public virtual void TestAMLaunchAndCleanup() { Logger rootLogger = LogManager.GetRootLogger(); rootLogger.SetLevel(Level.Debug); TestApplicationMasterLauncher.MyContainerManagerImpl containerManager = new TestApplicationMasterLauncher.MyContainerManagerImpl (); MockRMWithCustomAMLauncher rm = new MockRMWithCustomAMLauncher(containerManager); rm.Start(); MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 5120); RMApp app = rm.SubmitApp(2000); // kick the scheduling nm1.NodeHeartbeat(true); int waitCount = 0; while (containerManager.launched == false && waitCount++ < 20) { Log.Info("Waiting for AM Launch to happen.."); Sharpen.Thread.Sleep(1000); } NUnit.Framework.Assert.IsTrue(containerManager.launched); RMAppAttempt attempt = app.GetCurrentAppAttempt(); ApplicationAttemptId appAttemptId = attempt.GetAppAttemptId(); NUnit.Framework.Assert.AreEqual(appAttemptId.ToString(), containerManager.attemptIdAtContainerManager ); NUnit.Framework.Assert.AreEqual(app.GetSubmitTime(), containerManager.submitTimeAtContainerManager ); NUnit.Framework.Assert.AreEqual(app.GetRMAppAttempt(appAttemptId).GetMasterContainer ().GetId().ToString(), containerManager.containerIdAtContainerManager); NUnit.Framework.Assert.AreEqual(nm1.GetNodeId().ToString(), containerManager.nmHostAtContainerManager ); NUnit.Framework.Assert.AreEqual(YarnConfiguration.DefaultRmAmMaxAttempts, containerManager .maxAppAttempts); MockAM am = new MockAM(rm.GetRMContext(), rm.GetApplicationMasterService(), appAttemptId ); am.RegisterAppAttempt(); am.UnregisterAppAttempt(); //complete the AM container to finish the app normally nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState.Complete); am.WaitForState(RMAppAttemptState.Finished); waitCount = 0; while (containerManager.cleanedup == false && waitCount++ < 20) { Log.Info("Waiting for AM Cleanup to happen.."); Sharpen.Thread.Sleep(1000); } NUnit.Framework.Assert.IsTrue(containerManager.cleanedup); am.WaitForState(RMAppAttemptState.Finished); rm.Stop(); }
// Disable webapp /// <exception cref="System.Exception"/> public static void FinishAMAndVerifyAppState(RMApp rmApp, Org.Apache.Hadoop.Yarn.Server.Resourcemanager.MockRM rm, MockNM nm, MockAM am) { FinishApplicationMasterRequest req = FinishApplicationMasterRequest.NewInstance(FinalApplicationStatus .Succeeded, string.Empty, string.Empty); am.UnregisterAppAttempt(req, true); am.WaitForState(RMAppAttemptState.Finishing); nm.NodeHeartbeat(am.GetApplicationAttemptId(), 1, ContainerState.Complete); am.WaitForState(RMAppAttemptState.Finished); rm.WaitForState(rmApp.GetApplicationId(), RMAppState.Finished); }
/// <exception cref="System.Exception"/> public virtual void TestInvalidatedAMHostPortOnAMRestart() { MockRM rm1 = new MockRM(conf); rm1.Start(); MockNM nm1 = new MockNM("127.0.0.1:1234", 15120, rm1.GetResourceTrackerService()); nm1.RegisterNode(); // a failed app RMApp app2 = rm1.SubmitApp(200); MockAM am2 = MockRM.LaunchAndRegisterAM(app2, rm1, nm1); nm1.NodeHeartbeat(am2.GetApplicationAttemptId(), 1, ContainerState.Complete); am2.WaitForState(RMAppAttemptState.Failed); rm1.WaitForState(app2.GetApplicationId(), RMAppState.Accepted); // before new attempt is launched, the app report returns the invalid AM // host and port. GetApplicationReportRequest request1 = GetApplicationReportRequest.NewInstance(app2 .GetApplicationId()); ApplicationReport report1 = rm1.GetClientRMService().GetApplicationReport(request1 ).GetApplicationReport(); NUnit.Framework.Assert.AreEqual("N/A", report1.GetHost()); NUnit.Framework.Assert.AreEqual(-1, report1.GetRpcPort()); }
/// <exception cref="System.Exception"/> public virtual void SendAMLaunchFailed(ApplicationAttemptId appAttemptId) { MockAM am = new MockAM(GetRMContext(), masterService, appAttemptId); am.WaitForState(RMAppAttemptState.Allocated); GetRMContext().GetDispatcher().GetEventHandler().Handle(new RMAppAttemptEvent(appAttemptId , RMAppAttemptEventType.LaunchFailed, "Failed")); }
