Examples with ContainerLaunchContext org.apache.hadoop.yarn.api.records.ContainerLaunchContext used on opensource projects

Example 51 with ContainerLaunchContext

use of org.apache.hadoop.yarn.api.records.ContainerLaunchContext 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 52 with ContainerLaunchContext

use of org.apache.hadoop.yarn.api.records.ContainerLaunchContext in project hadoop by apache.

the class TestContainerManager method testContainerLaunchAndSignal.

// Verify signal container request can be delivered from
// NodeStatusUpdaterImpl to ContainerExecutor.
private void testContainerLaunchAndSignal(SignalContainerCommand command) throws IOException, InterruptedException, YarnException {
    Signal signal = ContainerLaunch.translateCommandToSignal(command);
    containerManager.start();
    File scriptFile = new File(tmpDir, "scriptFile.sh");
    PrintWriter fileWriter = new PrintWriter(scriptFile);
    File processStartFile = new File(tmpDir, "start_file.txt").getAbsoluteFile();
    // So that start file is readable by the test
    fileWriter.write("\numask 0");
    fileWriter.write("\necho Hello World! > " + processStartFile);
    fileWriter.write("\necho $$ >> " + processStartFile);
    fileWriter.write("\nexec sleep 1000s");
    fileWriter.close();
    ContainerLaunchContext containerLaunchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    // ////// Construct the Container-id
    ContainerId cId = createContainerId(0);
    URL resource_alpha = URL.fromPath(localFS.makeQualified(new Path(scriptFile.getAbsolutePath())));
    LocalResource rsrc_alpha = recordFactory.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<String, LocalResource>();
    localResources.put(destinationFile, rsrc_alpha);
    containerLaunchContext.setLocalResources(localResources);
    List<String> commands = new ArrayList<>();
    commands.add("/bin/bash");
    commands.add(scriptFile.getAbsolutePath());
    containerLaunchContext.setCommands(commands);
    StartContainerRequest scRequest = StartContainerRequest.newInstance(containerLaunchContext, createContainerToken(cId, DUMMY_RM_IDENTIFIER, context.getNodeId(), user, context.getContainerTokenSecretManager()));
    List<StartContainerRequest> list = new ArrayList<>();
    list.add(scRequest);
    StartContainersRequest allRequests = StartContainersRequest.newInstance(list);
    containerManager.startContainers(allRequests);
    int timeoutSecs = 0;
    while (!processStartFile.exists() && timeoutSecs++ < 20) {
        Thread.sleep(1000);
        LOG.info("Waiting for process start-file to be created");
    }
    Assert.assertTrue("ProcessStartFile doesn't exist!", processStartFile.exists());
    // Simulate NodeStatusUpdaterImpl sending CMgrSignalContainersEvent
    SignalContainerRequest signalReq = SignalContainerRequest.newInstance(cId, command);
    List<SignalContainerRequest> reqs = new ArrayList<>();
    reqs.add(signalReq);
    containerManager.handle(new CMgrSignalContainersEvent(reqs));
    final ArgumentCaptor<ContainerSignalContext> signalContextCaptor = ArgumentCaptor.forClass(ContainerSignalContext.class);
    if (signal.equals(Signal.NULL)) {
        verify(exec, never()).signalContainer(signalContextCaptor.capture());
    } else {
        verify(exec, timeout(10000).atLeastOnce()).signalContainer(signalContextCaptor.capture());
        ContainerSignalContext signalContext = signalContextCaptor.getAllValues().get(0);
        Assert.assertEquals(cId, signalContext.getContainer().getContainerId());
        Assert.assertEquals(signal, signalContext.getSignal());
    }
}

Example 53 with ContainerLaunchContext

use of org.apache.hadoop.yarn.api.records.ContainerLaunchContext in project hadoop by apache.

the class TestContainerManager method testLocalFilesCleanup.

