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

Example 1 with NMStateStoreService

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

the class ContainerManagerImpl method recover.

@SuppressWarnings("unchecked")
private void recover() throws IOException, URISyntaxException {
    NMStateStoreService stateStore = context.getNMStateStore();
    if (stateStore.canRecover()) {
        rsrcLocalizationSrvc.recoverLocalizedResources(stateStore.loadLocalizationState());
        RecoveredApplicationsState appsState = stateStore.loadApplicationsState();
        for (ContainerManagerApplicationProto proto : appsState.getApplications()) {
            if (LOG.isDebugEnabled()) {
                LOG.debug("Recovering application with state: " + proto.toString());
            }
            recoverApplication(proto);
        }
        for (RecoveredContainerState rcs : stateStore.loadContainersState()) {
            if (LOG.isDebugEnabled()) {
                LOG.debug("Recovering container with state: " + rcs);
            }
            recoverContainer(rcs);
        }
    } else {
        LOG.info("Not a recoverable state store. Nothing to recover.");
    }
}

Example 2 with NMStateStoreService

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

the class TestResourceLocalizationService method testPublicResourceInitializesLocalDir.

@Test
@SuppressWarnings("unchecked")
public void testPublicResourceInitializesLocalDir() throws Exception {
    // Setup state to simulate restart NM with existing state meaning no
    // directory creation during initialization
    NMStateStoreService spyStateStore = spy(nmContext.getNMStateStore());
    when(spyStateStore.canRecover()).thenReturn(true);
    NMContext spyContext = spy(nmContext);
    when(spyContext.getNMStateStore()).thenReturn(spyStateStore);
    List<Path> localDirs = new ArrayList<Path>();
    String[] sDirs = new String[4];
    for (int i = 0; i < 4; ++i) {
        localDirs.add(lfs.makeQualified(new Path(basedir, i + "")));
        sDirs[i] = localDirs.get(i).toString();
    }
    conf.setStrings(YarnConfiguration.NM_LOCAL_DIRS, sDirs);
    DrainDispatcher dispatcher = new DrainDispatcher();
    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);
    DeletionService delService = mock(DeletionService.class);
    LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
    dirsHandler.init(conf);
    dispatcher.init(conf);
    dispatcher.start();
    try {
        ResourceLocalizationService rawService = new ResourceLocalizationService(dispatcher, exec, delService, dirsHandler, spyContext);
        ResourceLocalizationService spyService = spy(rawService);
        doReturn(mockServer).when(spyService).createServer();
        doReturn(lfs).when(spyService).getLocalFileContext(isA(Configuration.class));
        spyService.init(conf);
        final FsPermission defaultPerm = new FsPermission((short) 0755);
        // verify directory is not created at initialization
        for (Path p : localDirs) {
            p = new Path((new URI(p.toString())).getPath());
            Path publicCache = new Path(p, ContainerLocalizer.FILECACHE);
            verify(spylfs, never()).mkdir(eq(publicCache), eq(defaultPerm), eq(true));
        }
        spyService.start();
        final String user = "user0";
        // init application
        final Application app = mock(Application.class);
        final ApplicationId appId = BuilderUtils.newApplicationId(314159265358979L, 3);
        when(app.getUser()).thenReturn(user);
        when(app.getAppId()).thenReturn(appId);
        spyService.handle(new ApplicationLocalizationEvent(LocalizationEventType.INIT_APPLICATION_RESOURCES, app));
        dispatcher.await();
        // init container.
        final Container c = getMockContainer(appId, 42, user);
        // init resources
        Random r = new Random();
        long seed = r.nextLong();
        System.out.println("SEED: " + seed);
        r.setSeed(seed);
        // Queue up public resource localization
        final LocalResource pubResource1 = getPublicMockedResource(r);
        final LocalResourceRequest pubReq1 = new LocalResourceRequest(pubResource1);
        LocalResource pubResource2 = null;
        do {
            pubResource2 = getPublicMockedResource(r);
        } while (pubResource2 == null || pubResource2.equals(pubResource1));
        // above call to make sure we don't get identical resources.
        final LocalResourceRequest pubReq2 = new LocalResourceRequest(pubResource2);
        Set<LocalResourceRequest> pubRsrcs = new HashSet<LocalResourceRequest>();
        pubRsrcs.add(pubReq1);
        pubRsrcs.add(pubReq2);
        Map<LocalResourceVisibility, Collection<LocalResourceRequest>> req = new HashMap<LocalResourceVisibility, Collection<LocalResourceRequest>>();
        req.put(LocalResourceVisibility.PUBLIC, pubRsrcs);
        spyService.handle(new ContainerLocalizationRequestEvent(c, req));
        dispatcher.await();
        verify(spyService, times(1)).checkAndInitializeLocalDirs();
        // verify directory creation
        for (Path p : localDirs) {
            p = new Path((new URI(p.toString())).getPath());
            Path publicCache = new Path(p, ContainerLocalizer.FILECACHE);
            verify(spylfs).mkdir(eq(publicCache), eq(defaultPerm), eq(true));
        }
    } finally {
        dispatcher.stop();
    }
}

