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

Example 91 with LocalResource

use of org.apache.hadoop.yarn.api.records.LocalResource 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 92 with LocalResource

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

the class TestContainerLaunch method internalKillTest.

private void internalKillTest(boolean delayed) throws Exception {
    conf.setLong(YarnConfiguration.NM_SLEEP_DELAY_BEFORE_SIGKILL_MS, delayed ? 1000 : 0);
    containerManager.start();
    // ////// Construct the Container-id
    ApplicationId appId = ApplicationId.newInstance(1, 1);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, 1);
    ContainerId cId = ContainerId.newContainerId(appAttemptId, 0);
    File processStartFile = new File(tmpDir, "pid.txt").getAbsoluteFile();
    // setup a script that can handle sigterm gracefully
    File scriptFile = Shell.appendScriptExtension(tmpDir, "testscript");
    PrintWriter writer = new PrintWriter(new FileOutputStream(scriptFile));
    if (Shell.WINDOWS) {
        writer.println("@echo \"Running testscript for delayed kill\"");
        writer.println("@echo \"Writing pid to start file\"");
        writer.println("@echo " + cId + "> " + processStartFile);
        writer.println("@ping -n 100 127.0.0.1 >nul");
    } else {
        writer.println("#!/bin/bash\n\n");
        writer.println("echo \"Running testscript for delayed kill\"");
        writer.println("hello=\"Got SIGTERM\"");
        writer.println("umask 0");
        writer.println("trap \"echo $hello >> " + processStartFile + "\" SIGTERM");
        writer.println("echo \"Writing pid to start file\"");
        writer.println("echo $$ >> " + processStartFile);
        writer.println("while true; do\nsleep 1s;\ndone");
    }
    writer.close();
    FileUtil.setExecutable(scriptFile, true);
    ContainerLaunchContext containerLaunchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    // upload the script file so that the container can run it
    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.sh";
    Map<String, LocalResource> localResources = new HashMap<String, LocalResource>();
    localResources.put(destinationFile, rsrc_alpha);
    containerLaunchContext.setLocalResources(localResources);
    // set up the rest of the container
    List<String> commands = Arrays.asList(Shell.getRunScriptCommand(scriptFile));
    containerLaunchContext.setCommands(commands);
    Priority priority = Priority.newInstance(10);
    long createTime = 1234;
    Token containerToken = createContainerToken(cId, priority, createTime);
    StartContainerRequest scRequest = StartContainerRequest.newInstance(containerLaunchContext, containerToken);
    List<StartContainerRequest> list = new ArrayList<StartContainerRequest>();
    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());
    NMContainerStatus nmContainerStatus = containerManager.getContext().getContainers().get(cId).getNMContainerStatus();
    Assert.assertEquals(priority, nmContainerStatus.getPriority());
    // Now test the stop functionality.
    List<ContainerId> containerIds = new ArrayList<ContainerId>();
    containerIds.add(cId);
    StopContainersRequest stopRequest = StopContainersRequest.newInstance(containerIds);
    containerManager.stopContainers(stopRequest);
    BaseContainerManagerTest.waitForContainerState(containerManager, cId, ContainerState.COMPLETE);
    // if delayed container stop sends a sigterm followed by a sigkill
    // otherwise sigkill is sent immediately 
    GetContainerStatusesRequest gcsRequest = GetContainerStatusesRequest.newInstance(containerIds);
    ContainerStatus containerStatus = containerManager.getContainerStatuses(gcsRequest).getContainerStatuses().get(0);
    Assert.assertEquals(ContainerExitStatus.KILLED_BY_APPMASTER, containerStatus.getExitStatus());
    // verify that the job object with ID matching container ID no longer exists.
    if (Shell.WINDOWS || !delayed) {
        Assert.assertFalse("Process is still alive!", DefaultContainerExecutor.containerIsAlive(cId.toString()));
    } else {
        BufferedReader reader = new BufferedReader(new FileReader(processStartFile));
        boolean foundSigTermMessage = false;
        while (true) {
            String line = reader.readLine();
            if (line == null) {
                break;
            }
            if (line.contains("SIGTERM")) {
                foundSigTermMessage = true;
                break;
            }
        }
        Assert.assertTrue("Did not find sigterm message", foundSigTermMessage);
        reader.close();
    }
}

Example 93 with LocalResource

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

the class TestContainerLaunch method testKillProcessGroup.

