Search in sources :

Example 1 with JobDescriptorImpl

use of org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl in project scheduling by ow2-proactive.

the class SchedulingMethodImpl method selectAndStartTasks.

private int selectAndStartTasks(Policy currentPolicy, Map<JobId, JobDescriptor> jobMap, Set<String> freeResources, LinkedList<EligibleTaskDescriptor> fullListOfTaskRetrievedFromPolicy) {
    int numberOfTaskStarted = 0;
    VariableBatchSizeIterator progressiveIterator = new VariableBatchSizeIterator(fullListOfTaskRetrievedFromPolicy);
    while (progressiveIterator.hasMoreElements() && !freeResources.isEmpty()) {
        LinkedList<EligibleTaskDescriptor> taskRetrievedFromPolicy = new LinkedList<>(progressiveIterator.getNextElements(freeResources.size()));
        if (logger.isDebugEnabled()) {
            loggingEligibleTasksDetails(fullListOfTaskRetrievedFromPolicy, taskRetrievedFromPolicy);
        }
        updateVariablesForTasksToSchedule(taskRetrievedFromPolicy);
        for (EligibleTaskDescriptor etd : taskRetrievedFromPolicy) {
            // load and Initialize the executable container
            loadAndInit(((EligibleTaskDescriptorImpl) etd).getInternal());
        }
        while (!taskRetrievedFromPolicy.isEmpty()) {
            if (freeResources.isEmpty()) {
                break;
            }
            // get the next compatible tasks from the whole returned policy tasks
            LinkedList<EligibleTaskDescriptor> tasksToSchedule = new LinkedList<>();
            int neededResourcesNumber = 0;
            while (!taskRetrievedFromPolicy.isEmpty() && neededResourcesNumber == 0) {
                // the loop will search for next compatible task until it find something
                neededResourcesNumber = getNextcompatibleTasks(jobMap, taskRetrievedFromPolicy, freeResources.size(), tasksToSchedule);
            }
            if (logger.isDebugEnabled()) {
                logger.debug("tasksToSchedule : " + tasksToSchedule);
            }
            logger.debug("required number of nodes : " + neededResourcesNumber);
            if (neededResourcesNumber == 0 || tasksToSchedule.isEmpty()) {
                break;
            }
            NodeSet nodeSet = getRMNodes(jobMap, neededResourcesNumber, tasksToSchedule, freeResources);
            if (nodeSet != null) {
                freeResources.removeAll(nodeSet.getAllNodesUrls());
            }
            // start selected tasks
            Node node = null;
            InternalJob currentJob = null;
            try {
                while (nodeSet != null && !nodeSet.isEmpty()) {
                    EligibleTaskDescriptor taskDescriptor = tasksToSchedule.removeFirst();
                    currentJob = ((JobDescriptorImpl) jobMap.get(taskDescriptor.getJobId())).getInternal();
                    InternalTask internalTask = ((EligibleTaskDescriptorImpl) taskDescriptor).getInternal();
                    if (currentPolicy.isTaskExecutable(nodeSet, taskDescriptor)) {
                        // create launcher and try to start the task
                        node = nodeSet.get(0);
                        if (createExecution(nodeSet, node, currentJob, internalTask, taskDescriptor)) {
                            numberOfTaskStarted++;
                        }
                    }
                    // if every task that should be launched have been removed
                    if (tasksToSchedule.isEmpty()) {
                        // get back unused nodes to the RManager
                        if (!nodeSet.isEmpty()) {
                            releaseNodes(currentJob, nodeSet);
                            freeResources.addAll(nodeSet.getAllNodesUrls());
                        }
                        // and leave the loop
                        break;
                    }
                }
            } catch (ActiveObjectCreationException e1) {
                // Something goes wrong with the active object creation (createLauncher)
                logger.warn("An exception occured while creating the task launcher.", e1);
                // so try to get back every remaining nodes to the resource manager
                try {
                    releaseNodes(currentJob, nodeSet);
                    freeResources.addAll(nodeSet.getAllNodesUrls());
                } catch (Exception e2) {
                    logger.info("Unable to get back the nodeSet to the RM", e2);
                }
                if (--activeObjectCreationRetryTimeNumber == 0) {
                    break;
                }
            } catch (Exception e1) {
                // if we are here, it is that something append while launching the current task.
                logger.warn("An exception occured while starting task.", e1);
                // so try to get back every remaining nodes to the resource manager
                try {
                    releaseNodes(currentJob, nodeSet);
                    freeResources.addAll(nodeSet.getAllNodesUrls());
                } catch (Exception e2) {
                    logger.info("Unable to get back the nodeSet to the RM", e2);
                }
            }
        }
        if (freeResources.isEmpty()) {
            break;
        }
        if (activeObjectCreationRetryTimeNumber == 0) {
            break;
        }
    }
    return numberOfTaskStarted;
}
Also used : NodeSet(org.ow2.proactive.utils.NodeSet) InternalJob(org.ow2.proactive.scheduler.job.InternalJob) EligibleTaskDescriptor(org.ow2.proactive.scheduler.descriptor.EligibleTaskDescriptor) InternalTask(org.ow2.proactive.scheduler.task.internal.InternalTask) Node(org.objectweb.proactive.core.node.Node) EligibleTaskDescriptorImpl(org.ow2.proactive.scheduler.descriptor.EligibleTaskDescriptorImpl) VariableBatchSizeIterator(org.ow2.proactive.scheduler.core.helpers.VariableBatchSizeIterator) LinkedList(java.util.LinkedList) ActiveObjectCreationException(org.objectweb.proactive.ActiveObjectCreationException) ActiveObjectCreationException(org.objectweb.proactive.ActiveObjectCreationException) TopologyDisabledException(org.ow2.proactive.resourcemanager.frontend.topology.TopologyDisabledException) InvalidScriptException(org.ow2.proactive.scripting.InvalidScriptException) RMProxyCreationException(org.ow2.proactive.scheduler.core.rmproxies.RMProxyCreationException) IOException(java.io.IOException)

