Examples with ResourceFailedLocalizationEvent org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ResourceFailedLocalizationEvent used on opensource projects

Example 1 with ResourceFailedLocalizationEvent

use of org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ResourceFailedLocalizationEvent in project hadoop by apache.

the class TestResourceLocalizationService method testParallelDownloadAttemptsForPublicResource.

@Test(timeout = 100000)
@SuppressWarnings("unchecked")
public void testParallelDownloadAttemptsForPublicResource() throws Exception {
    DrainDispatcher dispatcher1 = null;
    String user = "testuser";
    try {
        // creating one local directory
        List<Path> localDirs = new ArrayList<Path>();
        String[] sDirs = new String[1];
        for (int i = 0; i < 1; ++i) {
            localDirs.add(lfs.makeQualified(new Path(basedir, i + "")));
            sDirs[i] = localDirs.get(i).toString();
        }
        conf.setStrings(YarnConfiguration.NM_LOCAL_DIRS, sDirs);
        // Registering event handlers
        EventHandler<ApplicationEvent> applicationBus = mock(EventHandler.class);
        dispatcher1 = new DrainDispatcher();
        dispatcher1.register(ApplicationEventType.class, applicationBus);
        EventHandler<ContainerEvent> containerBus = mock(EventHandler.class);
        dispatcher1.register(ContainerEventType.class, containerBus);
        ContainerExecutor exec = mock(ContainerExecutor.class);
        DeletionService delService = mock(DeletionService.class);
        LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
        // initializing directory handler.
        dirsHandler.init(conf);
        dispatcher1.init(conf);
        dispatcher1.start();
        // Creating and initializing ResourceLocalizationService but not starting
        // it as otherwise it will remove requests from pending queue.
        ResourceLocalizationService rawService = new ResourceLocalizationService(dispatcher1, exec, delService, dirsHandler, nmContext);
        ResourceLocalizationService spyService = spy(rawService);
        dispatcher1.register(LocalizationEventType.class, spyService);
        spyService.init(conf);
        // Initially pending map should be empty for public localizer
        Assert.assertEquals(0, spyService.getPublicLocalizer().pending.size());
        LocalResourceRequest req = new LocalResourceRequest(new Path("/tmp"), 123L, LocalResourceType.FILE, LocalResourceVisibility.PUBLIC, "");
        // Initializing application
        ApplicationImpl app = mock(ApplicationImpl.class);
        ApplicationId appId = BuilderUtils.newApplicationId(1, 1);
        when(app.getAppId()).thenReturn(appId);
        when(app.getUser()).thenReturn(user);
        dispatcher1.getEventHandler().handle(new ApplicationLocalizationEvent(LocalizationEventType.INIT_APPLICATION_RESOURCES, app));
        // Container - 1
        // container requesting the resource
        ContainerImpl container1 = createMockContainer(user, 1);
        dispatcher1.getEventHandler().handle(createContainerLocalizationEvent(container1, LocalResourceVisibility.PUBLIC, req));
        // Waiting for resource to change into DOWNLOADING state.
        Assert.assertTrue(waitForResourceState(null, spyService, req, LocalResourceVisibility.PUBLIC, user, null, ResourceState.DOWNLOADING, 5000));
        // Waiting for download to start.
        Assert.assertTrue(waitForPublicDownloadToStart(spyService, 1, 5000));
        LocalizedResource lr = getLocalizedResource(spyService, req, LocalResourceVisibility.PUBLIC, user, null);
        // Resource would now have moved into DOWNLOADING state
        Assert.assertEquals(ResourceState.DOWNLOADING, lr.getState());
        // pending should have this resource now.
        Assert.assertEquals(1, spyService.getPublicLocalizer().pending.size());
        // Now resource should have 0 permit.
        Assert.assertEquals(0, lr.sem.availablePermits());
        // Container - 2
        // Container requesting the same resource.
        ContainerImpl container2 = createMockContainer(user, 2);
        dispatcher1.getEventHandler().handle(createContainerLocalizationEvent(container2, LocalResourceVisibility.PUBLIC, req));
        // Waiting for download to start. This should return false as new download
        // will not start
        Assert.assertFalse(waitForPublicDownloadToStart(spyService, 2, 5000));
        // Now Failing the resource download. As a part of it
        // resource state is changed and then lock is released.
        ResourceFailedLocalizationEvent locFailedEvent = new ResourceFailedLocalizationEvent(req, new Exception("test").toString());
        spyService.getLocalResourcesTracker(LocalResourceVisibility.PUBLIC, user, null).handle(locFailedEvent);
        // Waiting for resource to change into FAILED state.
        Assert.assertTrue(waitForResourceState(lr, spyService, req, LocalResourceVisibility.PUBLIC, user, null, ResourceState.FAILED, 5000));
        // releasing lock as a part of download failed process.
        lr.unlock();
        // removing pending download request.
        spyService.getPublicLocalizer().pending.clear();
        LocalizerContext lc = mock(LocalizerContext.class);
        when(lc.getContainerId()).thenReturn(ContainerId.newContainerId(ApplicationAttemptId.newInstance(ApplicationId.newInstance(1L, 1), 1), 1L));
        // Now I need to simulate a race condition wherein Event is added to
        // dispatcher before resource state changes to either FAILED or LOCALIZED
        // Hence sending event directly to dispatcher.
        LocalizerResourceRequestEvent localizerEvent = new LocalizerResourceRequestEvent(lr, null, lc, null);
        dispatcher1.getEventHandler().handle(localizerEvent);
        // Waiting for download to start. This should return false as new download
        // will not start
        Assert.assertFalse(waitForPublicDownloadToStart(spyService, 1, 5000));
        // Checking available permits now.
        Assert.assertEquals(1, lr.sem.availablePermits());
    } finally {
        if (dispatcher1 != null) {
            dispatcher1.stop();
        }
    }
}