Example 3 with NMStateStoreService

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

the class TestLocalResourcesTrackerImpl method testStateStoreSuccessfulLocalization.

@Test
@SuppressWarnings("unchecked")
public void testStateStoreSuccessfulLocalization() 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);
    DeletionService mockDelService = mock(DeletionService.class);
    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());
        // Simulate lr1 getting localized
        ResourceLocalizedEvent rle1 = new ResourceLocalizedEvent(lr1, pathCaptor.getValue(), 120);
        tracker.handle(rle1);
        dispatcher.await();
        ArgumentCaptor<LocalizedResourceProto> localizedProtoCaptor = ArgumentCaptor.forClass(LocalizedResourceProto.class);
        verify(stateStore).finishResourceLocalization(eq(user), eq(appId), localizedProtoCaptor.capture());
        LocalizedResourceProto localizedProto = localizedProtoCaptor.getValue();
        Assert.assertEquals(lr1, new LocalResourceRequest(new LocalResourcePBImpl(localizedProto.getResource())));
        Assert.assertEquals(localizedPath1.toString(), localizedProto.getLocalPath());
        LocalizedResource localizedRsrc1 = tracker.getLocalizedResource(lr1);
        Assert.assertNotNull(localizedRsrc1);
        // simulate release and retention processing
        tracker.handle(new ResourceReleaseEvent(lr1, cId1));
        dispatcher.await();
        boolean removeResult = tracker.remove(localizedRsrc1, mockDelService);
        Assert.assertTrue(removeResult);
        verify(stateStore).removeLocalizedResource(eq(user), eq(appId), eq(localizedPath1));
    } finally {
        if (dispatcher != null) {
            dispatcher.stop();
        }
    }
}

Example 4 with NMStateStoreService

use of org.apache.hadoop.yarn.server.nodemanager.recovery.NMStateStoreService 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 5 with NMStateStoreService

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

the class TestContainerManagerRecovery method testApplicationRecovery.