Example 2 with JobDescriptorImpl

use of org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl in project scheduling by ow2-proactive.

the class SchedulingMethodImpl method getNextcompatibleTasks.

/**
 * Extract the n first compatible tasks from the first argument list,
 * and return them according that the extraction is stopped when the maxResource number is reached.<br>
 * Two tasks are compatible if and only if they have the same list of selection script and
 * the same list of node exclusion.
 * The check of compliance is currently done by the {@link SchedulingTaskComparator} class.<br>
 * This method has two side effects : extracted tasks are removed from the bagOfTasks and put in the toFill list
 *
 * @param bagOfTasks the list of tasks form which to extract tasks
 * @param maxResource the limit number of resources that the extraction should not exceed
 * @param toFill the list that will contains the task to schedule at the end. This list must not be null but must be empty.<br>
 * 		  this list will be filled with the n first compatible tasks according that the number of resources needed
 * 		  by these tasks does not exceed the given max resource number.
 * @return the number of nodes needed to start every task present in the 'toFill' argument at the end of the method.
 */
protected int getNextcompatibleTasks(Map<JobId, JobDescriptor> jobsMap, LinkedList<EligibleTaskDescriptor> bagOfTasks, int maxResource, LinkedList<EligibleTaskDescriptor> toFill) {
    if (toFill == null || bagOfTasks == null) {
        throw new IllegalArgumentException("The two given lists must not be null !");
    }
    int neededResource = 0;
    if (!PASchedulerProperties.SCHEDULER_REST_URL.isSet()) {
        Iterator<EligibleTaskDescriptor> it = bagOfTasks.iterator();
        EligibleTaskDescriptor etd;
        while (it.hasNext()) {
            etd = it.next();
            if (checkEligibleTaskDescriptorScript.isTaskContainsAPIBinding(etd)) {
                // skip task here
                it.remove();
            }
        }
    }
    if (maxResource > 0 && !bagOfTasks.isEmpty()) {
        EligibleTaskDescriptor etd = bagOfTasks.removeFirst();
        ((EligibleTaskDescriptorImpl) etd).addAttempt();
        InternalJob currentJob = ((JobDescriptorImpl) jobsMap.get(etd.getJobId())).getInternal();
        InternalTask internalTask = currentJob.getIHMTasks().get(etd.getTaskId());
        int neededNodes = internalTask.getNumberOfNodesNeeded();
        SchedulingTaskComparator referent = new SchedulingTaskComparator(internalTask, currentJob);
        boolean firstLoop = true;
        do {
            if (!firstLoop) {
                // if bagOfTasks is not empty
                if (!bagOfTasks.isEmpty()) {
                    etd = bagOfTasks.removeFirst();
                    ((EligibleTaskDescriptorImpl) etd).addAttempt();
                    currentJob = ((JobDescriptorImpl) jobsMap.get(etd.getJobId())).getInternal();
                    internalTask = currentJob.getIHMTasks().get(etd.getTaskId());
                    neededNodes = internalTask.getNumberOfNodesNeeded();
                }
            } else {
                firstLoop = false;
            }
            if (neededNodes > maxResource) {
            // no instruction is important :
            // in this case, a multi node task leads the search to be stopped and the
            // the current task would be retried on the next step
            // we continue to start the maximum number of task in a single scheduling loop.
