Examples with StartContainersResponse org.apache.hadoop.yarn.api.protocolrecords.StartContainersResponse used on opensource projects

Example 1 with StartContainersResponse

use of org.apache.hadoop.yarn.api.protocolrecords.StartContainersResponse in project hadoop by apache.

the class NodeManager method startContainers.

@Override
public synchronized StartContainersResponse startContainers(StartContainersRequest requests) throws YarnException {
    for (StartContainerRequest request : requests.getStartContainerRequests()) {
        Token containerToken = request.getContainerToken();
        ContainerTokenIdentifier tokenId = null;
        try {
            tokenId = BuilderUtils.newContainerTokenIdentifier(containerToken);
        } catch (IOException e) {
            throw RPCUtil.getRemoteException(e);
        }
        ContainerId containerID = tokenId.getContainerID();
        ApplicationId applicationId = containerID.getApplicationAttemptId().getApplicationId();
        List<Container> applicationContainers = containers.get(applicationId);
        if (applicationContainers == null) {
            applicationContainers = new ArrayList<Container>();
            containers.put(applicationId, applicationContainers);
        }
        // Sanity check
        for (Container container : applicationContainers) {
            if (container.getId().compareTo(containerID) == 0) {
                throw new IllegalStateException("Container " + containerID + " already setup on node " + containerManagerAddress);
            }
        }
        Container container = BuilderUtils.newContainer(containerID, this.nodeId, nodeHttpAddress, // DKDC - Doesn't matter
        tokenId.getResource(), // DKDC - Doesn't matter
        null, // DKDC - Doesn't matter
        null);
        ContainerStatus containerStatus = BuilderUtils.newContainerStatus(container.getId(), ContainerState.NEW, "", -1000, container.getResource());
        applicationContainers.add(container);
        containerStatusMap.put(container, containerStatus);
        Resources.subtractFrom(available, tokenId.getResource());
        Resources.addTo(used, tokenId.getResource());
        if (LOG.isDebugEnabled()) {
            LOG.debug("startContainer:" + " node=" + containerManagerAddress + " application=" + applicationId + " container=" + container + " available=" + available + " used=" + used);
        }
    }
    StartContainersResponse response = StartContainersResponse.newInstance(null, null, null);
    return response;
}

Example 2 with StartContainersResponse

use of org.apache.hadoop.yarn.api.protocolrecords.StartContainersResponse in project hadoop by apache.

the class AMLauncher method launch.

private void launch() throws IOException, YarnException {
    connect();
    ContainerId masterContainerID = masterContainer.getId();
    ApplicationSubmissionContext applicationContext = application.getSubmissionContext();
    LOG.info("Setting up container " + masterContainer + " for AM " + application.getAppAttemptId());
    ContainerLaunchContext launchContext = createAMContainerLaunchContext(applicationContext, masterContainerID);
    StartContainerRequest scRequest = StartContainerRequest.newInstance(launchContext, masterContainer.getContainerToken());
    List<StartContainerRequest> list = new ArrayList<StartContainerRequest>();
    list.add(scRequest);
    StartContainersRequest allRequests = StartContainersRequest.newInstance(list);
    StartContainersResponse response = containerMgrProxy.startContainers(allRequests);
    if (response.getFailedRequests() != null && response.getFailedRequests().containsKey(masterContainerID)) {
        Throwable t = response.getFailedRequests().get(masterContainerID).deSerialize();
        parseAndThrowException(t);
    } else {
        LOG.info("Done launching container " + masterContainer + " for AM " + application.getAppAttemptId());
    }
}

Example 3 with StartContainersResponse

use of org.apache.hadoop.yarn.api.protocolrecords.StartContainersResponse in project hadoop by apache.

the class TestContainerManagerRecovery method testContainerCleanupOnShutdown.

@Test
public void testContainerCleanupOnShutdown() throws Exception {
    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 = Collections.emptyMap();
    ContainerLaunchContext clc = ContainerLaunchContext.newInstance(localResources, containerEnv, containerCmds, serviceData, containerTokens, acls);
    // create the logAggregationContext
    LogAggregationContext logAggregationContext = LogAggregationContext.newInstance("includePattern", "excludePattern");
    // verify containers are stopped on shutdown without recovery
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_ENABLED, false);
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_SUPERVISED, false);
    Context context = createContext(conf, new NMNullStateStoreService());
    ContainerManagerImpl cm = spy(createContainerManager(context));
    cm.init(conf);
    cm.start();
    StartContainersResponse startResponse = startContainer(context, cm, cid, clc, logAggregationContext);
    assertEquals(1, startResponse.getSuccessfullyStartedContainers().size());
    cm.stop();
    verify(cm).handle(isA(CMgrCompletedAppsEvent.class));
    // verify containers are stopped on shutdown with unsupervised recovery
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_ENABLED, true);
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_SUPERVISED, false);
    NMMemoryStateStoreService memStore = new NMMemoryStateStoreService();
    memStore.init(conf);
    memStore.start();
    context = createContext(conf, memStore);
    cm = spy(createContainerManager(context));
    cm.init(conf);
    cm.start();
    startResponse = startContainer(context, cm, cid, clc, logAggregationContext);
    assertEquals(1, startResponse.getSuccessfullyStartedContainers().size());
    cm.stop();
    memStore.close();
    verify(cm).handle(isA(CMgrCompletedAppsEvent.class));
    // verify containers are not stopped on shutdown with supervised recovery
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_ENABLED, true);
    conf.setBoolean(YarnConfiguration.NM_RECOVERY_SUPERVISED, true);
    memStore = new NMMemoryStateStoreService();
    memStore.init(conf);
    memStore.start();
    context = createContext(conf, memStore);
    cm = spy(createContainerManager(context));
    cm.init(conf);
    cm.start();
    startResponse = startContainer(context, cm, cid, clc, logAggregationContext);
    assertEquals(1, startResponse.getSuccessfullyStartedContainers().size());
    cm.stop();
    memStore.close();
    verify(cm, never()).handle(isA(CMgrCompletedAppsEvent.class));
}

Example 4 with StartContainersResponse

use of org.apache.hadoop.yarn.api.protocolrecords.StartContainersResponse 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 5 with StartContainersResponse

use of org.apache.hadoop.yarn.api.protocolrecords.StartContainersResponse 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());
        }
    }
}