List of usage examples for org.apache.hadoop.yarn.client.api AMRMClient createAMRMClient
@Public public static <T extends ContainerRequest> AMRMClient<T> createAMRMClient()
From source file:com.cloudera.llama.am.yarn.YarnRMConnector.java
License:Apache License
private void _registerSchedulerAndCreateNMClient(String queue) throws Exception { NMTokenCache nmTokenCache = new NMTokenCache(); nmClient = NMClient.createNMClient(); nmClient.setNMTokenCache(nmTokenCache); nmClient.init(yarnConf);/*ww w . j a v a 2s. c o m*/ nmClient.start(); LOG.debug("Started NMClient, AM '{}' with scheduler for '{}' queue", appId, queue); int heartbeatInterval = getConf().getInt(HEARTBEAT_INTERVAL_KEY, HEARTBEAT_INTERNAL_DEFAULT); AMRMClient<LlamaContainerRequest> amRmClient = AMRMClient.createAMRMClient(); amRmClient.setNMTokenCache(nmTokenCache); amRmClientAsync = AMRMClientAsync.createAMRMClientAsync(amRmClient, heartbeatInterval, YarnRMConnector.this); amRmClientAsync.init(yarnConf); amRmClientAsync.start(); String urlWithoutScheme = getConf().get(ADVERTISED_TRACKING_URL_KEY, "http://") .substring("http://".length()); RegisterApplicationMasterResponse response = amRmClientAsync.registerApplicationMaster( getConf().get(ADVERTISED_HOSTNAME_KEY, ""), getConf().getInt(ADVERTISED_PORT_KEY, 0), urlWithoutScheme); maxResource = response.getMaximumResourceCapability(); nodes = Collections.synchronizedMap(new HashMap<String, Resource>()); for (NodeReport nodeReport : yarnClient.getNodeReports()) { if (nodeReport.getNodeState() == NodeState.RUNNING) { String nodeKey = getNodeName(nodeReport.getNodeId()); nodes.put(nodeKey, nodeReport.getCapability()); LOG.debug("Added node '{}' with '{}' cpus and '{}' memory", nodeKey, nodeReport.getCapability().getVirtualCores(), nodeReport.getCapability().getMemory()); } } LOG.debug("Registered with scheduler, AM '{}' for '{}' queue", appId, queue); }
From source file:com.continuuity.weave.internal.yarn.Hadoop21YarnAMClient.java
License:Apache License
public Hadoop21YarnAMClient(Configuration conf) { String masterContainerId = System.getenv().get(ApplicationConstants.Environment.CONTAINER_ID.name()); Preconditions.checkArgument(masterContainerId != null, "Missing %s from environment", ApplicationConstants.Environment.CONTAINER_ID.name()); this.containerId = ConverterUtils.toContainerId(masterContainerId); this.containerRequests = ArrayListMultimap.create(); this.amrmClient = AMRMClient.createAMRMClient(); this.amrmClient.init(conf); this.nmClient = new Hadoop21YarnNMClient(conf); }
From source file:com.datatorrent.stram.StramMiniClusterTest.java
License:Apache License
@Ignore @Test//from w w w. ja va 2 s. c o m public void testUnmanagedAM2() throws Exception { new InlineAM(conf) { @Override @SuppressWarnings("SleepWhileInLoop") public void runAM(ApplicationAttemptId attemptId) throws Exception { LOG.debug("AM running {}", attemptId); AMRMClient<ContainerRequest> amRmClient = AMRMClient.createAMRMClient(); amRmClient.init(conf); amRmClient.start(); // register with the RM (LAUNCHED -> RUNNING) amRmClient.registerApplicationMaster("", 0, null); // AM specific logic String[] hosts = { "vm1" }; String[] racks = { "somerack" }; Resource capability = Records.newRecord(Resource.class); capability.setMemory(1000); Priority priority = Records.newRecord(Priority.class); priority.setPriority(10); AMRMClient.ContainerRequest req = new AMRMClient.ContainerRequest(capability, hosts, racks, priority); amRmClient.addContainerRequest(req); amRmClient.addContainerRequest(req); /* capability = Records.newRecord(Resource.class); capability.setMemory(5512); priority = Records.newRecord(Priority.class); priority.setPriority(11); req = new AMRMClient.ContainerRequest(capability, hosts, racks, priority, 3); amRmClient.addContainerRequest(req); */ for (int i = 0; i < 100; i++) { AllocateResponse ar = amRmClient.allocate(0); sleep(1000); LOG.debug("allocateResponse: {}", ar); for (Container c : ar.getAllocatedContainers()) { LOG.debug("*** allocated {}", c.getResource()); amRmClient.removeContainerRequest(req); } /* GetClusterNodesRequest request = Records.newRecord(GetClusterNodesRequest.class); ClientRMService clientRMService = yarnCluster.getResourceManager().getClientRMService(); GetClusterNodesResponse response = clientRMService.getClusterNodes(request); List<NodeReport> nodeReports = response.getNodeReports(); LOG.info(nodeReports); for (NodeReport nr: nodeReports) { LOG.info("Node: " + nr.getNodeId()); LOG.info("Total memory: " + nr.getCapability()); LOG.info("Used memory: " + nr.getUsed()); LOG.info("Number containers: " + nr.getNumContainers()); } */ } // unregister from RM amRmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "testUnmanagedAM finished", null); } }.run(); }