@Test
public void testApplicationRecovery() throws Exception {
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_ENABLED, true);
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_SUPERVISED, true);
    conf.setBoolean(YarnConfiguration.YARN_ACL_ENABLE, true);
    conf.set(YarnConfiguration.YARN_ADMIN_ACL, "yarn_admin_user");
    NMStateStoreService stateStore = new NMMemoryStateStoreService();
    stateStore.init(conf);
    stateStore.start();
    Context context = createContext(conf, stateStore);
    ContainerManagerImpl cm = createContainerManager(context);
    cm.init(conf);
    cm.start();
    // add an application by starting a container
    String appUser = "app_user1";
    String modUser = "modify_user1";
    String viewUser = "view_user1";
    String enemyUser = "enemy_user";
    ApplicationId appId = ApplicationId.newInstance(0, 1);
    ApplicationAttemptId attemptId = ApplicationAttemptId.newInstance(appId, 1);
    ContainerId cid = ContainerId.newContainerId(attemptId, 1);
    Map<String, LocalResource> localResources = Collections.emptyMap();
    Map<String, String> containerEnv = Collections.emptyMap();
    List<String> containerCmds = Collections.emptyList();
    Map<String, ByteBuffer> serviceData = Collections.emptyMap();
    Credentials containerCreds = new Credentials();
    DataOutputBuffer dob = new DataOutputBuffer();
    containerCreds.writeTokenStorageToStream(dob);
    ByteBuffer containerTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength());
    Map<ApplicationAccessType, String> acls = new HashMap<ApplicationAccessType, String>();
    acls.put(ApplicationAccessType.MODIFY_APP, modUser);
    acls.put(ApplicationAccessType.VIEW_APP, viewUser);
    ContainerLaunchContext clc = ContainerLaunchContext.newInstance(localResources, containerEnv, containerCmds, serviceData, containerTokens, acls);
    // create the logAggregationContext
    LogAggregationContext logAggregationContext = LogAggregationContext.newInstance("includePattern", "excludePattern", "includePatternInRollingAggregation", "excludePatternInRollingAggregation");
    StartContainersResponse startResponse = startContainer(context, cm, cid, clc, logAggregationContext);
    assertTrue(startResponse.getFailedRequests().isEmpty());
    assertEquals(1, context.getApplications().size());
    Application app = context.getApplications().get(appId);
    assertNotNull(app);
    waitForAppState(app, ApplicationState.INITING);
    assertTrue(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(modUser), ApplicationAccessType.MODIFY_APP, appUser, appId));
    assertFalse(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(viewUser), ApplicationAccessType.MODIFY_APP, appUser, appId));
    assertTrue(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(viewUser), ApplicationAccessType.VIEW_APP, appUser, appId));
    assertFalse(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(enemyUser), ApplicationAccessType.VIEW_APP, appUser, appId));
    // reset container manager and verify app recovered with proper acls
    cm.stop();
    context = createContext(conf, stateStore);
    cm = createContainerManager(context);
    cm.init(conf);
    cm.start();
    assertEquals(1, context.getApplications().size());
    app = context.getApplications().get(appId);
    assertNotNull(app);
    // check whether LogAggregationContext is recovered correctly
    LogAggregationContext recovered = ((ApplicationImpl) app).getLogAggregationContext();
    assertNotNull(recovered);
    assertEquals(logAggregationContext.getIncludePattern(), recovered.getIncludePattern());
    assertEquals(logAggregationContext.getExcludePattern(), recovered.getExcludePattern());
    assertEquals(logAggregationContext.getRolledLogsIncludePattern(), recovered.getRolledLogsIncludePattern());
    assertEquals(logAggregationContext.getRolledLogsExcludePattern(), recovered.getRolledLogsExcludePattern());
    waitForAppState(app, ApplicationState.INITING);
    assertTrue(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(modUser), ApplicationAccessType.MODIFY_APP, appUser, appId));
    assertFalse(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(viewUser), ApplicationAccessType.MODIFY_APP, appUser, appId));
    assertTrue(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(viewUser), ApplicationAccessType.VIEW_APP, appUser, appId));
    assertFalse(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(enemyUser), ApplicationAccessType.VIEW_APP, appUser, appId));
    // simulate application completion
    List<ApplicationId> finishedApps = new ArrayList<ApplicationId>();
    finishedApps.add(appId);
    app.handle(new ApplicationFinishEvent(appId, "Application killed by ResourceManager"));
    waitForAppState(app, ApplicationState.APPLICATION_RESOURCES_CLEANINGUP);
    // restart and verify app is marked for finishing
    cm.stop();
    context = createContext(conf, stateStore);
    cm = createContainerManager(context);
    cm.init(conf);
    cm.start();
    assertEquals(1, context.getApplications().size());
    app = context.getApplications().get(appId);
    assertNotNull(app);
    // no longer saving FINISH_APP event in NM stateStore,
    // simulate by resending FINISH_APP event
    app.handle(new ApplicationFinishEvent(appId, "Application killed by ResourceManager"));
    waitForAppState(app, ApplicationState.APPLICATION_RESOURCES_CLEANINGUP);
    assertTrue(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(modUser), ApplicationAccessType.MODIFY_APP, appUser, appId));
    assertFalse(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(viewUser), ApplicationAccessType.MODIFY_APP, appUser, appId));
    assertTrue(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(viewUser), ApplicationAccessType.VIEW_APP, appUser, appId));
    assertFalse(context.getApplicationACLsManager().checkAccess(UserGroupInformation.createRemoteUser(enemyUser), ApplicationAccessType.VIEW_APP, appUser, appId));
    // simulate log aggregation completion
    app.handle(new ApplicationEvent(app.getAppId(), ApplicationEventType.APPLICATION_RESOURCES_CLEANEDUP));
    assertEquals(app.getApplicationState(), ApplicationState.FINISHED);
    app.handle(new ApplicationEvent(app.getAppId(), ApplicationEventType.APPLICATION_LOG_HANDLING_FINISHED));
    // restart and verify app is no longer present after recovery
    cm.stop();
    context = createContext(conf, stateStore);
    cm = createContainerManager(context);
    cm.init(conf);
    cm.start();
    assertTrue(context.getApplications().isEmpty());
    cm.stop();
}