/// <exception cref="System.IO.IOException"/>
private NMTokenIdentifier GetNMTokenId(Org.Apache.Hadoop.Yarn.Api.Records.Token token
)
{
Org.Apache.Hadoop.Security.Token.Token <NMTokenIdentifier> convertedToken = ConverterUtils
.ConvertFromYarn(token, (Text)null);
return(convertedToken.DecodeIdentifier());
}
// To avoid using cached client
/// <exception cref="System.Exception"/>
public virtual void TestAMRMMasterKeysUpdate()
{
AtomicReference <AMRMTokenSecretManager> spySecretMgrRef = new AtomicReference <AMRMTokenSecretManager
>();
MockRM rm = new _MockRM_349(this, spySecretMgrRef, conf);
// Skip the login.
rm.Start();
MockNM nm = rm.RegisterNode("127.0.0.1:1234", 8000);
RMApp app = rm.SubmitApp(200);
MockAM am = MockRM.LaunchAndRegisterAM(app, rm, nm);
AMRMTokenSecretManager spySecretMgr = spySecretMgrRef.Get();
// Do allocate. Should not update AMRMToken
AllocateResponse response = am.Allocate(Org.Apache.Hadoop.Yarn.Util.Records.NewRecord
<AllocateRequest>());
NUnit.Framework.Assert.IsNull(response.GetAMRMToken());
Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> oldToken = rm.GetRMContext
().GetRMApps()[app.GetApplicationId()].GetRMAppAttempt(am.GetApplicationAttemptId
()).GetAMRMToken();
// roll over the master key
// Do allocate again. the AM should get the latest AMRMToken
rm.GetRMContext().GetAMRMTokenSecretManager().RollMasterKey();
response = am.Allocate(Org.Apache.Hadoop.Yarn.Util.Records.NewRecord <AllocateRequest
>());
NUnit.Framework.Assert.IsNotNull(response.GetAMRMToken());
Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> amrmToken = ConverterUtils
.ConvertFromYarn(response.GetAMRMToken(), new Text(response.GetAMRMToken().GetService
()));
NUnit.Framework.Assert.AreEqual(amrmToken.DecodeIdentifier().GetKeyId(), rm.GetRMContext
().GetAMRMTokenSecretManager().GetMasterKey().GetMasterKey().GetKeyId());
// Do allocate again with the same old token and verify the RM sends
// back the last generated token instead of generating it again.
Org.Mockito.Mockito.Reset(spySecretMgr);
UserGroupInformation ugi = UserGroupInformation.CreateUserForTesting(am.GetApplicationAttemptId
().ToString(), new string[0]);
ugi.AddTokenIdentifier(oldToken.DecodeIdentifier());
response = am.DoAllocateAs(ugi, Org.Apache.Hadoop.Yarn.Util.Records.NewRecord <AllocateRequest
>());
NUnit.Framework.Assert.IsNotNull(response.GetAMRMToken());
Org.Mockito.Mockito.Verify(spySecretMgr, Org.Mockito.Mockito.Never()).CreateAndGetAMRMToken
(Matchers.IsA <ApplicationAttemptId>());
// Do allocate again with the updated token and verify we do not
// receive a new token to use.
response = am.Allocate(Org.Apache.Hadoop.Yarn.Util.Records.NewRecord <AllocateRequest
>());
NUnit.Framework.Assert.IsNull(response.GetAMRMToken());
// Activate the next master key. Since there is new master key generated
// in AMRMTokenSecretManager. The AMRMToken will not get updated for AM
rm.GetRMContext().GetAMRMTokenSecretManager().ActivateNextMasterKey();
response = am.Allocate(Org.Apache.Hadoop.Yarn.Util.Records.NewRecord <AllocateRequest
>());
NUnit.Framework.Assert.IsNull(response.GetAMRMToken());
rm.Stop();
}
/// <summary>Populate persisted password of AMRMToken back to AMRMTokenSecretManager.
/// </summary>
/// <exception cref="System.IO.IOException"/>
public virtual void AddPersistedPassword(Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier
> token)
{
this.writeLock.Lock();
try
{
AMRMTokenIdentifier identifier = token.DecodeIdentifier();
Log.Debug("Adding password for " + identifier.GetApplicationAttemptId());
appAttemptSet.AddItem(identifier.GetApplicationAttemptId());
}
finally
{
this.writeLock.Unlock();
}
}
/// <summary>
/// Test if correct exception (StandbyException or RetriableException) can be
/// thrown during the NN failover.