From source file:com.datatorrent.stram.StreamingAppMasterService.java
License:Apache License
@Override protected void serviceInit(Configuration conf) throws Exception { LOG.info("Application master" + ", appId=" + appAttemptID.getApplicationId().getId() + ", clustertimestamp=" + appAttemptID.getApplicationId().getClusterTimestamp() + ", attemptId=" + appAttemptID.getAttemptId()); FileInputStream fis = new FileInputStream("./" + LogicalPlan.SER_FILE_NAME); try {//from w w w .j a v a 2s . com this.dag = LogicalPlan.read(fis); } finally { fis.close(); } // "debug" simply dumps all data using LOG.info if (dag.isDebug()) { dumpOutDebugInfo(); } FSRecoveryHandler recoveryHandler = new FSRecoveryHandler(dag.assertAppPath(), conf); this.dnmgr = StreamingContainerManager.getInstance(recoveryHandler, dag, true); dag = this.dnmgr.getLogicalPlan(); this.appContext = new ClusterAppContextImpl(dag.getAttributes()); Map<Class<?>, Class<? extends StringCodec<?>>> codecs = dag.getAttributes().get(DAG.STRING_CODECS); StringCodecs.loadConverters(codecs); LOG.info("Starting application with {} operators in {} containers", dnmgr.getPhysicalPlan().getAllOperators().size(), dnmgr.getPhysicalPlan().getContainers().size()); if (UserGroupInformation.isSecurityEnabled()) { // TODO :- Need to perform token renewal delegationTokenManager = new StramDelegationTokenManager(DELEGATION_KEY_UPDATE_INTERVAL, DELEGATION_TOKEN_MAX_LIFETIME, DELEGATION_TOKEN_RENEW_INTERVAL, DELEGATION_TOKEN_REMOVER_SCAN_INTERVAL); } this.nmClient = new NMClientAsyncImpl(new NMCallbackHandler()); addService(nmClient); this.amRmClient = AMRMClient.createAMRMClient(); addService(amRmClient); // start RPC server int rpcListenerCount = dag.getValue(DAGContext.HEARTBEAT_LISTENER_THREAD_COUNT); this.heartbeatListener = new StreamingContainerParent(this.getClass().getName(), dnmgr, delegationTokenManager, rpcListenerCount); addService(heartbeatListener); String appDataPushTransport = dag.getValue(LogicalPlan.METRICS_TRANSPORT); if (appDataPushTransport != null) { this.appDataPushAgent = new AppDataPushAgent(dnmgr, appContext); addService(this.appDataPushAgent); } // initialize all services added above super.serviceInit(conf); }
From source file:com.ibm.bi.dml.yarn.DMLAppMaster.java
License:Open Source License
/** * /* ww w . j av a 2s . c om*/ * @param args * @throws YarnException * @throws IOException */ public void runApplicationMaster(String[] args) throws YarnException, IOException { _conf = new YarnConfiguration(); //obtain application ID String containerIdString = System.getenv(Environment.CONTAINER_ID.name()); ContainerId containerId = ConverterUtils.toContainerId(containerIdString); _appId = containerId.getApplicationAttemptId().getApplicationId(); LOG.info("SystemML appplication master (applicationID: " + _appId + ")"); //initialize clients to ResourceManager AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(_conf); rmClient.start(); //register with ResourceManager rmClient.registerApplicationMaster("", 0, ""); //host, port for rm communication LOG.debug("Registered the SystemML application master with resource manager"); //start status reporter to ResourceManager DMLAppMasterStatusReporter reporter = new DMLAppMasterStatusReporter(rmClient, 10000); reporter.start(); LOG.debug("Started status reporter (heartbeat to resource manager)"); //set DMLscript app master context DMLScript.setActiveAM(); //parse input arguments String[] otherArgs = new GenericOptionsParser(_conf, args).getRemainingArgs(); //run SystemML CP FinalApplicationStatus status = null; try { //core dml script execution (equivalent to non-AM runtime) boolean success = DMLScript.executeScript(_conf, otherArgs); if (success) status = FinalApplicationStatus.SUCCEEDED; else status = FinalApplicationStatus.FAILED; } catch (DMLScriptException ex) { LOG.error(DMLYarnClient.APPMASTER_NAME + ": Failed to executed DML script due to stop call:\n\t" + ex.getMessage()); status = FinalApplicationStatus.FAILED; writeMessageToHDFSWorkingDir(ex.getMessage()); } catch (Exception ex) { LOG.error(DMLYarnClient.APPMASTER_NAME + ": Failed to executed DML script.", ex); status = FinalApplicationStatus.FAILED; } finally { //stop periodic status reports reporter.stopStatusReporter(); LOG.debug("Stopped status reporter"); //unregister resource manager client rmClient.unregisterApplicationMaster(status, "", ""); LOG.debug("Unregistered the SystemML application master"); } }