@Test
public void testKillProcessGroup() throws Exception {
    Assume.assumeTrue(Shell.isSetsidAvailable);
    containerManager.start();
    // Construct the Container-id
    ApplicationId appId = ApplicationId.newInstance(2, 2);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, 1);
    ContainerId cId = ContainerId.newContainerId(appAttemptId, 0);
    File processStartFile = new File(tmpDir, "pid.txt").getAbsoluteFile();
    File childProcessStartFile = new File(tmpDir, "child_pid.txt").getAbsoluteFile();
    // setup a script that can handle sigterm gracefully
    File scriptFile = Shell.appendScriptExtension(tmpDir, "testscript");
    PrintWriter writer = new PrintWriter(new FileOutputStream(scriptFile));
    writer.println("#!/bin/bash\n\n");
    writer.println("echo \"Running testscript for forked process\"");
    writer.println("umask 0");
    writer.println("echo $$ >> " + processStartFile);
    writer.println("while true;\ndo sleep 1s;\ndone > /dev/null 2>&1 &");
    writer.println("echo $! >> " + childProcessStartFile);
    writer.println("while true;\ndo sleep 1s;\ndone");
    writer.close();
    FileUtil.setExecutable(scriptFile, true);
    ContainerLaunchContext containerLaunchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    // upload the script file so that the container can run it
    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.sh";
    Map<String, LocalResource> localResources = new HashMap<String, LocalResource>();
    localResources.put(destinationFile, rsrc_alpha);
    containerLaunchContext.setLocalResources(localResources);
    // set up the rest of the container
    List<String> commands = Arrays.asList(Shell.getRunScriptCommand(scriptFile));
    containerLaunchContext.setCommands(commands);
    Priority priority = Priority.newInstance(10);
    long createTime = 1234;
    Token containerToken = createContainerToken(cId, priority, createTime);
    StartContainerRequest scRequest = StartContainerRequest.newInstance(containerLaunchContext, containerToken);
    List<StartContainerRequest> list = new ArrayList<StartContainerRequest>();
    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());
    BufferedReader reader = new BufferedReader(new FileReader(processStartFile));
    // Get the pid of the process
    String pid = reader.readLine().trim();
    // No more lines
    Assert.assertEquals(null, reader.readLine());
    reader.close();
    reader = new BufferedReader(new FileReader(childProcessStartFile));
    // Get the pid of the child process
    String child = reader.readLine().trim();
    // No more lines
    Assert.assertEquals(null, reader.readLine());
    reader.close();
    LOG.info("Manually killing pid " + pid + ", but not child pid " + child);
    Shell.execCommand(new String[] { "kill", "-9", pid });
    BaseContainerManagerTest.waitForContainerState(containerManager, cId, ContainerState.COMPLETE);
    Assert.assertFalse("Process is still alive!", DefaultContainerExecutor.containerIsAlive(pid));
    List<ContainerId> containerIds = new ArrayList<ContainerId>();
    containerIds.add(cId);
    GetContainerStatusesRequest gcsRequest = GetContainerStatusesRequest.newInstance(containerIds);
    ContainerStatus containerStatus = containerManager.getContainerStatuses(gcsRequest).getContainerStatuses().get(0);
    Assert.assertEquals(ExitCode.FORCE_KILLED.getExitCode(), containerStatus.getExitStatus());
}

Example 94 with LocalResource

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

the class TestContainerManager method testContainerSetup.

@Test
public void testContainerSetup() throws Exception {
    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);
    // ////// Construct the container-spec.
    ContainerLaunchContext containerLaunchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    URL resource_alpha = URL.fromPath(localFS.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, 40);
    // Now ascertain that the resources are localised correctly.
    ApplicationId appId = cId.getApplicationAttemptId().getApplicationId();
    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);
    for (File f : new File[] { localDir, sysDir, userCacheDir, appDir, appSysDir, containerDir, containerSysDir }) {
        Assert.assertTrue(f.getAbsolutePath() + " doesn't exist!!", f.exists());
        Assert.assertTrue(f.getAbsolutePath() + " is not a directory!!", f.isDirectory());
    }
    Assert.assertTrue(targetFile.getAbsolutePath() + " doesn't exist!!", targetFile.exists());
    // Now verify the contents of the file
    BufferedReader reader = new BufferedReader(new FileReader(targetFile));
    Assert.assertEquals("Hello World!", reader.readLine());
    Assert.assertEquals(null, reader.readLine());
}

Example 95 with LocalResource

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

the class TestContainerManager method testContainerLaunchFromPreviousRM.

@Test
public void testContainerLaunchFromPreviousRM() throws IOException, InterruptedException, YarnException {
    containerManager.start();
    ContainerLaunchContext containerLaunchContext = recordFactory.newRecordInstance(ContainerLaunchContext.class);
    ContainerId cId1 = createContainerId(0);
    ContainerId cId2 = createContainerId(0);
    containerLaunchContext.setLocalResources(new HashMap<String, LocalResource>());
    // Construct the Container with Invalid RMIdentifier
    StartContainerRequest startRequest1 = StartContainerRequest.newInstance(containerLaunchContext, createContainerToken(cId1, ResourceManagerConstants.RM_INVALID_IDENTIFIER, context.getNodeId(), user, context.getContainerTokenSecretManager()));
    List<StartContainerRequest> list = new ArrayList<>();
    list.add(startRequest1);
    StartContainersRequest allRequests = StartContainersRequest.newInstance(list);
    containerManager.startContainers(allRequests);
    boolean catchException = false;
    try {
        StartContainersResponse response = containerManager.startContainers(allRequests);
        if (response.getFailedRequests().containsKey(cId1)) {
            throw response.getFailedRequests().get(cId1).deSerialize();
        }
    } catch (Throwable e) {
        e.printStackTrace();
        catchException = true;
        Assert.assertTrue(e.getMessage().contains("Container " + cId1 + " rejected as it is allocated by a previous RM"));
        Assert.assertTrue(e.getClass().getName().equalsIgnoreCase(InvalidContainerException.class.getName()));
    }
    // Verify that startContainer fail because of invalid container request
    Assert.assertTrue(catchException);
    // Construct the Container with a RMIdentifier within current RM
    StartContainerRequest startRequest2 = StartContainerRequest.newInstance(containerLaunchContext, createContainerToken(cId2, DUMMY_RM_IDENTIFIER, context.getNodeId(), user, context.getContainerTokenSecretManager()));
    List<StartContainerRequest> list2 = new ArrayList<>();
    list.add(startRequest2);
    StartContainersRequest allRequests2 = StartContainersRequest.newInstance(list2);
    containerManager.startContainers(allRequests2);
    boolean noException = true;
    try {
        containerManager.startContainers(allRequests2);
    } catch (YarnException e) {
        noException = false;
    }
    // Verify that startContainer get no YarnException
    Assert.assertTrue(noException);
}