Examples with TaskInfo alluxio.job.wire.TaskInfo used on opensource projects

Example 11 with TaskInfo

use of alluxio.job.wire.TaskInfo in project alluxio by Alluxio.

the class TaskExecutorManager method cancelTask.

/**
 * Cancels the given task.
 *
 * @param jobId the job id
 * @param taskId the task id
 */
public synchronized void cancelTask(long jobId, long taskId) {
    Pair<Long, Long> id = new Pair<>(jobId, taskId);
    TaskInfo taskInfo = mUnfinishedTasks.get(id);
    if (!mTaskFutures.containsKey(id) || taskInfo.getStatus().equals(Status.CANCELED)) {
        // job has finished, or failed, or canceled
        return;
    }
    Future<?> future = mTaskFutures.get(id);
    if (!future.cancel(true)) {
        taskInfo.setStatus(Status.FAILED);
        taskInfo.setErrorType("FailedCancel");
        taskInfo.setErrorMessage("Failed to cancel the task");
        LOG.info("Failed to cancel task {} for job {}", taskId, jobId);
    } else {
        taskInfo.setStatus(Status.CANCELED);
        LOG.info("Task {} for job {} canceled", taskId, jobId);
    }
    finishTask(id);
}

Example 12 with TaskInfo

use of alluxio.job.wire.TaskInfo in project alluxio by Alluxio.

the class JobMaster method workerHeartbeat.

/**
 * Updates the tasks' status when a worker periodically heartbeats with the master, and sends the
 * commands for the worker to execute.
 *
 * @param jobWorkerHealth the job worker health info
 * @param taskInfoList the list of the task information
 * @return the list of {@link JobCommand} to the worker
 */
public List<JobCommand> workerHeartbeat(JobWorkerHealth jobWorkerHealth, List<TaskInfo> taskInfoList) throws ResourceExhaustedException {
    long workerId = jobWorkerHealth.getWorkerId();
    String hostname;
    // Run under shared lock for mWorkers
    try (LockResource workersLockShared = new LockResource(mWorkerRWLock.readLock())) {
        MasterWorkerInfo worker = mWorkers.getFirstByField(mIdIndex, workerId);
        if (worker == null) {
            return Collections.singletonList(JobCommand.newBuilder().setRegisterCommand(RegisterCommand.getDefaultInstance()).build());
        }
        hostname = worker.getWorkerAddress().getHost();
        // Update last-update-time of this particular worker under lock
        // to prevent lost worker detector clearing it under race
        worker.updateLastUpdatedTimeMs();
    }
    mWorkerHealth.put(workerId, jobWorkerHealth);
    // Update task infos for all jobs involved
    Map<Long, List<TaskInfo>> taskInfosPerJob = new HashMap<>();
    for (TaskInfo taskInfo : taskInfoList) {
        taskInfo.setWorkerHost(hostname);
        if (!taskInfosPerJob.containsKey(taskInfo.getJobId())) {
            taskInfosPerJob.put(taskInfo.getJobId(), new ArrayList());
        }
        taskInfosPerJob.get(taskInfo.getJobId()).add(taskInfo);
    }
    for (Map.Entry<Long, List<TaskInfo>> taskInfosPair : taskInfosPerJob.entrySet()) {
        PlanCoordinator planCoordinator = mPlanTracker.getCoordinator(taskInfosPair.getKey());
        if (planCoordinator != null) {
            planCoordinator.updateTasks(taskInfosPair.getValue());
        }
    }
    return mCommandManager.pollAllPendingCommands(workerId);
}

Example 13 with TaskInfo

use of alluxio.job.wire.TaskInfo in project alluxio by Alluxio.

the class PlanCoordinator method failTasksForWorker.

/**
 * Fails any incomplete tasks being run on the specified worker.
 *
 * @param workerId the id of the worker to fail tasks for
 */
public void failTasksForWorker(long workerId) {
    synchronized (mPlanInfo) {
        if (mPlanInfo.getStatus().isFinished()) {
            return;
        }
        List<Long> taskIds = mWorkerIdToTaskIds.get(workerId);
        if (taskIds == null) {
            return;
        }
        boolean statusChanged = false;
        for (Long taskId : taskIds) {
            TaskInfo taskInfo = mPlanInfo.getTaskInfo(taskId);
            if (taskInfo == null || taskInfo.getStatus().isFinished()) {
                continue;
            }
            taskInfo.setStatus(Status.FAILED);
            taskInfo.setErrorType("JobWorkerLost");
            taskInfo.setErrorMessage(String.format("Job worker(%s) was lost before " + "the task(%d) could complete", taskInfo.getWorkerHost(), taskId));
            statusChanged = true;
            break;
        }
        if (statusChanged) {
            updateStatus();
        }
    }
}

Example 14 with TaskInfo

use of alluxio.job.wire.TaskInfo in project alluxio by Alluxio.

the class PlanCoordinator method updateStatus.