From source file:edu.cmu.graphchi.toolkits.collaborative_filtering.yarn.ApplicationMaster.java
License:Apache License
/** * Main run function for the application master * * @throws YarnException/*from w w w.j a v a 2 s. co m*/ * @throws IOException */ @SuppressWarnings({ "unchecked" }) public boolean run() throws YarnException, IOException { yarnClient.start(); LOG.info("Starting ApplicationMaster"); Credentials credentials = UserGroupInformation.getCurrentUser().getCredentials(); DataOutputBuffer dob = new DataOutputBuffer(); credentials.writeTokenStorageToStream(dob); // Now remove the AM->RM token so that containers cannot access it. Iterator<Token<?>> iter = credentials.getAllTokens().iterator(); while (iter.hasNext()) { Token<?> token = iter.next(); if (token.getKind().equals(AMRMTokenIdentifier.KIND_NAME)) { iter.remove(); } } allTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength()); amRMClient = AMRMClient.createAMRMClient(); amRMClient.init(conf); amRMClient.start(); containerListener = createNMCallbackHandler(); nmClientAsync = new NMClientAsyncImpl(containerListener); nmClientAsync.init(conf); nmClientAsync.start(); // Register self with ResourceManager // This will start heartbeating to the RM appMasterHostname = NetUtils.getHostname(); RegisterApplicationMasterResponse response = amRMClient.registerApplicationMaster(appMasterHostname, appMasterRpcPort, appMasterTrackingUrl); //TODO: Figure out how to do this. List<NodeReport> reports = this.yarnClient.getNodeReports(); LOG.info("Cluster Status"); List<Resource> availableResources = new ArrayList<Resource>(); for (NodeReport nr : reports) { LOG.info(" NodeId: " + nr.getNodeId() + " Capabilities " + nr.getCapability() + " Used Resources " + nr.getUsed()); int availableMem = nr.getCapability().getMemory() - nr.getUsed().getMemory(); int availableVCores = nr.getCapability().getVirtualCores() - nr.getUsed().getVirtualCores(); Resource resource = Resource.newInstance(availableMem, availableVCores); availableResources.add(resource); } /*Based on available resources scheduler should decide the best allocation of recommenders to resources and return a list of resources that should be requested from the ResourceManager*/ DataSetDescription datasetDesc = new DataSetDescription(this.setup.dataMetadataFile); List<RecommenderPool> recommenderPools = RecommenderScheduler.splitRecommenderPool(availableResources, recommenders, datasetDesc, this.setup.nShards); for (RecommenderPool res : recommenderPools) { ContainerRequest containerAsk = setupContainerAskForRM(res.getTotalMemory(), requestPriority); LOG.info("CONTAINER ASK: " + containerAsk); amRMClient.addContainerRequest(containerAsk); } float progress = 0; List<RecommenderPool> pendingPools = new ArrayList<RecommenderPool>(); for (RecommenderPool p : recommenderPools) pendingPools.add(p); this.numTotalContainers = recommenderPools.size(); this.numCompletedContainers.set(0); this.numAllocatedContainers.set(0); while (numCompletedContainers.get() != numTotalContainers) { try { Thread.sleep(200); AllocateResponse allocResp = amRMClient.allocate(progress); List<Container> newContainers = allocResp.getAllocatedContainers(); List<ContainerStatus> completedContainers = allocResp.getCompletedContainersStatuses(); if (this.numAllocatedContainers.get() >= this.numTotalContainers && pendingPools.size() != 0) { //Ask for new containers for pending pools LOG.warn("The number of allocated containers has exceeded number of total containers, but " + "the pending pool size is still not 0. Asking for new containers from RM"); for (RecommenderPool res : pendingPools) { ContainerRequest containerAsk = setupContainerAskForRM(res.getTotalMemory(), requestPriority); LOG.info("NEW