        TaskFinishedEvent taskFinishedEvent = (TaskFinishedEvent) event;
                        VertexRecoveryData vertexRecoveryData = recoveredDAGData.vertexRecoveryDataMap.get(taskFinishedEvent.getTaskID().getVertexID());
                        Preconditions.checkArgument(vertexRecoveryData != null, "Invalid TaskFinishedEvent, its vertex does not exist:" + taskFinishedEvent.getTaskID().getVertexID());
                        TaskRecoveryData taskRecoveryData = vertexRecoveryData.maybeCreateTaskRecoveryData(taskFinishedEvent.getTaskID());
                        taskRecoveryData.taskFinishedEvent = taskFinishedEvent;
                        break;
                    }
                case TASK_ATTEMPT_STARTED:
                    {
                        TaskAttemptStartedEvent taStartedEvent = (TaskAttemptStartedEvent) event;
                        VertexRecoveryData vertexRecoveryData = recoveredDAGData.vertexRecoveryDataMap.get(taStartedEvent.getTaskAttemptID().getTaskID().getVertexID());
                        Preconditions.checkArgument(vertexRecoveryData != null, "Invalid TaskAttemptStartedEvent, its vertexId does not exist, taId=" + taStartedEvent.getTaskAttemptID());
                        TaskRecoveryData taskRecoveryData = vertexRecoveryData.taskRecoveryDataMap.get(taStartedEvent.getTaskAttemptID().getTaskID());
                        Preconditions.checkArgument(taskRecoveryData != null, "Invalid TaskAttemptStartedEvent, its taskId does not exist, taId=" + taStartedEvent.getTaskAttemptID());
                        TaskAttemptRecoveryData taRecoveryData = taskRecoveryData.maybeCreateTaskAttemptRecoveryData(taStartedEvent.getTaskAttemptID());
                        taRecoveryData.taStartedEvent = taStartedEvent;
                        break;
                    }
                case TASK_ATTEMPT_FINISHED:
                    {
                        TaskAttemptFinishedEvent taFinishedEvent = (TaskAttemptFinishedEvent) event;
                        VertexRecoveryData vertexRecoveryData = recoveredDAGData.vertexRecoveryDataMap.get(taFinishedEvent.getTaskAttemptID().getTaskID().getVertexID());
                        Preconditions.checkArgument(vertexRecoveryData != null, "Invalid TaskAttemtFinishedEvent, its vertexId does not exist, taId=" + taFinishedEvent.getTaskAttemptID());
                        TaskRecoveryData taskRecoveryData = vertexRecoveryData.taskRecoveryDataMap.get(taFinishedEvent.getTaskAttemptID().getTaskID());
                        Preconditions.checkArgument(taskRecoveryData != null, "Invalid TaskAttemptFinishedEvent, its taskId does not exist, taId=" + taFinishedEvent.getTaskAttemptID());
                        TaskAttemptRecoveryData taRecoveryData = taskRecoveryData.maybeCreateTaskAttemptRecoveryData(taFinishedEvent.getTaskAttemptID());
                        taRecoveryData.taFinishedEvent = taFinishedEvent;
                        break;
                    }
                default:
                    throw new RuntimeException("Invalid data found, unknown event type " + eventType);
            }
            if (LOG.isDebugEnabled()) {
                LOG.debug("[DAG RECOVERY]" + " dagId=" + lastInProgressDAG + ", eventType=" + eventType + ", event=" + event.toString());
            }
        }
        dagRecoveryStream.close();
    }
    recoveredDAGData.checkRecoverableNonSummary();
    return recoveredDAGData;
}

Example 17 with HistoryEvent

use of org.apache.tez.dag.history.HistoryEvent in project tez by apache.

the class RecoveryParser method readRecoveryEvents.

