Search in sources :

Example 61 with DrainDispatcher

use of org.apache.hadoop.yarn.event.DrainDispatcher in project hadoop by apache.

the class TestRMNMSecretKeys method validateRMNMKeyExchange.

private void validateRMNMKeyExchange(YarnConfiguration conf) throws Exception {
    // Default rolling and activation intervals are large enough, no need to
    // intervene
    final DrainDispatcher dispatcher = new DrainDispatcher();
    ResourceManager rm = new ResourceManager() {

        @Override
        protected void doSecureLogin() throws IOException {
        // Do nothing.
        }

        @Override
        protected Dispatcher createDispatcher() {
            return dispatcher;
        }

        @Override
        protected void startWepApp() {
        // Don't need it, skip.
        }
    };
    rm.init(conf);
    rm.start();
    // Testing ContainerToken and NMToken
    String containerToken = "Container Token : ";
    String nmToken = "NM Token : ";
    MockNM nm = new MockNM("host:1234", 3072, rm.getResourceTrackerService());
    RegisterNodeManagerResponse registrationResponse = nm.registerNode();
    MasterKey containerTokenMasterKey = registrationResponse.getContainerTokenMasterKey();
    Assert.assertNotNull(containerToken + "Registration should cause a key-update!", containerTokenMasterKey);
    MasterKey nmTokenMasterKey = registrationResponse.getNMTokenMasterKey();
    Assert.assertNotNull(nmToken + "Registration should cause a key-update!", nmTokenMasterKey);
    dispatcher.await();
    NodeHeartbeatResponse response = nm.nodeHeartbeat(true);
    Assert.assertNull(containerToken + "First heartbeat after registration shouldn't get any key updates!", response.getContainerTokenMasterKey());
    Assert.assertNull(nmToken + "First heartbeat after registration shouldn't get any key updates!", response.getNMTokenMasterKey());
    dispatcher.await();
    response = nm.nodeHeartbeat(true);
    Assert.assertNull(containerToken + "Even second heartbeat after registration shouldn't get any key updates!", response.getContainerTokenMasterKey());
    Assert.assertNull(nmToken + "Even second heartbeat after registration shouldn't get any key updates!", response.getContainerTokenMasterKey());
    dispatcher.await();
    // Let's force a roll-over
    rm.getRMContext().getContainerTokenSecretManager().rollMasterKey();
    rm.getRMContext().getNMTokenSecretManager().rollMasterKey();
    // Heartbeats after roll-over and before activation should be fine.
    response = nm.nodeHeartbeat(true);
    Assert.assertNotNull(containerToken + "Heartbeats after roll-over and before activation should not err out.", response.getContainerTokenMasterKey());
    Assert.assertNotNull(nmToken + "Heartbeats after roll-over and before activation should not err out.", response.getNMTokenMasterKey());
    Assert.assertEquals(containerToken + "Roll-over should have incremented the key-id only by one!", containerTokenMasterKey.getKeyId() + 1, response.getContainerTokenMasterKey().getKeyId());
    Assert.assertEquals(nmToken + "Roll-over should have incremented the key-id only by one!", nmTokenMasterKey.getKeyId() + 1, response.getNMTokenMasterKey().getKeyId());
    dispatcher.await();
    response = nm.nodeHeartbeat(true);
    Assert.assertNull(containerToken + "Second heartbeat after roll-over shouldn't get any key updates!", response.getContainerTokenMasterKey());
    Assert.assertNull(nmToken + "Second heartbeat after roll-over shouldn't get any key updates!", response.getNMTokenMasterKey());
    dispatcher.await();
    // Let's force activation
    rm.getRMContext().getContainerTokenSecretManager().activateNextMasterKey();
    rm.getRMContext().getNMTokenSecretManager().activateNextMasterKey();
    response = nm.nodeHeartbeat(true);
    Assert.assertNull(containerToken + "Activation shouldn't cause any key updates!", response.getContainerTokenMasterKey());
    Assert.assertNull(nmToken + "Activation shouldn't cause any key updates!", response.getNMTokenMasterKey());
    dispatcher.await();
    response = nm.nodeHeartbeat(true);
    Assert.assertNull(containerToken + "Even second heartbeat after activation shouldn't get any key updates!", response.getContainerTokenMasterKey());
    Assert.assertNull(nmToken + "Even second heartbeat after activation shouldn't get any key updates!", response.getNMTokenMasterKey());
    dispatcher.await();
    rm.stop();
}
Also used : DrainDispatcher(org.apache.hadoop.yarn.event.DrainDispatcher) NodeHeartbeatResponse(org.apache.hadoop.yarn.server.api.protocolrecords.NodeHeartbeatResponse) MockNM(org.apache.hadoop.yarn.server.resourcemanager.MockNM) MasterKey(org.apache.hadoop.yarn.server.api.records.MasterKey) RegisterNodeManagerResponse(org.apache.hadoop.yarn.server.api.protocolrecords.RegisterNodeManagerResponse) ResourceManager(org.apache.hadoop.yarn.server.resourcemanager.ResourceManager)

Example 62 with DrainDispatcher

use of org.apache.hadoop.yarn.event.DrainDispatcher in project hadoop by apache.

the class TestNodeBlacklistingOnAMFailures method testNoBlacklistingForNonSystemErrors.

