List of usage examples for org.apache.hadoop.yarn.api.records ContainerStatus getState
@Public @Stable public abstract ContainerState getState();
ContainerState
of the container. From source file:ApplicationMaster.java
License:Apache License
private static void publishContainerEndEvent(TimelineClient timelineClient, ContainerStatus container) throws IOException, YarnException { TimelineEntity entity = new TimelineEntity(); entity.setEntityId(container.getContainerId().toString()); entity.setEntityType(DSEntity.DS_CONTAINER.toString()); entity.addPrimaryFilter("user", UserGroupInformation.getCurrentUser().getShortUserName()); TimelineEvent event = new TimelineEvent(); event.setTimestamp(System.currentTimeMillis()); event.setEventType(DSEvent.DS_CONTAINER_END.toString()); event.addEventInfo("State", container.getState().name()); event.addEventInfo("Exit Status", container.getExitStatus()); entity.addEvent(event);/*www . ja v a 2 s.co m*/ timelineClient.putEntities(entity); }
From source file:com.datatorrent.stram.StreamingAppMasterService.java
License:Apache License
/** * Main run function for the application master * * @throws YarnException//from w w w. j av a 2 s . c o m */ @SuppressWarnings("SleepWhileInLoop") private void execute() throws YarnException, IOException { LOG.info("Starting ApplicationMaster"); final Credentials credentials = UserGroupInformation.getCurrentUser().getCredentials(); LOG.info("number of tokens: {}", credentials.getAllTokens().size()); Iterator<Token<?>> iter = credentials.getAllTokens().iterator(); while (iter.hasNext()) { Token<?> token = iter.next(); LOG.debug("token: {}", token); } final Configuration conf = getConfig(); long tokenLifeTime = (long) (dag.getValue(LogicalPlan.TOKEN_REFRESH_ANTICIPATORY_FACTOR) * Math .min(dag.getValue(LogicalPlan.HDFS_TOKEN_LIFE_TIME), dag.getValue(LogicalPlan.RM_TOKEN_LIFE_TIME))); long expiryTime = System.currentTimeMillis() + tokenLifeTime; LOG.debug(" expiry token time {}", tokenLifeTime); String hdfsKeyTabFile = dag.getValue(LogicalPlan.KEY_TAB_FILE); // Register self with ResourceManager RegisterApplicationMasterResponse response = amRmClient.registerApplicationMaster(appMasterHostname, 0, appMasterTrackingUrl); // Dump out information about cluster capability as seen by the resource manager int maxMem = response.getMaximumResourceCapability().getMemory(); int maxVcores = response.getMaximumResourceCapability().getVirtualCores(); LOG.info("Max mem {}m and vcores {} capabililty of resources in this cluster ", maxMem, maxVcores); // for locality relaxation fall back Map<StreamingContainerAgent.ContainerStartRequest, MutablePair<Integer, ContainerRequest>> requestedResources = Maps .newHashMap(); // Setup heartbeat emitter // TODO poll RM every now and then with an empty request to let RM know that we are alive // The heartbeat interval after which an AM is timed out by the RM is defined by a config setting: // RM_AM_EXPIRY_INTERVAL_MS with default defined by DEFAULT_RM_AM_EXPIRY_INTERVAL_MS // The allocate calls to the RM count as heartbeat so, for now, this additional heartbeat emitter // is not required. int loopCounter = -1; List<ContainerId> releasedContainers = new ArrayList<ContainerId>(); int numTotalContainers = 0; // keep track of already requested containers to not request them again while waiting for allocation int numRequestedContainers = 0; int numReleasedContainers = 0; int nextRequestPriority = 0; ResourceRequestHandler resourceRequestor = new ResourceRequestHandler(); YarnClient clientRMService = YarnClient.createYarnClient(); try { // YARN-435 // we need getClusterNodes to populate the initial node list, // subsequent updates come through the heartbeat response