Example 2 with ResourceFailedLocalizationEvent

use of org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ResourceFailedLocalizationEvent in project hadoop by apache.

the class TestLocalResourcesTrackerImpl method testLocalResourceCache.

@Test(timeout = 10000)
@SuppressWarnings("unchecked")
public void testLocalResourceCache() {
    String user = "testuser";
    DrainDispatcher dispatcher = null;
    try {
        Configuration conf = new Configuration();
        dispatcher = createDispatcher(conf);
        EventHandler<LocalizerEvent> localizerEventHandler = mock(EventHandler.class);
        EventHandler<ContainerEvent> containerEventHandler = mock(EventHandler.class);
        // Registering event handlers.
        dispatcher.register(LocalizerEventType.class, localizerEventHandler);
        dispatcher.register(ContainerEventType.class, containerEventHandler);
        ConcurrentMap<LocalResourceRequest, LocalizedResource> localrsrc = new ConcurrentHashMap<LocalResourceRequest, LocalizedResource>();
        LocalResourcesTracker tracker = new LocalResourcesTrackerImpl(user, null, dispatcher, localrsrc, true, conf, new NMNullStateStoreService(), null);
        LocalResourceRequest lr = createLocalResourceRequest(user, 1, 1, LocalResourceVisibility.PUBLIC);
        // Creating 2 containers for same application which will be requesting
        // same local resource.
        // Container 1 requesting local resource.
        ContainerId cId1 = BuilderUtils.newContainerId(1, 1, 1, 1);
        LocalizerContext lc1 = new LocalizerContext(user, cId1, null);
        ResourceEvent reqEvent1 = new ResourceRequestEvent(lr, LocalResourceVisibility.PRIVATE, lc1);
        // No resource request is initially present in local cache
        Assert.assertEquals(0, localrsrc.size());
        // Container-1 requesting local resource.
        tracker.handle(reqEvent1);
        dispatcher.await();
        // New localized Resource should have been added to local resource map
        // and the requesting container will be added to its waiting queue.
        Assert.assertEquals(1, localrsrc.size());
        Assert.assertTrue(localrsrc.containsKey(lr));
        Assert.assertEquals(1, localrsrc.get(lr).getRefCount());
        Assert.assertTrue(localrsrc.get(lr).ref.contains(cId1));
        Assert.assertEquals(ResourceState.DOWNLOADING, localrsrc.get(lr).getState());
        // Container 2 requesting the resource
        ContainerId cId2 = BuilderUtils.newContainerId(1, 1, 1, 2);
        LocalizerContext lc2 = new LocalizerContext(user, cId2, null);
        ResourceEvent reqEvent2 = new ResourceRequestEvent(lr, LocalResourceVisibility.PRIVATE, lc2);
        tracker.handle(reqEvent2);
        dispatcher.await();
        // Container 2 should have been added to the waiting queue of the local
        // resource
        Assert.assertEquals(2, localrsrc.get(lr).getRefCount());
        Assert.assertTrue(localrsrc.get(lr).ref.contains(cId2));
        // Failing resource localization
        ResourceEvent resourceFailedEvent = new ResourceFailedLocalizationEvent(lr, (new Exception("test").getMessage()));
        // Backing up the resource to track its state change as it will be
        // removed after the failed event.
        LocalizedResource localizedResource = localrsrc.get(lr);
        tracker.handle(resourceFailedEvent);
