Examples with NodeUpdateSchedulerEvent org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeUpdateSchedulerEvent used on opensource projects

Example 41 with NodeUpdateSchedulerEvent

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeUpdateSchedulerEvent in project hadoop by apache.

the class TestCapacityScheduler method testHeadRoomCalculationWithDRC.

@Test
public void testHeadRoomCalculationWithDRC() throws Exception {
    // test with total cluster resource of 20GB memory and 20 vcores.
    // the queue where two apps running has user limit 0.8
    // allocate 10GB memory and 1 vcore to app 1.
    // app 1 should have headroom
    // 20GB*0.8 - 10GB = 6GB memory available and 15 vcores.
    // allocate 1GB memory and 1 vcore to app2.
    // app 2 should have headroom 20GB - 10 - 1 = 1GB memory,
    // and 20*0.8 - 1 = 15 vcores.
    CapacitySchedulerConfiguration csconf = new CapacitySchedulerConfiguration();
    csconf.setResourceComparator(DominantResourceCalculator.class);
    YarnConfiguration conf = new YarnConfiguration(csconf);
    conf.setClass(YarnConfiguration.RM_SCHEDULER, CapacityScheduler.class, ResourceScheduler.class);
    MockRM rm = new MockRM(conf);
    rm.start();
    CapacityScheduler cs = (CapacityScheduler) rm.getResourceScheduler();
    LeafQueue qb = (LeafQueue) cs.getQueue("default");
    qb.setUserLimitFactor((float) 0.8);
    // add app 1
    ApplicationId appId = BuilderUtils.newApplicationId(100, 1);
    ApplicationAttemptId appAttemptId = BuilderUtils.newApplicationAttemptId(appId, 1);
    RMAppAttemptMetrics attemptMetric = new RMAppAttemptMetrics(appAttemptId, rm.getRMContext());
    RMAppImpl app = mock(RMAppImpl.class);
    when(app.getApplicationId()).thenReturn(appId);
    RMAppAttemptImpl attempt = mock(RMAppAttemptImpl.class);
    Container container = mock(Container.class);
    when(attempt.getMasterContainer()).thenReturn(container);
    ApplicationSubmissionContext submissionContext = mock(ApplicationSubmissionContext.class);
    when(attempt.getSubmissionContext()).thenReturn(submissionContext);
    when(attempt.getAppAttemptId()).thenReturn(appAttemptId);
    when(attempt.getRMAppAttemptMetrics()).thenReturn(attemptMetric);
    when(app.getCurrentAppAttempt()).thenReturn(attempt);
    rm.getRMContext().getRMApps().put(appId, app);
    SchedulerEvent addAppEvent = new AppAddedSchedulerEvent(appId, "default", "user1");
    cs.handle(addAppEvent);
    SchedulerEvent addAttemptEvent = new AppAttemptAddedSchedulerEvent(appAttemptId, false);
    cs.handle(addAttemptEvent);
    // add app 2
    ApplicationId appId2 = BuilderUtils.newApplicationId(100, 2);
    ApplicationAttemptId appAttemptId2 = BuilderUtils.newApplicationAttemptId(appId2, 1);
    RMAppAttemptMetrics attemptMetric2 = new RMAppAttemptMetrics(appAttemptId2, rm.getRMContext());
    RMAppImpl app2 = mock(RMAppImpl.class);
    when(app2.getApplicationId()).thenReturn(appId2);
    RMAppAttemptImpl attempt2 = mock(RMAppAttemptImpl.class);
    when(attempt2.getMasterContainer()).thenReturn(container);
    when(attempt2.getSubmissionContext()).thenReturn(submissionContext);
    when(attempt2.getAppAttemptId()).thenReturn(appAttemptId2);
    when(attempt2.getRMAppAttemptMetrics()).thenReturn(attemptMetric2);
    when(app2.getCurrentAppAttempt()).thenReturn(attempt2);
    rm.getRMContext().getRMApps().put(appId2, app2);
    addAppEvent = new AppAddedSchedulerEvent(appId2, "default", "user2");
    cs.handle(addAppEvent);
    addAttemptEvent = new AppAttemptAddedSchedulerEvent(appAttemptId2, false);
    cs.handle(addAttemptEvent);
    // add nodes  to cluster, so cluster have 20GB and 20 vcores
    Resource newResource = Resource.newInstance(10 * GB, 10);
    RMNode node = MockNodes.newNodeInfo(0, newResource, 1, "127.0.0.1");
    cs.handle(new NodeAddedSchedulerEvent(node));
    Resource newResource2 = Resource.newInstance(10 * GB, 10);
    RMNode node2 = MockNodes.newNodeInfo(0, newResource2, 1, "127.0.0.2");
    cs.handle(new NodeAddedSchedulerEvent(node2));
    FiCaSchedulerApp fiCaApp1 = cs.getSchedulerApplications().get(app.getApplicationId()).getCurrentAppAttempt();
    FiCaSchedulerApp fiCaApp2 = cs.getSchedulerApplications().get(app2.getApplicationId()).getCurrentAppAttempt();
    Priority u0Priority = TestUtils.createMockPriority(1);
    RecordFactory recordFactory = RecordFactoryProvider.getRecordFactory(null);
    // allocate container for app1 with 10GB memory and 1 vcore
    fiCaApp1.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 10 * GB, 1, true, u0Priority, recordFactory)));
    cs.handle(new NodeUpdateSchedulerEvent(node));
    cs.handle(new NodeUpdateSchedulerEvent(node2));
    assertEquals(6 * GB, fiCaApp1.getHeadroom().getMemorySize());
    assertEquals(15, fiCaApp1.getHeadroom().getVirtualCores());
    // allocate container for app2 with 1GB memory and 1 vcore
    fiCaApp2.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 1 * GB, 1, true, u0Priority, recordFactory)));
    cs.handle(new NodeUpdateSchedulerEvent(node));
    cs.handle(new NodeUpdateSchedulerEvent(node2));
    assertEquals(9 * GB, fiCaApp2.getHeadroom().getMemorySize());
    assertEquals(15, fiCaApp2.getHeadroom().getVirtualCores());
}