/// <exception cref="System.Exception"/> public virtual void TestAppOnMultiNode() { Logger rootLogger = LogManager.GetRootLogger(); rootLogger.SetLevel(Level.Debug); conf.Set("yarn.scheduler.capacity.node-locality-delay", "-1"); MockRM rm = new MockRM(conf); rm.Start(); MockNM nm1 = rm.RegisterNode("h1:1234", 5120); MockNM nm2 = rm.RegisterNode("h2:5678", 10240); RMApp app = rm.SubmitApp(2000); //kick the scheduling nm1.NodeHeartbeat(true); RMAppAttempt attempt = app.GetCurrentAppAttempt(); MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId()); am.RegisterAppAttempt(); //request for containers int request = 13; am.Allocate("h1", 1000, request, new AList <ContainerId>()); //kick the scheduler IList <Container> conts = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId >()).GetAllocatedContainers(); int contReceived = conts.Count; while (contReceived < 3) { //only 3 containers are available on node1 nm1.NodeHeartbeat(true); Sharpen.Collections.AddAll(conts, am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers()); contReceived = conts.Count; Log.Info("Got " + contReceived + " containers. Waiting to get " + 3); Sharpen.Thread.Sleep(WaitSleepMs); } NUnit.Framework.Assert.AreEqual(3, conts.Count); //send node2 heartbeat conts = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers (); contReceived = conts.Count; while (contReceived < 10) { nm2.NodeHeartbeat(true); Sharpen.Collections.AddAll(conts, am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers()); contReceived = conts.Count; Log.Info("Got " + contReceived + " containers. Waiting to get " + 10); Sharpen.Thread.Sleep(WaitSleepMs); } NUnit.Framework.Assert.AreEqual(10, conts.Count); am.UnregisterAppAttempt(); nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState.Complete); am.WaitForState(RMAppAttemptState.Finished); rm.Stop(); }
// This is to test AM Host and rpc port are invalidated after the am attempt // is killed or failed, so that client doesn't get the wrong information. /// <exception cref="System.Exception"/> public virtual void TestInvalidateAMHostPortWhenAMFailedOrKilled() { conf.SetInt(YarnConfiguration.RmAmMaxAttempts, 1); MockRM rm1 = new MockRM(conf); rm1.Start(); // a succeeded app RMApp app1 = rm1.SubmitApp(200); MockNM nm1 = new MockNM("127.0.0.1:1234", 15120, rm1.GetResourceTrackerService()); nm1.RegisterNode(); MockAM am1 = MockRM.LaunchAndRegisterAM(app1, rm1, nm1); MockRM.FinishAMAndVerifyAppState(app1, rm1, nm1, am1); // a failed app RMApp app2 = rm1.SubmitApp(200); MockAM am2 = MockRM.LaunchAndRegisterAM(app2, rm1, nm1); nm1.NodeHeartbeat(am2.GetApplicationAttemptId(), 1, ContainerState.Complete); am2.WaitForState(RMAppAttemptState.Failed); rm1.WaitForState(app2.GetApplicationId(), RMAppState.Failed); // a killed app RMApp app3 = rm1.SubmitApp(200); MockAM am3 = MockRM.LaunchAndRegisterAM(app3, rm1, nm1); rm1.KillApp(app3.GetApplicationId()); rm1.WaitForState(app3.GetApplicationId(), RMAppState.Killed); rm1.WaitForState(am3.GetApplicationAttemptId(), RMAppAttemptState.Killed); GetApplicationsRequest request1 = GetApplicationsRequest.NewInstance(EnumSet.Of(YarnApplicationState .Finished, YarnApplicationState.Killed, YarnApplicationState.Failed)); GetApplicationsResponse response1 = rm1.GetClientRMService().GetApplications(request1 ); IList <ApplicationReport> appList1 = response1.GetApplicationList(); NUnit.Framework.Assert.AreEqual(3, appList1.Count); foreach (ApplicationReport report in appList1) { // killed/failed apps host and rpc port are invalidated. if (report.GetApplicationId().Equals(app2.GetApplicationId()) || report.GetApplicationId ().Equals(app3.GetApplicationId())) { NUnit.Framework.Assert.AreEqual("N/A", report.GetHost()); NUnit.Framework.Assert.AreEqual(-1, report.GetRpcPort()); } // succeeded app's host and rpc port is not invalidated if (report.GetApplicationId().Equals(app1.GetApplicationId())) { NUnit.Framework.Assert.IsFalse(report.GetHost().Equals("N/A")); NUnit.Framework.Assert.IsTrue(report.GetRpcPort() != -1); } } }
