private static void AssertBlacklistAdditionsAndRemovals(int expectedAdditions, int
                                                         expectedRemovals, TestAMRMClientOnRMRestart.MyResourceManager rm)
 {
     NUnit.Framework.Assert.AreEqual(expectedAdditions, rm.GetMyFifoScheduler().lastBlacklistAdditions
                                     .Count);
     NUnit.Framework.Assert.AreEqual(expectedRemovals, rm.GetMyFifoScheduler().lastBlacklistRemovals
                                     .Count);
 }
 private static void AssertAsksAndReleases(int expectedAsk, int expectedRelease, TestAMRMClientOnRMRestart.MyResourceManager
                                           rm)
 {
     NUnit.Framework.Assert.AreEqual(expectedAsk, rm.GetMyFifoScheduler().lastAsk.Count
                                     );
     NUnit.Framework.Assert.AreEqual(expectedRelease, rm.GetMyFifoScheduler().lastRelease
                                     .Count);
 }
        // Test does major 6 steps verification.
        // Step-1 : AMRMClient send allocate request for 2 container requests
        // Step-2 : 2 containers are allocated by RM.
        // Step-3 : AM Send 1 containerRequest(cRequest3) and 1 releaseRequests to
        // RM
        // Step-4 : On RM restart, AM(does not know RM is restarted) sends additional
        // containerRequest(cRequest4) and blacklisted nodes.
        // Intern RM send resync command
        // Step-5 : Allocater after resync command & new containerRequest(cRequest5)
        // Step-6 : RM allocates containers i.e cRequest3,cRequest4 and cRequest5
        /// <exception cref="System.Exception"/>
        public virtual void TestAMRMClientResendsRequestsOnRMRestart()
        {
            UserGroupInformation.SetLoginUser(null);
            MemoryRMStateStore memStore = new MemoryRMStateStore();

            memStore.Init(conf);
            // Phase-1 Start 1st RM
            TestAMRMClientOnRMRestart.MyResourceManager rm1 = new TestAMRMClientOnRMRestart.MyResourceManager
                                                                  (conf, memStore);
            rm1.Start();
            DrainDispatcher dispatcher = (DrainDispatcher)rm1.GetRMContext().GetDispatcher();
            // Submit the application
            RMApp app = rm1.SubmitApp(1024);

            dispatcher.Await();
            MockNM nm1 = new MockNM("h1:1234", 15120, rm1.GetResourceTrackerService());

            nm1.RegisterNode();
            nm1.NodeHeartbeat(true);
            // Node heartbeat
            dispatcher.Await();
            ApplicationAttemptId appAttemptId = app.GetCurrentAppAttempt().GetAppAttemptId();

            rm1.SendAMLaunched(appAttemptId);
            dispatcher.Await();
            Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> token = rm1.GetRMContext
                                                                                     ().GetRMApps()[appAttemptId.GetApplicationId()].GetRMAppAttempt(appAttemptId).GetAMRMToken
                                                                                     ();
            UserGroupInformation ugi = UserGroupInformation.GetCurrentUser();

            ugi.AddTokenIdentifier(token.DecodeIdentifier());
            // Step-1 : AMRMClient send allocate request for 2 ContainerRequest
            // cRequest1 = h1 and cRequest2 = h1,h2
            // blacklisted nodes = h2
            AMRMClient <AMRMClient.ContainerRequest> amClient = new TestAMRMClientOnRMRestart.MyAMRMClientImpl
                                                                    (rm1);

            amClient.Init(conf);
            amClient.Start();
            amClient.RegisterApplicationMaster("Host", 10000, string.Empty);
            AMRMClient.ContainerRequest cRequest1 = CreateReq(1, 1024, new string[] { "h1" });
            amClient.AddContainerRequest(cRequest1);
            AMRMClient.ContainerRequest cRequest2 = CreateReq(1, 1024, new string[] { "h1", "h2" });
            amClient.AddContainerRequest(cRequest2);
            IList <string> blacklistAdditions = new AList <string>();
            IList <string> blacklistRemoval   = new AList <string>();

            blacklistAdditions.AddItem("h2");
            blacklistRemoval.AddItem("h10");
            amClient.UpdateBlacklist(blacklistAdditions, blacklistRemoval);
            blacklistAdditions.Remove("h2");
            // remove from local list
            AllocateResponse allocateResponse = amClient.Allocate(0.1f);

            dispatcher.Await();
            NUnit.Framework.Assert.AreEqual("No of assignments must be 0", 0, allocateResponse
                                            .GetAllocatedContainers().Count);
            // Why 4 ask, why not 3 ask even h2 is blacklisted?
            // On blacklisting host,applicationmaster has to remove ask request from
            // remoterequest table.Here,test does not remove explicitely
            AssertAsksAndReleases(4, 0, rm1);
            AssertBlacklistAdditionsAndRemovals(1, 1, rm1);
            // Step-2 : NM heart beat is sent.
            // On 2nd AM allocate request, RM allocates 2 containers to AM
            nm1.NodeHeartbeat(true);
            // Node heartbeat
            dispatcher.Await();
            allocateResponse = amClient.Allocate(0.2f);
            dispatcher.Await();
            // 2 containers are allocated i.e for cRequest1 and cRequest2.
            NUnit.Framework.Assert.AreEqual("No of assignments must be 0", 2, allocateResponse
                                            .GetAllocatedContainers().Count);
            AssertAsksAndReleases(0, 0, rm1);
            AssertBlacklistAdditionsAndRemovals(0, 0, rm1);
            IList <Container> allocatedContainers = allocateResponse.GetAllocatedContainers();

