List of usage examples for org.apache.hadoop.yarn.api.records Resource getVirtualCores
@Public @Evolving public abstract int getVirtualCores();
From source file:org.apache.drill.yarn.appMaster.ClusterControllerImpl.java
License:Apache License
private void yarnReport() { RegisterApplicationMasterResponse response = yarn.getRegistrationResponse(); LOG.info("YARN queue: " + response.getQueue()); Resource resource = response.getMaximumResourceCapability(); LOG.info("YARN max resource: " + resource.getMemory() + " MB, " + resource.getVirtualCores() + " cores"); EnumSet<SchedulerResourceTypes> types = response.getSchedulerResourceTypes(); StringBuilder buf = new StringBuilder(); String sep = ""; for (SchedulerResourceTypes type : types) { buf.append(sep);/*from w w w .j a v a 2s. c o m*/ buf.append(type.toString()); sep = ", "; } LOG.info("YARN scheduler resource types: " + buf.toString()); }
From source file:org.apache.flink.yarn.Client.java
License:Apache License
public void run(String[] args) throws Exception { if (UserGroupInformation.isSecurityEnabled()) { throw new RuntimeException("Flink YARN client does not have security support right now." + "File a bug, we will fix it asap"); }/* ww w . ja va 2s . c o m*/ //Utils.logFilesInCurrentDirectory(LOG); // // Command Line Options // Options options = new Options(); options.addOption(VERBOSE); options.addOption(FLINK_CONF_DIR); options.addOption(FLINK_JAR); options.addOption(JM_MEMORY); options.addOption(TM_MEMORY); options.addOption(TM_CORES); options.addOption(CONTAINER); options.addOption(GEN_CONF); options.addOption(QUEUE); options.addOption(QUERY); options.addOption(SHIP_PATH); CommandLineParser parser = new PosixParser(); CommandLine cmd = null; try { cmd = parser.parse(options, args); } catch (MissingOptionException moe) { System.out.println(moe.getMessage()); printUsage(); System.exit(1); } if (System.getProperty("log4j.configuration") == null) { Logger root = Logger.getRootLogger(); root.removeAllAppenders(); PatternLayout layout = new PatternLayout("%d{HH:mm:ss,SSS} %-5p %-60c %x - %m%n"); ConsoleAppender appender = new ConsoleAppender(layout, "System.err"); root.addAppender(appender); if (cmd.hasOption(VERBOSE.getOpt())) { root.setLevel(Level.DEBUG); LOG.debug("CLASSPATH: " + System.getProperty("java.class.path")); } else { root.setLevel(Level.INFO); } } // Jar Path Path localJarPath; if (cmd.hasOption(FLINK_JAR.getOpt())) { String userPath = cmd.getOptionValue(FLINK_JAR.getOpt()); if (!userPath.startsWith("file://")) { userPath = "file://" + userPath; } localJarPath = new Path(userPath); } else { localJarPath = new Path( "file://" + Client.class.getProtectionDomain().getCodeSource().getLocation().getPath()); } if (cmd.hasOption(GEN_CONF.getOpt())) { LOG.info("Placing default configuration in current directory"); File outFile = generateDefaultConf(localJarPath); LOG.info("File written to " + outFile.getAbsolutePath()); System.exit(0); } // Conf Path Path confPath = null; String confDirPath = ""; if (cmd.hasOption(FLINK_CONF_DIR.getOpt())) { confDirPath = cmd.getOptionValue(FLINK_CONF_DIR.getOpt()) + "/"; File confFile = new File(confDirPath + CONFIG_FILE_NAME); if (!confFile.exists()) { LOG.fatal("Unable to locate configuration file in " + confFile); System.exit(1); } confPath = new Path(confFile.getAbsolutePath()); } else { System.out.println("No configuration file has been specified"); // no configuration path given. // -> see if there is one in the current directory File currDir = new File("."); File[] candidates = currDir.listFiles(new FilenameFilter() { @Override public boolean accept(final File dir, final String name) { return name != null && name.endsWith(".yaml"); } }); if (candidates == null || candidates.length == 0) { System.out.println( "No configuration file has been found in current directory.