// from AMLauncher /// <exception cref="System.Exception"/> public virtual MockAM SendAMLaunched(ApplicationAttemptId appAttemptId) { MockAM am = new MockAM(GetRMContext(), masterService, appAttemptId); am.WaitForState(RMAppAttemptState.Allocated); //create and set AMRMToken Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> amrmToken = this.rmContext .GetAMRMTokenSecretManager().CreateAndGetAMRMToken(appAttemptId); ((RMAppAttemptImpl)this.rmContext.GetRMApps()[appAttemptId.GetApplicationId()].GetRMAppAttempt (appAttemptId)).SetAMRMToken(amrmToken); GetRMContext().GetDispatcher().GetEventHandler().Handle(new RMAppAttemptEvent(appAttemptId , RMAppAttemptEventType.Launched)); return(am); }
/// <exception cref="System.Exception"/> public virtual void TestFinishApplicationMasterBeforeRegistering() { MockRM rm = new MockRM(conf); try { rm.Start(); // Register node1 MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb); // Submit an application RMApp app1 = rm.SubmitApp(2048); MockAM am1 = MockRM.LaunchAM(app1, rm, nm1); FinishApplicationMasterRequest req = FinishApplicationMasterRequest.NewInstance(FinalApplicationStatus .Failed, string.Empty, string.Empty); try { am1.UnregisterAppAttempt(req, false); NUnit.Framework.Assert.Fail("ApplicationMasterNotRegisteredException should be thrown" ); } catch (ApplicationMasterNotRegisteredException e) { NUnit.Framework.Assert.IsNotNull(e); NUnit.Framework.Assert.IsNotNull(e.Message); NUnit.Framework.Assert.IsTrue(e.Message.Contains("Application Master is trying to unregister before registering for:" )); } catch (Exception) { NUnit.Framework.Assert.Fail("ApplicationMasterNotRegisteredException should be thrown" ); } am1.RegisterAppAttempt(); am1.UnregisterAppAttempt(req, false); am1.WaitForState(RMAppAttemptState.Finishing); } finally { if (rm != null) { rm.Stop(); } } }
/// <summary>Validate killing an application when it is at accepted state.</summary> /// <exception cref="System.Exception">exception</exception> public virtual void TestApplicationKillAtAcceptedState() { Dispatcher dispatcher = new _AsyncDispatcher_573(); MockRM rm = new _MockRM_596(dispatcher, conf); // test metrics QueueMetrics metrics = rm.GetResourceScheduler().GetRootQueueMetrics(); int appsKilled = metrics.GetAppsKilled(); int appsSubmitted = metrics.GetAppsSubmitted(); rm.Start(); MockNM nm1 = new MockNM("127.0.0.1:1234", 15120, rm.GetResourceTrackerService()); nm1.RegisterNode(); // a failed app RMApp application = rm.SubmitApp(200); MockAM am = MockRM.LaunchAM(application, rm, nm1); am.WaitForState(RMAppAttemptState.Launched); nm1.NodeHeartbeat(am.GetApplicationAttemptId(), 1, ContainerState.Running); rm.WaitForState(application.GetApplicationId(), RMAppState.Accepted); // Now kill the application before new attempt is launched, the app report // returns the invalid AM host and port. KillApplicationRequest request = KillApplicationRequest.NewInstance(application.GetApplicationId ()); rm.GetClientRMService().ForceKillApplication(request); // Specific test for YARN-1689 follows // Now let's say a race causes AM to register now. This should not crash RM. am.RegisterAppAttempt(false); // We explicitly intercepted the kill-event to RMAppAttempt, so app should // still be in KILLING state. rm.WaitForState(application.GetApplicationId(), RMAppState.Killing); // AM should now be in running rm.WaitForState(am.GetApplicationAttemptId(), RMAppAttemptState.Running); // Simulate that appAttempt is killed. rm.GetRMContext().GetDispatcher().GetEventHandler().Handle(new RMAppEvent(application .GetApplicationId(), RMAppEventType.AttemptKilled)); rm.WaitForState(application.GetApplicationId(), RMAppState.Killed); // test metrics metrics = rm.GetResourceScheduler().GetRootQueueMetrics(); NUnit.Framework.Assert.AreEqual(appsKilled + 1, metrics.GetAppsKilled()); NUnit.Framework.Assert.AreEqual(appsSubmitted + 1, metrics.GetAppsSubmitted()); }