@Test(timeout = 100000)
public void testNoBlacklistingForNonSystemErrors() throws Exception {
    YarnConfiguration conf = new YarnConfiguration();
    conf.setBoolean(YarnConfiguration.AM_SCHEDULING_NODE_BLACKLISTING_ENABLED, true);
    // disable the float so it is possible to blacklist the entire cluster
    conf.setFloat(YarnConfiguration.AM_SCHEDULING_NODE_BLACKLISTING_DISABLE_THRESHOLD, 1.5f);
    conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 100);
    DrainDispatcher dispatcher = new DrainDispatcher();
    MockRM rm = startRM(conf, dispatcher);
    MockNM node = new MockNM("127.0.0.1:1234", 8000, rm.getResourceTrackerService());
    node.registerNode();
    RMApp app = rm.submitApp(200);
    ApplicationId appId = app.getApplicationId();
    int numAppAttempts = 1;
    // Now the AM container should be allocated
    RMAppAttempt attempt = MockRM.waitForAttemptScheduled(app, rm);
    node.nodeHeartbeat(true);
    dispatcher.await();
    MockRM.waitForState(attempt, RMAppAttemptState.ALLOCATED, 20000);
    rm.sendAMLaunched(attempt.getAppAttemptId());
    rm.waitForState(attempt.getAppAttemptId(), RMAppAttemptState.LAUNCHED);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, numAppAttempts);
    ContainerId amContainerId = ContainerId.newContainerId(appAttemptId, 1);
    for (int containerExitStatus : new int[] { ContainerExitStatus.PREEMPTED, ContainerExitStatus.KILLED_BY_RESOURCEMANAGER, // ContainerExitStatus.KILLED_BY_APPMASTER,
    ContainerExitStatus.KILLED_AFTER_APP_COMPLETION, ContainerExitStatus.ABORTED, ContainerExitStatus.DISKS_FAILED, ContainerExitStatus.KILLED_EXCEEDED_VMEM, ContainerExitStatus.KILLED_EXCEEDED_PMEM }) {
        // Set the exist status to be containerExitStatus so that we can verify
        // that the system automatically blacklisting the node
        makeAMContainerExit(rm, amContainerId, node, containerExitStatus);
        // restart the am
        attempt = MockRM.waitForAttemptScheduled(app, rm);
        System.out.println("New AppAttempt launched " + attempt.getAppAttemptId());
        node.nodeHeartbeat(true);
        dispatcher.await();
        MockRM.waitForState(attempt, RMAppAttemptState.ALLOCATED, 20000);
        rm.sendAMLaunched(attempt.getAppAttemptId());
        rm.waitForState(attempt.getAppAttemptId(), RMAppAttemptState.LAUNCHED);
        numAppAttempts++;
        appAttemptId = ApplicationAttemptId.newInstance(appId, numAppAttempts);
        amContainerId = ContainerId.newContainerId(appAttemptId, 1);
        rm.waitForState(node, amContainerId, RMContainerState.ACQUIRED);
    }
}
Also used : DrainDispatcher(org.apache.hadoop.yarn.event.DrainDispatcher) RMApp(org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp) RMAppAttempt(org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) Test(org.junit.Test)

Example 63 with DrainDispatcher

use of org.apache.hadoop.yarn.event.DrainDispatcher in project hadoop by apache.

the class TestReservations method testGetAppToUnreserve.