\n" + "Copying default."); File outFile = generateDefaultConf(localJarPath); confPath = new Path(outFile.toURI()); } else { if (candidates.length > 1) { System.out.println("Multiple .yaml configuration files were found in the current directory\n" + "Please specify one explicitly"); System.exit(1); } else if (candidates.length == 1) { confPath = new Path(candidates[0].toURI()); } } } List<File> shipFiles = new ArrayList<File>(); // path to directory to ship if (cmd.hasOption(SHIP_PATH.getOpt())) { String shipPath = cmd.getOptionValue(SHIP_PATH.getOpt()); File shipDir = new File(shipPath); if (shipDir.isDirectory()) { shipFiles = new ArrayList<File>(Arrays.asList(shipDir.listFiles(new FilenameFilter() { @Override public boolean accept(File dir, String name) { return !(name.equals(".") || name.equals("..")); } }))); } else { LOG.warn("Ship directory is not a directory!"); } } boolean hasLog4j = false; //check if there is a log4j file if (confDirPath.length() > 0) { File l4j = new File(confDirPath + "/log4j.properties"); if (l4j.exists()) { shipFiles.add(l4j); hasLog4j = true; } } // queue String queue = "default"; if (cmd.hasOption(QUEUE.getOpt())) { queue = cmd.getOptionValue(QUEUE.getOpt()); } // JobManager Memory int jmMemory = 512; if (cmd.hasOption(JM_MEMORY.getOpt())) { jmMemory = Integer.valueOf(cmd.getOptionValue(JM_MEMORY.getOpt())); } // Task Managers memory int tmMemory = 1024; if (cmd.hasOption(TM_MEMORY.getOpt())) { tmMemory = Integer.valueOf(cmd.getOptionValue(TM_MEMORY.getOpt())); } // Task Managers vcores int tmCores = 1; if (cmd.hasOption(TM_CORES.getOpt())) { tmCores = Integer.valueOf(cmd.getOptionValue(TM_CORES.getOpt())); } Utils.getFlinkConfiguration(confPath.toUri().getPath()); int jmPort = GlobalConfiguration.getInteger(ConfigConstants.JOB_MANAGER_IPC_PORT_KEY, 0); if (jmPort == 0) { LOG.warn("Unable to find job manager port in configuration!"); jmPort = ConfigConstants.DEFAULT_JOB_MANAGER_IPC_PORT; } conf = Utils.initializeYarnConfiguration(); // intialize HDFS LOG.info("Copy App Master jar from local filesystem and add to local environment"); // Copy the application master jar to the filesystem // Create a local resource to point to the destination jar path final FileSystem fs = FileSystem.get(conf); if (fs.getScheme().startsWith("file")) { LOG.warn("The file system scheme is '" + fs.getScheme() + "'. This indicates that the " + "specified Hadoop configuration path is wrong and the sytem is using the default Hadoop configuration values." + "The Flink YARN client needs to store its files in a distributed file system"); } // Create yarnClient final YarnClient yarnClient = YarnClient.createYarnClient(); yarnClient.init(conf); yarnClient.start(); // Query cluster for metrics if (cmd.hasOption(QUERY.getOpt())) { showClusterMetrics(yarnClient); } if (!cmd.hasOption(CONTAINER.getOpt())) { LOG.fatal("Missing required argument " + CONTAINER.getOpt()); printUsage(); yarnClient.stop(); System.exit(1); } // TM Count final int taskManagerCount = Integer.valueOf(cmd.getOptionValue(CONTAINER.getOpt())); System.out.println("Using values:"); System.out.println("\tContainer Count = " + taskManagerCount); System.out.println("\tJar Path = " + localJarPath.toUri().getPath()); System.out.println("\tConfiguration file = " + confPath.toUri().getPath()); System.out.println("\tJobManager memory = " + jmMemory); System.out.println("\tTaskManager memory = " + tmMemory); System.out.println("\tTaskManager cores = " + tmCores); // Create application via yarnClient YarnClientApplication app = yarnClient.createApplication(); GetNewApplicationResponse appResponse = app.getNewApplicationResponse(); Resource maxRes = appResponse.getMaximumResourceCapability(); if (tmMemory > maxRes.getMemory() || tmCores > maxRes.getVirtualCores()) { LOG.fatal("The cluster does not have the requested resources for the TaskManagers available!