/**
 * Updates the status of the job. When all the tasks are completed, run the join method in the
 * definition.
 */
private synchronized void updateStatus() {
    int completed = 0;
    List<TaskInfo> taskInfoList = mPlanInfo.getTaskInfoList();
    JobConfig config = mPlanInfo.getJobConfig();
    Preconditions.checkNotNull(config);
    FileSystem fileSystem = mJobServerContext.getFileSystem();
    for (TaskInfo info : taskInfoList) {
        Status status = info.getStatus();
        switch(status) {
            case FAILED:
                setJobAsFailed(info.getErrorType(), "Task execution failed: " + info.getErrorMessage());
                return;
            case CANCELED:
                if (mPlanInfo.getStatus() != Status.FAILED) {
                    mPlanInfo.setStatus(Status.CANCELED);
                    DistributedCmdMetrics.incrementForAllConfigsCancelStatus(config);
                }
                return;
            case RUNNING:
                if (mPlanInfo.getStatus() != Status.FAILED && mPlanInfo.getStatus() != Status.CANCELED) {
                    mPlanInfo.setStatus(Status.RUNNING);
                }
                break;
            case COMPLETED:
                completed++;
                break;
            case CREATED:
                // do nothing
                break;
            default:
                throw new IllegalArgumentException("Unsupported status " + info.getStatus());
        }
    }
    if (completed == taskInfoList.size()) {
        if (mPlanInfo.getStatus() == Status.COMPLETED) {
            return;
        }
        // all the tasks completed, run join
        try {
            // Try to join first, so that in case of failure we don't move to a completed state yet
            mPlanInfo.setResult(join(taskInfoList));
            mPlanInfo.setStatus(Status.COMPLETED);
            // Increment the counter for Complete status when all the tasks in a job are completed.
            DistributedCmdMetrics.incrementForAllConfigsCompleteStatus(config, fileSystem, new CountingRetry(5));
        } catch (Exception e) {
            LOG.warn("Job error when joining tasks Job Id={} Config={}", mPlanInfo.getId(), mPlanInfo.getJobConfig(), e);
            setJobAsFailed(ErrorUtils.getErrorType(e), e.getMessage());
        }
    }
}

Example 15 with TaskInfo

use of alluxio.job.wire.TaskInfo in project alluxio by Alluxio.

the class StatCommand method formatOutput.

private String formatOutput(CommandLine cl, JobInfo info) {
    StringBuilder output = new StringBuilder();
    output.append("ID: ").append(info.getId()).append("\n");
    output.append("Name: ").append(info.getName()).append("\n");
    output.append("Description: ");
    if (cl.hasOption("v")) {
        output.append(info.getDescription());
    } else {
        output.append(StringUtils.abbreviate(info.getDescription(), 200));
    }
    output.append("\n");
    output.append("Status: ").append(info.getStatus()).append("\n");
    if (info.getErrorMessage() != null && !info.getErrorMessage().isEmpty()) {
        output.append("Error: ").append(info.getErrorMessage()).append("\n");
    }
    if (info.getResult() != null && !info.getResult().toString().isEmpty()) {
        output.append("Result: ").append(info.getResult().toString()).append("\n");
    }
    if (cl.hasOption("v")) {
        for (JobInfo childInfo : info.getChildren()) {
            output.append("Task ").append(childInfo.getId()).append("\n");
            if (childInfo instanceof TaskInfo) {
                TaskInfo taskInfo = (TaskInfo) childInfo;
                if (taskInfo.getWorkerHost() != null) {
                    output.append("\t").append("Worker: ").append(taskInfo.getWorkerHost()).append("\n");
                }
            }
            if (!childInfo.getDescription().isEmpty()) {
                output.append("\t").append("Description: ").append(StringUtils.abbreviate(childInfo.getDescription(), 200)).append("\n");
            }
            output.append("\t").append("Status: ").append(childInfo.getStatus()).append("\n");
            if (childInfo.getErrorMessage() != null && !childInfo.getErrorMessage().isEmpty()) {
                output.append("\t").append("Error: ").append(childInfo.getErrorMessage()).append("\n");
            }
            if (childInfo.getResult() != null) {
                output.append("\t").append("Result: ").append(childInfo.getResult()).append("\n");
            }
        }
    }
    return output.toString();
}