@Test
public void testGetAppToUnreserve() throws Exception {
    CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
    setup(csConf);
    final String user_0 = "user_0";
    final ApplicationAttemptId appAttemptId_0 = TestUtils.getMockApplicationAttemptId(0, 0);
    LeafQueue a = stubLeafQueue((LeafQueue) queues.get(A));
    FiCaSchedulerApp app_0 = new FiCaSchedulerApp(appAttemptId_0, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    String host_0 = "host_0";
    FiCaSchedulerNode node_0 = TestUtils.getMockNode(host_0, DEFAULT_RACK, 0, 8 * GB);
    String host_1 = "host_1";
    FiCaSchedulerNode node_1 = TestUtils.getMockNode(host_1, DEFAULT_RACK, 0, 8 * GB);
    Resource clusterResource = Resources.createResource(2 * 8 * GB);
    // Setup resource-requests
    Priority p = TestUtils.createMockPriority(5);
    SchedulerRequestKey priorityMap = toSchedulerKey(p);
    Resource capability = Resources.createResource(2 * GB, 0);
    RMApplicationHistoryWriter writer = mock(RMApplicationHistoryWriter.class);
    SystemMetricsPublisher publisher = mock(SystemMetricsPublisher.class);
    RMContext rmContext = mock(RMContext.class);
    ContainerAllocationExpirer expirer = mock(ContainerAllocationExpirer.class);
    DrainDispatcher drainDispatcher = new DrainDispatcher();
    when(rmContext.getContainerAllocationExpirer()).thenReturn(expirer);
    when(rmContext.getDispatcher()).thenReturn(drainDispatcher);
    when(rmContext.getRMApplicationHistoryWriter()).thenReturn(writer);
    when(rmContext.getSystemMetricsPublisher()).thenReturn(publisher);
    when(rmContext.getYarnConfiguration()).thenReturn(new YarnConfiguration());
    ApplicationAttemptId appAttemptId = BuilderUtils.newApplicationAttemptId(app_0.getApplicationId(), 1);
    ContainerId containerId = BuilderUtils.newContainerId(appAttemptId, 1);
    Container container = TestUtils.getMockContainer(containerId, node_1.getNodeID(), Resources.createResource(2 * GB), priorityMap.getPriority());
    RMContainer rmContainer = new RMContainerImpl(container, SchedulerRequestKey.extractFrom(container), appAttemptId, node_1.getNodeID(), "user", rmContext);
    Container container_1 = TestUtils.getMockContainer(containerId, node_0.getNodeID(), Resources.createResource(1 * GB), priorityMap.getPriority());
    RMContainer rmContainer_1 = new RMContainerImpl(container_1, SchedulerRequestKey.extractFrom(container_1), appAttemptId, node_0.getNodeID(), "user", rmContext);
    // no reserved containers
    NodeId unreserveId = app_0.getNodeIdToUnreserve(priorityMap, capability, cs.getResourceCalculator(), clusterResource);
    assertEquals(null, unreserveId);
    // no reserved containers - reserve then unreserve
    app_0.reserve(node_0, priorityMap, rmContainer_1, container_1);
    app_0.unreserve(priorityMap, node_0, rmContainer_1);
    unreserveId = app_0.getNodeIdToUnreserve(priorityMap, capability, cs.getResourceCalculator(), clusterResource);
    assertEquals(null, unreserveId);
    // no container large enough is reserved
    app_0.reserve(node_0, priorityMap, rmContainer_1, container_1);
    unreserveId = app_0.getNodeIdToUnreserve(priorityMap, capability, cs.getResourceCalculator(), clusterResource);
    assertEquals(null, unreserveId);
    // reserve one that is now large enough
    app_0.reserve(node_1, priorityMap, rmContainer, container);
    unreserveId = app_0.getNodeIdToUnreserve(priorityMap, capability, cs.getResourceCalculator(), clusterResource);
    assertEquals(node_1.getNodeID(), unreserveId);
}
Also used : DrainDispatcher(org.apache.hadoop.yarn.event.DrainDispatcher) RMContext(org.apache.hadoop.yarn.server.resourcemanager.RMContext) FiCaSchedulerNode(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode) ContainerAllocationExpirer(org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.ContainerAllocationExpirer) Priority(org.apache.hadoop.yarn.api.records.Priority) Resource(org.apache.hadoop.yarn.api.records.Resource) RMApplicationHistoryWriter(org.apache.hadoop.yarn.server.resourcemanager.ahs.RMApplicationHistoryWriter) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) RMContainer(org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer) SchedulerRequestKey(org.apache.hadoop.yarn.server.scheduler.SchedulerRequestKey) RMContainer(org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer) Container(org.apache.hadoop.yarn.api.records.Container) RMContainerImpl(org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainerImpl) SystemMetricsPublisher(org.apache.hadoop.yarn.server.resourcemanager.metrics.SystemMetricsPublisher) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) FiCaSchedulerApp(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp) NodeId(org.apache.hadoop.yarn.api.records.NodeId) ActiveUsersManager(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ActiveUsersManager) Test(org.junit.Test)

Example 64 with DrainDispatcher

use of org.apache.hadoop.yarn.event.DrainDispatcher in project hadoop by apache.

the class TestResourceLocalizationService method testLocalizerHeartbeatWhenAppCleaningUp.