\n" + "Maximum Memory: " + maxRes.getMemory() + ", Maximum Cores: " + tmCores); yarnClient.stop(); System.exit(1); } if (jmMemory > maxRes.getMemory()) { LOG.fatal("The cluster does not have the requested resources for the JobManager available!\n" + "Maximum Memory: " + maxRes.getMemory()); yarnClient.stop(); System.exit(1); } int totalMemoryRequired = jmMemory + tmMemory * taskManagerCount; ClusterResourceDescription freeClusterMem = getCurrentFreeClusterResources(yarnClient); if (freeClusterMem.totalFreeMemory < totalMemoryRequired) { LOG.fatal("This YARN session requires " + totalMemoryRequired + "MB of memory in the cluster. " + "There are currently only " + freeClusterMem.totalFreeMemory + "MB available."); yarnClient.stop(); System.exit(1); } if (tmMemory > freeClusterMem.containerLimit) { LOG.fatal("The requested amount of memory for the TaskManagers (" + tmMemory + "MB) is more than " + "the largest possible YARN container: " + freeClusterMem.containerLimit); yarnClient.stop(); System.exit(1); } if (jmMemory > freeClusterMem.containerLimit) { LOG.fatal("The requested amount of memory for the JobManager (" + jmMemory + "MB) is more than " + "the largest possible YARN container: " + freeClusterMem.containerLimit); yarnClient.stop(); System.exit(1); } // respect custom JVM options in the YAML file final String javaOpts = GlobalConfiguration.getString(ConfigConstants.FLINK_JVM_OPTIONS, ""); // Set up the container launch context for the application master ContainerLaunchContext amContainer = Records.newRecord(ContainerLaunchContext.class); String amCommand = "$JAVA_HOME/bin/java" + " -Xmx" + Utils.calculateHeapSize(jmMemory) + "M " + javaOpts; if (hasLog4j) { amCommand += " -Dlog.file=\"" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/jobmanager-log4j.log\" -Dlog4j.configuration=file:log4j.properties"; } amCommand += " org.apache.flink.yarn.ApplicationMaster" + " " + " 1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/jobmanager-stdout.log" + " 2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/jobmanager-stderr.log"; amContainer.setCommands(Collections.singletonList(amCommand)); System.err.println("amCommand=" + amCommand); // Set-up ApplicationSubmissionContext for the application ApplicationSubmissionContext appContext = app.getApplicationSubmissionContext(); final ApplicationId appId = appContext.getApplicationId(); // Setup jar for ApplicationMaster LocalResource appMasterJar = Records.newRecord(LocalResource.class); LocalResource flinkConf = Records.newRecord(LocalResource.class); Path remotePathJar = Utils.setupLocalResource(conf, fs, appId.toString(), localJarPath, appMasterJar, fs.getHomeDirectory()); Path remotePathConf = Utils.setupLocalResource(conf, fs, appId.toString(), confPath, flinkConf, fs.getHomeDirectory()); Map<String, LocalResource> localResources = new HashMap<String, LocalResource>(2); localResources.put("flink.jar", appMasterJar); localResources.put("flink-conf.yaml", flinkConf); // setup security tokens (code from apache storm) final Path[] paths = new Path[3 + shipFiles.size()]; StringBuffer envShipFileList = new StringBuffer(); // upload ship files for (int i = 0; i < shipFiles.size(); i++) { File shipFile = shipFiles.get(i); LocalResource shipResources = Records.newRecord(LocalResource.class); Path shipLocalPath = new Path("file://" + shipFile.getAbsolutePath()); paths[3 + i] = Utils.setupLocalResource(conf, fs, appId.toString(), shipLocalPath, shipResources, fs.getHomeDirectory()); localResources.put(shipFile.getName(), shipResources); envShipFileList.append(paths[3 + i]); if (i + 1 < shipFiles.size()) { envShipFileList.append(','); } } paths[0] = remotePathJar; paths[1] = remotePathConf; paths[2] = new Path(fs.getHomeDirectory(), ".flink/" + appId.toString() + "/"); FsPermission permission = new FsPermission(FsAction.ALL, FsAction.ALL, FsAction.ALL); fs.setPermission(paths[2], permission); // set permission for path. Utils.setTokensFor(amContainer, paths, this.conf); amContainer.setLocalResources(localResources); fs.close(); // Setup CLASSPATH for ApplicationMaster Map<String, String> appMasterEnv = new HashMap<String, String>(); Utils.setupEnv(conf, appMasterEnv); // set configuration values appMasterEnv.put(Client.ENV_TM_COUNT, String.valueOf(taskManagerCount)); appMasterEnv.put(Client.ENV_TM_CORES, String.valueOf(tmCores)); appMasterEnv.put(Client.ENV_TM_MEMORY, String.valueOf(tmMemory)); appMasterEnv.put(Client.FLINK_JAR_PATH, remotePathJar.toString()); appMasterEnv.put(Client.ENV_APP_ID, appId.toString()); appMasterEnv.put(Client.ENV_CLIENT_HOME_DIR, fs.getHomeDirectory().toString()); appMasterEnv.put(Client.ENV_CLIENT_SHIP_FILES, envShipFileList.toString()); appMasterEnv.put(Client.ENV_CLIENT_USERNAME, UserGroupInformation.getCurrentUser().getShortUserName()); amContainer.setEnvironment(appMasterEnv); // Set up resource type requirements for ApplicationMaster Resource capability = Records.newRecord(Resource.class); capability.setMemory(jmMemory); capability.setVirtualCores(1); appContext.setApplicationName("Flink"); // application name appContext.setAMContainerSpec(amContainer); appContext.setResource(capability); appContext.setQueue(queue); // file that we write into the conf/ dir containing the jobManager address. final File addrFile = new File(confDirPath + CliFrontend.JOBMANAGER_ADDRESS_FILE); Runtime.getRuntime().addShutdownHook(new Thread() { @Override public void run() { try { LOG.info("Killing the Flink-YARN application."); yarnClient.killApplication(appId); LOG.info("Deleting files in " + paths[2]); FileSystem shutFS = FileSystem.get(conf); shutFS.delete(paths[2], true); // delete conf and jar file. shutFS.close(); } catch (Exception e) { LOG.warn("Exception while killing the YARN application", e); } try { addrFile.delete(); } catch (Exception e) { LOG.warn("Exception while deleting the jobmanager address file", e); } LOG.info("YARN Client is shutting down"); yarnClient.stop(); } }); LOG.info("Submitting application master " + appId); yarnClient.submitApplication(appContext); ApplicationReport appReport = yarnClient.getApplicationReport(appId); YarnApplicationState appState = appReport.getYarnApplicationState(); boolean told = false; char[] el = { '/', '|', '\\', '-' }; int i = 0; while (appState != YarnApplicationState.FINISHED && appState != YarnApplicationState.KILLED && appState != YarnApplicationState.FAILED) { if (!told && appState == YarnApplicationState.RUNNING) { System.err.println("Flink JobManager is now running on " + appReport.getHost() + ":" + jmPort); System.err.println("JobManager Web Interface: " + appReport.getTrackingUrl()); // write jobmanager connect information PrintWriter out = new PrintWriter(addrFile); out.println(appReport.getHost() + ":" + jmPort); out.close(); addrFile.setReadable(true, false); // readable for all. told = true; } if (!told) { System.err.print(el[i++] + "\r"); if (i == el.length) { i = 0; } Thread.sleep(500); // wait for the application to switch to RUNNING } else { Thread.sleep(5000); } appReport = yarnClient.getApplicationReport(appId); appState = appReport.getYarnApplicationState(); } LOG.info("Application " + appId + " finished with" + " state " + appState + " at " + appReport.getFinishTime()); if (appState == YarnApplicationState.FAILED || appState == YarnApplicationState.KILLED) { LOG.warn("Application failed. Diagnostics " + appReport.getDiagnostics()); } }
