Examples with NMNullStateStoreService org.apache.hadoop.yarn.server.nodemanager.recovery.NMNullStateStoreService used on opensource projects

Example 1 with NMNullStateStoreService

use of org.apache.hadoop.yarn.server.nodemanager.recovery.NMNullStateStoreService in project hadoop by apache.

the class TestEventFlow method testSuccessfulContainerLaunch.

@Test
public void testSuccessfulContainerLaunch() throws InterruptedException, IOException, YarnException {
    FileContext localFS = FileContext.getLocalFSFileContext();
    localFS.delete(new Path(localDir.getAbsolutePath()), true);
    localFS.delete(new Path(localLogDir.getAbsolutePath()), true);
    localFS.delete(new Path(remoteLogDir.getAbsolutePath()), true);
    localDir.mkdir();
    localLogDir.mkdir();
    remoteLogDir.mkdir();
    YarnConfiguration conf = new YarnConfiguration();
    Context context = new NMContext(new NMContainerTokenSecretManager(conf), new NMTokenSecretManagerInNM(), null, null, new NMNullStateStoreService(), false, conf) {

        @Override
        public int getHttpPort() {
            return 1234;
        }
    };
    conf.set(YarnConfiguration.NM_LOCAL_DIRS, localDir.getAbsolutePath());
    conf.set(YarnConfiguration.NM_LOG_DIRS, localLogDir.getAbsolutePath());
    conf.set(YarnConfiguration.NM_REMOTE_APP_LOG_DIR, remoteLogDir.getAbsolutePath());
    conf.set(YarnConfiguration.NM_LOCALIZER_ADDRESS, "0.0.0.0:" + ServerSocketUtil.getPort(8040, 10));
    ContainerExecutor exec = new DefaultContainerExecutor();
    exec.setConf(conf);
    DeletionService del = new DeletionService(exec);
    Dispatcher dispatcher = new AsyncDispatcher();
    LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
    NodeHealthCheckerService healthChecker = new NodeHealthCheckerService(NodeManager.getNodeHealthScriptRunner(conf), dirsHandler);
    healthChecker.init(conf);
    NodeManagerMetrics metrics = NodeManagerMetrics.create();
    NodeStatusUpdater nodeStatusUpdater = new NodeStatusUpdaterImpl(context, dispatcher, healthChecker, metrics) {

        @Override
        protected ResourceTracker getRMClient() {
            return new LocalRMInterface();
        }

        ;

        @Override
        protected void stopRMProxy() {
            return;
        }

        @Override
        protected void startStatusUpdater() {
            // Don't start any updating thread.
            return;
        }

        @Override
        public long getRMIdentifier() {
            return SIMULATED_RM_IDENTIFIER;
        }
    };
    DummyContainerManager containerManager = new DummyContainerManager(context, exec, del, nodeStatusUpdater, metrics, dirsHandler);
    nodeStatusUpdater.init(conf);
    ((NMContext) context).setContainerManager(containerManager);
    nodeStatusUpdater.start();
    ((NMContext) context).setNodeStatusUpdater(nodeStatusUpdater);
    containerManager.init(conf);
    containerManager.start();
    ContainerLaunchContext launchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    ApplicationId applicationId = ApplicationId.newInstance(0, 0);
    ApplicationAttemptId applicationAttemptId = ApplicationAttemptId.newInstance(applicationId, 0);
    ContainerId cID = ContainerId.newContainerId(applicationAttemptId, 0);
    String user = "testing";
    StartContainerRequest scRequest = StartContainerRequest.newInstance(launchContext, TestContainerManager.createContainerToken(cID, SIMULATED_RM_IDENTIFIER, context.getNodeId(), user, context.getContainerTokenSecretManager()));
    List<StartContainerRequest> list = new ArrayList<StartContainerRequest>();
    list.add(scRequest);
    StartContainersRequest allRequests = StartContainersRequest.newInstance(list);
    containerManager.startContainers(allRequests);
    BaseContainerManagerTest.waitForContainerState(containerManager, cID, Arrays.asList(ContainerState.RUNNING, ContainerState.SCHEDULED), 20);
    List<ContainerId> containerIds = new ArrayList<ContainerId>();
    containerIds.add(cID);
    StopContainersRequest stopRequest = StopContainersRequest.newInstance(containerIds);
    containerManager.stopContainers(stopRequest);
    BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.COMPLETE);
    containerManager.stop();
}

Example 2 with NMNullStateStoreService

use of org.apache.hadoop.yarn.server.nodemanager.recovery.NMNullStateStoreService in project hadoop by apache.

the class TestContainerLogsPage method testContainerLogDirs.