CONTAINER ASK: " + containerAsk); amRMClient.addContainerRequest(containerAsk); } } if (newContainers.size() > 0) { LOG.info("Allocated " + newContainers.size() + " new containers"); numAllocatedContainers.addAndGet(newContainers.size()); for (Container container : newContainers) { //Find matching recommender pool from pendingRecommender pools. RecommenderPool pool = null; for (RecommenderPool p : pendingPools) { if (p.getTotalMemory() == container.getResource().getMemory()) { pool = p; break; } } if (pool == null) { LOG.warn("No Takers for Container " + container + " Releasing container"); amRMClient.releaseAssignedContainer(container.getId()); } else { startContainer(container, pool); //This pool has now got a container. Remove it from pending pools pendingPools.remove(pool); } } } onContainersCompleted(completedContainers); } catch (InterruptedException ex) { } } finish(); return success; }
From source file:eu.stratosphere.yarn.ApplicationMaster.java
License:Apache License
private void run() throws Exception { //Utils.logFilesInCurrentDirectory(LOG); // Initialize clients to ResourceManager and NodeManagers Configuration conf = Utils.initializeYarnConfiguration(); FileSystem fs = FileSystem.get(conf); Map<String, String> envs = System.getenv(); final String currDir = envs.get(Environment.PWD.key()); final String logDirs = envs.get(Environment.LOG_DIRS.key()); final String ownHostname = envs.get(Environment.NM_HOST.key()); final String appId = envs.get(Client.ENV_APP_ID); final String clientHomeDir = envs.get(Client.ENV_CLIENT_HOME_DIR); final String applicationMasterHost = envs.get(Environment.NM_HOST.key()); final String remoteStratosphereJarPath = envs.get(Client.STRATOSPHERE_JAR_PATH); final String shipListString = envs.get(Client.ENV_CLIENT_SHIP_FILES); final String yarnClientUsername = envs.get(Client.ENV_CLIENT_USERNAME); final int taskManagerCount = Integer.valueOf(envs.get(Client.ENV_TM_COUNT)); final int memoryPerTaskManager = Integer.valueOf(envs.get(Client.ENV_TM_MEMORY)); final int coresPerTaskManager = Integer.valueOf(envs.get(Client.ENV_TM_CORES)); int heapLimit = Utils.calculateHeapSize(memoryPerTaskManager); if (currDir == null) { throw new RuntimeException("Current directory unknown"); }/*from w w w .jav a 2 s . c o m*/ if (ownHostname == null) { throw new RuntimeException("Own hostname (" + Environment.NM_HOST + ") not set."); } LOG.info("Working directory " + currDir); // load Stratosphere configuration. Utils.getStratosphereConfiguration(currDir); final String localWebInterfaceDir = currDir + "/resources/" + ConfigConstants.DEFAULT_JOB_MANAGER_WEB_PATH_NAME; // Update yaml conf -> set jobManager address to this machine's address. FileInputStream fis = new FileInputStream(currDir + "/stratosphere-conf.yaml"); BufferedReader br = new BufferedReader(new InputStreamReader(fis)); Writer output = new BufferedWriter(new FileWriter(currDir + "/stratosphere-conf-modified.yaml")); String line; while ((line = br.readLine()) != null) { if (line.contains(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY)) { output.append(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY + ": " + ownHostname + "\n"); } else if (line.contains(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY)) { output.append(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY + ": " + "\n"); } else { output.append(line + "\n"); } } // just to make sure. output.append(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY + ": " + ownHostname + "\n"); output.append(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY + ": " + localWebInterfaceDir + "\n"); output.append(ConfigConstants.JOB_MANAGER_WEB_LOG_PATH_KEY + ": " + logDirs + "\n"); output.close(); br.close(); File newConf = new File(currDir + "/stratosphere-conf-modified.yaml"); if (!newConf.exists()) { LOG.warn("modified yaml does not exist!"); } Utils.copyJarContents("resources/" + ConfigConstants.DEFAULT_JOB_MANAGER_WEB_PATH_NAME, ApplicationMaster.class.getProtectionDomain().getCodeSource().getLocation().getPath()); JobManager jm; { String pathToNepheleConfig = currDir + "/stratosphere-conf-modified.yaml"; String[] args = { "-executionMode", "cluster", "-configDir", pathToNepheleConfig }; // start the job manager jm = JobManager.initialize(args); // Start info server for jobmanager jm.startInfoServer(); } AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf); rmClient.start(); NMClient nmClient = NMClient.createNMClient(); nmClient.init(conf); nmClient.start(); // Register with ResourceManager LOG.info("registering ApplicationMaster"); rmClient.registerApplicationMaster(applicationMasterHost, 0, "http://" + applicationMasterHost + ":" + GlobalConfiguration.getString(ConfigConstants.JOB_MANAGER_WEB_PORT_KEY, "undefined")); // Priority for worker containers - priorities are intra-application Priority priority = Records.newRecord(Priority.class); priority.setPriority(0); // Resource requirements for worker containers Resource capability = Records.newRecord(Resource.class); capability.setMemory(memoryPerTaskManager); capability.setVirtualCores(coresPerTaskManager); // Make container requests to ResourceManager for (int i = 0; i < taskManagerCount; ++i) { ContainerRequest containerAsk = new ContainerRequest(capability, null, null, priority); LOG.info("Requesting TaskManager container " + i); rmClient.addContainerRequest(containerAsk); } LocalResource stratosphereJar = Records.newRecord(LocalResource.class); LocalResource stratosphereConf = Records.newRecord(LocalResource.class); // register Stratosphere Jar with remote HDFS final Path remoteJarPath = new Path(remoteStratosphereJarPath); Utils.registerLocalResource(fs, remoteJarPath, stratosphereJar); // register conf with local fs. Path remoteConfPath = Utils.setupLocalResource(conf, fs, appId, new Path("file://" + currDir + "/stratosphere-conf-modified.yaml"), stratosphereConf, new Path(clientHomeDir)); LOG.info("Prepared localresource for modified yaml: " + stratosphereConf); boolean hasLog4j = new File(currDir + "/log4j.properties").exists(); // prepare the files to ship LocalResource[] remoteShipRsc = null; String[] remoteShipPaths = shipListString.split(","); if (!shipListString.isEmpty()) { remoteShipRsc = new LocalResource[remoteShipPaths.length]; { // scope for i int i = 0; for (String remoteShipPathStr : remoteShipPaths) { if (remoteShipPathStr == null || remoteShipPathStr.isEmpty()) { continue; } remoteShipRsc[i] = Records.newRecord(LocalResource.class); Path remoteShipPath = new Path(remoteShipPathStr); Utils.registerLocalResource(fs, remoteShipPath, remoteShipRsc[i]); i++; } } } // respect custom JVM options in the YAML file final String javaOpts = GlobalConfiguration.getString(ConfigConstants.STRATOSPHERE_JVM_OPTIONS, ""); // Obtain allocated containers and launch int allocatedContainers = 0; int completedContainers = 0; while (allocatedContainers < taskManagerCount) { AllocateResponse response = rmClient.allocate(0); for (Container container : response.getAllocatedContainers()) { LOG.info("Got new Container for TM " + container.getId() + " on host " + container.getNodeId().getHost()); ++allocatedContainers; // Launch container by create ContainerLaunchContext ContainerLaunchContext ctx = Records.newRecord(ContainerLaunchContext.class); String tmCommand = "$JAVA_HOME/bin/java -Xmx" + heapLimit + "m " + javaOpts; if (hasLog4j) { tmCommand += " -Dlog.file=\"" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-log4j.log\" -Dlog4j.configuration=file:log4j.properties"; } tmCommand += " eu.stratosphere.yarn.YarnTaskManagerRunner -configDir . " + " 1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-stdout.log" + " 2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-stderr.log"; ctx.setCommands(Collections.singletonList(tmCommand)); LOG.info("Starting TM with command=" + tmCommand); // copy resources to the TaskManagers. Map<String, LocalResource> localResources = new HashMap<String, LocalResource>(2); localResources.put("stratosphere.jar", stratosphereJar); localResources.put("stratosphere-conf.yaml", stratosphereConf); // add ship resources if (!shipListString.isEmpty()) { Preconditions.checkNotNull(remoteShipRsc); for (int i = 0; i < remoteShipPaths.length; i++) { localResources.put(new Path(remoteShipPaths[i]).getName(), remoteShipRsc[i]); } } ctx.setLocalResources(localResources); // Setup CLASSPATH for Container (=TaskTracker) Map<String, String> containerEnv = new HashMap<String, String>(); Utils.setupEnv(conf, containerEnv); //add stratosphere.jar to class