clientRMService.init(conf); clientRMService.start(); ApplicationReport ar = StramClientUtils.getStartedAppInstanceByName(clientRMService, dag.getAttributes().get(DAG.APPLICATION_NAME), UserGroupInformation.getLoginUser().getUserName(), dag.getAttributes().get(DAG.APPLICATION_ID)); if (ar != null) { appDone = true; dnmgr.shutdownDiagnosticsMessage = String.format( "Application master failed due to application %s with duplicate application name \"%s\" by the same user \"%s\" is already started.", ar.getApplicationId().toString(), ar.getName(), ar.getUser()); LOG.info("Forced shutdown due to {}", dnmgr.shutdownDiagnosticsMessage); finishApplication(FinalApplicationStatus.FAILED, numTotalContainers); return; } resourceRequestor.updateNodeReports(clientRMService.getNodeReports()); } catch (Exception e) { throw new RuntimeException("Failed to retrieve cluster nodes report.", e); } finally { clientRMService.stop(); } // check for previously allocated containers // as of 2.2, containers won't survive AM restart, but this will change in the future - YARN-1490 checkContainerStatus(); FinalApplicationStatus finalStatus = FinalApplicationStatus.SUCCEEDED; final InetSocketAddress rmAddress = conf.getSocketAddr(YarnConfiguration.RM_ADDRESS, YarnConfiguration.DEFAULT_RM_ADDRESS, YarnConfiguration.DEFAULT_RM_PORT); while (!appDone) { loopCounter++; if (UserGroupInformation.isSecurityEnabled() && System.currentTimeMillis() >= expiryTime && hdfsKeyTabFile != null) { String applicationId = appAttemptID.getApplicationId().toString(); expiryTime = StramUserLogin.refreshTokens(tokenLifeTime, "." + File.separator + "tmp", applicationId, conf, hdfsKeyTabFile, credentials, rmAddress, true); } Runnable r; while ((r = this.pendingTasks.poll()) != null) { r.run(); } // log current state /* * LOG.info("Current application state: loop=" + loopCounter + ", appDone=" + appDone + ", total=" + * numTotalContainers + ", requested=" + numRequestedContainers + ", completed=" + numCompletedContainers + * ", failed=" + numFailedContainers + ", currentAllocated=" + this.allAllocatedContainers.size()); */ // Sleep before each loop when asking RM for containers // to avoid flooding RM with spurious requests when it // need not have any available containers try { sleep(1000); } catch (InterruptedException e) { LOG.info("Sleep interrupted " + e.getMessage()); } // Setup request to be sent to RM to allocate containers List<ContainerRequest> containerRequests = new ArrayList<ContainerRequest>(); List<ContainerRequest> removedContainerRequests = new ArrayList<ContainerRequest>(); // request containers for pending deploy requests if (!dnmgr.containerStartRequests.isEmpty()) { StreamingContainerAgent.ContainerStartRequest csr; while ((csr = dnmgr.containerStartRequests.poll()) != null) { if (csr.container.getRequiredMemoryMB() > maxMem) { LOG.warn("Container memory {}m above max threshold of cluster. Using max value {}m.", csr.container.getRequiredMemoryMB(), maxMem); csr.container.setRequiredMemoryMB(maxMem); } if (csr.container.getRequiredVCores() > maxVcores) { LOG.warn("Container vcores {} above max threshold of cluster. Using max value {}.", csr.container.getRequiredVCores(), maxVcores); csr.container.setRequiredVCores(maxVcores); } csr.container.setResourceRequestPriority(nextRequestPriority++); ContainerRequest cr = resourceRequestor.createContainerRequest(csr, true); MutablePair<Integer, ContainerRequest> pair = new MutablePair<Integer, ContainerRequest>( loopCounter, cr); requestedResources.put(csr, pair); containerRequests.add(cr); } } if (!requestedResources.isEmpty()) { //resourceRequestor.clearNodeMapping(); for (Map.Entry<StreamingContainerAgent.ContainerStartRequest, MutablePair<Integer, ContainerRequest>> entry : requestedResources .entrySet()) { if ((loopCounter - entry.getValue().getKey()) > NUMBER_MISSED_HEARTBEATS) { StreamingContainerAgent.ContainerStartRequest csr = entry.getKey(); removedContainerRequests.add(entry.getValue().getRight()); ContainerRequest cr = resourceRequestor.createContainerRequest(csr, false); entry.getValue().setLeft(loopCounter); entry.getValue().setRight(cr); containerRequests.add(cr); } } } numTotalContainers += containerRequests.size(); numRequestedContainers += containerRequests.size(); AllocateResponse amResp = sendContainerAskToRM(containerRequests, removedContainerRequests, releasedContainers); if (amResp.getAMCommand() != null) { LOG.info(" statement executed:{}", amResp.getAMCommand()); switch (amResp.getAMCommand()) { case AM_RESYNC: case AM_SHUTDOWN: throw new YarnRuntimeException("Received the " + amResp.getAMCommand() + " command from RM"); default: throw new YarnRuntimeException("Received the " + amResp.getAMCommand() + " command from RM"); } } releasedContainers.clear(); // Retrieve list of allocated containers from the response List<Container> newAllocatedContainers = amResp.getAllocatedContainers(); // LOG.info("Got response from RM for container ask, allocatedCnt=" + newAllocatedContainers.size()); numRequestedContainers -= newAllocatedContainers.size(); long timestamp = System.currentTimeMillis(); for (Container allocatedContainer : newAllocatedContainers) { LOG.info("Got new container." + ", containerId=" + allocatedContainer.getId() + ", containerNode=" + allocatedContainer.getNodeId() + ", containerNodeURI=" + allocatedContainer.getNodeHttpAddress() + ", containerResourceMemory" + allocatedContainer.getResource().getMemory() + ", priority" + allocatedContainer.getPriority()); // + ", containerToken" + allocatedContainer.getContainerToken().getIdentifier().toString()); boolean alreadyAllocated = true; StreamingContainerAgent.ContainerStartRequest csr = null; for (Map.Entry<StreamingContainerAgent.ContainerStartRequest, MutablePair<Integer, ContainerRequest>> entry : requestedResources .entrySet()) { if (entry.getKey().container.getResourceRequestPriority() == allocatedContainer.getPriority() .getPriority()) { alreadyAllocated = false; csr = entry.getKey(); break; } } if (alreadyAllocated) { LOG.info("Releasing {} as resource with priority {} was already assigned", allocatedContainer.getId(), allocatedContainer.getPriority()); releasedContainers.add(allocatedContainer.getId()); numReleasedContainers++; numRequestedContainers++; continue; } if (csr != null) { requestedResources.remove(csr); } // allocate resource to container ContainerResource resource = new ContainerResource(allocatedContainer.getPriority().getPriority(), allocatedContainer.getId().toString(), allocatedContainer.getNodeId().toString(), allocatedContainer.getResource().getMemory(), allocatedContainer.getResource().getVirtualCores(), allocatedContainer.getNodeHttpAddress()); StreamingContainerAgent sca = dnmgr.assignContainer(resource, null); if (sca == null) { // allocated container no longer needed, add release request LOG.warn("Container {} allocated but nothing to deploy, going to release this container.", allocatedContainer.getId()); releasedContainers.add(allocatedContainer.getId()); } else { AllocatedContainer allocatedContainerHolder = new AllocatedContainer(allocatedContainer); this.allocatedContainers.put(allocatedContainer.getId().toString(), allocatedContainerHolder); ByteBuffer tokens = null; if (UserGroupInformation.isSecurityEnabled()) { UserGroupInformation ugi = UserGroupInformation.getLoginUser(); Token<StramDelegationTokenIdentifier> delegationToken = allocateDelegationToken( ugi.getUserName(), heartbeatListener.getAddress()); allocatedContainerHolder.delegationToken = delegationToken; //ByteBuffer tokens = LaunchContainerRunnable.getTokens(delegationTokenManager, heartbeatListener.getAddress()); tokens = LaunchContainerRunnable.getTokens(ugi, delegationToken); } LaunchContainerRunnable launchContainer = new LaunchContainerRunnable(allocatedContainer, nmClient, sca, tokens); // Thread launchThread = new Thread(runnableLaunchContainer); // launchThreads.add(launchThread); // launchThread.start(); launchContainer.run(); // communication with NMs is now async // record container start event StramEvent ev = new StramEvent.StartContainerEvent(allocatedContainer.getId().toString(), allocatedContainer.getNodeId().toString()); ev.setTimestamp(timestamp); dnmgr.recordEventAsync(ev); } } // track node updates for future locality constraint allocations // TODO: it seems 2.0.4-alpha doesn't give us any updates resourceRequestor.updateNodeReports(amResp.getUpdatedNodes()); // Check the completed containers List<ContainerStatus> completedContainers = amResp.getCompletedContainersStatuses(); // LOG.debug("Got response from RM for container ask, completedCnt=" + completedContainers.size()); for (ContainerStatus containerStatus : completedContainers) { LOG.info("Completed containerId=" + containerStatus.getContainerId() + ", state=" + containerStatus.getState() + ", exitStatus=" + containerStatus.getExitStatus() + ", diagnostics=" + containerStatus.getDiagnostics()); // non complete containers should not be here assert (containerStatus.getState() == ContainerState.COMPLETE); AllocatedContainer allocatedContainer = allocatedContainers .remove(containerStatus.getContainerId().toString()); if (allocatedContainer != null && allocatedContainer.delegationToken != null) { UserGroupInformation ugi = UserGroupInformation.getLoginUser(); delegationTokenManager.cancelToken(allocatedContainer.delegationToken, ugi.getUserName()); } int exitStatus = containerStatus.getExitStatus(); if (0 != exitStatus) { if (allocatedContainer != null) { numFailedContainers.incrementAndGet(); } // if (exitStatus == 1) { // // non-recoverable StreamingContainer failure // appDone = true; // finalStatus = FinalApplicationStatus.FAILED; // dnmgr.shutdownDiagnosticsMessage = "Unrecoverable failure " + containerStatus.getContainerId(); // LOG.info("Exiting due to: {}", dnmgr.shutdownDiagnosticsMessage); // } // else { // Recoverable failure or process killed (externally or via stop request by AM) // also occurs when a container was released by the application but never assigned/launched LOG.debug("Container {} failed or killed.", containerStatus.getContainerId()); dnmgr.scheduleContainerRestart(containerStatus.getContainerId().toString()); // } } else { // container completed successfully numCompletedContainers.incrementAndGet(); LOG.info("Container completed successfully." + ", containerId=" + containerStatus.getContainerId()); } String containerIdStr = containerStatus.getContainerId().toString(); dnmgr.removeContainerAgent(containerIdStr); // record container stop event StramEvent ev = new StramEvent.StopContainerEvent(containerIdStr, containerStatus.getExitStatus()); ev.setReason(containerStatus.getDiagnostics()); dnmgr.recordEventAsync(ev); } if (dnmgr.forcedShutdown) { LOG.info("Forced shutdown due to {}", dnmgr.shutdownDiagnosticsMessage); finalStatus = FinalApplicationStatus.FAILED; appDone = true; } else if (allocatedContainers.isEmpty() && numRequestedContainers == 0 && dnmgr.containerStartRequests.isEmpty()) { LOG.debug("Exiting as no more containers are allocated or requested"); finalStatus = FinalApplicationStatus.SUCCEEDED; appDone = true; } LOG.debug("Current application state: loop=" + loopCounter + ", appDone=" + appDone + ", total=" + numTotalContainers + ", requested=" + numRequestedContainers + ", released=" + numReleasedContainers + ", completed=" + numCompletedContainers + ", failed=" + numFailedContainers + ", currentAllocated=" + allocatedContainers.size()); // monitor child containers dnmgr.monitorHeartbeat(); } finishApplication(finalStatus, numTotalContainers); }
