/// <exception cref="System.IO.IOException"/> private void DoneApplicationAttempt(ApplicationAttemptId applicationAttemptId, RMAppAttemptState rmAppAttemptFinalState, bool keepContainers) { lock (this) { FiCaSchedulerApp attempt = GetApplicationAttempt(applicationAttemptId); SchedulerApplication <FiCaSchedulerApp> application = applications[applicationAttemptId .GetApplicationId()]; if (application == null || attempt == null) { throw new IOException("Unknown application " + applicationAttemptId + " has completed!" ); } // Kill all 'live' containers foreach (RMContainer container in attempt.GetLiveContainers()) { if (keepContainers && container.GetState().Equals(RMContainerState.Running)) { // do not kill the running container in the case of work-preserving AM // restart. Log.Info("Skip killing " + container.GetContainerId()); continue; } CompletedContainer(container, SchedulerUtils.CreateAbnormalContainerStatus(container .GetContainerId(), SchedulerUtils.CompletedApplication), RMContainerEventType.Kill ); } // Clean up pending requests, metrics etc. attempt.Stop(rmAppAttemptFinalState); } }
private void RemoveNode(RMNode nodeInfo) { lock (this) { FiCaSchedulerNode node = GetNode(nodeInfo.GetNodeID()); if (node == null) { return; } // Kill running containers foreach (RMContainer container in node.GetRunningContainers()) { CompletedContainer(container, SchedulerUtils.CreateAbnormalContainerStatus(container .GetContainerId(), SchedulerUtils.LostContainer), RMContainerEventType.Kill); } //Remove the node Sharpen.Collections.Remove(this.nodes, nodeInfo.GetNodeID()); UpdateMaximumAllocation(node, false); // Update cluster metrics Resources.SubtractFrom(clusterResource, node.GetTotalResource()); } }
public override void Handle(SchedulerEvent @event) { switch (@event.GetType()) { case SchedulerEventType.NodeAdded: { NodeAddedSchedulerEvent nodeAddedEvent = (NodeAddedSchedulerEvent)@event; AddNode(nodeAddedEvent.GetAddedRMNode()); RecoverContainersOnNode(nodeAddedEvent.GetContainerReports(), nodeAddedEvent.GetAddedRMNode ()); break; } case SchedulerEventType.NodeRemoved: { NodeRemovedSchedulerEvent nodeRemovedEvent = (NodeRemovedSchedulerEvent)@event; RemoveNode(nodeRemovedEvent.GetRemovedRMNode()); break; } case SchedulerEventType.NodeResourceUpdate: { NodeResourceUpdateSchedulerEvent nodeResourceUpdatedEvent = (NodeResourceUpdateSchedulerEvent )@event; UpdateNodeResource(nodeResourceUpdatedEvent.GetRMNode(), nodeResourceUpdatedEvent .GetResourceOption()); break; } case SchedulerEventType.NodeUpdate: { NodeUpdateSchedulerEvent nodeUpdatedEvent = (NodeUpdateSchedulerEvent)@event; NodeUpdate(nodeUpdatedEvent.GetRMNode()); break; } case SchedulerEventType.AppAdded: { AppAddedSchedulerEvent appAddedEvent = (AppAddedSchedulerEvent)@event; AddApplication(appAddedEvent.GetApplicationId(), appAddedEvent.GetQueue(), appAddedEvent .GetUser(), appAddedEvent.GetIsAppRecovering()); break; } case SchedulerEventType.AppRemoved: { AppRemovedSchedulerEvent appRemovedEvent = (AppRemovedSchedulerEvent)@event; DoneApplication(appRemovedEvent.GetApplicationID(), appRemovedEvent.GetFinalState ()); break; } case SchedulerEventType.AppAttemptAdded: { AppAttemptAddedSchedulerEvent appAttemptAddedEvent = (AppAttemptAddedSchedulerEvent )@event; AddApplicationAttempt(appAttemptAddedEvent.GetApplicationAttemptId(), appAttemptAddedEvent .GetTransferStateFromPreviousAttempt(), appAttemptAddedEvent.GetIsAttemptRecovering ()); break; } case SchedulerEventType.AppAttemptRemoved: { AppAttemptRemovedSchedulerEvent appAttemptRemovedEvent = (AppAttemptRemovedSchedulerEvent )@event; try { DoneApplicationAttempt(appAttemptRemovedEvent.GetApplicationAttemptID(), appAttemptRemovedEvent .GetFinalAttemptState(), appAttemptRemovedEvent.GetKeepContainersAcrossAppAttempts ()); } catch (IOException ie) { Log.Error("Unable to remove application " + appAttemptRemovedEvent.GetApplicationAttemptID (), ie); } break; } case SchedulerEventType.ContainerExpired: { ContainerExpiredSchedulerEvent containerExpiredEvent = (ContainerExpiredSchedulerEvent )@event; ContainerId containerid = containerExpiredEvent.GetContainerId(); CompletedContainer(GetRMContainer(containerid), SchedulerUtils.CreateAbnormalContainerStatus (containerid, SchedulerUtils.ExpiredContainer), RMContainerEventType.Expire); break; } case SchedulerEventType.ContainerRescheduled: { ContainerRescheduledEvent containerRescheduledEvent = (ContainerRescheduledEvent) @event; RMContainer container = containerRescheduledEvent.GetContainer(); RecoverResourceRequestForContainer(container); break; } default: { Log.Error("Invalid eventtype " + @event.GetType() + ". Ignoring!"); break; } } }
public virtual void TestReleaseWhileRunning() { DrainDispatcher drainDispatcher = new DrainDispatcher(); EventHandler <RMAppAttemptEvent> appAttemptEventHandler = Org.Mockito.Mockito.Mock <EventHandler>(); EventHandler generic = Org.Mockito.Mockito.Mock <EventHandler>(); drainDispatcher.Register(typeof(RMAppAttemptEventType), appAttemptEventHandler); drainDispatcher.Register(typeof(RMNodeEventType), generic); drainDispatcher.Init(new YarnConfiguration()); drainDispatcher.Start(); NodeId nodeId = BuilderUtils.NewNodeId("host", 3425); ApplicationId appId = BuilderUtils.NewApplicationId(1, 1); ApplicationAttemptId appAttemptId = BuilderUtils.NewApplicationAttemptId(appId, 1 ); ContainerId containerId = BuilderUtils.NewContainerId(appAttemptId, 1); ContainerAllocationExpirer expirer = Org.Mockito.Mockito.Mock <ContainerAllocationExpirer >(); Resource resource = BuilderUtils.NewResource(512, 1); Priority priority = BuilderUtils.NewPriority(5); Container container = BuilderUtils.NewContainer(containerId, nodeId, "host:3465", resource, priority, null); ConcurrentMap <ApplicationId, RMApp> rmApps = Org.Mockito.Mockito.Spy(new ConcurrentHashMap <ApplicationId, RMApp>()); RMApp rmApp = Org.Mockito.Mockito.Mock <RMApp>(); Org.Mockito.Mockito.When(rmApp.GetRMAppAttempt((ApplicationAttemptId)Matchers.Any ())).ThenReturn(null); Org.Mockito.Mockito.DoReturn(rmApp).When(rmApps)[(ApplicationId)Matchers.Any()]; RMApplicationHistoryWriter writer = Org.Mockito.Mockito.Mock <RMApplicationHistoryWriter >(); SystemMetricsPublisher publisher = Org.Mockito.Mockito.Mock <SystemMetricsPublisher >(); RMContext rmContext = Org.Mockito.Mockito.Mock <RMContext>(); Org.Mockito.Mockito.When(rmContext.GetDispatcher()).ThenReturn(drainDispatcher); Org.Mockito.Mockito.When(rmContext.GetContainerAllocationExpirer()).ThenReturn(expirer ); Org.Mockito.Mockito.When(rmContext.GetRMApplicationHistoryWriter()).ThenReturn(writer ); Org.Mockito.Mockito.When(rmContext.GetRMApps()).ThenReturn(rmApps); Org.Mockito.Mockito.When(rmContext.GetSystemMetricsPublisher()).ThenReturn(publisher ); Org.Mockito.Mockito.When(rmContext.GetYarnConfiguration()).ThenReturn(new YarnConfiguration ()); RMContainer