From source file:org.apache.flink.yarn.FlinkYarnClient.java
License:Apache License
public String getClusterDescription() throws Exception { ByteArrayOutputStream baos = new ByteArrayOutputStream(); PrintStream ps = new PrintStream(baos); YarnClusterMetrics metrics = yarnClient.getYarnClusterMetrics(); ps.append("NodeManagers in the Cluster " + metrics.getNumNodeManagers()); List<NodeReport> nodes = yarnClient.getNodeReports(NodeState.RUNNING); final String format = "|%-16s |%-16s %n"; ps.printf("|Property |Value %n"); ps.println("+---------------------------------------+"); int totalMemory = 0; int totalCores = 0; for (NodeReport rep : nodes) { final Resource res = rep.getCapability(); totalMemory += res.getMemory();/* w w w .j a va 2 s.c o m*/ totalCores += res.getVirtualCores(); ps.format(format, "NodeID", rep.getNodeId()); ps.format(format, "Memory", res.getMemory() + " MB"); ps.format(format, "vCores", res.getVirtualCores()); ps.format(format, "HealthReport", rep.getHealthReport()); ps.format(format, "Containers", rep.getNumContainers()); ps.println("+---------------------------------------+"); } ps.println("Summary: totalMemory " + totalMemory + " totalCores " + totalCores); List<QueueInfo> qInfo = yarnClient.getAllQueues(); for (QueueInfo q : qInfo) { ps.println("Queue: " + q.getQueueName() + ", Current Capacity: " + q.getCurrentCapacity() + " Max Capacity: " + q.getMaximumCapacity() + " Applications: " + q.getApplications().size()); } yarnClient.stop(); return baos.toString(); }
From source file:org.apache.flink.yarn.FlinkYarnClientBase.java
License:Apache License
@Override public String getClusterDescription() throws Exception { ByteArrayOutputStream baos = new ByteArrayOutputStream(); PrintStream ps = new PrintStream(baos); YarnClusterMetrics metrics = yarnClient.getYarnClusterMetrics(); ps.append("NodeManagers in the Cluster " + metrics.getNumNodeManagers()); List<NodeReport> nodes = yarnClient.getNodeReports(NodeState.RUNNING); final String format = "|%-16s |%-16s %n"; ps.printf("|Property |Value %n"); ps.println("+---------------------------------------+"); int totalMemory = 0; int totalCores = 0; for (NodeReport rep : nodes) { final Resource res = rep.getCapability(); totalMemory += res.getMemory();/*from w w w . j av a 2 s .c om*/ totalCores += res.getVirtualCores(); ps.format(format, "NodeID", rep.getNodeId()); ps.format(format, "Memory", res.getMemory() + " MB"); ps.format(format, "vCores", res.getVirtualCores()); ps.format(format, "HealthReport", rep.getHealthReport()); ps.format(format, "Containers", rep.getNumContainers()); ps.println("+---------------------------------------+"); } ps.println("Summary: totalMemory " + totalMemory + " totalCores " + totalCores); List<QueueInfo> qInfo = yarnClient.getAllQueues(); for (QueueInfo q : qInfo) { ps.println("Queue: " + q.getQueueName() + ", Current Capacity: " + q.getCurrentCapacity() + " Max Capacity: " + q.getMaximumCapacity() + " Applications: " + q.getApplications().size()); } yarnClient.stop(); return baos.toString(); }
From source file:org.apache.hoya.core.launch.AbstractLauncher.java
License:Apache License
/** * Extract any resource requirements from this component's settings. * All fields that are set will override the existing values -if * unset that resource field will be left unchanged. * * Important: the configuration must already be fully resolved * in order to pick up global options./*from w w w . j a v a2 s . c o m*/ * @param resource resource to configure * @param map map of options */ public void extractResourceRequirements(Resource resource, Map<String, String> map) { if (map != null) { MapOperations options = new MapOperations("", map); resource.setMemory(options.getOptionInt(ResourceKeys.YARN_MEMORY, resource.getMemory())); resource.setVirtualCores(options.getOptionInt(ResourceKeys.YARN_CORES, resource.getVirtualCores())); } }