/// <exception cref="System.Exception"/> public virtual void TestAppWithNoContainers() { Logger rootLogger = LogManager.GetRootLogger(); rootLogger.SetLevel(Level.Debug); MockRM rm = new MockRM(conf); rm.Start(); MockNM nm1 = rm.RegisterNode("h1:1234", 5120); RMApp app = rm.SubmitApp(2000); //kick the scheduling nm1.NodeHeartbeat(true); RMAppAttempt attempt = app.GetCurrentAppAttempt(); MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId()); am.RegisterAppAttempt(); am.UnregisterAppAttempt(); nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState.Complete); am.WaitForState(RMAppAttemptState.Finished); rm.Stop(); }
/// <exception cref="System.Exception"/> private void AmRestartTests(bool keepRunningContainers) { MockRM rm = new MockRM(conf); rm.Start(); RMApp app = rm.SubmitApp(200, "name", "user", new Dictionary <ApplicationAccessType , string>(), false, "default", -1, null, "MAPREDUCE", false, keepRunningContainers ); MockNM nm = new MockNM("127.0.0.1:1234", 10240, rm.GetResourceTrackerService()); nm.RegisterNode(); MockAM am0 = MockRM.LaunchAndRegisterAM(app, rm, nm); int NumContainers = 1; // allocate NUM_CONTAINERS containers am0.Allocate("127.0.0.1", 1024, NumContainers, new AList <ContainerId>()); nm.NodeHeartbeat(true); // wait for containers to be allocated. IList <Container> containers = am0.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers(); while (containers.Count != NumContainers) { nm.NodeHeartbeat(true); Sharpen.Collections.AddAll(containers, am0.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers()); Sharpen.Thread.Sleep(200); } // launch the 2nd container. ContainerId containerId2 = ContainerId.NewContainerId(am0.GetApplicationAttemptId (), 2); nm.NodeHeartbeat(am0.GetApplicationAttemptId(), containerId2.GetContainerId(), ContainerState .Running); rm.WaitForState(nm, containerId2, RMContainerState.Running); // Capture the containers here so the metrics can be calculated after the // app has completed. ICollection <RMContainer> rmContainers = rm.scheduler.GetSchedulerAppInfo(am0.GetApplicationAttemptId ()).GetLiveContainers(); // fail the first app attempt by sending CONTAINER_FINISHED event without // registering. ContainerId amContainerId = app.GetCurrentAppAttempt().GetMasterContainer().GetId (); nm.NodeHeartbeat(am0.GetApplicationAttemptId(), amContainerId.GetContainerId(), ContainerState .Complete); am0.WaitForState(RMAppAttemptState.Failed); long memorySeconds = 0; long vcoreSeconds = 0; // Calculate container usage metrics for first attempt. if (keepRunningContainers) { // Only calculate the usage for the one container that has completed. foreach (RMContainer c in rmContainers) { if (c.GetContainerId().Equals(amContainerId)) { AggregateAppResourceUsage ru = CalculateContainerResourceMetrics(c); memorySeconds += ru.GetMemorySeconds(); vcoreSeconds += ru.GetVcoreSeconds(); } else { // The remaining container should be RUNNING. NUnit.Framework.Assert.IsTrue("After first attempt failed, remaining container " + "should still be running. ", c.GetContainerState().Equals(ContainerState.Running )); } } } else { // If keepRunningContainers is false, all live containers should now // be completed. Calculate the resource usage metrics for all of them. foreach (RMContainer c in rmContainers) { AggregateAppResourceUsage ru = CalculateContainerResourceMetrics(c); memorySeconds += ru.GetMemorySeconds(); vcoreSeconds += ru.GetVcoreSeconds(); } } // wait for app to start a new attempt. rm.WaitForState(app.GetApplicationId(), RMAppState.Accepted); // assert this is a new AM. RMAppAttempt attempt2 = app.GetCurrentAppAttempt(); NUnit.Framework.Assert.IsFalse(attempt2.GetAppAttemptId().Equals(am0.GetApplicationAttemptId ())); // launch the new AM nm.NodeHeartbeat(true); MockAM am1 = rm.SendAMLaunched(attempt2.GetAppAttemptId()); am1.RegisterAppAttempt(); // allocate NUM_CONTAINERS containers am1.Allocate("127.0.0.1", 1024, NumContainers, new AList <ContainerId>()); nm.NodeHeartbeat(true); // wait for containers to be allocated. containers = am1.