            // removed allocated container requests
            amClient.RemoveContainerRequest(cRequest1);
            amClient.RemoveContainerRequest(cRequest2);
            allocateResponse = amClient.Allocate(0.2f);
            dispatcher.Await();
            NUnit.Framework.Assert.AreEqual("No of assignments must be 0", 0, allocateResponse
                                            .GetAllocatedContainers().Count);
            AssertAsksAndReleases(4, 0, rm1);
            AssertBlacklistAdditionsAndRemovals(0, 0, rm1);
            // Step-3 : Send 1 containerRequest and 1 releaseRequests to RM
            AMRMClient.ContainerRequest cRequest3 = CreateReq(1, 1024, new string[] { "h1" });
            amClient.AddContainerRequest(cRequest3);
            int pendingRelease         = 0;
            IEnumerator <Container> it = allocatedContainers.GetEnumerator();

            while (it.HasNext())
            {
                amClient.ReleaseAssignedContainer(it.Next().GetId());
                pendingRelease++;
                it.Remove();
                break;
            }
            // remove one container
            allocateResponse = amClient.Allocate(0.3f);
            dispatcher.Await();
            NUnit.Framework.Assert.AreEqual("No of assignments must be 0", 0, allocateResponse
                                            .GetAllocatedContainers().Count);
            AssertAsksAndReleases(3, pendingRelease, rm1);
            AssertBlacklistAdditionsAndRemovals(0, 0, rm1);
            int completedContainer = allocateResponse.GetCompletedContainersStatuses().Count;

            pendingRelease -= completedContainer;
            // Phase-2 start 2nd RM is up
            TestAMRMClientOnRMRestart.MyResourceManager rm2 = new TestAMRMClientOnRMRestart.MyResourceManager
                                                                  (conf, memStore);
            rm2.Start();
            nm1.SetResourceTrackerService(rm2.GetResourceTrackerService());
            ((TestAMRMClientOnRMRestart.MyAMRMClientImpl)amClient).UpdateRMProxy(rm2);
            dispatcher = (DrainDispatcher)rm2.GetRMContext().GetDispatcher();
            // NM should be rebooted on heartbeat, even first heartbeat for nm2
            NodeHeartbeatResponse hbResponse = nm1.NodeHeartbeat(true);

            NUnit.Framework.Assert.AreEqual(NodeAction.Resync, hbResponse.GetNodeAction());
            // new NM to represent NM re-register
            nm1 = new MockNM("h1:1234", 10240, rm2.GetResourceTrackerService());
            nm1.RegisterNode();
            nm1.NodeHeartbeat(true);
            dispatcher.Await();
            blacklistAdditions.AddItem("h3");
            amClient.UpdateBlacklist(blacklistAdditions, null);
            blacklistAdditions.Remove("h3");
            it = allocatedContainers.GetEnumerator();
            while (it.HasNext())
            {
                amClient.ReleaseAssignedContainer(it.Next().GetId());
                pendingRelease++;
                it.Remove();
            }
            AMRMClient.ContainerRequest cRequest4 = CreateReq(1, 1024, new string[] { "h1", "h2" });
            amClient.AddContainerRequest(cRequest4);
            // Step-4 : On RM restart, AM(does not know RM is restarted) sends
            // additional
            // containerRequest and blacklisted nodes.
            // Intern RM send resync command,AMRMClient resend allocate request
            allocateResponse = amClient.Allocate(0.3f);
            dispatcher.Await();
            completedContainer = allocateResponse.GetCompletedContainersStatuses().Count;
            pendingRelease    -= completedContainer;
            AssertAsksAndReleases(4, pendingRelease, rm2);
            AssertBlacklistAdditionsAndRemovals(2, 0, rm2);
            AMRMClient.ContainerRequest cRequest5 = CreateReq(1, 1024, new string[] { "h1", "h2"
                                                                                      , "h3" });
            amClient.AddContainerRequest(cRequest5);
            // Step-5 : Allocater after resync command
            allocateResponse = amClient.Allocate(0.5f);
            dispatcher.Await();
            NUnit.Framework.Assert.AreEqual("No of assignments must be 0", 0, allocateResponse
                                            .GetAllocatedContainers().Count);
            AssertAsksAndReleases(5, 0, rm2);
            AssertBlacklistAdditionsAndRemovals(0, 0, rm2);
            int noAssignedContainer = 0;
            int count = 5;

