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

Example 6 with NMMemoryStateStoreService

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

Example 7 with NMMemoryStateStoreService

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

the class TestContainerManagerRecovery method testContainerResizeRecovery.

@Test
public void testContainerResizeRecovery() throws Exception {
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_ENABLED, true);
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_SUPERVISED, true);
    NMStateStoreService stateStore = new NMMemoryStateStoreService();
    stateStore.init(conf);
    stateStore.start();
    Context context = createContext(conf, stateStore);
    ContainerManagerImpl cm = createContainerManager(context, delSrvc);
    cm.init(conf);
    cm.start();
    // add an application by starting a container
    ApplicationId appId = ApplicationId.newInstance(0, 1);
    ApplicationAttemptId attemptId = ApplicationAttemptId.newInstance(appId, 1);
    ContainerId cid = ContainerId.newContainerId(attemptId, 1);
    Map<String, String> containerEnv = Collections.emptyMap();
    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 = Collections.emptyMap();
    File tmpDir = new File("target", this.getClass().getSimpleName() + "-tmpDir");
    File scriptFile = Shell.appendScriptExtension(tmpDir, "scriptFile");
    PrintWriter fileWriter = new PrintWriter(scriptFile);
    if (Shell.WINDOWS) {
        fileWriter.println("@ping -n 100 127.0.0.1 >nul");
    } else {
        fileWriter.write("\numask 0");
        fileWriter.write("\nexec sleep 100");
    }
    fileWriter.close();
    FileContext localFS = FileContext.getLocalFSFileContext();
    URL resource_alpha = URL.fromPath(localFS.makeQualified(new Path(scriptFile.getAbsolutePath())));
    LocalResource rsrc_alpha = RecordFactoryProvider.getRecordFactory(null).newRecordInstance(LocalResource.class);
    rsrc_alpha.setResource(resource_alpha);
    rsrc_alpha.setSize(-1);
    rsrc_alpha.setVisibility(LocalResourceVisibility.APPLICATION);
    rsrc_alpha.setType(LocalResourceType.FILE);
    rsrc_alpha.setTimestamp(scriptFile.lastModified());
    String destinationFile = "dest_file";
    Map<String, LocalResource> localResources = new HashMap<>();
    localResources.put(destinationFile, rsrc_alpha);
    List<String> commands = Arrays.asList(Shell.getRunScriptCommand(scriptFile));
    ContainerLaunchContext clc = ContainerLaunchContext.newInstance(localResources, containerEnv, commands, serviceData, containerTokens, acls);
    StartContainersResponse startResponse = startContainer(context, cm, cid, clc, null);
    assertTrue(startResponse.getFailedRequests().isEmpty());
    assertEquals(1, context.getApplications().size());
    Application app = context.getApplications().get(appId);
    assertNotNull(app);
    // make sure the container reaches RUNNING state
    waitForNMContainerState(cm, cid, org.apache.hadoop.yarn.server.nodemanager.containermanager.container.ContainerState.RUNNING);
    Resource targetResource = Resource.newInstance(2048, 2);
    IncreaseContainersResourceResponse increaseResponse = increaseContainersResource(context, cm, cid, targetResource);
    assertTrue(increaseResponse.getFailedRequests().isEmpty());
    // check status
    ContainerStatus containerStatus = getContainerStatus(context, cm, cid);
    assertEquals(targetResource, containerStatus.getCapability());
    // restart and verify container is running and recovered
    // to the correct size
    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);
    containerStatus = getContainerStatus(context, cm, cid);
    assertEquals(targetResource, containerStatus.getCapability());
}

Example 8 with NMMemoryStateStoreService

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

the class TestContainerManagerRecovery method testNMRecoveryForAppFinishedWithLogAggregationFailure.