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()) .GetAllocatedContainers(); while (containers.Count != NumContainers) { nm.NodeHeartbeat(true); Sharpen.Collections.AddAll(containers, am1.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers()); Sharpen.Thread.Sleep(200); } rm.WaitForState(app.GetApplicationId(), RMAppState.Running); // Capture running containers for later use by metrics calculations. rmContainers = rm.scheduler.GetSchedulerAppInfo(attempt2.GetAppAttemptId()).GetLiveContainers (); // complete container by sending the container complete event which has // earlier attempt's attemptId amContainerId = app.GetCurrentAppAttempt().GetMasterContainer().GetId(); nm.NodeHeartbeat(am0.GetApplicationAttemptId(), amContainerId.GetContainerId(), ContainerState .Complete); MockRM.FinishAMAndVerifyAppState(app, rm, nm, am1); // Calculate container usage metrics for second attempt. foreach (RMContainer c_1 in rmContainers) { AggregateAppResourceUsage ru = CalculateContainerResourceMetrics(c_1); memorySeconds += ru.GetMemorySeconds(); vcoreSeconds += ru.GetVcoreSeconds(); } RMAppMetrics rmAppMetrics = app.GetRMAppMetrics(); NUnit.Framework.Assert.AreEqual("Unexcpected MemorySeconds value", memorySeconds, rmAppMetrics.GetMemorySeconds()); NUnit.Framework.Assert.AreEqual("Unexpected VcoreSeconds value", vcoreSeconds, rmAppMetrics .GetVcoreSeconds()); rm.Stop(); return; }
public virtual void TestAppCleanup() { Logger rootLogger = LogManager.GetRootLogger(); rootLogger.SetLevel(Level.Debug); MockRM rm = new MockRM(); rm.Start(); MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 5000); RMApp app = rm.SubmitApp(2000); //kick the scheduling nm1.NodeHeartbeat(true); RMAppAttempt attempt = app.GetCurrentAppAttempt(); MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId()); am.RegisterAppAttempt(); //request for containers int request = 2; am.Allocate("127.0.0.1", 1000, request, new AList <ContainerId>()); //kick the scheduler nm1.NodeHeartbeat(true); IList <Container> conts = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId >()).GetAllocatedContainers(); int contReceived = conts.Count; int waitCount = 0; while (contReceived < request && waitCount++ < 200) { Log.Info("Got " + contReceived + " containers. Waiting to get " + request); Sharpen.Thread.Sleep(100); conts = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers (); contReceived += conts.Count; nm1.NodeHeartbeat(true); } NUnit.Framework.Assert.AreEqual(request, contReceived); am.UnregisterAppAttempt(); NodeHeartbeatResponse resp = nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState .Complete); am.WaitForState(RMAppAttemptState.Finished); //currently only containers are cleaned via this //AM container is cleaned via container launcher resp = nm1.NodeHeartbeat(true); IList <ContainerId> containersToCleanup = resp.GetContainersToCleanup(); IList <ApplicationId> appsToCleanup = resp.GetApplicationsToCleanup(); int numCleanedContainers = containersToCleanup.Count; int numCleanedApps = appsToCleanup.Count; waitCount = 0; while ((numCleanedContainers < 2 || numCleanedApps < 1) && waitCount++ < 200) { Log.Info("Waiting to get cleanup events.. cleanedConts: " + numCleanedContainers + " cleanedApps: " + numCleanedApps); Sharpen.Thread.Sleep(100); resp = nm1.NodeHeartbeat(true); IList <ContainerId> deltaContainersToCleanup = resp.GetContainersToCleanup(); IList <ApplicationId> deltaAppsToCleanup = resp.GetApplicationsToCleanup(); // Add the deltas to the global list Sharpen.Collections.AddAll(containersToCleanup, deltaContainersToCleanup); Sharpen.Collections.AddAll(appsToCleanup, deltaAppsToCleanup); // Update counts now numCleanedContainers = containersToCleanup.Count; numCleanedApps = appsToCleanup.Count; } NUnit.Framework.Assert.AreEqual(1, appsToCleanup.Count); NUnit.Framework.Assert.AreEqual(app.GetApplicationId(), appsToCleanup[0]); NUnit.Framework.Assert.AreEqual(1, numCleanedApps); NUnit.Framework.Assert.AreEqual(2, numCleanedContainers); rm.Stop(); }
/// <exception cref="System.Exception"/> public virtual void TestallocateBeforeAMRegistration() { Logger rootLogger = LogManager.GetRootLogger(); bool thrown = false; rootLogger.SetLevel(Level.Debug); MockRM rm = new MockRM(); rm.Start(); MockNM nm1 = rm.RegisterNode("h1:1234", 5000); RMApp app = rm.SubmitApp(2000); // kick the scheduling nm1.NodeHeartbeat(true); RMAppAttempt attempt = app.GetCurrentAppAttempt(); MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId()); // request for containers int request = 2; AllocateResponse ar = null; try { ar = am.Allocate("h1", 1000, request, new AList <ContainerId>()); NUnit.Framework.Assert.Fail(); } catch (ApplicationMasterNotRegisteredException) { } // kick the scheduler nm1.NodeHeartbeat(true); AllocateResponse amrs = null; try { amrs = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()); NUnit.Framework.Assert.Fail(); } catch (ApplicationMasterNotRegisteredException) { } am.RegisterAppAttempt(); try { am.RegisterAppAttempt(false); NUnit.Framework.Assert.Fail(); } catch (Exception e) { NUnit.Framework.Assert.AreEqual("Application Master is already registered : " + attempt .GetAppAttemptId().GetApplicationId(), e.Message); } // Simulate an AM that was disconnected and app attempt was removed // (responseMap does not contain attemptid) am.UnregisterAppAttempt(); nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState.Complete); am.WaitForState(RMAppAttemptState.Finished); try { amrs = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()); NUnit.Framework.Assert.Fail(); } catch (ApplicationAttemptNotFoundException) { } }
/// <exception cref="System.Exception"/> public virtual void TestNMToken() { MockRM rm = new MockRM(conf); try { rm.Start(); MockNM nm1 = rm.RegisterNode("h1:1234", 10000); NMTokenSecretManagerInRM nmTokenSecretManager = rm.GetRMContext().GetNMTokenSecretManager (); // submitting new application RMApp app = rm.SubmitApp(1000); // start scheduling. nm1.NodeHeartbeat(true); // Starting application attempt and launching // It should get registered with NMTokenSecretManager. RMAppAttempt attempt = app.GetCurrentAppAttempt(); MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId()); NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptRegistered (attempt.GetAppAttemptId())); // This will register application master. am.RegisterAppAttempt(); AList <Container> containersReceivedForNM1 = new AList <Container>(); IList <ContainerId> releaseContainerList = new AList <ContainerId>(); Dictionary <string, Token> nmTokens = new Dictionary <string, Token>(); // initially requesting 2 containers. AllocateResponse response = am.Allocate("h1", 1000, 2, releaseContainerList); NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count); AllocateContainersAndValidateNMTokens(am, containersReceivedForNM1, 2, nmTokens, nm1); NUnit.Framework.Assert.AreEqual(1, nmTokens.Count); // requesting 2 more containers. response = am.Allocate("h1", 1000, 2, releaseContainerList); NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count); AllocateContainersAndValidateNMTokens(am, containersReceivedForNM1, 