public static List<HistoryEvent> readRecoveryEvents(TezConfiguration tezConf, ApplicationId appId, int attempt) throws IOException {
    Path tezSystemStagingDir = TezCommonUtils.getTezSystemStagingPath(tezConf, appId.toString());
    Path recoveryDataDir = TezCommonUtils.getRecoveryPath(tezSystemStagingDir, tezConf);
    FileSystem fs = tezSystemStagingDir.getFileSystem(tezConf);
    List<HistoryEvent> historyEvents = new ArrayList<HistoryEvent>();
    for (int i = 1; i <= attempt; ++i) {
        Path currentAttemptRecoveryDataDir = TezCommonUtils.getAttemptRecoveryPath(recoveryDataDir, i);
        Path recoveryFilePath = new Path(currentAttemptRecoveryDataDir, appId.toString().replace("application", "dag") + "_1" + TezConstants.DAG_RECOVERY_RECOVER_FILE_SUFFIX);
        if (fs.exists(recoveryFilePath)) {
            LOG.info("Read recovery file:" + recoveryFilePath);
            FSDataInputStream in = null;
            try {
                in = fs.open(recoveryFilePath);
                historyEvents.addAll(RecoveryParser.parseDAGRecoveryFile(in));
            } catch (IOException e) {
                throw e;
            } finally {
                if (in != null) {
                    in.close();
                }
            }
        }
    }
    return historyEvents;
}

Example 18 with HistoryEvent

use of org.apache.tez.dag.history.HistoryEvent in project tez by apache.

the class RecoveryParser method getNextEvent.

private static HistoryEvent getNextEvent(FSDataInputStream inputStream) throws IOException {
    int eventTypeOrdinal = -1;
    try {
        eventTypeOrdinal = inputStream.readInt();
    } catch (EOFException eof) {
        return null;
    }
    if (eventTypeOrdinal < 0 || eventTypeOrdinal >= HistoryEventType.values().length) {
        // reached end
        throw new IOException("Corrupt data found when trying to read next event type" + ", eventTypeOrdinal=" + eventTypeOrdinal);
    }
    HistoryEventType eventType = HistoryEventType.values()[eventTypeOrdinal];
    HistoryEvent event;
    switch(eventType) {
        case AM_LAUNCHED:
            event = new AMLaunchedEvent();
            break;
        case AM_STARTED:
            event = new AMStartedEvent();
            break;
        case DAG_SUBMITTED:
            event = new DAGSubmittedEvent();
            break;
        case DAG_INITIALIZED:
            event = new DAGInitializedEvent();
            break;
        case DAG_STARTED:
            event = new DAGStartedEvent();
            break;
        case DAG_COMMIT_STARTED:
            event = new DAGCommitStartedEvent();
            break;
        case DAG_FINISHED:
            event = new DAGFinishedEvent();
            break;
        case DAG_KILL_REQUEST:
            event = new DAGKillRequestEvent();
            break;
        case CONTAINER_LAUNCHED:
            event = new ContainerLaunchedEvent();
            break;
        case CONTAINER_STOPPED:
            event = new ContainerStoppedEvent();
            break;
        case VERTEX_INITIALIZED:
            event = new VertexInitializedEvent();
            break;
        case VERTEX_CONFIGURE_DONE:
            event = new VertexConfigurationDoneEvent();
            break;
        case VERTEX_STARTED:
            event = new VertexStartedEvent();
            break;
        case VERTEX_COMMIT_STARTED:
            event = new VertexCommitStartedEvent();
            break;
        case VERTEX_GROUP_COMMIT_STARTED:
            event = new VertexGroupCommitStartedEvent();
            break;
        case VERTEX_GROUP_COMMIT_FINISHED:
            event = new VertexGroupCommitFinishedEvent();
            break;
        case VERTEX_FINISHED:
            event = new VertexFinishedEvent();
            break;
        case TASK_STARTED:
            event = new TaskStartedEvent();
            break;
        case TASK_FINISHED:
            event = new TaskFinishedEvent();
            break;
        case TASK_ATTEMPT_STARTED:
            event = new TaskAttemptStartedEvent();
            break;
        case TASK_ATTEMPT_FINISHED:
            event = new TaskAttemptFinishedEvent();
            break;
        default:
            throw new IOException("Invalid data found, unknown event type " + eventType);
    }
    try {
        event.fromProtoStream(inputStream);
    } catch (EOFException eof) {
        return null;
    }
    if (LOG.isDebugEnabled()) {
        LOG.debug("Parsed event from input stream" + ", eventType=" + eventType + ", event=" + event.toString());
    }
    return event;
}

Example 19 with HistoryEvent

use of org.apache.tez.dag.history.HistoryEvent in project tez by apache.

the class TestAMRecovery method testVertexPartiallyFinished_ScatterGather.

