List of usage examples for org.apache.hadoop.yarn.api.records NodeReport getCapability
@Public @Stable public abstract Resource getCapability();
Resource on the node. From source file:com.ibm.bi.dml.yarn.ropt.YarnClusterAnalyzer.java
License:Open Source License
/** * //from ww w .jav a 2s . c o m * @return * @throws YarnException * @throws IOException */ public static double getClusterUtilization() throws IOException { double util = 0; try { if (_client == null) _client = createYarnClient(); List<NodeReport> nodesReport = _client.getNodeReports(); double maxMem = 0; double currMem = 0; long maxCores = 0; long currCores = 0; for (NodeReport node : nodesReport) { Resource max = node.getCapability(); Resource used = node.getUsed(); maxMem += max.getMemory(); currMem += used.getMemory(); maxCores += max.getVirtualCores(); currCores += used.getVirtualCores(); } util = Math.max(Math.min(1, currMem / maxMem), //memory util Math.min(1, (double) currCores / maxCores)); //vcore util } catch (Exception ex) { throw new IOException(ex); } return util; }
From source file:com.ibm.bi.dml.yarn.ropt.YarnClusterAnalyzer.java
License:Open Source License
/** * Analyzes properties of Yarn cluster and Hadoop configurations. *///from w w w. j a va2 s. c o m public static void analyzeYarnCluster(YarnClient yarnClient, YarnConfiguration conf, boolean verbose) { try { List<NodeReport> nodesReport = yarnClient.getNodeReports(); if (verbose) System.out.println("There are " + nodesReport.size() + " nodes in the cluster"); if (nodesReport.isEmpty()) throw new YarnException("There are zero available nodes in the yarn cluster"); nodesMaxPhySorted = new ArrayList<Long>(nodesReport.size()); clusterTotalMem = 0; clusterTotalCores = 0; clusterTotalNodes = 0; minimumMRContainerPhyMB = -1; for (NodeReport node : nodesReport) { Resource resource = node.getCapability(); Resource used = node.getUsed(); if (used == null) used = Resource.newInstance(0, 0); int mb = resource.getMemory(); int cores = resource.getVirtualCores(); if (mb <= 0) throw new YarnException("A node has non-positive memory " + mb); int myMinMRPhyMB = mb / cores / CPU_HYPER_FACTOR; if (minimumMRContainerPhyMB < myMinMRPhyMB) minimumMRContainerPhyMB = myMinMRPhyMB; // minimumMRContainerPhyMB needs to be the largest among the mins clusterTotalMem += (long) mb * 1024 * 1024; nodesMaxPhySorted.add((long) mb * 1024 * 1024); clusterTotalCores += cores; clusterTotalNodes++; if (verbose) System.out.println("\t" + node.getNodeId() + " has " + mb + " MB (" + used.getMemory() + " MB used) memory and " + resource.getVirtualCores() + " (" + used.getVirtualCores() + " used) cores"); } Collections.sort(nodesMaxPhySorted, Collections.reverseOrder()); nodesMaxBudgetSorted = new ArrayList<Double>(nodesMaxPhySorted.size()); for (int i = 0; i < nodesMaxPhySorted.size(); i++) nodesMaxBudgetSorted.add(ResourceOptimizer.phyToBudget(nodesMaxPhySorted.get(i))); _remotePar = nodesReport.size(); if (_remotePar == 0) throw new YarnException("There are no available nodes in the yarn cluster"); // Now get the default cluster settings _remoteMRSortMem = (1024 * 1024) * conf.getLong("io.sort.mb", 100); //100MB //handle jvm max mem (map mem budget is relevant for map-side distcache and parfor) //(for robustness we probe both: child and map configuration parameters) String javaOpts1 = conf.get("mapred.child.java.opts"); //internally mapred/mapreduce synonym String javaOpts2 = conf.get("mapreduce.map.java.opts", null); //internally mapred/mapreduce synonym String javaOpts3 = conf.get("mapreduce.reduce.java.opts", null); //internally mapred/mapreduce synonym if (javaOpts2 != null) //specific value overrides generic _remoteJVMMaxMemMap = extractMaxMemoryOpt(javaOpts2); else _remoteJVMMaxMemMap = extractMaxMemoryOpt(javaOpts1); if (javaOpts3 != null) //specific value overrides generic _remoteJVMMaxMemReduce = extractMaxMemoryOpt(javaOpts3); else _remoteJVMMaxMemReduce = extractMaxMemoryOpt(javaOpts1); //HDFS blocksize String blocksize = conf.get(MRConfigurationNames.DFS_BLOCK_SIZE, "134217728"); _blocksize = Long.parseLong(blocksize); minimalPhyAllocate = (long) 1024 * 1024 * conf.getInt(YarnConfiguration.RM_SCHEDULER_MINIMUM_ALLOCATION_MB, YarnConfiguration.DEFAULT_RM_SCHEDULER_MINIMUM_ALLOCATION_MB); maximumPhyAllocate = (long) 1024 * 1024 * conf.getInt(YarnConfiguration.RM_SCHEDULER_MAXIMUM_ALLOCATION_MB, YarnConfiguration.DEFAULT_RM_SCHEDULER_MAXIMUM_ALLOCATION_MB); mrAMPhy = (long) conf.getInt("yarn.app.mapreduce.am.resource.mb", 1536) * 1024 * 1024; } catch (Exception e) { throw new RuntimeException("Unable to analyze yarn cluster ", e); } /* * This is for AppMaster to query available resource in the cluster during heartbeat * AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf); rmClient.start(); AllocateResponse response = rmClient.allocate(0); int nodeCount = response.getNumClusterNodes(); Resource resource = response.getAvailableResources(); List<NodeReport> nodeUpdate = response.getUpdatedNodes(); LOG.info("This is a " + nodeCount + " node cluster with totally " + resource.getMemory() + " memory and " + resource.getVirtualCores() + " cores"); LOG.info(nodereport.size() + " updatedNode reports received"); for (NodeReport node : nodeUpdate) { resource = node.getCapability(); LOG.info(node.getNodeId() + " updated with " + resource.getMemory() + " memory and " + resource.getVirtualCores() + " cores"); }*/ }
From source file:edu.cmu.graphchi.toolkits.collaborative_filtering.yarn.ApplicationMaster.java
License:Apache License
/** * Main run function for the application master * * @throws YarnException/*w w w .j a v a 2s . c o 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.Client.java
License:Apache License
private ClusterResourceDescription getCurrentFreeClusterResources(YarnClient yarnClient) throws YarnException, IOException { ClusterResourceDescription crd = new ClusterResourceDescription(); crd.totalFreeMemory = 0;/*from ww w . jav a2 s.com*/ crd.containerLimit = 0; List<NodeReport> nodes = yarnClient.getNodeReports(NodeState.RUNNING); for (NodeReport rep : nodes) { int free = rep.getCapability().getMemory() - (rep.getUsed() != null ? rep.getUsed().getMemory() : 0); crd.totalFreeMemory += free; if (free > crd.containerLimit) { crd.containerLimit = free; } } return crd; }
From source file:eu.stratosphere.yarn.Client.java
License:Apache License
private void showClusterMetrics(YarnClient yarnClient) throws YarnException, IOException { YarnClusterMetrics metrics = yarnClient.getYarnClusterMetrics(); System.out.println("NodeManagers in the Cluster " + metrics.getNumNodeManagers()); List<NodeReport> nodes = yarnClient.getNodeReports(NodeState.RUNNING); final String format = "|%-16s |%-16s %n"; System.out.printf("|Property |Value %n"); System.out.println("+---------------------------------------+"); int totalMemory = 0; int totalCores = 0; for (NodeReport rep : nodes) { final Resource res = rep.getCapability(); totalMemory += res.getMemory();//from ww w .j a v a2 s . c o m totalCores += res.getVirtualCores(); System.out.format(format, "NodeID", rep.getNodeId()); System.out.format(format, "Memory", res.getMemory() + " MB"); System.out.format(format, "vCores", res.getVirtualCores()); System.out.format(format, "HealthReport", rep.getHealthReport()); System.out.format(format, "Containers", rep.getNumContainers()); System.out.println("+---------------------------------------+"); } System.out.println("Summary: totalMemory " + totalMemory + " totalCores " + totalCores); List<QueueInfo> qInfo = yarnClient.getAllQueues(); for (QueueInfo q : qInfo) { System.out.println("Queue: " + q.getQueueName() + ", Current Capacity: " + q.getCurrentCapacity() + " Max Capacity: " + q.getMaximumCapacity() + " Applications: " + q.getApplications().size()); } yarnClient.stop(); System.exit(0); }