From source file:org.apache.hoya.providers.hoyaam.HoyaAMClientProvider.java
License:Apache License
/** * Update the AM resource with any local needs * @param capability capability to update *//*from w ww.j av a2 s .c o m*/ public void prepareAMResourceRequirements(MapOperations hoyaAM, Resource capability) { capability.setMemory(hoyaAM.getOptionInt(ResourceKeys.YARN_MEMORY, capability.getMemory())); capability.setVirtualCores(hoyaAM.getOptionInt(ResourceKeys.YARN_CORES, capability.getVirtualCores())); }
From source file:org.apache.hoya.yarn.appmaster.HoyaAppMaster.java
License:Apache License
/** * Create and run the cluster./*from w w w . j ava 2 s . c om*/ * @return exit code * @throws Throwable on a failure */ private int createAndRunCluster(String clustername) throws Throwable { HoyaVersionInfo.loadAndPrintVersionInfo(log); //load the cluster description from the cd argument String hoyaClusterDir = serviceArgs.getHoyaClusterURI(); URI hoyaClusterURI = new URI(hoyaClusterDir); Path clusterDirPath = new Path(hoyaClusterURI); HoyaFileSystem fs = getClusterFS(); // build up information about the running application -this // will be passed down to the cluster status MapOperations appInformation = new MapOperations(); AggregateConf instanceDefinition = InstanceIO.loadInstanceDefinitionUnresolved(fs, clusterDirPath); log.info("Deploying cluster {}:", instanceDefinition); //REVISIT: why is this done? appState.updateInstanceDefinition(instanceDefinition); File confDir = getLocalConfDir(); if (!confDir.exists() || !confDir.isDirectory()) { log.error("Bad conf dir {}", confDir); File parentFile = confDir.getParentFile(); log.error("Parent dir {}:\n{}", parentFile, HoyaUtils.listDir(parentFile)); throw new BadCommandArgumentsException("Configuration directory %s doesn't exist", confDir); } Configuration serviceConf = getConfig(); // Try to get the proper filtering of static resources through the yarn proxy working serviceConf.set("hadoop.http.filter.initializers", "org.apache.hadoop.yarn.server.webproxy.amfilter.AmFilterInitializer"); conf = new YarnConfiguration(serviceConf); //get our provider MapOperations globalOptions = instanceDefinition.getInternalOperations().getGlobalOptions(); String providerType = globalOptions.getMandatoryOption(OptionKeys.INTERNAL_PROVIDER_NAME); log.info("Cluster provider type is {}", providerType); HoyaProviderFactory factory = HoyaProviderFactory.createHoyaProviderFactory(providerType); providerService = factory.createServerProvider(); // init the provider BUT DO NOT START IT YET providerService.init(getConfig()); addService(providerService); InetSocketAddress address = HoyaUtils.getRmSchedulerAddress(conf); log.info("RM is at {}", address); yarnRPC = YarnRPC.create(conf); /* * Extract the container ID. This is then * turned into an (incompete) container */ appMasterContainerID = ConverterUtils.toContainerId( HoyaUtils.mandatoryEnvVariable(ApplicationConstants.Environment.CONTAINER_ID.name())); appAttemptID = appMasterContainerID.getApplicationAttemptId(); ApplicationId appid = appAttemptID.getApplicationId(); log.info("Hoya AM for ID {}", appid.getId()); appInformation.put(StatusKeys.INFO_AM_CONTAINER_ID, appMasterContainerID.toString()); appInformation.put(StatusKeys.INFO_AM_APP_ID, appid.toString()); appInformation.put(StatusKeys.INFO_AM_ATTEMPT_ID, appAttemptID.toString()); UserGroupInformation currentUser = UserGroupInformation.getCurrentUser(); Credentials