@Test(timeout = 30000)
public void testContainerLogDirs() throws IOException, YarnException {
    File absLogDir = new File("target", TestNMWebServer.class.getSimpleName() + "LogDir").getAbsoluteFile();
    String logdirwithFile = absLogDir.toURI().toString();
    Configuration conf = new Configuration();
    conf.set(YarnConfiguration.NM_LOG_DIRS, logdirwithFile);
    NodeHealthCheckerService healthChecker = createNodeHealthCheckerService(conf);
    healthChecker.init(conf);
    LocalDirsHandlerService dirsHandler = healthChecker.getDiskHandler();
    NMContext nmContext = new NodeManager.NMContext(null, null, dirsHandler, new ApplicationACLsManager(conf), new NMNullStateStoreService(), false, conf);
    // Add an application and the corresponding containers
    RecordFactory recordFactory = RecordFactoryProvider.getRecordFactory(conf);
    String user = "nobody";
    long clusterTimeStamp = 1234;
    ApplicationId appId = BuilderUtils.newApplicationId(recordFactory, clusterTimeStamp, 1);
    Application app = mock(Application.class);
    when(app.getUser()).thenReturn(user);
    when(app.getAppId()).thenReturn(appId);
    ApplicationAttemptId appAttemptId = BuilderUtils.newApplicationAttemptId(appId, 1);
    ContainerId container1 = BuilderUtils.newContainerId(recordFactory, appId, appAttemptId, 0);
    nmContext.getApplications().put(appId, app);
    MockContainer container = new MockContainer(appAttemptId, new AsyncDispatcher(), conf, user, appId, 1);
    container.setState(ContainerState.RUNNING);
    nmContext.getContainers().put(container1, container);
    List<File> files = null;
    files = ContainerLogsUtils.getContainerLogDirs(container1, user, nmContext);
    Assert.assertTrue(!(files.get(0).toString().contains("file:")));
    // After container is completed, it is removed from nmContext
    nmContext.getContainers().remove(container1);
    Assert.assertNull(nmContext.getContainers().get(container1));
    files = ContainerLogsUtils.getContainerLogDirs(container1, user, nmContext);
    Assert.assertTrue(!(files.get(0).toString().contains("file:")));
    // Create a new context to check if correct container log dirs are fetched
    // on full disk.
    LocalDirsHandlerService dirsHandlerForFullDisk = spy(dirsHandler);
    // good log dirs are empty and nm log dir is in the full log dir list.
    when(dirsHandlerForFullDisk.getLogDirs()).thenReturn(new ArrayList<String>());
    when(dirsHandlerForFullDisk.getLogDirsForRead()).thenReturn(Arrays.asList(new String[] { absLogDir.getAbsolutePath() }));
    nmContext = new NodeManager.NMContext(null, null, dirsHandlerForFullDisk, new ApplicationACLsManager(conf), new NMNullStateStoreService(), false, conf);
    nmContext.getApplications().put(appId, app);
    container.setState(ContainerState.RUNNING);
    nmContext.getContainers().put(container1, container);
    List<File> dirs = ContainerLogsUtils.getContainerLogDirs(container1, user, nmContext);
    File containerLogDir = new File(absLogDir, appId + "/" + container1);
    Assert.assertTrue(dirs.contains(containerLogDir));
}

Example 3 with NMNullStateStoreService

use of org.apache.hadoop.yarn.server.nodemanager.recovery.NMNullStateStoreService in project hadoop by apache.

the class TestContainerLogsPage method testContainerLogFile.