rmContainer = new RMContainerImpl(container, appAttemptId, nodeId, "user" , rmContext); NUnit.Framework.Assert.AreEqual(RMContainerState.New, rmContainer.GetState()); NUnit.Framework.Assert.AreEqual(resource, rmContainer.GetAllocatedResource()); NUnit.Framework.Assert.AreEqual(nodeId, rmContainer.GetAllocatedNode()); NUnit.Framework.Assert.AreEqual(priority, rmContainer.GetAllocatedPriority()); Org.Mockito.Mockito.Verify(writer).ContainerStarted(Matchers.Any <RMContainer>()); Org.Mockito.Mockito.Verify(publisher).ContainerCreated(Matchers.Any <RMContainer>( ), Matchers.AnyLong()); rmContainer.Handle(new RMContainerEvent(containerId, RMContainerEventType.Start)); drainDispatcher.Await(); NUnit.Framework.Assert.AreEqual(RMContainerState.Allocated, rmContainer.GetState( )); rmContainer.Handle(new RMContainerEvent(containerId, RMContainerEventType.Acquired )); drainDispatcher.Await(); NUnit.Framework.Assert.AreEqual(RMContainerState.Acquired, rmContainer.GetState() ); rmContainer.Handle(new RMContainerEvent(containerId, RMContainerEventType.Launched )); drainDispatcher.Await(); NUnit.Framework.Assert.AreEqual(RMContainerState.Running, rmContainer.GetState()); NUnit.Framework.Assert.AreEqual("http://host:3465/node/containerlogs/container_1_0001_01_000001/user" , rmContainer.GetLogURL()); // In RUNNING state. Verify RELEASED and associated actions. Org.Mockito.Mockito.Reset(appAttemptEventHandler); ContainerStatus containerStatus = SchedulerUtils.CreateAbnormalContainerStatus(containerId , SchedulerUtils.ReleasedContainer); rmContainer.Handle(new RMContainerFinishedEvent(containerId, containerStatus, RMContainerEventType .Released)); drainDispatcher.Await(); NUnit.Framework.Assert.AreEqual(RMContainerState.Released, rmContainer.GetState() ); NUnit.Framework.Assert.AreEqual(SchedulerUtils.ReleasedContainer, rmContainer.GetDiagnosticsInfo ()); NUnit.Framework.Assert.AreEqual(ContainerExitStatus.Aborted, rmContainer.GetContainerExitStatus ()); NUnit.Framework.Assert.AreEqual(ContainerState.Complete, rmContainer.GetContainerState ()); Org.Mockito.Mockito.Verify(writer).ContainerFinished(Matchers.Any <RMContainer>()); Org.Mockito.Mockito.Verify(publisher).ContainerFinished(Matchers.Any <RMContainer> (), Matchers.AnyLong()); ArgumentCaptor <RMAppAttemptContainerFinishedEvent> captor = ArgumentCaptor.ForClass <RMAppAttemptContainerFinishedEvent>(); Org.Mockito.Mockito.Verify(appAttemptEventHandler).Handle(captor.Capture()); RMAppAttemptContainerFinishedEvent cfEvent = captor.GetValue(); NUnit.Framework.Assert.AreEqual(appAttemptId, cfEvent.GetApplicationAttemptId()); NUnit.Framework.Assert.AreEqual(containerStatus, cfEvent.GetContainerStatus()); NUnit.Framework.Assert.AreEqual(RMAppAttemptEventType.ContainerFinished, cfEvent. GetType()); // In RELEASED state. A FINIHSED event may come in. rmContainer.Handle(new RMContainerFinishedEvent(containerId, SchedulerUtils.CreateAbnormalContainerStatus (containerId, "FinishedContainer"), RMContainerEventType.Finished)); NUnit.Framework.Assert.AreEqual(RMContainerState.Released, rmContainer.GetState() ); }