Example 42 with NodeUpdateSchedulerEvent

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeUpdateSchedulerEvent in project hadoop by apache.

the class TestCapacitySchedulerLazyPreemption method testSimplePreemption.

@Test(timeout = 60000)
public void testSimplePreemption() throws Exception {
    /**
     * Test case: Submit two application (app1/app2) to different queues, queue
     * structure:
     *
     * <pre>
     *             Root
     *            /  |  \
     *           a   b   c
     *          10   20  70
     * </pre>
     *
     * 1) Two nodes in the cluster, each of them has 4G.
     *
     * 2) app1 submit to queue-a first, it asked 7 * 1G containers, so there's no
     * more resource available.
     *
     * 3) app2 submit to queue-c, ask for one 1G container (for AM)
     *
     * Now the cluster is fulfilled.
     *
     * 4) app2 asks for another 1G container, system will preempt one container
     * from app1, and app2 will receive the preempted container
     */
    MockRM rm1 = new MockRM(conf);
    rm1.getRMContext().setNodeLabelManager(mgr);
    rm1.start();
    MockNM nm1 = rm1.registerNode("h1:1234", 4 * GB);
    MockNM nm2 = rm1.registerNode("h2:1234", 4 * GB);
    CapacityScheduler cs = (CapacityScheduler) rm1.getResourceScheduler();
    RMNode rmNode1 = rm1.getRMContext().getRMNodes().get(nm1.getNodeId());
    RMNode rmNode2 = rm1.getRMContext().getRMNodes().get(nm2.getNodeId());
    // launch an app to queue, AM container should be launched in nm1
    RMApp app1 = rm1.submitApp(1 * GB, "app", "user", null, "a");
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    am1.allocate("*", 1 * GB, 7, new ArrayList<ContainerId>());
    // Do allocation 3 times for node1/node2
    for (int i = 0; i < 3; i++) {
        cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
        cs.handle(new NodeUpdateSchedulerEvent(rmNode2));
    }
    // App1 should have 7 containers now, and no available resource for cluster
    FiCaSchedulerApp schedulerApp1 = cs.getApplicationAttempt(am1.getApplicationAttemptId());
    Assert.assertEquals(7, schedulerApp1.getLiveContainers().size());
    // Submit app2 to queue-c and asks for a 1G container for AM
    RMApp app2 = rm1.submitApp(1 * GB, "app", "user", null, "c");
    MockAM am2 = MockRM.launchAndRegisterAM(app2, rm1, nm2);
    // NM1/NM2 has available resource = 0G
    Assert.assertEquals(0 * GB, cs.getNode(nm1.getNodeId()).getUnallocatedResource().getMemorySize());
    Assert.assertEquals(0 * GB, cs.getNode(nm2.getNodeId()).getUnallocatedResource().getMemorySize());
    // AM asks for a 1 * GB container
    am2.allocate(Arrays.asList(ResourceRequest.newInstance(Priority.newInstance(1), ResourceRequest.ANY, Resources.createResource(1 * GB), 1)), null);
    // Get edit policy and do one update
    SchedulingEditPolicy editPolicy = getSchedulingEditPolicy(rm1);
    // Call edit schedule twice, and check if one container from app1 marked
    // to be "killable"
    editPolicy.editSchedule();
    editPolicy.editSchedule();
    PreemptionManager pm = cs.getPreemptionManager();
    Map<ContainerId, RMContainer> killableContainers = waitKillableContainersSize(pm, "a", RMNodeLabelsManager.NO_LABEL, 1);
    Assert.assertEquals(1, killableContainers.size());
    Assert.assertEquals(killableContainers.entrySet().iterator().next().getKey().getApplicationAttemptId(), am1.getApplicationAttemptId());
    // Call CS.handle once to see if container preempted
    cs.handle(new NodeUpdateSchedulerEvent(rmNode2));
    FiCaSchedulerApp schedulerApp2 = cs.getApplicationAttempt(am2.getApplicationAttemptId());
    // App1 has 6 containers, and app2 has 2 containers
    Assert.assertEquals(6, schedulerApp1.getLiveContainers().size());
    Assert.assertEquals(2, schedulerApp2.getLiveContainers().size());
    rm1.close();
}