            while (count-- > 0)
            {
                nm1.NodeHeartbeat(true);
                dispatcher.Await();
                allocateResponse = amClient.Allocate(0.5f);
                dispatcher.Await();
                noAssignedContainer += allocateResponse.GetAllocatedContainers().Count;
                if (noAssignedContainer == 3)
                {
                    break;
                }
                Sharpen.Thread.Sleep(1000);
            }
            // Step-6 : RM allocates containers i.e cRequest3,cRequest4 and cRequest5
            NUnit.Framework.Assert.AreEqual("Number of container should be 3", 3, noAssignedContainer
                                            );
            amClient.Stop();
            rm1.Stop();
            rm2.Stop();
        }
 public virtual void UpdateRMProxy(TestAMRMClientOnRMRestart.MyResourceManager rm)
 {
     rmClient = rm.GetApplicationMasterService();
 }
 public MyAMRMClientImpl(TestAMRMClientOnRMRestart.MyResourceManager rm)
 {
     this.rm = rm;
 }
        // Test verify for
        // 1. AM try to unregister without registering
        // 2. AM register to RM, and try to unregister immediately after RM restart
        /// <exception cref="System.Exception"/>
        public virtual void TestAMRMClientForUnregisterAMOnRMRestart()
        {
            MemoryRMStateStore memStore = new MemoryRMStateStore();

            memStore.Init(conf);
            // Phase-1 Start 1st RM
            TestAMRMClientOnRMRestart.MyResourceManager rm1 = new TestAMRMClientOnRMRestart.MyResourceManager
                                                                  (conf, memStore);
            rm1.Start();
            DrainDispatcher dispatcher = (DrainDispatcher)rm1.GetRMContext().GetDispatcher();
            // Submit the application
            RMApp app = rm1.SubmitApp(1024);

            dispatcher.Await();
            MockNM nm1 = new MockNM("h1:1234", 15120, rm1.GetResourceTrackerService());

            nm1.RegisterNode();
            nm1.NodeHeartbeat(true);
            // Node heartbeat
            dispatcher.Await();
            ApplicationAttemptId appAttemptId = app.GetCurrentAppAttempt().GetAppAttemptId();

            rm1.SendAMLaunched(appAttemptId);
            dispatcher.Await();
            Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> token = rm1.GetRMContext
                                                                                     ().GetRMApps()[appAttemptId.GetApplicationId()].GetRMAppAttempt(appAttemptId).GetAMRMToken
                                                                                     ();
            UserGroupInformation ugi = UserGroupInformation.GetCurrentUser();

            ugi.AddTokenIdentifier(token.DecodeIdentifier());
            AMRMClient <AMRMClient.ContainerRequest> amClient = new TestAMRMClientOnRMRestart.MyAMRMClientImpl
                                                                    (rm1);

            amClient.Init(conf);
            amClient.Start();
            amClient.RegisterApplicationMaster("h1", 10000, string.Empty);
            amClient.Allocate(0.1f);
            // Phase-2 start 2nd RM is up
            TestAMRMClientOnRMRestart.MyResourceManager rm2 = new TestAMRMClientOnRMRestart.MyResourceManager
                                                                  (conf, memStore);
            rm2.Start();
            nm1.SetResourceTrackerService(rm2.GetResourceTrackerService());
            ((TestAMRMClientOnRMRestart.MyAMRMClientImpl)amClient).UpdateRMProxy(rm2);
            dispatcher = (DrainDispatcher)rm2.GetRMContext().GetDispatcher();
            // NM should be rebooted on heartbeat, even first heartbeat for nm2
            NodeHeartbeatResponse hbResponse = nm1.NodeHeartbeat(true);

            NUnit.Framework.Assert.AreEqual(NodeAction.Resync, hbResponse.GetNodeAction());
            // new NM to represent NM re-register
            nm1 = new MockNM("h1:1234", 10240, rm2.GetResourceTrackerService());
            ContainerId       containerId     = ContainerId.NewContainerId(appAttemptId, 1);
            NMContainerStatus containerReport = NMContainerStatus.NewInstance(containerId, ContainerState
                                                                              .Running, Resource.NewInstance(1024, 1), "recover container", 0, Priority.NewInstance
                                                                                  (0), 0);

            nm1.RegisterNode(Arrays.AsList(containerReport), null);
            nm1.NodeHeartbeat(true);
            dispatcher.Await();
            amClient.UnregisterApplicationMaster(FinalApplicationStatus.Succeeded, null, null
                                                 );
            rm2.WaitForState(appAttemptId, RMAppAttemptState.Finishing);
            nm1.NodeHeartbeat(appAttemptId, 1, ContainerState.Complete);
            rm2.WaitForState(appAttemptId, RMAppAttemptState.Finished);
            rm2.WaitForState(app.GetApplicationId(), RMAppState.Finished);
            amClient.Stop();
            rm1.Stop();
            rm2.Stop();
        }