@Test(timeout = 20000)
@SuppressWarnings("unchecked")
public void testLocalizerHeartbeatWhenAppCleaningUp() throws Exception {
    conf.set(YarnConfiguration.NM_LOCAL_DIRS, lfs.makeQualified(new Path(basedir, 0 + "")).toString());
    // Start dispatcher.
    DrainDispatcher dispatcher = new DrainDispatcher();
    dispatcher.init(conf);
    dispatcher.start();
    dispatcher.register(ApplicationEventType.class, mock(EventHandler.class));
    dispatcher.register(ContainerEventType.class, mock(EventHandler.class));
    DummyExecutor exec = new DummyExecutor();
    LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
    dirsHandler.init(conf);
    // Start resource localization service.
    ResourceLocalizationService rawService = new ResourceLocalizationService(dispatcher, exec, mock(DeletionService.class), dirsHandler, nmContext);
    ResourceLocalizationService spyService = spy(rawService);
    doReturn(mockServer).when(spyService).createServer();
    doReturn(lfs).when(spyService).getLocalFileContext(isA(Configuration.class));
    try {
        spyService.init(conf);
        spyService.start();
        // Init application resources.
        final Application app = mock(Application.class);
        final ApplicationId appId = BuilderUtils.newApplicationId(1234567890L, 3);
        when(app.getUser()).thenReturn("user0");
        when(app.getAppId()).thenReturn(appId);
        when(app.toString()).thenReturn(appId.toString());
        spyService.handle(new ApplicationLocalizationEvent(LocalizationEventType.INIT_APPLICATION_RESOURCES, app));
        dispatcher.await();
        // Initialize localizer.
        Random r = new Random();
        long seed = r.nextLong();
        System.out.println("SEED: " + seed);
        r.setSeed(seed);
        final Container c = getMockContainer(appId, 46, "user0");
        FSDataOutputStream out = new FSDataOutputStream(new DataOutputBuffer(), null);
        doReturn(out).when(spylfs).createInternal(isA(Path.class), isA(EnumSet.class), isA(FsPermission.class), anyInt(), anyShort(), anyLong(), isA(Progressable.class), isA(ChecksumOpt.class), anyBoolean());
        final LocalResource resource1 = getAppMockedResource(r);
        final LocalResource resource2 = getAppMockedResource(r);
        // Send localization requests for container.
        // 2 resources generated with APPLICATION visibility.
        final LocalResourceRequest req1 = new LocalResourceRequest(resource1);
        final LocalResourceRequest req2 = new LocalResourceRequest(resource2);
        Map<LocalResourceVisibility, Collection<LocalResourceRequest>> rsrcs = new HashMap<LocalResourceVisibility, Collection<LocalResourceRequest>>();
        List<LocalResourceRequest> appResourceList = Arrays.asList(req1, req2);
        rsrcs.put(LocalResourceVisibility.APPLICATION, appResourceList);
        spyService.handle(new ContainerLocalizationRequestEvent(c, rsrcs));
        dispatcher.await();
        // Wait for localization to begin.
        exec.waitForLocalizers(1);
        final String containerIdStr = c.getContainerId().toString();
        LocalizerRunner locRunnerForContainer = spyService.getLocalizerRunner(containerIdStr);
        // Heartbeats from container localizer
        LocalResourceStatus rsrcSuccess = mock(LocalResourceStatus.class);
        LocalizerStatus stat = mock(LocalizerStatus.class);
        when(stat.getLocalizerId()).thenReturn(containerIdStr);
        when(rsrcSuccess.getResource()).thenReturn(resource1);
        when(rsrcSuccess.getLocalSize()).thenReturn(4344L);
        when(rsrcSuccess.getLocalPath()).thenReturn(getPath("/some/path"));
        when(rsrcSuccess.getStatus()).thenReturn(ResourceStatusType.FETCH_SUCCESS);
        when(stat.getResources()).thenReturn(Collections.<LocalResourceStatus>emptyList());
        // First heartbeat which schedules first resource.
        LocalizerHeartbeatResponse response = spyService.heartbeat(stat);
        assertEquals("NM should tell localizer to be LIVE in Heartbeat.", LocalizerAction.LIVE, response.getLocalizerAction());
        // Cleanup application.
        spyService.handle(new ContainerLocalizationCleanupEvent(c, rsrcs));
        spyService.handle(new ApplicationLocalizationEvent(LocalizationEventType.DESTROY_APPLICATION_RESOURCES, app));
        dispatcher.await();
        try {
            // Directly send heartbeat to introduce race as app is being cleaned up.
            locRunnerForContainer.processHeartbeat(Collections.singletonList(rsrcSuccess));
        } catch (Exception e) {
            fail("Exception should not have been thrown on processing heartbeat");
        }
        // Send another heartbeat.
        response = spyService.heartbeat(stat);
        assertEquals("NM should tell localizer to DIE in Heartbeat.", LocalizerAction.DIE, response.getLocalizerAction());
        exec.setStopLocalization();
    } finally {
        spyService.stop();
        dispatcher.stop();
    }
}
Also used : DrainDispatcher(org.apache.hadoop.yarn.event.DrainDispatcher) ContainerLocalizationRequestEvent(org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ContainerLocalizationRequestEvent) Configuration(org.apache.hadoop.conf.Configuration) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) HashMap(java.util.HashMap) EventHandler(org.apache.hadoop.yarn.event.EventHandler) LocalResourceVisibility(org.apache.hadoop.yarn.api.records.LocalResourceVisibility) Container(org.apache.hadoop.yarn.server.nodemanager.containermanager.container.Container) Random(java.util.Random) LocalizerRunner(org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.ResourceLocalizationService.LocalizerRunner) DataOutputBuffer(org.apache.hadoop.io.DataOutputBuffer) ApplicationLocalizationEvent(org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ApplicationLocalizationEvent) FSDataOutputStream(org.apache.hadoop.fs.FSDataOutputStream) FsPermission(org.apache.hadoop.fs.permission.FsPermission) Path(org.apache.hadoop.fs.Path) EnumSet(java.util.EnumSet) DeletionService(org.apache.hadoop.yarn.server.nodemanager.DeletionService) ContainerLocalizationCleanupEvent(org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ContainerLocalizationCleanupEvent) LocalizerStatus(org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerStatus) LocalDirsHandlerService(org.apache.hadoop.yarn.server.nodemanager.LocalDirsHandlerService) IOException(java.io.IOException) BrokenBarrierException(java.util.concurrent.BrokenBarrierException) UnsupportedFileSystemException(org.apache.hadoop.fs.UnsupportedFileSystemException) URISyntaxException(java.net.URISyntaxException) YarnException(org.apache.hadoop.yarn.exceptions.YarnException) SerializedException(org.apache.hadoop.yarn.api.records.SerializedException) NotSerializableException(java.io.NotSerializableException) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) ChecksumOpt(org.apache.hadoop.fs.Options.ChecksumOpt) Progressable(org.apache.hadoop.util.Progressable) Collection(java.util.Collection) LocalizerHeartbeatResponse(org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerHeartbeatResponse) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) Application(org.apache.hadoop.yarn.server.nodemanager.containermanager.application.Application) LocalResourceStatus(org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalResourceStatus) Test(org.junit.Test)