@Test
public void testLocalFilesCleanup() throws InterruptedException, IOException, YarnException {
    // Real del service
    delSrvc = new DeletionService(exec);
    delSrvc.init(conf);
    containerManager = createContainerManager(delSrvc);
    containerManager.init(conf);
    containerManager.start();
    // ////// Create the resources for the container
    File dir = new File(tmpDir, "dir");
    dir.mkdirs();
    File file = new File(dir, "file");
    PrintWriter fileWriter = new PrintWriter(file);
    fileWriter.write("Hello World!");
    fileWriter.close();
    // ////// Construct the Container-id
    ContainerId cId = createContainerId(0);
    ApplicationId appId = cId.getApplicationAttemptId().getApplicationId();
    // ////// Construct the container-spec.
    ContainerLaunchContext containerLaunchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    //    containerLaunchContext.resources =
    //        new HashMap<CharSequence, LocalResource>();
    URL resource_alpha = URL.fromPath(FileContext.getLocalFSFileContext().makeQualified(new Path(file.getAbsolutePath())));
    LocalResource rsrc_alpha = recordFactory.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(file.lastModified());
    String destinationFile = "dest_file";
    Map<String, LocalResource> localResources = new HashMap<String, LocalResource>();
    localResources.put(destinationFile, rsrc_alpha);
    containerLaunchContext.setLocalResources(localResources);
    StartContainerRequest scRequest = StartContainerRequest.newInstance(containerLaunchContext, createContainerToken(cId, DUMMY_RM_IDENTIFIER, context.getNodeId(), user, context.getContainerTokenSecretManager()));
    List<StartContainerRequest> list = new ArrayList<>();
    list.add(scRequest);
    StartContainersRequest allRequests = StartContainersRequest.newInstance(list);
    containerManager.startContainers(allRequests);
    BaseContainerManagerTest.waitForContainerState(containerManager, cId, ContainerState.COMPLETE);
    BaseContainerManagerTest.waitForApplicationState(containerManager, cId.getApplicationAttemptId().getApplicationId(), ApplicationState.RUNNING);
    // Now ascertain that the resources are localised correctly.
    String appIDStr = appId.toString();
    String containerIDStr = cId.toString();
    File userCacheDir = new File(localDir, ContainerLocalizer.USERCACHE);
    File userDir = new File(userCacheDir, user);
    File appCache = new File(userDir, ContainerLocalizer.APPCACHE);
    File appDir = new File(appCache, appIDStr);
    File containerDir = new File(appDir, containerIDStr);
    File targetFile = new File(containerDir, destinationFile);
    File sysDir = new File(localDir, ResourceLocalizationService.NM_PRIVATE_DIR);
    File appSysDir = new File(sysDir, appIDStr);
    File containerSysDir = new File(appSysDir, containerIDStr);
    // AppDir should still exist
    Assert.assertTrue("AppDir " + appDir.getAbsolutePath() + " doesn't exist!!", appDir.exists());
    Assert.assertTrue("AppSysDir " + appSysDir.getAbsolutePath() + " doesn't exist!!", appSysDir.exists());
    for (File f : new File[] { containerDir, containerSysDir }) {
        Assert.assertFalse(f.getAbsolutePath() + " exists!!", f.exists());
    }
    Assert.assertFalse(targetFile.getAbsolutePath() + " exists!!", targetFile.exists());
    // Simulate RM sending an AppFinish event.
    containerManager.handle(new CMgrCompletedAppsEvent(Arrays.asList(new ApplicationId[] { appId }), CMgrCompletedAppsEvent.Reason.ON_SHUTDOWN));
    BaseContainerManagerTest.waitForApplicationState(containerManager, cId.getApplicationAttemptId().getApplicationId(), ApplicationState.FINISHED);
    // Now ascertain that the resources are localised correctly.
    for (File f : new File[] { appDir, containerDir, appSysDir, containerSysDir }) {
        // Wait for deletion. Deletion can happen long after AppFinish because of
        // the async DeletionService
        int timeout = 0;
        while (f.exists() && timeout++ < 15) {
            Thread.sleep(1000);
        }
        Assert.assertFalse(f.getAbsolutePath() + " exists!!", f.exists());
    }
    // Wait for deletion
    int timeout = 0;
    while (targetFile.exists() && timeout++ < 15) {
        Thread.sleep(1000);
    }
    Assert.assertFalse(targetFile.getAbsolutePath() + " exists!!", targetFile.exists());
}

Example 54 with ContainerLaunchContext

use of org.apache.hadoop.yarn.api.records.ContainerLaunchContext in project hadoop by apache.

the class TestContainerManager method testIncreaseContainerResourceWithInvalidRequests.

