Examples with UserPayload org.apache.tez.dag.api.UserPayload used on opensource projects

Example 11 with UserPayload

use of org.apache.tez.dag.api.UserPayload in project hive by apache.

the class TezSessionState method openInternal.

protected void openInternal(final HiveConf conf, Collection<String> additionalFiles, boolean isAsync, LogHelper console, Path scratchDir) throws IOException, LoginException, IllegalArgumentException, URISyntaxException, TezException {
    this.conf = conf;
    // TODO Why is the queue name set again. It has already been setup via setQueueName. Do only one of the two.
    String confQueueName = conf.get(TezConfiguration.TEZ_QUEUE_NAME);
    if (queueName != null && !queueName.equals(confQueueName)) {
        LOG.warn("Resetting a queue name that was already set: was " + queueName + ", now " + confQueueName);
    }
    this.queueName = confQueueName;
    this.doAsEnabled = conf.getBoolVar(HiveConf.ConfVars.HIVE_SERVER2_ENABLE_DOAS);
    final boolean llapMode = "llap".equalsIgnoreCase(HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_EXECUTION_MODE));
    // TODO This - at least for the session pool - will always be the hive user. How does doAs above this affect things ?
    UserGroupInformation ugi = Utils.getUGI();
    user = ugi.getShortUserName();
    LOG.info("User of session id " + sessionId + " is " + user);
    // create the tez tmp dir
    tezScratchDir = scratchDir == null ? createTezDir(sessionId) : scratchDir;
    additionalFilesNotFromConf.clear();
    if (additionalFiles != null) {
        additionalFilesNotFromConf.addAll(additionalFiles);
    }
    refreshLocalResourcesFromConf(conf);
    // unless already installed on all the cluster nodes, we'll have to
    // localize hive-exec.jar as well.
    appJarLr = createJarLocalResource(utils.getExecJarPathLocal());
    // configuration for the application master
    final Map<String, LocalResource> commonLocalResources = new HashMap<String, LocalResource>();
    commonLocalResources.put(utils.getBaseName(appJarLr), appJarLr);
    for (LocalResource lr : localizedResources) {
        commonLocalResources.put(utils.getBaseName(lr), lr);
    }
    if (llapMode) {
        // localize llap client jars
        addJarLRByClass(LlapTaskSchedulerService.class, commonLocalResources);
        addJarLRByClass(LlapProtocolClientImpl.class, commonLocalResources);
        addJarLRByClass(LlapProtocolClientProxy.class, commonLocalResources);
        addJarLRByClassName("org.apache.hadoop.registry.client.api.RegistryOperations", commonLocalResources);
    }
    // Create environment for AM.
    Map<String, String> amEnv = new HashMap<String, String>();
    MRHelpers.updateEnvBasedOnMRAMEnv(conf, amEnv);
    // and finally we're ready to create and start the session
    // generate basic tez config
    final TezConfiguration tezConfig = new TezConfiguration(conf);
    // set up the staging directory to use
    tezConfig.set(TezConfiguration.TEZ_AM_STAGING_DIR, tezScratchDir.toUri().toString());
    conf.stripHiddenConfigurations(tezConfig);
    ServicePluginsDescriptor servicePluginsDescriptor;
    Credentials llapCredentials = null;
    if (llapMode) {
        if (UserGroupInformation.isSecurityEnabled()) {
            llapCredentials = new Credentials();
            llapCredentials.addToken(LlapTokenIdentifier.KIND_NAME, getLlapToken(user, tezConfig));
        }
        // TODO Change this to not serialize the entire Configuration - minor.
        UserPayload servicePluginPayload = TezUtils.createUserPayloadFromConf(tezConfig);
        // we need plugins to handle llap and uber mode
        servicePluginsDescriptor = ServicePluginsDescriptor.create(true, new TaskSchedulerDescriptor[] { TaskSchedulerDescriptor.create(LLAP_SERVICE, LLAP_SCHEDULER).setUserPayload(servicePluginPayload) }, new ContainerLauncherDescriptor[] { ContainerLauncherDescriptor.create(LLAP_SERVICE, LLAP_LAUNCHER) }, new TaskCommunicatorDescriptor[] { TaskCommunicatorDescriptor.create(LLAP_SERVICE, LLAP_TASK_COMMUNICATOR).setUserPayload(servicePluginPayload) });
    } else {
        servicePluginsDescriptor = ServicePluginsDescriptor.create(true);
    }
    // container prewarming. tell the am how many containers we need
    if (HiveConf.getBoolVar(conf, ConfVars.HIVE_PREWARM_ENABLED)) {
        int n = HiveConf.getIntVar(conf, ConfVars.HIVE_PREWARM_NUM_CONTAINERS);
        n = Math.max(tezConfig.getInt(TezConfiguration.TEZ_AM_SESSION_MIN_HELD_CONTAINERS, TezConfiguration.TEZ_AM_SESSION_MIN_HELD_CONTAINERS_DEFAULT), n);
        tezConfig.setInt(TezConfiguration.TEZ_AM_SESSION_MIN_HELD_CONTAINERS, n);
    }
    setupSessionAcls(tezConfig, conf);
    final TezClient session = TezClient.newBuilder("HIVE-" + sessionId, tezConfig).setIsSession(true).setLocalResources(commonLocalResources).setCredentials(llapCredentials).setServicePluginDescriptor(servicePluginsDescriptor).build();
    LOG.info("Opening new Tez Session (id: " + sessionId + ", scratch dir: " + tezScratchDir + ")");
    TezJobMonitor.initShutdownHook();
    if (!isAsync) {
        startSessionAndContainers(session, conf, commonLocalResources, tezConfig, false);
        this.session = session;
    } else {
        FutureTask<TezClient> sessionFuture = new FutureTask<>(new Callable<TezClient>() {