From source file:com.hazelcast.yarn.ApplicationMaster.java
License:Open Source License
@Override public void onContainersCompleted(List<ContainerStatus> statuses) { for (ContainerStatus status : statuses) { this.containers.remove(status.getContainerId()); LOG.log(Level.INFO, "Container completed. Container id: {0}. State: {1}.", new Object[] { status.getContainerId(), status.getState() }); }/*from ww w. ja va2s. com*/ }
From source file:com.inforefiner.hdata.ApplicationMaster.java
License:Apache License
private static void publishContainerEndEvent(final TimelineClient timelineClient, ContainerStatus container, String domainId, UserGroupInformation ugi) { final TimelineEntity entity = new TimelineEntity(); entity.setEntityId(container.getContainerId().toString()); entity.setEntityType(DSEntity.DS_CONTAINER.toString()); entity.setDomainId(domainId);//from w w w. ja v a 2s . co m entity.addPrimaryFilter("user", ugi.getShortUserName()); TimelineEvent event = new TimelineEvent(); event.setTimestamp(System.currentTimeMillis()); event.setEventType(DSEvent.DS_CONTAINER_END.toString()); event.addEventInfo("State", container.getState().name()); event.addEventInfo("Exit Status", container.getExitStatus()); entity.addEvent(event); try { timelineClient.putEntities(entity); } catch (YarnException | IOException e) { LOG.error("Container end event could not be published for " + container.getContainerId().toString(), e); } }
From source file:com.scistor.dshell.ScistorApplicationMaster.java
License:Apache License
private static void publishContainerEndEvent(TimelineClient timelineClient, ContainerStatus container) throws IOException, YarnException { TimelineEntity entity = new TimelineEntity(); entity.setEntityId(container.getContainerId().toString()); entity.setEntityType(DSEntity.DS_CONTAINER.toString()); entity.addPrimaryFilter("user", UserGroupInformation.getCurrentUser().toString()); TimelineEvent event = new TimelineEvent(); event.setTimestamp(System.currentTimeMillis()); event.setEventType(DSEvent.DS_CONTAINER_END.toString()); event.addEventInfo("State", container.getState().name()); event.addEventInfo("Exit Status", container.getExitStatus()); entity.addEvent(event);// w w w . j a va2 s . c om timelineClient.putEntities(entity); }
From source file:edu.cmu.graphchi.toolkits.collaborative_filtering.yarn.ApplicationMaster.java
License:Apache License
public void onContainersCompleted(List<ContainerStatus> completedContainers) { LOG.info("Got response from RM for container ask, completedCnt=" + completedContainers.size()); for (ContainerStatus containerStatus : completedContainers) { LOG.info("Got container status for containerID=" + containerStatus.getContainerId() + ", state=" + containerStatus.getState() + ", exitStatus=" + containerStatus.getExitStatus() + ", diagnostics=" + containerStatus.getDiagnostics()); // non complete containers should not be here assert (containerStatus.getState() == ContainerState.COMPLETE); // increment counters for completed/failed containers int exitStatus = containerStatus.getExitStatus(); if (0 != exitStatus) { // container failed if (ContainerExitStatus.ABORTED != exitStatus) { // shell script failed // counts as completed numCompletedContainers.incrementAndGet(); numFailedContainers.incrementAndGet(); } else { // container was killed by framework, possibly preempted // we should re-try as the container was lost for some reason //TODO: Add retry numCompletedContainers.incrementAndGet(); numFailedContainers.incrementAndGet(); // we do not need to release the container as it would be done // by the RM }//from w w w .j a va 2 s. com } else { //nothing to do // container completed successfully numCompletedContainers.incrementAndGet(); LOG.info("Container completed successfully." + ", containerId=" + containerStatus.getContainerId()); } } }