@Test
public void testIncreaseContainerResourceWithInvalidRequests() throws Exception {
    containerManager.start();
    // Start 4 containers 0..4 with default resource (1024, 1)
    List<StartContainerRequest> list = new ArrayList<>();
    ContainerLaunchContext containerLaunchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    for (int i = 0; i < 4; i++) {
        ContainerId cId = createContainerId(i);
        long identifier = DUMMY_RM_IDENTIFIER;
        Token containerToken = createContainerToken(cId, identifier, context.getNodeId(), user, context.getContainerTokenSecretManager());
        StartContainerRequest request = StartContainerRequest.newInstance(containerLaunchContext, containerToken);
        list.add(request);
    }
    StartContainersRequest requestList = StartContainersRequest.newInstance(list);
    StartContainersResponse response = containerManager.startContainers(requestList);
    Assert.assertEquals(4, response.getSuccessfullyStartedContainers().size());
    int i = 0;
    for (ContainerId id : response.getSuccessfullyStartedContainers()) {
        Assert.assertEquals(i, id.getContainerId());
        i++;
    }
    Thread.sleep(2000);
    // Construct container resource increase request,
    List<Token> increaseTokens = new ArrayList<>();
    // Add increase request for container-0, the request will fail as the
    // container will have exited, and won't be in RUNNING state
    ContainerId cId0 = createContainerId(0);
    Token containerToken = createContainerToken(cId0, DUMMY_RM_IDENTIFIER, context.getNodeId(), user, Resource.newInstance(1234, 3), context.getContainerTokenSecretManager(), null);
    increaseTokens.add(containerToken);
    // Add increase request for container-7, the request will fail as the
    // container does not exist
    ContainerId cId7 = createContainerId(7);
    containerToken = createContainerToken(cId7, DUMMY_RM_IDENTIFIER, context.getNodeId(), user, Resource.newInstance(1234, 3), context.getContainerTokenSecretManager(), null);
    increaseTokens.add(containerToken);
    IncreaseContainersResourceRequest increaseRequest = IncreaseContainersResourceRequest.newInstance(increaseTokens);
    IncreaseContainersResourceResponse increaseResponse = containerManager.increaseContainersResource(increaseRequest);
    // Check response
    Assert.assertEquals(0, increaseResponse.getSuccessfullyIncreasedContainers().size());
    Assert.assertEquals(2, increaseResponse.getFailedRequests().size());
    for (Map.Entry<ContainerId, SerializedException> entry : increaseResponse.getFailedRequests().entrySet()) {
        Assert.assertNotNull("Failed message", entry.getValue().getMessage());
        if (cId0.equals(entry.getKey())) {
            Assert.assertTrue(entry.getValue().getMessage().contains("Resource can only be changed when a " + "container is in RUNNING state"));
        } else if (cId7.equals(entry.getKey())) {
            Assert.assertTrue(entry.getValue().getMessage().contains("Container " + cId7.toString() + " is not handled by this NodeManager"));
        } else {
            throw new YarnException("Received failed request from wrong" + " container: " + entry.getKey().toString());
        }
    }
}

Example 55 with ContainerLaunchContext

use of org.apache.hadoop.yarn.api.records.ContainerLaunchContext in project hadoop by apache.

the class TestContainerManager method testStartContainerFailureWithUnknownAuxService.

@Test
public void testStartContainerFailureWithUnknownAuxService() throws Exception {
    conf.setStrings(YarnConfiguration.NM_AUX_SERVICES, new String[] { "existService" });
    conf.setClass(String.format(YarnConfiguration.NM_AUX_SERVICE_FMT, "existService"), ServiceA.class, Service.class);
    containerManager.start();
    List<StartContainerRequest> startRequest = new ArrayList<>();
    ContainerLaunchContext containerLaunchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    Map<String, ByteBuffer> serviceData = new HashMap<String, ByteBuffer>();
    String serviceName = "non_exist_auxService";
    serviceData.put(serviceName, ByteBuffer.wrap(serviceName.getBytes()));
    containerLaunchContext.setServiceData(serviceData);
    ContainerId cId = createContainerId(0);
    String user = "start_container_fail";
    Token containerToken = createContainerToken(cId, DUMMY_RM_IDENTIFIER, context.getNodeId(), user, context.getContainerTokenSecretManager());
    StartContainerRequest request = StartContainerRequest.newInstance(containerLaunchContext, containerToken);
    // start containers
    startRequest.add(request);
    StartContainersRequest requestList = StartContainersRequest.newInstance(startRequest);
    StartContainersResponse response = containerManager.startContainers(requestList);
    Assert.assertEquals(1, response.getFailedRequests().size());
    Assert.assertEquals(0, response.getSuccessfullyStartedContainers().size());
    Assert.assertTrue(response.getFailedRequests().containsKey(cId));
    Assert.assertTrue(response.getFailedRequests().get(cId).getMessage().contains("The auxService:" + serviceName + " does not exist"));
}