            // this case will focus on starting single node task first if lot of resources are busy.
            // (multi-nodes starvation may occurs)
            } else {
                // check if the task is compatible with the other previous one
                if (referent.equals(new SchedulingTaskComparator(internalTask, currentJob))) {
                    tlogger.debug(internalTask.getId(), "scheduling");
                    neededResource += neededNodes;
                    maxResource -= neededNodes;
                    toFill.add(etd);
                } else {
                    bagOfTasks.addFirst(etd);
                    break;
                }
            }
        } while (maxResource > 0 && !bagOfTasks.isEmpty());
    }
    return neededResource;
}
Also used : InternalJob(org.ow2.proactive.scheduler.job.InternalJob) EligibleTaskDescriptor(org.ow2.proactive.scheduler.descriptor.EligibleTaskDescriptor) InternalTask(org.ow2.proactive.scheduler.task.internal.InternalTask) EligibleTaskDescriptorImpl(org.ow2.proactive.scheduler.descriptor.EligibleTaskDescriptorImpl) JobDescriptorImpl(org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl)

Example 3 with JobDescriptorImpl

use of org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl in project scheduling by ow2-proactive.

the class SchedulingServiceTest2 method testFailedTask.

private void testFailedTask(boolean failNativeTask) throws Exception {
    service.submitJob(createJob(createTestJob()));
    listener.assertEvents(SchedulerEvent.JOB_SUBMITTED);
    Map<JobId, JobDescriptor> jobsMap;
    JobDescriptor jobDesc;
    jobsMap = service.lockJobsToSchedule();
    assertEquals(1, jobsMap.size());
    jobDesc = jobsMap.values().iterator().next();
    Assert.assertEquals(2, jobDesc.getEligibleTasks().size());
    for (TaskDescriptor taskDesc : jobDesc.getEligibleTasks()) {
        taskStarted(jobDesc, (EligibleTaskDescriptor) taskDesc);
    }
    service.unlockJobsToSchedule(jobsMap.values());
    if (failNativeTask) {
        InternalTask nativeTask = ((JobDescriptorImpl) jobDesc).getInternal().getTask("nativeTask");
        // native task terminates with code 1, flag 'cancelJobOnError' was
        // set so job should be cancelled
        service.taskTerminatedWithResult(nativeTask.getId(), new TaskResultImpl(nativeTask.getId(), new RuntimeException(), null, 0));
    } else {
        InternalTask javaTask = ((JobDescriptorImpl) jobDesc).getInternal().getTask("javaTask");
        // java task terminates with exception, flag 'cancelJobOnError' was
        // set so job should be cancelled
        service.taskTerminatedWithResult(javaTask.getId(), new TaskResultImpl(javaTask.getId(), new RuntimeException(), null, 0));
    }
    jobsMap = service.lockJobsToSchedule();
    assertEquals(0, jobsMap.size());
    listener.assertEvents(SchedulerEvent.JOB_PENDING_TO_RUNNING, SchedulerEvent.JOB_UPDATED, SchedulerEvent.TASK_PENDING_TO_RUNNING, SchedulerEvent.TASK_PENDING_TO_RUNNING, SchedulerEvent.TASK_RUNNING_TO_FINISHED, SchedulerEvent.TASK_RUNNING_TO_FINISHED, SchedulerEvent.JOB_RUNNING_TO_FINISHED, SchedulerEvent.JOB_UPDATED);
    infrastructure.assertRequests(2);
}
Also used : TaskDescriptor(org.ow2.proactive.scheduler.common.TaskDescriptor) EligibleTaskDescriptor(org.ow2.proactive.scheduler.descriptor.EligibleTaskDescriptor) TaskResultImpl(org.ow2.proactive.scheduler.task.TaskResultImpl) JobDescriptor(org.ow2.proactive.scheduler.common.JobDescriptor) InternalTask(org.ow2.proactive.scheduler.task.internal.InternalTask) JobId(org.ow2.proactive.scheduler.common.job.JobId)