Example 43 with NodeUpdateSchedulerEvent

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeUpdateSchedulerEvent in project hadoop by apache.

the class TestAMRMClient method triggerSchedulingWithNMHeartBeat.

/**
   * Make sure we get allocations regardless of timing issues.
   */
private void triggerSchedulingWithNMHeartBeat() {
    // Simulate fair scheduler update thread
    RMContext context = yarnCluster.getResourceManager().getRMContext();
    if (context.getScheduler() instanceof FairScheduler) {
        FairScheduler scheduler = (FairScheduler) context.getScheduler();
        scheduler.update();
    }
    // Trigger NM's heartbeat to RM and trigger allocations
    for (RMNode rmNode : context.getRMNodes().values()) {
        context.getScheduler().handle(new NodeUpdateSchedulerEvent(rmNode));
    }
    if (context.getScheduler() instanceof FairScheduler) {
        FairScheduler scheduler = (FairScheduler) context.getScheduler();
        scheduler.update();
    }
}

Example 44 with NodeUpdateSchedulerEvent

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeUpdateSchedulerEvent in project hadoop by apache.

the class SLSCapacityScheduler method handle.

@Override
public void handle(SchedulerEvent schedulerEvent) {
    // metrics off
    if (!metricsON) {
        super.handle(schedulerEvent);
        return;
    }
    if (!running)
        running = true;
    // metrics on
    Timer.Context handlerTimer = null;
    Timer.Context operationTimer = null;
    NodeUpdateSchedulerEventWrapper eventWrapper;
    try {
        //if (schedulerEvent instanceof NodeUpdateSchedulerEvent) {
        if (schedulerEvent.getType() == SchedulerEventType.NODE_UPDATE && schedulerEvent instanceof NodeUpdateSchedulerEvent) {
            eventWrapper = new NodeUpdateSchedulerEventWrapper((NodeUpdateSchedulerEvent) schedulerEvent);
            schedulerEvent = eventWrapper;
            updateQueueWithNodeUpdate(eventWrapper);
        } else if (schedulerEvent.getType() == SchedulerEventType.APP_ATTEMPT_REMOVED && schedulerEvent instanceof AppAttemptRemovedSchedulerEvent) {
            // check if having AM Container, update resource usage information
            AppAttemptRemovedSchedulerEvent appRemoveEvent = (AppAttemptRemovedSchedulerEvent) schedulerEvent;
            ApplicationAttemptId appAttemptId = appRemoveEvent.getApplicationAttemptID();
            String queue = appQueueMap.get(appAttemptId);
            SchedulerAppReport app = super.getSchedulerAppInfo(appAttemptId);
            if (!app.getLiveContainers().isEmpty()) {
                // have 0 or 1
                // should have one container which is AM container
                RMContainer rmc = app.getLiveContainers().iterator().next();
                updateQueueMetrics(queue, rmc.getContainer().getResource().getMemorySize(), rmc.getContainer().getResource().getVirtualCores());
            }
        }
        handlerTimer = schedulerHandleTimer.time();
        operationTimer = schedulerHandleTimerMap.get(schedulerEvent.getType()).time();
        super.handle(schedulerEvent);
    } finally {
        if (handlerTimer != null)
            handlerTimer.stop();
        if (operationTimer != null)
            operationTimer.stop();
        schedulerHandleCounter.inc();
        schedulerHandleCounterMap.get(schedulerEvent.getType()).inc();
        if (schedulerEvent.getType() == SchedulerEventType.APP_ATTEMPT_REMOVED && schedulerEvent instanceof AppAttemptRemovedSchedulerEvent) {
            SLSRunner.decreaseRemainingApps();
            AppAttemptRemovedSchedulerEvent appRemoveEvent = (AppAttemptRemovedSchedulerEvent) schedulerEvent;
            ApplicationAttemptId appAttemptId = appRemoveEvent.getApplicationAttemptID();
            appQueueMap.remove(appRemoveEvent.getApplicationAttemptID());
        } else if (schedulerEvent.getType() == SchedulerEventType.APP_ATTEMPT_ADDED && schedulerEvent instanceof AppAttemptAddedSchedulerEvent) {
            AppAttemptAddedSchedulerEvent appAddEvent = (AppAttemptAddedSchedulerEvent) schedulerEvent;
            SchedulerApplication app = applications.get(appAddEvent.getApplicationAttemptId().getApplicationId());
            appQueueMap.put(appAddEvent.getApplicationAttemptId(), app.getQueue().getQueueName());
        }
    }
}