Example 65 with DrainDispatcher

use of org.apache.hadoop.yarn.event.DrainDispatcher in project hadoop by apache.

the class TestResourceLocalizationService method testLocalizationHeartbeat.

@Test(timeout = 10000)
// mocked generics
@SuppressWarnings("unchecked")
public void testLocalizationHeartbeat() throws Exception {
    List<Path> localDirs = new ArrayList<Path>();
    String[] sDirs = new String[1];
    // Making sure that we have only one local disk so that it will only be
    // selected for consecutive resource localization calls.  This is required
    // to test LocalCacheDirectoryManager.
    localDirs.add(lfs.makeQualified(new Path(basedir, 0 + "")));
    sDirs[0] = localDirs.get(0).toString();
    conf.setStrings(YarnConfiguration.NM_LOCAL_DIRS, sDirs);
    // Adding configuration to make sure there is only one file per
    // directory
    conf.set(YarnConfiguration.NM_LOCAL_CACHE_MAX_FILES_PER_DIRECTORY, "37");
    DrainDispatcher dispatcher = new DrainDispatcher();
    dispatcher.init(conf);
    dispatcher.start();
    EventHandler<ApplicationEvent> applicationBus = mock(EventHandler.class);
    dispatcher.register(ApplicationEventType.class, applicationBus);
    EventHandler<ContainerEvent> containerBus = mock(EventHandler.class);
    dispatcher.register(ContainerEventType.class, containerBus);
    ContainerExecutor exec = mock(ContainerExecutor.class);
    LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
    dirsHandler.init(conf);
    DeletionService delServiceReal = new DeletionService(exec);
    DeletionService delService = spy(delServiceReal);
    delService.init(new Configuration());
    delService.start();
    ResourceLocalizationService rawService = new ResourceLocalizationService(dispatcher, exec, delService, dirsHandler, nmContext);
    ResourceLocalizationService spyService = spy(rawService);
    doReturn(mockServer).when(spyService).createServer();
    doReturn(lfs).when(spyService).getLocalFileContext(isA(Configuration.class));
    FsPermission defaultPermission = FsPermission.getDirDefault().applyUMask(lfs.getUMask());
    FsPermission nmPermission = ResourceLocalizationService.NM_PRIVATE_PERM.applyUMask(lfs.getUMask());
    final Path userDir = new Path(sDirs[0].substring("file:".length()), ContainerLocalizer.USERCACHE);
    final Path fileDir = new Path(sDirs[0].substring("file:".length()), ContainerLocalizer.FILECACHE);
    final Path sysDir = new Path(sDirs[0].substring("file:".length()), ResourceLocalizationService.NM_PRIVATE_DIR);
    final FileStatus fs = new FileStatus(0, true, 1, 0, System.currentTimeMillis(), 0, defaultPermission, "", "", new Path(sDirs[0]));
    final FileStatus nmFs = new FileStatus(0, true, 1, 0, System.currentTimeMillis(), 0, nmPermission, "", "", sysDir);
    doAnswer(new Answer<FileStatus>() {