Example 4 with JobDescriptorImpl

use of org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl in project scheduling by ow2-proactive.

the class SchedulingServiceTest3 method testTaskKill.

@Test
public void testTaskKill() throws Exception {
    service.submitJob(createJob(createTestJob(false)));
    listener.assertEvents(SchedulerEvent.JOB_SUBMITTED);
    Map<JobId, JobDescriptor> jobsMap;
    JobDescriptor jobDesc;
    jobsMap = service.lockJobsToSchedule();
    assertEquals(1, jobsMap.size());
    jobDesc = jobsMap.values().iterator().next();
    Assert.assertEquals(2, jobDesc.getEligibleTasks().size());
    for (TaskDescriptor taskDesc : jobDesc.getEligibleTasks()) {
        taskStarted(jobDesc, (EligibleTaskDescriptor) taskDesc);
    }
    service.unlockJobsToSchedule(jobsMap.values());
    try {
        service.killTask(jobDesc.getJobId(), "invalid task name");
        Assert.fail();
    } catch (UnknownTaskException e) {
    }
    try {
        service.killTask(JobIdImpl.makeJobId("1234567"), "javaTask");
        Assert.fail();
    } catch (UnknownJobException e) {
    }
    Assert.assertTrue(service.killTask(jobDesc.getJobId(), "javaTask"));
    listener.assertEvents(SchedulerEvent.JOB_PENDING_TO_RUNNING, SchedulerEvent.JOB_UPDATED, SchedulerEvent.TASK_PENDING_TO_RUNNING, SchedulerEvent.TASK_PENDING_TO_RUNNING, SchedulerEvent.TASK_RUNNING_TO_FINISHED);
    infrastructure.assertRequests(1);
    Assert.assertFalse(service.killTask(jobDesc.getJobId(), "javaTask"));
    TaskId nativeTaskId = ((JobDescriptorImpl) jobDesc).getInternal().getTask("nativeTask").getId();
    service.taskTerminatedWithResult(nativeTaskId, new TaskResultImpl(nativeTaskId, new Integer(0), null, 0));
    listener.assertEvents(SchedulerEvent.TASK_RUNNING_TO_FINISHED, SchedulerEvent.JOB_RUNNING_TO_FINISHED, SchedulerEvent.JOB_UPDATED);
    infrastructure.assertRequests(1);
    try {
        service.killTask(jobDesc.getJobId(), "javaTask");
    } catch (UnknownJobException e) {
        Assert.fail("The job should still exist in the memory context as the auto Job removal feature isn't enabled");
    }
}
Also used : UnknownTaskException(org.ow2.proactive.scheduler.common.exception.UnknownTaskException) TaskDescriptor(org.ow2.proactive.scheduler.common.TaskDescriptor) EligibleTaskDescriptor(org.ow2.proactive.scheduler.descriptor.EligibleTaskDescriptor) TaskId(org.ow2.proactive.scheduler.common.task.TaskId) TaskResultImpl(org.ow2.proactive.scheduler.task.TaskResultImpl) JobDescriptor(org.ow2.proactive.scheduler.common.JobDescriptor) UnknownJobException(org.ow2.proactive.scheduler.common.exception.UnknownJobException) JobDescriptorImpl(org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl) JobId(org.ow2.proactive.scheduler.common.job.JobId) Test(org.junit.Test)