/**
 * Fine-grained recovery task-level, In a vertex (v1), task 0 is done task 1
 * is not started. History flush happens. AM dies. Once AM is recovered, task 0 is
 * not re-run. Task 1 is re-run. (SCATTER_GATHER)
 *
 * @throws Exception
 */
@Test(timeout = 120000)
public void testVertexPartiallyFinished_ScatterGather() throws Exception {
    DAG dag = createDAG("VertexPartiallyFinished_ScatterGather", ControlledShuffleVertexManager.class, DataMovementType.SCATTER_GATHER, true);
    TezCounters counters = runDAGAndVerify(dag, DAGStatus.State.SUCCEEDED);
    assertEquals(4, counters.findCounter(DAGCounter.NUM_SUCCEEDED_TASKS).getValue());
    assertEquals(2, counters.findCounter(TestCounter.Counter_1).getValue());
    List<HistoryEvent> historyEvents1 = readRecoveryLog(1);
    List<HistoryEvent> historyEvents2 = readRecoveryLog(2);
    printHistoryEvents(historyEvents1, 1);
    printHistoryEvents(historyEvents1, 2);
    // task_0 of v1 is finished in attempt 1, task_1 of v1 is not finished in
    // attempt 1
    assertEquals(1, findTaskAttemptFinishedEvent(historyEvents1, 0, 0).size());
    assertEquals(0, findTaskAttemptFinishedEvent(historyEvents1, 0, 1).size());
    // task_0 of v1 is finished in attempt 1 and not rerun, task_1 of v1 is
    // finished in attempt 2
    assertEquals(1, findTaskAttemptFinishedEvent(historyEvents2, 0, 0).size());
    assertEquals(1, findTaskAttemptFinishedEvent(historyEvents2, 0, 1).size());
}

Example 20 with HistoryEvent

use of org.apache.tez.dag.history.HistoryEvent in project tez by apache.

the class TestAMRecovery method testVertexPartialFinished_One2One.

/**
 * Fine-grained recovery task-level, In a vertex (v1), task 0 is done task 1
 * is not started. History flush happens. AM dies. Once AM is recovered, task 0 is
 * not re-run. Task 1 is re-run. (ONE_TO_ONE)
 *
 * @throws Exception
 */
@Test(timeout = 120000)
public void testVertexPartialFinished_One2One() throws Exception {
    DAG dag = createDAG("VertexPartialFinished_One2One", ControlledInputReadyVertexManager.class, DataMovementType.ONE_TO_ONE, true);
    TezCounters counters = runDAGAndVerify(dag, DAGStatus.State.SUCCEEDED);
    assertEquals(4, counters.findCounter(DAGCounter.NUM_SUCCEEDED_TASKS).getValue());
    assertEquals(2, counters.findCounter(TestCounter.Counter_1).getValue());
    List<HistoryEvent> historyEvents1 = readRecoveryLog(1);
    List<HistoryEvent> historyEvents2 = readRecoveryLog(2);
    printHistoryEvents(historyEvents1, 1);
    printHistoryEvents(historyEvents1, 2);
    // task_0 of v1 is finished in attempt 1, task_1 of v1 is not finished in
    // attempt 1
    assertEquals(1, findTaskAttemptFinishedEvent(historyEvents1, 0, 0).size());
    assertEquals(0, findTaskAttemptFinishedEvent(historyEvents1, 0, 1).size());
    // task_0 of v1 is finished in attempt 1 and not rerun, task_1 of v1 is
    // finished in attempt 2
    assertEquals(1, findTaskAttemptFinishedEvent(historyEvents2, 0, 0).size());
    assertEquals(1, findTaskAttemptFinishedEvent(historyEvents2, 0, 1).size());
}