4, nmTokens, nm1); NUnit.Framework.Assert.AreEqual(1, nmTokens.Count); // We will be simulating NM restart so restarting newly added h2:1234 // NM 2 now registers. MockNM nm2 = rm.RegisterNode("h2:1234", 10000); nm2.NodeHeartbeat(true); AList <Container> containersReceivedForNM2 = new AList <Container>(); response = am.Allocate("h2", 1000, 2, releaseContainerList); NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count); AllocateContainersAndValidateNMTokens(am, containersReceivedForNM2, 2, nmTokens, nm2); NUnit.Framework.Assert.AreEqual(2, nmTokens.Count); // Simulating NM-2 restart. nm2 = rm.RegisterNode("h2:1234", 10000); // Wait for reconnect to make it through the RM and create a new RMNode IDictionary <NodeId, RMNode> nodes = rm.GetRMContext().GetRMNodes(); while (nodes[nm2.GetNodeId()].GetLastNodeHeartBeatResponse().GetResponseId() > 0) { Sharpen.Thread.Sleep(WaitSleepMs); } int interval = 40; // Wait for nm Token to be cleared. while (nmTokenSecretManager.IsApplicationAttemptNMTokenPresent(attempt.GetAppAttemptId (), nm2.GetNodeId()) && interval-- > 0) { Log.Info("waiting for nmToken to be cleared for : " + nm2.GetNodeId()); Sharpen.Thread.Sleep(WaitSleepMs); } NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptRegistered (attempt.GetAppAttemptId())); // removing NMToken for h2:1234 Sharpen.Collections.Remove(nmTokens, nm2.GetNodeId().ToString()); NUnit.Framework.Assert.AreEqual(1, nmTokens.Count); // We should again receive the NMToken. response = am.Allocate("h2", 1000, 2, releaseContainerList); NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count); AllocateContainersAndValidateNMTokens(am, containersReceivedForNM2, 4, nmTokens, nm2); NUnit.Framework.Assert.AreEqual(2, nmTokens.Count); // Now rolling over NMToken masterKey. it should resend the NMToken in // next allocate call. NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent (attempt.GetAppAttemptId(), nm1.GetNodeId())); NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent (attempt.GetAppAttemptId(), nm2.GetNodeId())); nmTokenSecretManager.RollMasterKey(); nmTokenSecretManager.ActivateNextMasterKey(); NUnit.Framework.Assert.IsFalse(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent (attempt.GetAppAttemptId(), nm1.GetNodeId())); NUnit.Framework.Assert.IsFalse(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent (attempt.GetAppAttemptId(), nm2.GetNodeId())); // It should not remove application attempt entry. NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptRegistered (attempt.GetAppAttemptId())); nmTokens.Clear(); NUnit.Framework.Assert.AreEqual(0, nmTokens.Count); // We should again receive the NMToken. response = am.Allocate("h2", 1000, 1, releaseContainerList); NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count); AllocateContainersAndValidateNMTokens(am, containersReceivedForNM2, 5, nmTokens, nm2); NUnit.Framework.Assert.AreEqual(1, nmTokens.Count); NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent (attempt.GetAppAttemptId(), nm2.GetNodeId())); // After AM is finished making sure that nmtoken entry for app NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptRegistered (attempt.GetAppAttemptId())); am.UnregisterAppAttempt(); // marking all the containers as finished. foreach (Container container in containersReceivedForNM1) { nm1.NodeHeartbeat(attempt.GetAppAttemptId(), container.GetId().GetContainerId(), ContainerState.Complete); } foreach (Container container_1 in containersReceivedForNM2) { nm2.NodeHeartbeat(attempt.GetAppAttemptId(), container_1.GetId().GetContainerId() , ContainerState.Complete); } nm1.NodeHeartbeat(am.GetApplicationAttemptId(), 1, ContainerState.Complete); am.WaitForState(RMAppAttemptState.Finished); NUnit.Framework.Assert.IsFalse(nmTokenSecretManager.IsApplicationAttemptRegistered (attempt.GetAppAttemptId())); } finally { rm.Stop(); } }