Example 45 with NodeUpdateSchedulerEvent

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeUpdateSchedulerEvent in project hadoop by apache.

the class ResourceSchedulerWrapper method handle.

@Override
public void handle(SchedulerEvent schedulerEvent) {
    // metrics off
    if (!metricsON) {
        scheduler.handle(schedulerEvent);
        return;
    }
    if (!running)
        running = true;
    // metrics on
    Timer.Context handlerTimer = null;
    Timer.Context operationTimer = null;
    NodeUpdateSchedulerEventWrapper eventWrapper;
    try {
        //if (schedulerEvent instanceof NodeUpdateSchedulerEvent) {
        if (schedulerEvent.getType() == SchedulerEventType.NODE_UPDATE && schedulerEvent instanceof NodeUpdateSchedulerEvent) {
            eventWrapper = new NodeUpdateSchedulerEventWrapper((NodeUpdateSchedulerEvent) schedulerEvent);
            schedulerEvent = eventWrapper;
            updateQueueWithNodeUpdate(eventWrapper);
        } else if (schedulerEvent.getType() == SchedulerEventType.APP_ATTEMPT_REMOVED && schedulerEvent instanceof AppAttemptRemovedSchedulerEvent) {
            // check if having AM Container, update resource usage information
            AppAttemptRemovedSchedulerEvent appRemoveEvent = (AppAttemptRemovedSchedulerEvent) schedulerEvent;
            ApplicationAttemptId appAttemptId = appRemoveEvent.getApplicationAttemptID();
            String queue = appQueueMap.get(appAttemptId.getApplicationId());
            SchedulerAppReport app = scheduler.getSchedulerAppInfo(appAttemptId);
            if (!app.getLiveContainers().isEmpty()) {
                // have 0 or 1
                // should have one container which is AM container
                RMContainer rmc = app.getLiveContainers().iterator().next();
                updateQueueMetrics(queue, rmc.getContainer().getResource().getMemorySize(), rmc.getContainer().getResource().getVirtualCores());
            }
        }
        handlerTimer = schedulerHandleTimer.time();
        operationTimer = schedulerHandleTimerMap.get(schedulerEvent.getType()).time();
        scheduler.handle(schedulerEvent);
    } finally {
        if (handlerTimer != null)
            handlerTimer.stop();
        if (operationTimer != null)
            operationTimer.stop();
        schedulerHandleCounter.inc();
        schedulerHandleCounterMap.get(schedulerEvent.getType()).inc();
        if (schedulerEvent.getType() == SchedulerEventType.APP_REMOVED && schedulerEvent instanceof AppRemovedSchedulerEvent) {
            SLSRunner.decreaseRemainingApps();
            AppRemovedSchedulerEvent appRemoveEvent = (AppRemovedSchedulerEvent) schedulerEvent;
            appQueueMap.remove(appRemoveEvent.getApplicationID());
        } else if (schedulerEvent.getType() == SchedulerEventType.APP_ADDED && schedulerEvent instanceof AppAddedSchedulerEvent) {
            AppAddedSchedulerEvent appAddEvent = (AppAddedSchedulerEvent) schedulerEvent;
            String queueName = appAddEvent.getQueue();
            appQueueMap.put(appAddEvent.getApplicationId(), queueName);
        }
    }
}