credentials = currentUser.getCredentials(); DataOutputBuffer dob = new DataOutputBuffer(); credentials.writeTokenStorageToStream(dob); dob.close(); // 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(); log.info("Token {}", token.getKind()); if (token.getKind().equals(AMRMTokenIdentifier.KIND_NAME)) { iter.remove(); } } allTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength()); // set up secret manager secretManager = new ClientToAMTokenSecretManager(appAttemptID, null); // if not a secure cluster, extract the username -it will be // propagated to workers if (!UserGroupInformation.isSecurityEnabled()) { hoyaUsername = System.getenv(HADOOP_USER_NAME); log.info(HADOOP_USER_NAME + "='{}'", hoyaUsername); } Map<String, String> envVars; /** * It is critical this section is synchronized, to stop async AM events * arriving while registering a restarting AM. */ synchronized (appState) { int heartbeatInterval = HEARTBEAT_INTERVAL; //add the RM client -this brings the callbacks in asyncRMClient = AMRMClientAsync.createAMRMClientAsync(heartbeatInterval, this); addService(asyncRMClient); //wrap it for the app state model rmOperationHandler = new AsyncRMOperationHandler(asyncRMClient); //now bring it up runChildService(asyncRMClient); //nmclient relays callbacks back to this class nmClientAsync = new NMClientAsyncImpl("nmclient", this); runChildService(nmClientAsync); //bring up the Hoya RPC service startHoyaRPCServer(); InetSocketAddress rpcServiceAddr = rpcService.getConnectAddress(); appMasterHostname = rpcServiceAddr.getHostName(); appMasterRpcPort = rpcServiceAddr.getPort(); appMasterTrackingUrl = null; log.info("AM Server is listening at {}:{}", appMasterHostname, appMasterRpcPort); appInformation.put(StatusKeys.INFO_AM_HOSTNAME, appMasterHostname); appInformation.set(StatusKeys.INFO_AM_RPC_PORT, appMasterRpcPort); //build the role map List<ProviderRole> providerRoles = new ArrayList<ProviderRole>(providerService.getRoles()); providerRoles.addAll(HoyaAMClientProvider.ROLES); // Start up the WebApp and track the URL for it webApp = new HoyaAMWebApp(); WebApps.$for("hoyaam", WebAppApi.class, new WebAppApiImpl(this, appState, providerService), "ws") .with(serviceConf).start(webApp); appMasterTrackingUrl = "http://" + appMasterHostname + ":" + webApp.port(); WebAppService<HoyaAMWebApp> webAppService = new WebAppService<HoyaAMWebApp>("hoya", webApp); webAppService.init(conf); webAppService.start(); addService(webAppService); appInformation.put(StatusKeys.INFO_AM_WEB_URL, appMasterTrackingUrl + "/"); appInformation.set(StatusKeys.INFO_AM_WEB_PORT, webApp.port()); // Register self with ResourceManager // This will start heartbeating to the RM // address = HoyaUtils.getRmSchedulerAddress(asyncRMClient.getConfig()); log.info("Connecting to RM at {},address tracking URL={}", appMasterRpcPort, appMasterTrackingUrl); RegisterApplicationMasterResponse response = asyncRMClient.registerApplicationMaster(appMasterHostname, appMasterRpcPort, appMasterTrackingUrl); Resource maxResources = response.getMaximumResourceCapability(); containerMaxMemory = maxResources.getMemory(); containerMaxCores = maxResources.getVirtualCores(); appState.setContainerLimits(maxResources.getMemory(), maxResources.getVirtualCores()); // set the RM-defined maximum cluster values appInformation.put(ResourceKeys.YARN_CORES, Integer.toString(containerMaxCores)); appInformation.put(ResourceKeys.YARN_MEMORY, Integer.toString(containerMaxMemory)); boolean securityEnabled = UserGroupInformation.isSecurityEnabled(); if (securityEnabled) { secretManager.setMasterKey(response.getClientToAMTokenMasterKey().array()); applicationACLs = response.getApplicationACLs(); //tell the server what the ACLs are