public virtual void TestExpireWhileRunning() { DrainDispatcher drainDispatcher = new DrainDispatcher(); EventHandler <RMAppAttemptEvent> appAttemptEventHandler = Org.Mockito.Mockito.Mock <EventHandler>(); EventHandler generic = Org.Mockito.Mockito.Mock <EventHandler>(); drainDispatcher.Register(typeof(RMAppAttemptEventType), appAttemptEventHandler); drainDispatcher.Register(typeof(RMNodeEventType), generic); drainDispatcher.Init(new YarnConfiguration()); drainDispatcher.Start(); NodeId nodeId = BuilderUtils.NewNodeId("host", 3425); ApplicationId appId = BuilderUtils.NewApplicationId(1, 1); ApplicationAttemptId appAttemptId = BuilderUtils.NewApplicationAttemptId(appId, 1 ); ContainerId containerId = BuilderUtils.NewContainerId(appAttemptId, 1); ContainerAllocationExpirer expirer = Org.Mockito.Mockito.Mock <ContainerAllocationExpirer >(); Resource resource = BuilderUtils.NewResource(512, 1); Priority priority = BuilderUtils.NewPriority(5); Container container = BuilderUtils.NewContainer(containerId, nodeId, "host:3465", resource, priority, null); RMApplicationHistoryWriter writer = Org.Mockito.Mockito.Mock <RMApplicationHistoryWriter >(); SystemMetricsPublisher publisher = Org.Mockito.Mockito.Mock <SystemMetricsPublisher >(); RMContext rmContext = Org.Mockito.Mockito.Mock <RMContext>(); Org.Mockito.Mockito.When(rmContext.GetDispatcher()).ThenReturn(drainDispatcher); Org.Mockito.Mockito.When(rmContext.GetContainerAllocationExpirer()).ThenReturn(expirer ); Org.Mockito.Mockito.When(rmContext.GetRMApplicationHistoryWriter()).ThenReturn(writer ); Org.Mockito.Mockito.When(rmContext.GetSystemMetricsPublisher()).ThenReturn(publisher ); Org.Mockito.Mockito.When(rmContext.GetYarnConfiguration()).ThenReturn(new YarnConfiguration ()); RMContainer rmContainer = new RMContainerImpl(container, appAttemptId, nodeId, "user" , rmContext); NUnit.Framework.Assert.AreEqual(RMContainerState.New, rmContainer.GetState()); NUnit.Framework.Assert.AreEqual(resource, rmContainer.GetAllocatedResource()); NUnit.Framework.Assert.AreEqual(nodeId, rmContainer.GetAllocatedNode()); NUnit.Framework.Assert.AreEqual(priority, rmContainer.GetAllocatedPriority()); Org.Mockito.Mockito.Verify(writer).ContainerStarted(Matchers.Any <RMContainer>()); Org.Mockito.Mockito.Verify(publisher).ContainerCreated(Matchers.Any <RMContainer>( ), Matchers.AnyLong()); rmContainer.Handle(new RMContainerEvent(containerId, RMContainerEventType.Start)); drainDispatcher.Await(); NUnit.Framework.Assert.AreEqual(RMContainerState.Allocated, rmContainer.GetState( )); rmContainer.Handle(new RMContainerEvent(containerId, RMContainerEventType.Acquired )); drainDispatcher.Await(); NUnit.Framework.Assert.AreEqual(RMContainerState.Acquired, rmContainer.GetState() ); rmContainer.Handle(new RMContainerEvent(containerId, RMContainerEventType.Launched )); drainDispatcher.Await(); NUnit.Framework.Assert.AreEqual(RMContainerState.Running, rmContainer.GetState()); NUnit.Framework.Assert.AreEqual("http://host:3465/node/containerlogs/container_1_0001_01_000001/user" , rmContainer.GetLogURL()); // In RUNNING state. Verify EXPIRE and associated actions. Org.Mockito.Mockito.Reset(appAttemptEventHandler); ContainerStatus containerStatus = SchedulerUtils.CreateAbnormalContainerStatus(containerId , SchedulerUtils.ExpiredContainer); rmContainer.Handle(new RMContainerFinishedEvent(containerId, containerStatus, RMContainerEventType .Expire)); drainDispatcher.Await(); NUnit.Framework.Assert.AreEqual(RMContainerState.Running, rmContainer.GetState()); Org.Mockito.Mockito.Verify(writer, Org.Mockito.Mockito.Never()).ContainerFinished (Matchers.Any <RMContainer>()); Org.Mockito.Mockito.Verify(publisher, Org.Mockito.Mockito.Never()).ContainerFinished (Matchers.Any <RMContainer>(), Matchers.AnyLong()); }