From source file:io.amient.yarn1.YarnClient.java
License:Open Source License
/** * This method should be called by the implementing application static main * method. It does all the work around creating a yarn application and * submitting the request to the yarn resource manager. The class given in * the appClass argument will be run inside the yarn-allocated master * container./*w w w . ja v a2 s. c o m*/ */ public static void submitApplicationMaster(Properties appConfig, Class<? extends YarnMaster> masterClass, String[] args, Boolean awaitCompletion) throws Exception { log.info("Yarn1 App Configuration:"); for (Object param : appConfig.keySet()) { log.info(param.toString() + " = " + appConfig.get(param).toString()); } String yarnConfigPath = appConfig.getProperty("yarn1.site", "/etc/hadoop"); String masterClassName = masterClass.getName(); appConfig.setProperty("yarn1.master.class", masterClassName); String applicationName = appConfig.getProperty("yarn1.application.name", masterClassName); log.info("--------------------------------------------------------------"); if (Boolean.valueOf(appConfig.getProperty("yarn1.local.mode", "false"))) { YarnMaster.run(appConfig, args); return; } int masterPriority = Integer.valueOf( appConfig.getProperty("yarn1.master.priority", String.valueOf(YarnMaster.DEFAULT_MASTER_PRIORITY))); int masterMemoryMb = Integer.valueOf(appConfig.getProperty("yarn1.master.memory.mb", String.valueOf(YarnMaster.DEFAULT_MASTER_MEMORY_MB))); int masterNumCores = Integer.valueOf( appConfig.getProperty("yarn1.master.num.cores", String.valueOf(YarnMaster.DEFAULT_MASTER_CORES))); String queue = appConfig.getProperty("yarn1.queue"); Configuration yarnConfig = new YarnConfiguration(); yarnConfig.addResource(new FileInputStream(yarnConfigPath + "/core-site.xml")); yarnConfig.addResource(new FileInputStream(yarnConfigPath + "/hdfs-site.xml")); yarnConfig.addResource(new FileInputStream(yarnConfigPath + "/yarn-site.xml")); for (Map.Entry<Object, Object> entry : appConfig.entrySet()) { yarnConfig.set(entry.getKey().toString(), entry.getValue().toString()); } final org.apache.hadoop.yarn.client.api.YarnClient yarnClient = org.apache.hadoop.yarn.client.api.YarnClient .createYarnClient(); yarnClient.init(yarnConfig); yarnClient.start(); for (NodeReport report : yarnClient.getNodeReports(NodeState.RUNNING)) { log.debug("Node report:" + report.getNodeId() + " @ " + report.getHttpAddress() + " | " + report.getCapability()); } log.info("Submitting application master class " + masterClassName); YarnClientApplication app = yarnClient.createApplication(); GetNewApplicationResponse appResponse = app.getNewApplicationResponse(); final ApplicationId appId = appResponse.getApplicationId(); if (appId == null) { System.exit(111); } else { appConfig.setProperty("am.timestamp", String.valueOf(appId.getClusterTimestamp())); appConfig.setProperty("am.id", String.valueOf(appId.getId())); } YarnClient.distributeResources(yarnConfig, appConfig, applicationName); String masterJvmArgs = appConfig.getProperty("yarn1.master.jvm.args", ""); YarnContainerContext masterContainer = new YarnContainerContext(yarnConfig, appConfig, masterJvmArgs, masterPriority, masterMemoryMb, masterNumCores, applicationName, YarnMaster.class, args); ApplicationSubmissionContext appContext = app.getApplicationSubmissionContext(); appContext.setApplicationName(masterClassName); appContext.setResource(masterContainer.capability); appContext.setPriority(masterContainer.priority); appContext.setQueue(queue); appContext.setApplicationType(appConfig.getProperty("yarn1.application.type", "YARN")); appContext.setAMContainerSpec(masterContainer.createContainerLaunchContext()); log.info("Master container