@Test(timeout = 30000)
public void testContainerLogFile() throws IOException, YarnException {
    File absLogDir = new File("target", TestNMWebServer.class.getSimpleName() + "LogDir").getAbsoluteFile();
    String logdirwithFile = absLogDir.toURI().toString();
    Configuration conf = new Configuration();
    conf.set(YarnConfiguration.NM_LOG_DIRS, logdirwithFile);
    conf.setFloat(YarnConfiguration.NM_MAX_PER_DISK_UTILIZATION_PERCENTAGE, 0.0f);
    LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
    dirsHandler.init(conf);
    NMContext nmContext = new NodeManager.NMContext(null, null, dirsHandler, new ApplicationACLsManager(conf), new NMNullStateStoreService(), false, conf);
    // Add an application and the corresponding containers
    String user = "nobody";
    long clusterTimeStamp = 1234;
    ApplicationId appId = BuilderUtils.newApplicationId(clusterTimeStamp, 1);
    Application app = mock(Application.class);
    when(app.getUser()).thenReturn(user);
    when(app.getAppId()).thenReturn(appId);
    ApplicationAttemptId appAttemptId = BuilderUtils.newApplicationAttemptId(appId, 1);
    ContainerId containerId = BuilderUtils.newContainerId(appAttemptId, 1);
    nmContext.getApplications().put(appId, app);
    MockContainer container = new MockContainer(appAttemptId, new AsyncDispatcher(), conf, user, appId, 1);
    container.setState(ContainerState.RUNNING);
    nmContext.getContainers().put(containerId, container);
    File containerLogDir = new File(absLogDir, ContainerLaunch.getRelativeContainerLogDir(appId.toString(), containerId.toString()));
    containerLogDir.mkdirs();
    String fileName = "fileName";
    File containerLogFile = new File(containerLogDir, fileName);
    containerLogFile.createNewFile();
    File file = ContainerLogsUtils.getContainerLogFile(containerId, fileName, user, nmContext);
    Assert.assertEquals(containerLogFile.toURI().toString(), file.toURI().toString());
    FileUtil.fullyDelete(absLogDir);
}

Example 4 with NMNullStateStoreService

use of org.apache.hadoop.yarn.server.nodemanager.recovery.NMNullStateStoreService in project hadoop by apache.

the class TestNonAggregatingLogHandler method testLogDeletion.

@Test
public void testLogDeletion() throws IOException {
    File[] localLogDirs = getLocalLogDirFiles(this.getClass().getName(), 2);
    String localLogDirsString = localLogDirs[0].getAbsolutePath() + "," + localLogDirs[1].getAbsolutePath();
    conf.set(YarnConfiguration.NM_LOG_DIRS, localLogDirsString);
    conf.setBoolean(YarnConfiguration.LOG_AGGREGATION_ENABLED, false);
    conf.setLong(YarnConfiguration.NM_LOG_RETAIN_SECONDS, 0l);
    dirsHandler.init(conf);
    NonAggregatingLogHandler rawLogHandler = new NonAggregatingLogHandler(dispatcher, mockDelService, dirsHandler, new NMNullStateStoreService());
    NonAggregatingLogHandler logHandler = spy(rawLogHandler);
    AbstractFileSystem spylfs = spy(FileContext.getLocalFSFileContext().getDefaultFileSystem());
    FileContext lfs = FileContext.getFileContext(spylfs, conf);
    doReturn(lfs).when(logHandler).getLocalFileContext(isA(Configuration.class));
    FsPermission defaultPermission = FsPermission.getDirDefault().applyUMask(lfs.getUMask());
    final FileStatus fs = new FileStatus(0, true, 1, 0, System.currentTimeMillis(), 0, defaultPermission, "", "", new Path(localLogDirs[0].getAbsolutePath()));
    doReturn(fs).when(spylfs).getFileStatus(isA(Path.class));
    logHandler.init(conf);
    logHandler.start();
    logHandler.handle(new LogHandlerAppStartedEvent(appId, user, null, null));
    logHandler.handle(new LogHandlerContainerFinishedEvent(container11, 0));
    logHandler.handle(new LogHandlerAppFinishedEvent(appId));
    Path[] localAppLogDirs = new Path[2];
    localAppLogDirs[0] = new Path(localLogDirs[0].getAbsolutePath(), appId.toString());
    localAppLogDirs[1] = new Path(localLogDirs[1].getAbsolutePath(), appId.toString());
    testDeletionServiceCall(mockDelService, user, 5000, localAppLogDirs);
    logHandler.close();
    for (int i = 0; i < localLogDirs.length; i++) {
        FileUtils.deleteDirectory(localLogDirs[i]);
    }
}

Example 5 with NMNullStateStoreService

use of org.apache.hadoop.yarn.server.nodemanager.recovery.NMNullStateStoreService 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();
        }
    }
}