        @Override
        public FileStatus answer(InvocationOnMock invocation) throws Throwable {
            Object[] args = invocation.getArguments();
            if (args.length > 0) {
                if (args[0].equals(userDir) || args[0].equals(fileDir)) {
                    return fs;
                }
            }
            return nmFs;
        }
    }).when(spylfs).getFileStatus(isA(Path.class));
    try {
        spyService.init(conf);
        spyService.start();
        // init application
        final Application app = mock(Application.class);
        final ApplicationId appId = BuilderUtils.newApplicationId(314159265358979L, 3);
        when(app.getUser()).thenReturn("user0");
        when(app.getAppId()).thenReturn(appId);
        spyService.handle(new ApplicationLocalizationEvent(LocalizationEventType.INIT_APPLICATION_RESOURCES, app));
        ArgumentMatcher<ApplicationEvent> matchesAppInit = new ArgumentMatcher<ApplicationEvent>() {

            @Override
            public boolean matches(Object o) {
                ApplicationEvent evt = (ApplicationEvent) o;
                return evt.getType() == ApplicationEventType.APPLICATION_INITED && appId == evt.getApplicationID();
            }
        };
        dispatcher.await();
        verify(applicationBus).handle(argThat(matchesAppInit));
        // init container rsrc, localizer
        Random r = new Random();
        long seed = r.nextLong();
        System.out.println("SEED: " + seed);
        r.setSeed(seed);
        final Container c = getMockContainer(appId, 42, "user0");
        FSDataOutputStream out = new FSDataOutputStream(new DataOutputBuffer(), null);
        doReturn(out).when(spylfs).createInternal(isA(Path.class), isA(EnumSet.class), isA(FsPermission.class), anyInt(), anyShort(), anyLong(), isA(Progressable.class), isA(ChecksumOpt.class), anyBoolean());
        final LocalResource resource1 = getPrivateMockedResource(r);
        LocalResource resource2 = null;
        do {
            resource2 = getPrivateMockedResource(r);
        } while (resource2 == null || resource2.equals(resource1));
        LocalResource resource3 = null;
        do {
            resource3 = getPrivateMockedResource(r);
        } while (resource3 == null || resource3.equals(resource1) || resource3.equals(resource2));
        // above call to make sure we don't get identical resources.
        final LocalResourceRequest req1 = new LocalResourceRequest(resource1);
        final LocalResourceRequest req2 = new LocalResourceRequest(resource2);
        final LocalResourceRequest req3 = new LocalResourceRequest(resource3);
        Map<LocalResourceVisibility, Collection<LocalResourceRequest>> rsrcs = new HashMap<LocalResourceVisibility, Collection<LocalResourceRequest>>();
        List<LocalResourceRequest> privateResourceList = new ArrayList<LocalResourceRequest>();
        privateResourceList.add(req1);
        privateResourceList.add(req2);
        privateResourceList.add(req3);
        rsrcs.put(LocalResourceVisibility.PRIVATE, privateResourceList);
        spyService.handle(new ContainerLocalizationRequestEvent(c, rsrcs));
        // Sigh. Thread init of private localizer not accessible
        Thread.sleep(1000);
        dispatcher.await();
        String appStr = appId.toString();
        String ctnrStr = c.getContainerId().toString();
        ArgumentCaptor<LocalizerStartContext> contextCaptor = ArgumentCaptor.forClass(LocalizerStartContext.class);
        verify(exec).startLocalizer(contextCaptor.capture());
        LocalizerStartContext context = contextCaptor.getValue();
        Path localizationTokenPath = context.getNmPrivateContainerTokens();
        assertEquals("user0", context.getUser());
        assertEquals(appStr, context.getAppId());
        assertEquals(ctnrStr, context.getLocId());
        // heartbeat from localizer
        LocalResourceStatus rsrc1success = mock(LocalResourceStatus.class);
        LocalResourceStatus rsrc2pending = mock(LocalResourceStatus.class);
        LocalResourceStatus rsrc2success = mock(LocalResourceStatus.class);
        LocalResourceStatus rsrc3success = mock(LocalResourceStatus.class);
        LocalizerStatus stat = mock(LocalizerStatus.class);
        when(stat.getLocalizerId()).thenReturn(ctnrStr);