        dispatcher.await();
        // After receiving failed resource event; all waiting containers will be
        // notified with Container Resource Failed Event.
        Assert.assertEquals(0, localrsrc.size());
        verify(containerEventHandler, timeout(1000).times(2)).handle(isA(ContainerResourceFailedEvent.class));
        Assert.assertEquals(ResourceState.FAILED, localizedResource.getState());
        // Container 1 trying to release the resource (This resource is already
        // deleted from the cache. This call should return silently without
        // exception.
        ResourceReleaseEvent relEvent1 = new ResourceReleaseEvent(lr, cId1);
        tracker.handle(relEvent1);
        dispatcher.await();
        // Container-3 now requests for the same resource. This request call
        // is coming prior to Container-2's release call.
        ContainerId cId3 = BuilderUtils.newContainerId(1, 1, 1, 3);
        LocalizerContext lc3 = new LocalizerContext(user, cId3, null);
        ResourceEvent reqEvent3 = new ResourceRequestEvent(lr, LocalResourceVisibility.PRIVATE, lc3);
        tracker.handle(reqEvent3);
        dispatcher.await();
        // Local resource cache now should have the requested resource and the
        // number of waiting containers should be 1.
        Assert.assertEquals(1, localrsrc.size());
        Assert.assertTrue(localrsrc.containsKey(lr));
        Assert.assertEquals(1, localrsrc.get(lr).getRefCount());
        Assert.assertTrue(localrsrc.get(lr).ref.contains(cId3));
        // Container-2 Releases the resource
        ResourceReleaseEvent relEvent2 = new ResourceReleaseEvent(lr, cId2);
        tracker.handle(relEvent2);
        dispatcher.await();
        // Making sure that there is no change in the cache after the release.
        Assert.assertEquals(1, localrsrc.size());
        Assert.assertTrue(localrsrc.containsKey(lr));
        Assert.assertEquals(1, localrsrc.get(lr).getRefCount());
        Assert.assertTrue(localrsrc.get(lr).ref.contains(cId3));
        // Sending ResourceLocalizedEvent to tracker. In turn resource should
        // send Container Resource Localized Event to waiting containers.
        Path localizedPath = new Path("/tmp/file1");
        ResourceLocalizedEvent localizedEvent = new ResourceLocalizedEvent(lr, localizedPath, 123L);
        tracker.handle(localizedEvent);
        dispatcher.await();
        // Verifying ContainerResourceLocalizedEvent .
        verify(containerEventHandler, timeout(1000).times(1)).handle(isA(ContainerResourceLocalizedEvent.class));
        Assert.assertEquals(ResourceState.LOCALIZED, localrsrc.get(lr).getState());
        Assert.assertEquals(1, localrsrc.get(lr).getRefCount());
        // Container-3 releasing the resource.
        ResourceReleaseEvent relEvent3 = new ResourceReleaseEvent(lr, cId3);
        tracker.handle(relEvent3);
        dispatcher.await();
        Assert.assertEquals(0, localrsrc.get(lr).getRefCount());
    } finally {
        if (dispatcher != null) {
            dispatcher.stop();
        }
    }
}

Example 3 with ResourceFailedLocalizationEvent

use of org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ResourceFailedLocalizationEvent in project hadoop by apache.

the class TestLocalResourcesTrackerImpl method testStateStoreFailedLocalization.