@Test
public void testNMRecoveryForAppFinishedWithLogAggregationFailure() throws Exception {
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_ENABLED, true);
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_SUPERVISED, true);
    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
    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();
    ContainerLaunchContext clc = ContainerLaunchContext.newInstance(localResources, containerEnv, containerCmds, serviceData, null, null);
    StartContainersResponse startResponse = startContainer(context, cm, cid, clc, null);
    assertTrue(startResponse.getFailedRequests().isEmpty());
    assertEquals(1, context.getApplications().size());
    Application app = context.getApplications().get(appId);
    assertNotNull(app);
    waitForAppState(app, ApplicationState.INITING);
    // 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);
    app.handle(new ApplicationEvent(app.getAppId(), ApplicationEventType.APPLICATION_RESOURCES_CLEANEDUP));
    assertEquals(app.getApplicationState(), ApplicationState.FINISHED);
    // application is still in NM context.
    assertEquals(1, context.getApplications().size());
    // restart and verify app is still there and marked as finished.
    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);
    // TODO need to figure out why additional APPLICATION_RESOURCES_CLEANEDUP
    // is needed.
    app.handle(new ApplicationEvent(app.getAppId(), ApplicationEventType.APPLICATION_RESOURCES_CLEANEDUP));
    assertEquals(app.getApplicationState(), ApplicationState.FINISHED);
    // simulate log aggregation failed.
    app.handle(new ApplicationEvent(app.getAppId(), ApplicationEventType.APPLICATION_LOG_HANDLING_FAILED));
    // 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();
}

Example 9 with NMMemoryStateStoreService

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

the class TestNonAggregatingLogHandler method testRecovery.

@Test
public void testRecovery() throws Exception {
    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, YarnConfiguration.DEFAULT_NM_LOG_RETAIN_SECONDS);
    dirsHandler.init(conf);
    appEventHandler.resetLogHandlingEvent();
    assertFalse(appEventHandler.receiveLogHandlingFinishEvent());
    NMStateStoreService stateStore = new NMMemoryStateStoreService();
    stateStore.init(conf);
    stateStore.start();
    NonAggregatingLogHandlerWithMockExecutor logHandler = new NonAggregatingLogHandlerWithMockExecutor(dispatcher, mockDelService, dirsHandler, stateStore);
    logHandler.init(conf);
    logHandler.start();
    logHandler.handle(new LogHandlerAppStartedEvent(appId, user, null, null));
    logHandler.handle(new LogHandlerContainerFinishedEvent(container11, 0));
    logHandler.handle(new LogHandlerAppFinishedEvent(appId));
    // simulate a restart and verify deletion is rescheduled
    logHandler.close();
    logHandler = new NonAggregatingLogHandlerWithMockExecutor(dispatcher, mockDelService, dirsHandler, stateStore);
    logHandler.init(conf);
    logHandler.start();
    ArgumentCaptor<Runnable> schedArg = ArgumentCaptor.forClass(Runnable.class);
    verify(logHandler.mockSched).schedule(schedArg.capture(), anyLong(), eq(TimeUnit.MILLISECONDS));
    // execute the runnable and verify another restart has nothing scheduled
    schedArg.getValue().run();
    logHandler.close();
    logHandler = new NonAggregatingLogHandlerWithMockExecutor(dispatcher, mockDelService, dirsHandler, stateStore);
    logHandler.init(conf);
    logHandler.start();
    verify(logHandler.mockSched, never()).schedule(any(Runnable.class), anyLong(), any(TimeUnit.class));
    // wait events get drained.
    this.dispatcher.await();
    assertTrue(appEventHandler.receiveLogHandlingFinishEvent());
    appEventHandler.resetLogHandlingEvent();
    assertFalse(appEventHandler.receiveLogHandlingFailedEvent());
    // send an app finish event against a removed app
    logHandler.handle(new LogHandlerAppFinishedEvent(appId));
    this.dispatcher.await();
    // verify to receive a log failed event.
    assertTrue(appEventHandler.receiveLogHandlingFailedEvent());
    assertFalse(appEventHandler.receiveLogHandlingFinishEvent());
    logHandler.close();
}