Search in sources :

Example 1 with WallTimer

use of org.ow2.proactive.scheduler.task.utils.WallTimer in project scheduling by ow2-proactive.

the class TaskLauncher method doTask.

public void doTask(ExecutableContainer executableContainer, TaskResult[] previousTasksResults, TaskTerminateNotification terminateNotification, String terminateNotificationNodeURL, boolean taskRecoverable) {
    TaskResultImpl taskResult;
    WallTimer wallTimer = null;
    TaskContext context = null;
    Stopwatch taskStopwatchForFailures = null;
    TaskDataspaces dataspaces = null;
    File taskLogFile = null;
    try {
        taskStarted.set(true);
        logger.info("Task started " + taskId.getJobId().getReadableName() + " : " + taskId.getReadableName());
        wallTimer = new WallTimer(initializer.getWalltime(), taskKiller);
        taskStopwatchForFailures = Stopwatch.createUnstarted();
        taskLauncherRebinder = new TaskLauncherRebinder(taskId, terminateNotificationNodeURL, taskRecoverable);
        addShutdownHook();
        if (decrypter != null) {
            decrypter.setCredentials(executableContainer.getCredentials());
        }
        // lock the cache space cleaning mechanism
        DataSpaceNodeConfigurationAgent.lockCacheSpaceCleaning();
        dataspaces = factory.createTaskDataspaces(taskId, initializer.getNamingService(), executableContainer.isRunAsUser(), decrypter, taskLogger);
        copyTaskLogsFromUserSpace(taskLogger.createLogFilePath(dataspaces.getScratchFolder()), dataspaces);
        taskLogFile = taskLogger.createFileAppender(dataspaces.getScratchFolder());
        progressFileReader.start(dataspaces.getScratchFolder(), taskId);
        context = new TaskContext(executableContainer, initializer, previousTasksResults, new NodeDataSpacesURIs(dataspaces.getScratchURI(), dataspaces.getCacheURI(), dataspaces.getInputURI(), dataspaces.getOutputURI(), dataspaces.getUserURI(), dataspaces.getGlobalURI()), progressFileReader.getProgressFile().toString(), new NodeInfo(getHostName(), getNodeUrl(), getNodeName(), getNodeSourceName()), decrypter);
        File workingDir = getTaskWorkingDir(context, dataspaces);
        logger.info("Task working dir: " + workingDir);
        logger.info("Cache space: " + context.getNodeDataSpaceURIs().getCacheURI());
        logger.info("Input space: " + context.getNodeDataSpaceURIs().getInputURI());
        logger.info("Output space: " + context.getNodeDataSpaceURIs().getOutputURI());
        logger.info("User space: " + context.getNodeDataSpaceURIs().getUserURI());
        logger.info("Global space: " + context.getNodeDataSpaceURIs().getGlobalURI());
        logger.info("Scheduler rest url: " + context.getSchedulerRestUrl());
        wallTimer.start();
        // should handle interrupt
        dataspaces.copyInputDataToScratch(initializer.getFilteredInputFiles(fileSelectorsFilters(context)));
        TaskExecutor taskExecutor = factory.createTaskExecutor(workingDir);
        initializer.fetchUrlsIfNeeded();
        ((ScriptExecutableContainer) executableContainer).getScript().fetchUrlIfNeeded();
        taskStopwatchForFailures.start();
        taskResult = taskExecutor.execute(context, taskLogger.getOutputSink(), taskLogger.getErrorSink());
        taskStopwatchForFailures.stop();
        // by the time the task finishes, the scheduler might have had a
        // transient failure, so we need to make sure that the placeholder
        // for the task's result still exists, or get the new place for
        // the result if it does not exist anymore.
        TaskTerminateNotification rebindedTerminateNotification = taskLauncherRebinder.makeSureSchedulerIsConnected(terminateNotification);
        switch(taskKiller.getStatus()) {