@Test
@SuppressWarnings("unchecked")
public void testStateStoreFailedLocalization() throws Exception {
    final String user = "someuser";
    final ApplicationId appId = ApplicationId.newInstance(1, 1);
    // This is a random path. NO File creation will take place at this place.
    final Path localDir = new Path("/tmp");
    Configuration conf = new YarnConfiguration();
    DrainDispatcher dispatcher = null;
    dispatcher = createDispatcher(conf);
    EventHandler<LocalizerEvent> localizerEventHandler = mock(EventHandler.class);
    EventHandler<LocalizerEvent> containerEventHandler = mock(EventHandler.class);
    dispatcher.register(LocalizerEventType.class, localizerEventHandler);
    dispatcher.register(ContainerEventType.class, containerEventHandler);
    NMStateStoreService stateStore = mock(NMStateStoreService.class);
    try {
        LocalResourcesTracker tracker = new LocalResourcesTrackerImpl(user, appId, dispatcher, false, conf, stateStore);
        // Container 1 needs lr1 resource
        ContainerId cId1 = BuilderUtils.newContainerId(1, 1, 1, 1);
        LocalResourceRequest lr1 = createLocalResourceRequest(user, 1, 1, LocalResourceVisibility.APPLICATION);
        LocalizerContext lc1 = new LocalizerContext(user, cId1, null);
        // Container 1 requests lr1 to be localized
        ResourceEvent reqEvent1 = new ResourceRequestEvent(lr1, LocalResourceVisibility.APPLICATION, lc1);
        tracker.handle(reqEvent1);
        dispatcher.await();
        // Simulate the process of localization of lr1
        Path hierarchicalPath1 = tracker.getPathForLocalization(lr1, localDir, null);
        ArgumentCaptor<LocalResourceProto> localResourceCaptor = ArgumentCaptor.forClass(LocalResourceProto.class);
        ArgumentCaptor<Path> pathCaptor = ArgumentCaptor.forClass(Path.class);
        verify(stateStore).startResourceLocalization(eq(user), eq(appId), localResourceCaptor.capture(), pathCaptor.capture());
        LocalResourceProto lrProto = localResourceCaptor.getValue();
        Path localizedPath1 = pathCaptor.getValue();
        Assert.assertEquals(lr1, new LocalResourceRequest(new LocalResourcePBImpl(lrProto)));
        Assert.assertEquals(hierarchicalPath1, localizedPath1.getParent());
        ResourceFailedLocalizationEvent rfe1 = new ResourceFailedLocalizationEvent(lr1, new Exception("Test").toString());
        tracker.handle(rfe1);
        dispatcher.await();
        verify(stateStore).removeLocalizedResource(eq(user), eq(appId), eq(localizedPath1));
    } finally {
        if (dispatcher != null) {
            dispatcher.stop();
        }
    }
}

Example 4 with ResourceFailedLocalizationEvent

use of org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ResourceFailedLocalizationEvent in project hadoop by apache.

the class TestLocalResourcesTrackerImpl method testHierarchicalLocalCacheDirectories.