spec: " + masterContainer.capability); yarnClient.submitApplication(appContext); ApplicationReport report = yarnClient.getApplicationReport(appId); log.info("Tracking URL: " + report.getTrackingUrl()); if (awaitCompletion) { Runtime.getRuntime().addShutdownHook(new Thread() { @Override public void run() { if (!yarnClient.isInState(Service.STATE.STOPPED)) { log.info("Killing yarn application in shutdown hook"); try { yarnClient.killApplication(appId); } catch (Throwable e) { log.error("Failed to kill yarn application - please check YARN Resource Manager", e); } } } }); float lastProgress = -0.0f; while (true) { try { Thread.sleep(10000); report = yarnClient.getApplicationReport(appId); if (lastProgress != report.getProgress()) { lastProgress = report.getProgress(); log.info(report.getApplicationId() + " " + (report.getProgress() * 100.00) + "% " + (System.currentTimeMillis() - report.getStartTime()) + "(ms) " + report.getDiagnostics()); } if (!report.getFinalApplicationStatus().equals(FinalApplicationStatus.UNDEFINED)) { log.info(report.getApplicationId() + " " + report.getFinalApplicationStatus()); log.info("Tracking url: " + report.getTrackingUrl()); log.info("Finish time: " + ((System.currentTimeMillis() - report.getStartTime()) / 1000) + "(s)"); break; } } catch (Throwable e) { log.error("Master Heart Beat Error - terminating", e); yarnClient.killApplication(appId); Thread.sleep(2000); } } yarnClient.stop(); if (!report.getFinalApplicationStatus().equals(FinalApplicationStatus.SUCCEEDED)) { System.exit(112); } } yarnClient.stop(); }
From source file:io.hops.hopsworks.common.jobs.flink.AbstractYarnClusterDescriptor.java
License:Apache License
private ClusterResourceDescription getCurrentFreeClusterResources(YarnClient yarnClient) throws YarnException, IOException { List<NodeReport> nodes = yarnClient.getNodeReports(NodeState.RUNNING); int totalFreeMemory = 0; int containerLimit = 0; int[] nodeManagersFree = new int[nodes.size()]; for (int i = 0; i < nodes.size(); i++) { NodeReport rep = nodes.get(i); int free = rep.getCapability().getMemory() - (rep.getUsed() != null ? rep.getUsed().getMemory() : 0); nodeManagersFree[i] = free;// w w w .j a v a2 s. c om totalFreeMemory += free; if (free > containerLimit) { containerLimit = free; } } return new ClusterResourceDescription(totalFreeMemory, containerLimit, nodeManagersFree); }
From source file:io.hops.hopsworks.common.jobs.flink.AbstractYarnClusterDescriptor.java
License:Apache License
@Override public String getClusterDescription() { try {/*from ww w.j a v a2 s .com*/ ByteArrayOutputStream baos = new ByteArrayOutputStream(); PrintStream ps = new PrintStream(baos); YarnClient yarnClient = getYarnClient(); YarnClusterMetrics metrics = yarnClient.getYarnClusterMetrics(); ps.append("NodeManagers in the ClusterClient " + 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(); 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(); } catch (Exception e) { throw new RuntimeException("Couldn't get cluster description", e); } }
From source file:org.apache.drill.yarn.appMaster.NodeInventory.java
License:Apache License
private void buildNodeMap() throws YarnFacadeException { List<NodeReport> nodes = yarn.getNodeReports(); for (NodeReport node : nodes) { String hostName = node.getNodeId().getHost(); nodeMap.put(hostName, node.getHttpAddress()); yarnNodes.put(hostName, node);/* w w w . j a va2 s . c o m*/ } if (LOG.isInfoEnabled()) { LOG.info("YARN Node report"); for (NodeReport node : nodes) { LOG.info("Node: " + node.getHttpAddress() + ", Rack: " + node.getRackName() + " has " + node.getCapability().getMemory() + " MB, " + node.getCapability().getVirtualCores() + " vcores, labels: " + node.getNodeLabels()); } } }
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 v a 2 s . com 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(); }