            case WALLTIME_REACHED:
                taskResult = getWalltimedTaskResult(context, taskStopwatchForFailures);
                copyTaskLogsToUserSpace(taskLogFile, dataspaces);
                sendResultToScheduler(rebindedTerminateNotification, taskResult);
                return;
            case KILLED_MANUALLY:
                // killed by Scheduler, no need to send results back
                copyTaskLogsToUserSpace(taskLogFile, dataspaces);
                return;
        }
        dataspaces.copyScratchDataToOutput(initializer.getFilteredOutputFiles(fileSelectorsFilters(context, taskResult)));
        wallTimer.stop();
        copyTaskLogsToUserSpace(taskLogFile, dataspaces);
        sendResultToScheduler(rebindedTerminateNotification, taskResult);
    } catch (Throwable taskFailure) {
        if (wallTimer != null) {
            wallTimer.stop();
        }
        switch(taskKiller.getStatus()) {
            case WALLTIME_REACHED:
                taskResult = getWalltimedTaskResult(context, taskStopwatchForFailures);
                copyTaskLogsToUserSpace(taskLogFile, dataspaces);
                sendResultToScheduler(terminateNotification, taskResult);
                break;
            case KILLED_MANUALLY:
                // killed by Scheduler, no need to send results back
                copyTaskLogsToUserSpace(taskLogFile, dataspaces);
                return;
            default:
                logger.info("Failed to execute task", taskFailure);
                long elapsedTime = 0;
                if (taskStopwatchForFailures != null) {
                    elapsedTime = taskStopwatchForFailures.elapsed(TimeUnit.MILLISECONDS);
                }
                taskFailure.printStackTrace(taskLogger.getErrorSink());
                Map<String, byte[]> serializedVariables = extractVariablesFromContext(context);
                taskResult = new TaskResultImpl(taskId, taskFailure, taskLogger.getLogs(), elapsedTime);
                taskResult.setPropagatedVariables(serializedVariables);
                sendResultToScheduler(terminateNotification, taskResult);
        }
    } finally {
        try {
            progressFileReader.stop();
            taskLogger.close();
            if (dataspaces != null) {
                dataspaces.close();
            }
            // unlocks the cache space cleaning thread
            DataSpaceNodeConfigurationAgent.unlockCacheSpaceCleaning();
            removeShutdownHook();
        } finally {
            terminate();
        }
    }
}
Also used : TaskContext(org.ow2.proactive.scheduler.task.context.TaskContext) TaskDataspaces(org.ow2.proactive.scheduler.task.data.TaskDataspaces) Stopwatch(com.google.common.base.Stopwatch) WallTimer(org.ow2.proactive.scheduler.task.utils.WallTimer) NodeDataSpacesURIs(org.ow2.proactive.scheduler.task.context.NodeDataSpacesURIs) TaskTerminateNotification(org.ow2.proactive.scheduler.common.TaskTerminateNotification) TaskExecutor(org.ow2.proactive.scheduler.task.executors.TaskExecutor) NodeInfo(org.ow2.proactive.scheduler.task.context.NodeInfo) File(java.io.File) Map(java.util.Map)

Aggregations

Stopwatch (com.google.common.base.Stopwatch)1 File (java.io.File)1 Map (java.util.Map)1 TaskTerminateNotification (org.ow2.proactive.scheduler.common.TaskTerminateNotification)1 NodeDataSpacesURIs (org.ow2.proactive.scheduler.task.context.NodeDataSpacesURIs)1 NodeInfo (org.ow2.proactive.scheduler.task.context.NodeInfo)1 TaskContext (org.ow2.proactive.scheduler.task.context.TaskContext)1 TaskDataspaces (org.ow2.proactive.scheduler.task.data.TaskDataspaces)1 TaskExecutor (org.ow2.proactive.scheduler.task.executors.TaskExecutor)1 WallTimer (org.ow2.proactive.scheduler.task.utils.WallTimer)1