Search in sources :

Example 1 with TaskTerminateNotification

use of org.ow2.proactive.scheduler.common.TaskTerminateNotification 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;
    try {
        taskStarted.set(true);
        logger.info("Task started " + taskId.getJobId().getReadableName() + " : " + taskId.getReadableName());
        this.taskKiller = this.replaceTaskKillerWithDoubleTimeoutValueIfRunAsMe(executableContainer.isRunAsUser());
        wallTimer = new WallTimer(initializer.getWalltime(), taskKiller);
        taskStopwatchForFailures = Stopwatch.createUnstarted();
        taskLauncherRebinder = new TaskLauncherRebinder(taskId, terminateNotificationNodeURL, taskRecoverable);
        addShutdownHook();
        // lock the cache space cleaning mechanism
        DataSpaceNodeConfigurationAgent.lockCacheSpaceCleaning();
        dataspaces = factory.createTaskDataspaces(taskId, initializer.getNamingService(), executableContainer.isRunAsUser());
        File 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(), getHostname(), 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)));
        if (decrypter != null) {
            decrypter.setCredentials(executableContainer.getCredentials());
        }
        TaskExecutor taskExecutor = factory.createTaskExecutor(workingDir);
        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);
                sendResultToScheduler(rebindedTerminateNotification, taskResult);
                return;
            case KILLED_MANUALLY:
                // killed by Scheduler, no need to send results back
                return;
        }
        dataspaces.copyScratchDataToOutput(initializer.getFilteredOutputFiles(fileSelectorsFilters(context, taskResult)));
        wallTimer.stop();
        copyTaskLogsToUserSpace(taskLogFile, dataspaces);
        taskResult.setLogs(taskLogger.getLogs());
        sendResultToScheduler(rebindedTerminateNotification, taskResult);
    } catch (Throwable taskFailure) {
        if (wallTimer != null) {
            wallTimer.stop();
        }
        switch(taskKiller.getStatus()) {
            case WALLTIME_REACHED:
                taskResult = getWalltimedTaskResult(context, taskStopwatchForFailures);
                sendResultToScheduler(terminateNotification, taskResult);
                break;
            case KILLED_MANUALLY:
                // killed by Scheduler, no need to send results back
                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 : TaskTerminateNotification(org.ow2.proactive.scheduler.common.TaskTerminateNotification) TaskExecutor(org.ow2.proactive.scheduler.task.executors.TaskExecutor) 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) File(java.io.File) Map(java.util.Map)

Example 2 with TaskTerminateNotification

use of org.ow2.proactive.scheduler.common.TaskTerminateNotification in project scheduling by ow2-proactive.

the class TaskLauncher method sendResultToScheduler.

private void sendResultToScheduler(TaskTerminateNotification terminateNotification, TaskResultImpl taskResult) {
    if (isNodeShuttingDown()) {
        return;
    }
    int pingAttempts = initializer.getPingAttempts();
    int pingPeriodMs = initializer.getPingPeriod() * 1000;
    // We are going to contact the recipient of the task result. This
    // recipient is the TaskTerminateNotification, an active object on the
    // scheduler side. If the scheduler experienced a transient failure
    // while the task was computing, then the reference to this
    // TaskTerminateNotification is obsolete and we need to update it. This
    // is what the following code does.
    TaskTerminateNotification currentTerminateNotification = terminateNotification;
    for (int i = 0; i < pingAttempts; i++) {
        try {
            currentTerminateNotification.terminate(taskId, taskResult);
            logger.debug("Successfully notified task termination " + taskId);
            // termination has succeeded, exit the method
            return;
        } catch (Throwable t) {
            logger.warn("Cannot notify task termination, trying to rebind to the task termination handler");
            TaskTerminateNotification rebindedTerminateNotification = taskLauncherRebinder.getReboundTaskTerminateNotificationHandler(t);
            if (rebindedTerminateNotification != null) {
                currentTerminateNotification = rebindedTerminateNotification;
            } else {
                decreasePingAttemptsAndWait(pingAttempts, pingPeriodMs, i, t);
            }
        }
    }
    logger.error("Cannot notify task termination " + taskId + " after " + pingAttempts + " attempts, terminating task launcher now");
}
Also used : TaskTerminateNotification(org.ow2.proactive.scheduler.common.TaskTerminateNotification)

Example 3 with TaskTerminateNotification

use of org.ow2.proactive.scheduler.common.TaskTerminateNotification in project scheduling by ow2-proactive.

the class TaskLauncherRebinder method getReboundTaskTerminateNotificationHandler.

/**
 * Attempt to reaquire a correct reference to the TaskTerminateNotification
 * active object from a previously saved URL for this object.
 *
 * @return a reference to a TaskTerminateNotification that is bounded to the
 * terminateNotificationNodeURL, or null if it cannot be retrieved or if
 * task is not recoverable
 */
TaskTerminateNotification getReboundTaskTerminateNotificationHandler(Throwable error) {
    if (taskRecoverable) {
        try {
            logger.debug("List AOs on " + terminateNotificationNodeURL + " (expect only one): " + Arrays.toString(NodeFactory.getNode(terminateNotificationNodeURL).getActiveObjects(TaskTerminateNotification.class.getName())));
            Node node = NodeFactory.getNode(terminateNotificationNodeURL);
            Object[] aos = node.getActiveObjects(TaskTerminateNotification.class.getName());
            logger.info("On node " + node.getNodeInformation().getName() + " number of active objects found is " + aos.length + " and the first one " + aos[0] + " will be used to send back the task result");
            return (TaskTerminateNotification) aos[0];
        } catch (Throwable t) {
            // error when retrieving the termination handler after reconnection
            logger.error("Failed to rebind TaskTerminatedNotification handler of task " + taskId.getReadableName() + " from URL " + terminateNotificationNodeURL + " after exception " + error.getMessage(), t);
            return null;
        }
    } else {
        return null;
    }
}
Also used : TaskTerminateNotification(org.ow2.proactive.scheduler.common.TaskTerminateNotification) Node(org.objectweb.proactive.core.node.Node) PAActiveObject(org.objectweb.proactive.api.PAActiveObject)

Aggregations

TaskTerminateNotification (org.ow2.proactive.scheduler.common.TaskTerminateNotification)3 Stopwatch (com.google.common.base.Stopwatch)1 File (java.io.File)1 Map (java.util.Map)1 PAActiveObject (org.objectweb.proactive.api.PAActiveObject)1 Node (org.objectweb.proactive.core.node.Node)1 NodeDataSpacesURIs (org.ow2.proactive.scheduler.task.context.NodeDataSpacesURIs)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