rpcService.getServer().refreshServiceAcl(conf, new HoyaAMPolicyProvider()); } // extract container list List<Container> liveContainers = AMRestartSupport.retrieveContainersFromPreviousAttempt(response); String amRestartSupported = Boolean.toString(liveContainers != null); appInformation.put(StatusKeys.INFO_AM_RESTART_SUPPORTED, amRestartSupported); //now validate the installation Configuration providerConf = providerService.loadProviderConfigurationInformation(confDir); providerService.validateApplicationConfiguration(instanceDefinition, confDir, securityEnabled); //determine the location for the role history data Path historyDir = new Path(clusterDirPath, HISTORY_DIR_NAME); //build the instance appState.buildInstance(instanceDefinition, providerConf, providerRoles, fs.getFileSystem(), historyDir, liveContainers, appInformation); // add the AM to the list of nodes in the cluster appState.buildAppMasterNode(appMasterContainerID, appMasterHostname, webApp.port(), appMasterHostname + ":" + webApp.port()); // build up environment variables that the AM wants set in every container // irrespective of provider and role. envVars = new HashMap<String, String>(); if (hoyaUsername != null) { envVars.put(HADOOP_USER_NAME, hoyaUsername); } } String rolesTmpSubdir = appMasterContainerID.toString() + "/roles"; String amTmpDir = globalOptions.getMandatoryOption(OptionKeys.INTERNAL_AM_TMP_DIR); Path tmpDirPath = new Path(amTmpDir); Path launcherTmpDirPath = new Path(tmpDirPath, rolesTmpSubdir); fs.getFileSystem().mkdirs(launcherTmpDirPath); //launcher service launchService = new RoleLaunchService(this, providerService, fs, new Path(getGeneratedConfDir()), envVars, launcherTmpDirPath); runChildService(launchService); appState.noteAMLaunched(); //Give the provider restricted access to the state providerService.bind(appState); // launch the provider; this is expected to trigger a callback that // brings up the service launchProviderService(instanceDefinition, confDir); try { //now block waiting to be told to exit the process waitForAMCompletionSignal(); //shutdown time } finally { finish(); } return amExitCode; }
From source file:org.apache.ignite.yarn.ApplicationMaster.java
License:Apache License
/** * @return {@code True} if cluster contains available resources. *///www.j a v a 2 s.co m private boolean checkAvailableResource() { Resource availableRes = rmClient.getAvailableResources(); return availableRes == null || availableRes.getMemory() >= props.totalMemoryPerNode() && availableRes.getVirtualCores() >= props.cpusPerNode(); }
From source file:org.apache.myriad.scheduler.fgs.NMHeartBeatHandler.java
License:Apache License
@Override public void beforeRMNodeEventHandled(RMNodeEvent event, RMContext context) { switch (event.getType()) { case STARTED: // Since the RMNode was just started, it should not have a non-zero capacity RMNode rmNode = context.getRMNodes().get(event.getNodeId()); if (isNonZeroCapacityNode(rmNode)) { Resource totalCapability = rmNode.getTotalCapability(); logger.warn(/*from ww w. j av a 2s . c o m*/ "FineGrainedScaling feature got invoked for a NM with non-zero capacity. Host: {}, Mem: {}, CPU: {}. Setting the " + "NM's capacity to (0G,0CPU)", rmNode.getHostName(), totalCapability.getMemory(), totalCapability.getVirtualCores()); totalCapability.setMemory(0); totalCapability.setVirtualCores(0); } break; case STATUS_UPDATE: handleStatusUpdate(event, context); break; default: break; } }
From source file:org.apache.myriad.scheduler.fgs.NMHeartBeatHandler.java
License:Apache License
@VisibleForTesting protected boolean isNonZeroCapacityNode(RMNode node) { Resource resource = node.getTotalCapability(); return (resource.getMemory() != 0 || resource.getVirtualCores() != 0); }