            @Override
            public TezClient call() throws Exception {
                try {
                    return startSessionAndContainers(session, conf, commonLocalResources, tezConfig, true);
                } catch (Throwable t) {
                    LOG.error("Failed to start Tez session", t);
                    throw (t instanceof Exception) ? (Exception) t : new Exception(t);
                }
            }
        });
        new Thread(sessionFuture, "Tez session start thread").start();
        // We assume here nobody will try to get session before open() returns.
        this.console = console;
        this.sessionFuture = sessionFuture;
    }
}

Example 12 with UserPayload

use of org.apache.tez.dag.api.UserPayload in project hive by apache.

the class DagUtils method createVertex.

/*
   * Helper function to create Vertex from MapWork.
   */
private Vertex createVertex(JobConf conf, MapWork mapWork, LocalResource appJarLr, List<LocalResource> additionalLr, FileSystem fs, Path mrScratchDir, Context ctx, VertexType vertexType) throws Exception {
    Path tezDir = getTezDir(mrScratchDir);
    // set up the operator plan
    Utilities.cacheMapWork(conf, mapWork, mrScratchDir);
    // create the directories FileSinkOperators need
    Utilities.createTmpDirs(conf, mapWork);
    // finally create the vertex
    Vertex map = null;
    // use tez to combine splits
    boolean groupSplitsInInputInitializer;
    DataSourceDescriptor dataSource;
    int numTasks = -1;
    @SuppressWarnings("rawtypes") Class inputFormatClass = conf.getClass("mapred.input.format.class", InputFormat.class);
    boolean vertexHasCustomInput = VertexType.isCustomInputType(vertexType);
    LOG.info("Vertex has custom input? " + vertexHasCustomInput);
    if (vertexHasCustomInput) {
        groupSplitsInInputInitializer = false;
        // grouping happens in execution phase. The input payload should not enable grouping here,
        // it will be enabled in the CustomVertex.
        inputFormatClass = HiveInputFormat.class;
        conf.setClass("mapred.input.format.class", HiveInputFormat.class, InputFormat.class);
        // mapreduce.tez.input.initializer.serialize.event.payload should be set to false when using
        // this plug-in to avoid getting a serialized event at run-time.
        conf.setBoolean("mapreduce.tez.input.initializer.serialize.event.payload", false);
    } else {
        // is HiveInputFormat
        if (inputFormatClass == HiveInputFormat.class) {
            groupSplitsInInputInitializer = true;
        } else {
            groupSplitsInInputInitializer = false;
        }
    }
    if (mapWork instanceof MergeFileWork) {
        Path outputPath = ((MergeFileWork) mapWork).getOutputDir();
        // prepare the tmp output directory. The output tmp directory should
        // exist before jobClose (before renaming after job completion)
        Path tempOutPath = Utilities.toTempPath(outputPath);
        try {
            FileSystem tmpOutFS = tempOutPath.getFileSystem(conf);
            if (!tmpOutFS.exists(tempOutPath)) {
                tmpOutFS.mkdirs(tempOutPath);
            }
        } catch (IOException e) {
            throw new RuntimeException("Can't make path " + outputPath + " : " + e.getMessage(), e);
        }
    }
    // remember mapping of plan to input