/// </summary>
/// <exception cref="System.Exception"/>
public virtual void TestDelegationTokenDuringNNFailover()
{
EditLogTailer editLogTailer = nn1.GetNamesystem().GetEditLogTailer();
// stop the editLogTailer of nn1
editLogTailer.Stop();
Configuration conf = (Configuration)Whitebox.GetInternalState(editLogTailer, "conf"
);
nn1.GetNamesystem().SetEditLogTailerForTests(new TestDelegationTokensWithHA.EditLogTailerForTest
(this, nn1.GetNamesystem(), conf));
// create token
Org.Apache.Hadoop.Security.Token.Token <DelegationTokenIdentifier> token = GetDelegationToken
(fs, "JobTracker");
DelegationTokenIdentifier identifier = new DelegationTokenIdentifier();
byte[] tokenId = token.GetIdentifier();
identifier.ReadFields(new DataInputStream(new ByteArrayInputStream(tokenId)));
// Ensure that it's present in the nn0 secret manager and can
// be renewed directly from there.
Log.Info("A valid token should have non-null password, " + "and should be renewed successfully"
);
NUnit.Framework.Assert.IsTrue(null != dtSecretManager.RetrievePassword(identifier
));
dtSecretManager.RenewToken(token, "JobTracker");
// transition nn0 to standby
cluster.TransitionToStandby(0);
try
{
cluster.GetNameNodeRpc(0).RenewDelegationToken(token);
NUnit.Framework.Assert.Fail("StandbyException is expected since nn0 is in standby state"
);
}
catch (StandbyException e)
{
GenericTestUtils.AssertExceptionContains(HAServiceProtocol.HAServiceState.Standby
.ToString(), e);
}
new _Thread_220().Start();
Sharpen.Thread.Sleep(1000);
try
{
nn1.GetNamesystem().VerifyToken(token.DecodeIdentifier(), token.GetPassword());
NUnit.Framework.Assert.Fail("RetriableException/StandbyException is expected since nn1 is in transition"
);
}
catch (IOException e)
{
NUnit.Framework.Assert.IsTrue(e is StandbyException || e is RetriableException);
Log.Info("Got expected exception", e);
}
catchup = true;
lock (this)
{
Sharpen.Runtime.NotifyAll(this);
}
Configuration clientConf = dfs.GetConf();
DoRenewOrCancel(token, clientConf, TestDelegationTokensWithHA.TokenTestAction.Renew
);
DoRenewOrCancel(token, clientConf, TestDelegationTokensWithHA.TokenTestAction.Cancel
);
}
/// <exception cref="System.IO.IOException"/>
/// <exception cref="System.Exception"/>
private void VerifyNewVersionToken(Configuration conf, TestClientToAMTokens.CustomAM
am, Org.Apache.Hadoop.Security.Token.Token <ClientToAMTokenIdentifier> token, MockRM
rm)
{
UserGroupInformation ugi;
ugi = UserGroupInformation.CreateRemoteUser("me");
Org.Apache.Hadoop.Security.Token.Token <ClientToAMTokenIdentifier> newToken = new
Org.Apache.Hadoop.Security.Token.Token <ClientToAMTokenIdentifier>(new ClientToAMTokenIdentifierForTest
(token.DecodeIdentifier(), "message"), am.GetClientToAMTokenSecretManager());
newToken.SetService(token.GetService());
ugi.AddToken(newToken);
ugi.DoAs(new _PrivilegedExceptionAction_386(am, conf));
}
/// <summary>Decode the token identifier.</summary>
/// <remarks>
/// Decode the token identifier. The subclass can customize the way to decode
/// the token identifier.
/// </remarks>
/// <param name="token">the token where to extract the identifier</param>
/// <returns>the delegation token identifier</returns>
/// <exception cref="System.IO.IOException"/>
public virtual TokenIdent DecodeTokenIdentifier(Org.Apache.Hadoop.Security.Token.Token
<TokenIdent> token)
{
return(token.DecodeIdentifier());
}
// 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();
}
// Test verify for AM issued with rolled-over AMRMToken
// is still able to communicate with restarted RM.