From source file:gobblin.yarn.YarnService.java
License:Apache License
private ImmutableMap.Builder<String, String> buildContainerStatusEventMetadata( ContainerStatus containerStatus) { ImmutableMap.Builder<String, String> eventMetadataBuilder = new ImmutableMap.Builder<>(); eventMetadataBuilder.put(GobblinYarnMetricTagNames.CONTAINER_ID, containerStatus.getContainerId().toString()); eventMetadataBuilder.put(GobblinYarnEventConstants.EventMetadata.CONTAINER_STATUS_CONTAINER_STATE, containerStatus.getState().toString()); if (ContainerExitStatus.INVALID != containerStatus.getExitStatus()) { eventMetadataBuilder.put(GobblinYarnEventConstants.EventMetadata.CONTAINER_STATUS_EXIT_STATUS, containerStatus.getExitStatus() + ""); }//from w w w.j a v a 2 s. c o m if (!Strings.isNullOrEmpty(containerStatus.getDiagnostics())) { eventMetadataBuilder.put(GobblinYarnEventConstants.EventMetadata.CONTAINER_STATUS_EXIT_DIAGNOSTICS, containerStatus.getDiagnostics()); } return eventMetadataBuilder; }
From source file:husky.server.HuskyRMCallbackHandler.java
License:Apache License
public void onContainersCompleted(List<ContainerStatus> completedContainerStatus) { LOG.info("Get response from RM for container request, completedCnt = " + completedContainerStatus.size()); mNumCompletedContainers += completedContainerStatus.size(); for (ContainerStatus status : completedContainerStatus) { LOG.info(String.format("Container %s: %s, exit status: %d", status.getContainerId().toString(), status.getState().toString(), status.getExitStatus())); if (status.getExitStatus() == 0) { mNumSuccess += 1;/*from ww w . j a v a 2 s . c om*/ } } LOG.info("Total containers: " + mNumContainers + ", completed containers: " + mNumCompletedContainers); if (mNumContainers == mNumCompletedContainers) { // If all workers and master finish synchronized (finalResultLock) { finalResultLock.unlock(); finalResultLock.notifyAll(); } } }
From source file:io.hops.tensorflow.TimelineHandler.java
License:Apache License
public void publishContainerEndEvent(ContainerStatus container) { final TimelineEntity entity = new TimelineEntity(); entity.setEntityId(container.getContainerId().toString()); entity.setEntityType(ApplicationMaster.YarntfEntity.YARNTF_CONTAINER.toString()); entity.setDomainId(domainId);/*from ww w .ja v a 2 s . c om*/ entity.addPrimaryFilter("user", ugi.getShortUserName()); TimelineEvent event = new TimelineEvent(); event.setTimestamp(System.currentTimeMillis()); event.setEventType(ApplicationMaster.YarntfEvent.YARNTF_CONTAINER_END.toString()); event.addEventInfo("State", container.getState().name()); event.addEventInfo("Exit Status", container.getExitStatus()); entity.addEvent(event); try { timelineClient.putEntities(entity); } catch (YarnException | IOException e) { LOG.error("Container end event could not be published for " + container.getContainerId().toString(), e); } }
From source file:org.apache.flink.yarn.appMaster.ApplicationMaster.java
License:Apache License
private void logDeadContainer(ContainerStatus status, StringBuffer containerDiag) { String msg = "Diagnostics for containerId=" + status.getContainerId() + " in state=" + status.getState() + "\n" + status.getDiagnostics(); messages.add(new Message(msg)); containerDiag.append("\n\n"); containerDiag.append(msg);//www . j av a2 s.c o m }