        when(rsrc1success.getResource()).thenReturn(resource1);
        when(rsrc2pending.getResource()).thenReturn(resource2);
        when(rsrc2success.getResource()).thenReturn(resource2);
        when(rsrc3success.getResource()).thenReturn(resource3);
        when(rsrc1success.getLocalSize()).thenReturn(4344L);
        when(rsrc2success.getLocalSize()).thenReturn(2342L);
        when(rsrc3success.getLocalSize()).thenReturn(5345L);
        URL locPath = getPath("/cache/private/blah");
        when(rsrc1success.getLocalPath()).thenReturn(locPath);
        when(rsrc2success.getLocalPath()).thenReturn(locPath);
        when(rsrc3success.getLocalPath()).thenReturn(locPath);
        when(rsrc1success.getStatus()).thenReturn(ResourceStatusType.FETCH_SUCCESS);
        when(rsrc2pending.getStatus()).thenReturn(ResourceStatusType.FETCH_PENDING);
        when(rsrc2success.getStatus()).thenReturn(ResourceStatusType.FETCH_SUCCESS);
        when(rsrc3success.getStatus()).thenReturn(ResourceStatusType.FETCH_SUCCESS);
        // Four heartbeats with sending:
        // 1 - empty
        // 2 - resource1 FETCH_SUCCESS
        // 3 - resource2 FETCH_PENDING
        // 4 - resource2 FETCH_SUCCESS, resource3 FETCH_SUCCESS
        List<LocalResourceStatus> rsrcs4 = new ArrayList<LocalResourceStatus>();
        rsrcs4.add(rsrc2success);
        rsrcs4.add(rsrc3success);
        when(stat.getResources()).thenReturn(Collections.<LocalResourceStatus>emptyList()).thenReturn(Collections.singletonList(rsrc1success)).thenReturn(Collections.singletonList(rsrc2pending)).thenReturn(rsrcs4).thenReturn(Collections.<LocalResourceStatus>emptyList());
        String localPath = Path.SEPARATOR + ContainerLocalizer.USERCACHE + Path.SEPARATOR + "user0" + Path.SEPARATOR + ContainerLocalizer.FILECACHE;
        // First heartbeat
        LocalizerHeartbeatResponse response = spyService.heartbeat(stat);
        assertEquals(LocalizerAction.LIVE, response.getLocalizerAction());
        assertEquals(1, response.getResourceSpecs().size());
        assertEquals(req1, new LocalResourceRequest(response.getResourceSpecs().get(0).getResource()));
        URL localizedPath = response.getResourceSpecs().get(0).getDestinationDirectory();
        // Appending to local path unique number(10) generated as a part of
        // LocalResourcesTracker
        assertTrue(localizedPath.getFile().endsWith(localPath + Path.SEPARATOR + "10"));
        // Second heartbeat
        response = spyService.heartbeat(stat);
        assertEquals(LocalizerAction.LIVE, response.getLocalizerAction());
        assertEquals(1, response.getResourceSpecs().size());
        assertEquals(req2, new LocalResourceRequest(response.getResourceSpecs().get(0).getResource()));
        localizedPath = response.getResourceSpecs().get(0).getDestinationDirectory();
        // Resource's destination path should be now inside sub directory 0 as
        // LocalCacheDirectoryManager will be used and we have restricted number
        // of files per directory to 1.
        assertTrue(localizedPath.getFile().endsWith(localPath + Path.SEPARATOR + "0" + Path.SEPARATOR + "11"));
        // Third heartbeat
        response = spyService.heartbeat(stat);
        assertEquals(LocalizerAction.LIVE, response.getLocalizerAction());
        assertEquals(1, response.getResourceSpecs().size());
        assertEquals(req3, new LocalResourceRequest(response.getResourceSpecs().get(0).getResource()));
        localizedPath = response.getResourceSpecs().get(0).getDestinationDirectory();
        assertTrue(localizedPath.getFile().endsWith(localPath + Path.SEPARATOR + "1" + Path.SEPARATOR + "12"));
        response = spyService.heartbeat(stat);
        assertEquals(LocalizerAction.LIVE, response.getLocalizerAction());
        spyService.handle(new ContainerLocalizationEvent(LocalizationEventType.CONTAINER_RESOURCES_LOCALIZED, c));
        // get shutdown after receive CONTAINER_RESOURCES_LOCALIZED event
        response = spyService.heartbeat(stat);
        assertEquals(LocalizerAction.DIE, response.getLocalizerAction());
        dispatcher.await();
        // verify container notification
        ArgumentMatcher<ContainerEvent> matchesContainerLoc = new ArgumentMatcher<ContainerEvent>() {