Example 5 with JobDescriptorImpl

use of org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl in project scheduling by ow2-proactive.

the class TaskResultCreatorTest method testPausedTasksAreCheckedForATaskId.

@Test
public void testPausedTasksAreCheckedForATaskId() throws UnknownTaskException {
    TaskResultCreator taskResultCreator = new TaskResultCreator();
    TaskResultImpl mockedTaskResultImpl = mock(TaskResultImpl.class);
    Map<TaskId, TaskResult> loadTaskResultsValue = new HashMap<>();
    loadTaskResultsValue.put(this.createTaskID(), mockedTaskResultImpl);
    SchedulerDBManager mockedschedulerDbManager = mock(SchedulerDBManager.class);
    when(mockedschedulerDbManager.loadTasksResults(any(JobId.class), any(List.class))).thenReturn(loadTaskResultsValue);
    JobDescriptorImpl mockedJobDescriptorHasPausedTask = this.getMockedJobDescriptorWithPausedTask();
    taskResultCreator.getTaskResult(mockedschedulerDbManager, this.getMockedInternalJob(mockedJobDescriptorHasPausedTask), this.getMockedInternalTask());
    verify(mockedJobDescriptorHasPausedTask, atLeastOnce()).getPausedTasks();
}
Also used : TaskId(org.ow2.proactive.scheduler.common.task.TaskId) TaskResultImpl(org.ow2.proactive.scheduler.task.TaskResultImpl) HashMap(java.util.HashMap) SchedulerDBManager(org.ow2.proactive.scheduler.core.db.SchedulerDBManager) TaskResult(org.ow2.proactive.scheduler.common.task.TaskResult) List(java.util.List) JobDescriptorImpl(org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl) JobId(org.ow2.proactive.scheduler.common.job.JobId) Test(org.junit.Test)

Aggregations

JobDescriptorImpl (org.ow2.proactive.scheduler.descriptor.JobDescriptorImpl)16 Test (org.junit.Test)12 JobDescriptor (org.ow2.proactive.scheduler.common.JobDescriptor)10 EligibleTaskDescriptor (org.ow2.proactive.scheduler.descriptor.EligibleTaskDescriptor)9 InternalTask (org.ow2.proactive.scheduler.task.internal.InternalTask)9 TaskId (org.ow2.proactive.scheduler.common.task.TaskId)8 TaskResultImpl (org.ow2.proactive.scheduler.task.TaskResultImpl)8 HashMap (java.util.HashMap)6 JobId (org.ow2.proactive.scheduler.common.job.JobId)6 InternalJob (org.ow2.proactive.scheduler.job.InternalJob)5 TaskDescriptor (org.ow2.proactive.scheduler.common.TaskDescriptor)4 ArrayList (java.util.ArrayList)3 Map (java.util.Map)3 UnknownJobException (org.ow2.proactive.scheduler.common.exception.UnknownJobException)3 UnknownTaskException (org.ow2.proactive.scheduler.common.exception.UnknownTaskException)3 EligibleTaskDescriptorImpl (org.ow2.proactive.scheduler.descriptor.EligibleTaskDescriptorImpl)3 IOException (java.io.IOException)2 List (java.util.List)2 TopologyDisabledException (org.ow2.proactive.resourcemanager.frontend.topology.TopologyDisabledException)2 TaskResult (org.ow2.proactive.scheduler.common.task.TaskResult)2