@Test(timeout = 10000)
@SuppressWarnings("unchecked")
public void testHierarchicalLocalCacheDirectories() {
    String user = "testuser";
    DrainDispatcher dispatcher = null;
    try {
        Configuration conf = new Configuration();
        // setting per directory file limit to 1.
        conf.set(YarnConfiguration.NM_LOCAL_CACHE_MAX_FILES_PER_DIRECTORY, "37");
        dispatcher = createDispatcher(conf);
        EventHandler<LocalizerEvent> localizerEventHandler = mock(EventHandler.class);
        EventHandler<LocalizerEvent> containerEventHandler = mock(EventHandler.class);
        dispatcher.register(LocalizerEventType.class, localizerEventHandler);
        dispatcher.register(ContainerEventType.class, containerEventHandler);
        DeletionService mockDelService = mock(DeletionService.class);
        ConcurrentMap<LocalResourceRequest, LocalizedResource> localrsrc = new ConcurrentHashMap<LocalResourceRequest, LocalizedResource>();
        LocalResourcesTracker tracker = new LocalResourcesTrackerImpl(user, null, dispatcher, localrsrc, true, conf, new NMNullStateStoreService(), null);
        // This is a random path. NO File creation will take place at this place.
        Path localDir = new Path("/tmp");
        // Container 1 needs lr1 resource
        ContainerId cId1 = BuilderUtils.newContainerId(1, 1, 1, 1);
        LocalResourceRequest lr1 = createLocalResourceRequest(user, 1, 1, LocalResourceVisibility.PUBLIC);
        LocalizerContext lc1 = new LocalizerContext(user, cId1, null);
        // Container 1 requests lr1 to be localized
        ResourceEvent reqEvent1 = new ResourceRequestEvent(lr1, LocalResourceVisibility.PUBLIC, lc1);
        tracker.handle(reqEvent1);
        // Simulate the process of localization of lr1
        // NOTE: Localization path from tracker has resource ID at end
        Path hierarchicalPath1 = tracker.getPathForLocalization(lr1, localDir, null).getParent();
        // Simulate lr1 getting localized
        ResourceLocalizedEvent rle1 = new ResourceLocalizedEvent(lr1, new Path(hierarchicalPath1.toUri().toString() + Path.SEPARATOR + "file1"), 120);
        tracker.handle(rle1);
        // Localization successful.
        LocalResourceRequest lr2 = createLocalResourceRequest(user, 3, 3, LocalResourceVisibility.PUBLIC);
        // Container 1 requests lr2 to be localized.
        ResourceEvent reqEvent2 = new ResourceRequestEvent(lr2, LocalResourceVisibility.PUBLIC, lc1);
        tracker.handle(reqEvent2);
        Path hierarchicalPath2 = tracker.getPathForLocalization(lr2, localDir, null).getParent();
        // localization failed.
        ResourceFailedLocalizationEvent rfe2 = new ResourceFailedLocalizationEvent(lr2, new Exception("Test").toString());
        tracker.handle(rfe2);
        /*
       * The path returned for two localization should be different because we
       * are limiting one file per sub-directory.
       */
        Assert.assertNotSame(hierarchicalPath1, hierarchicalPath2);
        LocalResourceRequest lr3 = createLocalResourceRequest(user, 2, 2, LocalResourceVisibility.PUBLIC);
        ResourceEvent reqEvent3 = new ResourceRequestEvent(lr3, LocalResourceVisibility.PUBLIC, lc1);
        tracker.handle(reqEvent3);
        Path hierarchicalPath3 = tracker.getPathForLocalization(lr3, localDir, null).getParent();
        // localization successful
        ResourceLocalizedEvent rle3 = new ResourceLocalizedEvent(lr3, new Path(hierarchicalPath3.toUri().toString() + Path.SEPARATOR + "file3"), 120);
        tracker.handle(rle3);
        // Verifying that path created is inside the subdirectory
        Assert.assertEquals(hierarchicalPath3.toUri().toString(), hierarchicalPath1.toUri().toString() + Path.SEPARATOR + "0");
        // Container 1 releases resource lr1
        ResourceEvent relEvent1 = new ResourceReleaseEvent(lr1, cId1);
        tracker.handle(relEvent1);
        // Validate the file counts now
        int resources = 0;
        Iterator<LocalizedResource> iter = tracker.iterator();
        while (iter.hasNext()) {
            iter.next();
            resources++;
        }
        // There should be only two resources lr1 and lr3 now.
        Assert.assertEquals(2, resources);
        // Now simulate cache cleanup - removes unused resources.
        iter = tracker.iterator();
        while (iter.hasNext()) {
            LocalizedResource rsrc = iter.next();
            if (rsrc.getRefCount() == 0) {
                Assert.assertTrue(tracker.remove(rsrc, mockDelService));
                resources--;
            }
        }
        // lr1 is not used by anyone and will be removed, only lr3 will hang
        // around
        Assert.assertEquals(1, resources);
    } finally {
        if (dispatcher != null) {
            dispatcher.stop();
        }
    }
}