            @Override
            public boolean matches(Object o) {
                ContainerEvent evt = (ContainerEvent) o;
                return evt.getType() == ContainerEventType.RESOURCE_LOCALIZED && c.getContainerId() == evt.getContainerID();
            }
        };
        // total 3 resource localzation calls. one for each resource.
        verify(containerBus, times(3)).handle(argThat(matchesContainerLoc));
        // Verify deletion of localization token.
        verify(delService).delete((String) isNull(), eq(localizationTokenPath));
    } finally {
        spyService.stop();
        dispatcher.stop();
        delService.stop();
    }
}
Also used : ContainerExecutor(org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor) DefaultContainerExecutor(org.apache.hadoop.yarn.server.nodemanager.DefaultContainerExecutor) FileStatus(org.apache.hadoop.fs.FileStatus) ContainerLocalizationRequestEvent(org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ContainerLocalizationRequestEvent) Configuration(org.apache.hadoop.conf.Configuration) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) HashMap(java.util.HashMap) ArrayList(java.util.ArrayList) Container(org.apache.hadoop.yarn.server.nodemanager.containermanager.container.Container) Random(java.util.Random) ArgumentMatcher(org.mockito.ArgumentMatcher) DataOutputBuffer(org.apache.hadoop.io.DataOutputBuffer) ApplicationLocalizationEvent(org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ApplicationLocalizationEvent) FSDataOutputStream(org.apache.hadoop.fs.FSDataOutputStream) ContainerLocalizationEvent(org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.ContainerLocalizationEvent) ApplicationEvent(org.apache.hadoop.yarn.server.nodemanager.containermanager.application.ApplicationEvent) DeletionService(org.apache.hadoop.yarn.server.nodemanager.DeletionService) LocalizerStatus(org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerStatus) LocalDirsHandlerService(org.apache.hadoop.yarn.server.nodemanager.LocalDirsHandlerService) ChecksumOpt(org.apache.hadoop.fs.Options.ChecksumOpt) Progressable(org.apache.hadoop.util.Progressable) Collection(java.util.Collection) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) Application(org.apache.hadoop.yarn.server.nodemanager.containermanager.application.Application) DrainDispatcher(org.apache.hadoop.yarn.event.DrainDispatcher) URL(org.apache.hadoop.yarn.api.records.URL) LocalResourceVisibility(org.apache.hadoop.yarn.api.records.LocalResourceVisibility) LocalizerStartContext(org.apache.hadoop.yarn.server.nodemanager.executor.LocalizerStartContext) FsPermission(org.apache.hadoop.fs.permission.FsPermission) Path(org.apache.hadoop.fs.Path) ContainerEvent(org.apache.hadoop.yarn.server.nodemanager.containermanager.container.ContainerEvent) EnumSet(java.util.EnumSet) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) InvocationOnMock(org.mockito.invocation.InvocationOnMock) LocalizerHeartbeatResponse(org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerHeartbeatResponse) LocalResourceStatus(org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalResourceStatus) Test(org.junit.Test)

Aggregations

DrainDispatcher (org.apache.hadoop.yarn.event.DrainDispatcher)84 Test (org.junit.Test)73 YarnConfiguration (org.apache.hadoop.yarn.conf.YarnConfiguration)54 Configuration (org.apache.hadoop.conf.Configuration)50 RMApp (org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp)41 ApplicationAttemptId (org.apache.hadoop.yarn.api.records.ApplicationAttemptId)35 MockNM (org.apache.hadoop.yarn.server.resourcemanager.MockNM)32 Path (org.apache.hadoop.fs.Path)24 ApplicationId (org.apache.hadoop.yarn.api.records.ApplicationId)24 Job (org.apache.hadoop.mapreduce.v2.app.job.Job)23 JobId (org.apache.hadoop.mapreduce.v2.api.records.JobId)21 ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)20 IOException (java.io.IOException)16 ArrayList (java.util.ArrayList)16 DeletionService (org.apache.hadoop.yarn.server.nodemanager.DeletionService)16 LocalizerEvent (org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.event.LocalizerEvent)15 HashMap (java.util.HashMap)14 LocalDirsHandlerService (org.apache.hadoop.yarn.server.nodemanager.LocalDirsHandlerService)14 LocalResourceVisibility (org.apache.hadoop.yarn.api.records.LocalResourceVisibility)12 Container (org.apache.hadoop.yarn.server.nodemanager.containermanager.container.Container)12