    conf.set(Utilities.INPUT_NAME, mapWork.getName());
    if (HiveConf.getBoolVar(conf, ConfVars.HIVE_AM_SPLIT_GENERATION)) {
        // set up the operator plan. (before setting up splits on the AM)
        Utilities.setMapWork(conf, mapWork, mrScratchDir, false);
        // the correct plugin.
        if (groupSplitsInInputInitializer) {
            // Not setting a payload, since the MRInput payload is the same and can be accessed.
            InputInitializerDescriptor descriptor = InputInitializerDescriptor.create(HiveSplitGenerator.class.getName());
            dataSource = MRInputLegacy.createConfigBuilder(conf, inputFormatClass).groupSplits(true).setCustomInitializerDescriptor(descriptor).build();
        } else {
            // Not HiveInputFormat, or a custom VertexManager will take care of grouping splits
            if (vertexHasCustomInput) {
                dataSource = MultiMRInput.createConfigBuilder(conf, inputFormatClass).groupSplits(false).build();
            } else {
                dataSource = MRInputLegacy.createConfigBuilder(conf, inputFormatClass).groupSplits(false).build();
            }
        }
    } else {
        // Setup client side split generation.
        // we need to set this, because with HS2 and client side split
        // generation we end up not finding the map work. This is
        // because of thread local madness (tez split generation is
        // multi-threaded - HS2 plan cache uses thread locals). Setting
        // VECTOR_MODE/USE_VECTORIZED_INPUT_FILE_FORMAT causes the split gen code to use the conf instead
        // of the map work.
        conf.setBoolean(Utilities.VECTOR_MODE, mapWork.getVectorMode());
        conf.setBoolean(Utilities.USE_VECTORIZED_INPUT_FILE_FORMAT, mapWork.getUseVectorizedInputFileFormat());
        dataSource = MRInputHelpers.configureMRInputWithLegacySplitGeneration(conf, new Path(tezDir, "split_" + mapWork.getName().replaceAll(" ", "_")), true);
        numTasks = dataSource.getNumberOfShards();
        // set up the operator plan. (after generating splits - that changes configs)
        Utilities.setMapWork(conf, mapWork, mrScratchDir, false);
    }
    UserPayload serializedConf = TezUtils.createUserPayloadFromConf(conf);
    String procClassName = MapTezProcessor.class.getName();
    if (mapWork instanceof MergeFileWork) {
        procClassName = MergeFileTezProcessor.class.getName();
    }
    VertexExecutionContext executionContext = createVertexExecutionContext(mapWork);
    map = Vertex.create(mapWork.getName(), ProcessorDescriptor.create(procClassName).setUserPayload(serializedConf), numTasks, getContainerResource(conf));
    map.setTaskEnvironment(getContainerEnvironment(conf, true));
    map.setExecutionContext(executionContext);
    map.setTaskLaunchCmdOpts(getContainerJavaOpts(conf));
    assert mapWork.getAliasToWork().keySet().size() == 1;
    // Add the actual source input
    String alias = mapWork.getAliasToWork().keySet().iterator().next();
    map.addDataSource(alias, dataSource);
    Map<String, LocalResource> localResources = new HashMap<String, LocalResource>();
    localResources.put(getBaseName(appJarLr), appJarLr);
    for (LocalResource lr : additionalLr) {
        localResources.put(getBaseName(lr), lr);
    }
    map.addTaskLocalFiles(localResources);
    return map;
}