path. containerEnv.put(Client.ENV_CLIENT_USERNAME, yarnClientUsername); ctx.setEnvironment(containerEnv); UserGroupInformation user = UserGroupInformation.getCurrentUser(); try { Credentials credentials = user.getCredentials(); DataOutputBuffer dob = new DataOutputBuffer(); credentials.writeTokenStorageToStream(dob); ByteBuffer securityTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength()); ctx.setTokens(securityTokens); } catch (IOException e) { LOG.warn("Getting current user info failed when trying to launch the container" + e.getMessage()); } LOG.info("Launching container " + allocatedContainers); nmClient.startContainer(container, ctx); } for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; LOG.info("Completed container (while allocating) " + status.getContainerId() + ". Total Completed:" + completedContainers); LOG.info("Diagnostics " + status.getDiagnostics()); } Thread.sleep(100); } // Now wait for containers to complete while (completedContainers < taskManagerCount) { AllocateResponse response = rmClient.allocate(completedContainers / taskManagerCount); for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; LOG.info("Completed container " + status.getContainerId() + ". Total Completed:" + completedContainers); LOG.info("Diagnostics " + status.getDiagnostics()); } Thread.sleep(5000); } LOG.info("Shutting down JobManager"); jm.shutdown(); // Un-register with ResourceManager rmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "", ""); }
From source file:org.apache.flink.yarn.ApplicationMaster.java
License:Apache License
private void run() throws Exception { //Utils.logFilesInCurrentDirectory(LOG); // Initialize clients to ResourceManager and NodeManagers Configuration conf = Utils.initializeYarnConfiguration(); FileSystem fs = FileSystem.get(conf); Map<String, String> envs = System.getenv(); final String currDir = envs.get(Environment.PWD.key()); final String logDirs = envs.get(Environment.LOG_DIRS.key()); final String ownHostname = envs.get(Environment.NM_HOST.key()); final String appId = envs.get(Client.ENV_APP_ID); final String clientHomeDir = envs.get(Client.ENV_CLIENT_HOME_DIR); final String applicationMasterHost = envs.get(Environment.NM_HOST.key()); final String remoteFlinkJarPath = envs.get(Client.FLINK_JAR_PATH); final String shipListString = envs.get(Client.ENV_CLIENT_SHIP_FILES); final String yarnClientUsername = envs.get(Client.ENV_CLIENT_USERNAME); final int taskManagerCount = Integer.valueOf(envs.get(Client.ENV_TM_COUNT)); final int memoryPerTaskManager = Integer.valueOf(envs.get(Client.ENV_TM_MEMORY)); final int coresPerTaskManager = Integer.valueOf(envs.get(Client.ENV_TM_CORES)); int heapLimit = Utils.calculateHeapSize(memoryPerTaskManager); if (currDir == null) { throw new RuntimeException("Current directory unknown"); }//from w w w . j a v a 2s. c om if (ownHostname == null) { throw new RuntimeException("Own hostname (" + Environment.NM_HOST + ") not set."); } LOG.info("Working directory " + currDir); // load Flink configuration. Utils.getFlinkConfiguration(currDir); final String localWebInterfaceDir = currDir + "/resources/" + ConfigConstants.DEFAULT_JOB_MANAGER_WEB_PATH_NAME; // Update yaml conf -> set jobManager address to this machine's address. FileInputStream fis = new FileInputStream(currDir + "/flink-conf.yaml"); BufferedReader br = new BufferedReader(new InputStreamReader(fis)); Writer output = new BufferedWriter(new FileWriter(currDir + "/flink-conf-modified.yaml")); String line; while ((line = br.readLine()) != null) { if (line.contains(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY)) { output.append(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY + ": " + ownHostname + "\n"); } else if (line.contains(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY)) { output.append(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY + ": " + "\n"); } else { output.append(line + "\n"); } } // just to make sure. output.append(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY + ": " + ownHostname + "\n"); output.append(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY + ": " + localWebInterfaceDir + "\n"); output.append(ConfigConstants.JOB_MANAGER_WEB_LOG_PATH_KEY + ": " + logDirs + "\n"); output.close(); br.close(); File newConf = new File(currDir + "/flink-conf-modified.yaml"); if (!newConf.exists()) { LOG.warn("modified yaml does not exist!"); } Utils.copyJarContents("resources/" + ConfigConstants.DEFAULT_JOB_MANAGER_WEB_PATH_NAME, ApplicationMaster.class.getProtectionDomain().getCodeSource().getLocation().getPath()); JobManager jm; { String pathToNepheleConfig = currDir + "/flink-conf-modified.yaml"; String[] args = { "-executionMode", "cluster", "-configDir", pathToNepheleConfig }; // start the job manager jm = JobManager.initialize(args); // Start info server for jobmanager jm.startInfoServer(); } AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf); rmClient.start(); NMClient nmClient = NMClient.createNMClient(); nmClient.init(conf); nmClient.start(); // Register with ResourceManager LOG.info("registering ApplicationMaster"); rmClient.registerApplicationMaster(applicationMasterHost, 0, "http://" + applicationMasterHost + ":" + GlobalConfiguration.getString(ConfigConstants.JOB_MANAGER_WEB_PORT_KEY, "undefined")); // Priority for worker containers - priorities are intra-application Priority priority = Records.newRecord(Priority.class); priority.setPriority(0); // Resource requirements for worker containers Resource capability = Records.newRecord(Resource.class); capability.setMemory(memoryPerTaskManager); capability.setVirtualCores(coresPerTaskManager); // Make container requests to ResourceManager for (int i = 0; i < taskManagerCount; ++i) { ContainerRequest containerAsk = new ContainerRequest(capability, null, null, priority); LOG.info("Requesting TaskManager container " + i); rmClient.addContainerRequest(containerAsk); } LocalResource flinkJar = Records.newRecord(LocalResource.class); LocalResource flinkConf = Records.newRecord(LocalResource.class); // register Flink Jar with remote HDFS final Path remoteJarPath = new Path(remoteFlinkJarPath); Utils.registerLocalResource(fs, remoteJarPath, flinkJar); // register conf with local fs. Path remoteConfPath = Utils.setupLocalResource(conf, fs, appId, new Path("file://" + currDir + "/flink-conf-modified.yaml"), flinkConf, new Path(clientHomeDir)); LOG.info("Prepared localresource for modified yaml: " + flinkConf); boolean hasLog4j = new File(currDir + "/log4j.properties").exists(); // prepare the files to ship LocalResource[] remoteShipRsc = null; String[] remoteShipPaths = shipListString.split(","); if (!shipListString.isEmpty()) { remoteShipRsc = new LocalResource[remoteShipPaths.length]; { // scope for i int i = 0; for (String remoteShipPathStr : remoteShipPaths) { if (remoteShipPathStr == null || remoteShipPathStr.isEmpty()) { continue; } remoteShipRsc[i] = Records.newRecord(LocalResource.class); Path remoteShipPath = new Path(remoteShipPathStr); Utils.registerLocalResource(fs, remoteShipPath, remoteShipRsc[i]); i++; } } } // respect custom JVM options in the YAML file final String javaOpts = GlobalConfiguration.getString(ConfigConstants.FLINK_JVM_OPTIONS, ""); // Obtain allocated containers and launch int allocatedContainers = 0; int completedContainers = 0; while (allocatedContainers < taskManagerCount) { AllocateResponse response = rmClient.allocate(0); for (Container container : response.getAllocatedContainers()) { LOG.info("Got new Container for TM " + container.getId() + " on host " + container.getNodeId().getHost()); ++allocatedContainers; // Launch container by create ContainerLaunchContext ContainerLaunchContext ctx = Records.newRecord(ContainerLaunchContext.class); String tmCommand = "$JAVA_HOME/bin/java -Xmx" + heapLimit + "m " + javaOpts; if (hasLog4j) { tmCommand += " -Dlog.file=\"" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-log4j.log\" -Dlog4j.configuration=file:log4j.properties"; } tmCommand += " org.apache.flink.yarn.YarnTaskManagerRunner -configDir . " + " 1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-stdout.log" + " 2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-stderr.log"; ctx.setCommands(Collections.singletonList(tmCommand)); LOG.info("Starting TM with command=" + tmCommand); // copy resources to the TaskManagers. Map<String, LocalResource> localResources = new HashMap<String, LocalResource>(2); localResources.put("flink.jar", flinkJar); localResources.put("flink-conf.yaml", flinkConf); // add ship resources if (!shipListString.isEmpty()) { Preconditions.checkNotNull(remoteShipRsc); for (int i = 0; i < remoteShipPaths.length; i++) { localResources.put(new Path(remoteShipPaths[i]).getName(), remoteShipRsc[i]); } } ctx.setLocalResources(localResources); // Setup CLASSPATH for Container (=TaskTracker) Map<String, String> containerEnv = new HashMap<String, String>(); Utils.setupEnv(conf, containerEnv); //add flink.jar to class path. containerEnv.put(Client.ENV_CLIENT_USERNAME, yarnClientUsername); ctx.setEnvironment(containerEnv); UserGroupInformation user = UserGroupInformation.getCurrentUser(); try { Credentials credentials = user.getCredentials(); DataOutputBuffer dob = new DataOutputBuffer(); credentials.writeTokenStorageToStream(dob); ByteBuffer securityTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength()); ctx.setTokens(securityTokens); } catch (IOException e) { LOG.warn("Getting current user info failed when trying to launch the container" + e.getMessage()); } LOG.info("Launching container " + allocatedContainers); nmClient.startContainer(container, ctx); } for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; LOG.info("Completed container (while allocating) " + status.getContainerId() + ". Total Completed:" + completedContainers); LOG.info("Diagnostics " + status.getDiagnostics()); } Thread.sleep(100); } // Now wait for containers to complete while (completedContainers < taskManagerCount) { AllocateResponse response = rmClient.allocate(completedContainers / taskManagerCount); for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; LOG.info("Completed container " + status.getContainerId() + ". Total Completed:" + completedContainers); LOG.info("Diagnostics " + status.getDiagnostics()); } Thread.sleep(5000); } LOG.info("Shutting down JobManager"); jm.shutdown(); // Un-register with ResourceManager rmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "", ""); }
From source file:org.apache.flink.yarn.appMaster.ApplicationMaster.java
License:Apache License
public static void main(String[] args) throws Exception { // execute Application Master using the client's user final String yarnClientUsername = System.getenv(Client.ENV_CLIENT_USERNAME); LOG.info("YARN daemon runs as '" + UserGroupInformation.getCurrentUser().getShortUserName() + "' setting" + " user to execute Flink ApplicationMaster/JobManager to '" + yarnClientUsername + "'"); UserGroupInformation ugi = UserGroupInformation.createRemoteUser(yarnClientUsername); for (Token<? extends TokenIdentifier> toks : UserGroupInformation.getCurrentUser().getTokens()) { ugi.addToken(toks);// w w w . j a v a 2s .c om } ugi.doAs(new PrivilegedAction<Object>() { @Override public Object run() { AMRMClient<ContainerRequest> rmClient = null; ApplicationMaster am = null; try { Configuration conf = Utils.initializeYarnConfiguration(); rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf); rmClient.start(); // run the actual Application Master am = new ApplicationMaster(conf); am.generateConfigurationFile(); am.startJobManager(); am.setRMClient(rmClient); am.run(); } catch (Throwable e) { LOG.error("Error while running the application master", e); // the AM is not available. Report error through the unregister function. if (rmClient != null && am == null) { try { rmClient.unregisterApplicationMaster(FinalApplicationStatus.FAILED, "Flink YARN Application master" + " stopped unexpectedly with an exception.\n" + StringUtils.stringifyException(e), ""); } catch (Exception e1) { LOG.error("Unable to fail the application master", e1); } LOG.info("AM unregistered from RM"); return null; } if (rmClient == null) { LOG.error("Unable to unregister AM since the RM client is not available"); } if (am != null) { LOG.info("Writing error into internal message system"); am.setFailed(true); am.addMessage(new Message("The application master failed with an exception:\n" + StringUtils.stringifyException(e))); am.keepRPCAlive(); } } return null; } }); }
From source file:org.apache.oozie.action.hadoop.AMRMClientAsyncFactory.java
License:Apache License
public AMRMClientAsync<?> createAMRMClientAsync(int intervalMs, AMRMCallBackHandler callBackHandler) { AMRMClient<?> amRmClient = AMRMClient.createAMRMClient(); AMRMClientAsync<?> amRmClientAsync = AMRMClientAsync.createAMRMClientAsync(amRmClient, intervalMs, callBackHandler);/* w w w. jav a2 s . c om*/ return amRmClientAsync; }