/// <exception cref="System.Exception"/>
public virtual void TestAMRMClientOnAMRMTokenRollOverOnRMRestart()
{
conf.SetLong(YarnConfiguration.RmAmrmTokenMasterKeyRollingIntervalSecs, rolling_interval_sec
);
conf.SetLong(YarnConfiguration.RmAmExpiryIntervalMs, am_expire_ms);
MemoryRMStateStore memStore = new MemoryRMStateStore();
memStore.Init(conf);
// start first RM
TestAMRMClientOnRMRestart.MyResourceManager2 rm1 = new TestAMRMClientOnRMRestart.MyResourceManager2
(conf, memStore);
rm1.Start();
DrainDispatcher dispatcher = (DrainDispatcher)rm1.GetRMContext().GetDispatcher();
long startTime = Runtime.CurrentTimeMillis();
// 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();
AMRMTokenSecretManager amrmTokenSecretManagerForRM1 = rm1.GetRMContext().GetAMRMTokenSecretManager
();
Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> token = amrmTokenSecretManagerForRM1
.CreateAndGetAMRMToken(appAttemptId);
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);
// Wait for enough time and make sure the roll_over happens
// At mean time, the old AMRMToken should continue to work
while (Runtime.CurrentTimeMillis() - startTime < rolling_interval_sec * 1000)
{
amClient.Allocate(0.1f);
try
{
Sharpen.Thread.Sleep(1000);
}
catch (Exception)
{
}
}
// DO NOTHING
NUnit.Framework.Assert.IsTrue(amrmTokenSecretManagerForRM1.GetMasterKey().GetMasterKey
().GetKeyId() != token.DecodeIdentifier().GetKeyId());
amClient.Allocate(0.1f);
// active the nextMasterKey, and replace the currentMasterKey
Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> newToken = amrmTokenSecretManagerForRM1
.CreateAndGetAMRMToken(appAttemptId);
int waitCount = 0;
while (waitCount++ <= 50)
{
if (amrmTokenSecretManagerForRM1.GetCurrnetMasterKeyData().GetMasterKey().GetKeyId
() != token.DecodeIdentifier().GetKeyId())
{
break;
}
try
{
amClient.Allocate(0.1f);
}
catch (Exception)
{
break;
}
Sharpen.Thread.Sleep(500);
}
NUnit.Framework.Assert.IsTrue(amrmTokenSecretManagerForRM1.GetNextMasterKeyData()
== null);
NUnit.Framework.Assert.IsTrue(amrmTokenSecretManagerForRM1.GetCurrnetMasterKeyData
().GetMasterKey().GetKeyId() == newToken.DecodeIdentifier().GetKeyId());
// start 2nd RM
conf.Set(YarnConfiguration.RmSchedulerAddress, "0.0.0.0:9030");
TestAMRMClientOnRMRestart.MyResourceManager2 rm2 = new TestAMRMClientOnRMRestart.MyResourceManager2
(conf, memStore);
rm2.Start();
nm1.SetResourceTrackerService(rm2.GetResourceTrackerService());
((TestAMRMClientOnRMRestart.MyAMRMClientImpl)amClient).UpdateRMProxy(rm2);
dispatcher = (DrainDispatcher)rm2.GetRMContext().GetDispatcher();
AMRMTokenSecretManager amrmTokenSecretManagerForRM2 = rm2.GetRMContext().GetAMRMTokenSecretManager
();
NUnit.Framework.Assert.IsTrue(amrmTokenSecretManagerForRM2.GetCurrnetMasterKeyData
().GetMasterKey().GetKeyId() == newToken.DecodeIdentifier().GetKeyId());
NUnit.Framework.Assert.IsTrue(amrmTokenSecretManagerForRM2.GetNextMasterKeyData()
== null);
try
{
UserGroupInformation testUser = UserGroupInformation.CreateRemoteUser("testUser");
SecurityUtil.SetTokenService(token, rm2.GetApplicationMasterService().GetBindAddress
());
testUser.AddToken(token);
testUser.DoAs(new _PrivilegedAction_480(rm2)).Allocate(Org.Apache.Hadoop.Yarn.Util.Records
.NewRecord <AllocateRequest>());
NUnit.Framework.Assert.Fail("The old Token should not work");
}
catch (Exception ex)
{
NUnit.Framework.Assert.IsTrue(ex is SecretManager.InvalidToken);
NUnit.Framework.Assert.IsTrue(ex.Message.Contains("Invalid AMRMToken from " + token
.DecodeIdentifier().GetApplicationAttemptId()));
}
// make sure the recovered AMRMToken works for new RM
amClient.Allocate(0.1f);
amClient.UnregisterApplicationMaster(FinalApplicationStatus.Succeeded, null, null
);
amClient.Stop();
rm1.Stop();
rm2.Stop();
}
// 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();
}
