/// <exception cref="System.Exception"/>
public virtual RegisterApplicationMasterResponse RegisterAppAttempt(bool wait)
{
if (wait)
{
WaitForState(RMAppAttemptState.Launched);
}
responseId = 0;
RegisterApplicationMasterRequest req = Org.Apache.Hadoop.Yarn.Util.Records.NewRecord
<RegisterApplicationMasterRequest>();
req.SetHost(string.Empty);
req.SetRpcPort(1);
req.SetTrackingUrl(string.Empty);
if (ugi == null)
{
ugi = UserGroupInformation.CreateRemoteUser(attemptId.ToString());
Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> token = context.GetRMApps
()[attemptId.GetApplicationId()].GetRMAppAttempt(attemptId).GetAMRMToken();
ugi.AddTokenIdentifier(token.DecodeIdentifier());
}
try
{
return(ugi.DoAs(new _PrivilegedExceptionAction_117(this, req)));
}
catch (UndeclaredThrowableException e)
{
throw (Exception)e.InnerException;
}
}
// 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();
}
/// <exception cref="System.Exception"/>
public virtual AllocateResponse Allocate(AllocateRequest allocateRequest)
{
UserGroupInformation ugi = UserGroupInformation.CreateRemoteUser(attemptId.ToString
());
Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> token = context.GetRMApps
()[attemptId.GetApplicationId()].GetRMAppAttempt(attemptId).GetAMRMToken();
ugi.AddTokenIdentifier(token.DecodeIdentifier());
lastResponse = DoAllocateAs(ugi, allocateRequest);
return(lastResponse);
}
// do nothing
/// <exception cref="Org.Apache.Hadoop.Yarn.Exceptions.YarnException"/>
protected internal override UserGroupInformation GetRemoteUgi()
{
ApplicationId appId = ApplicationId.NewInstance(0, 0);
ApplicationAttemptId appAttemptId = ApplicationAttemptId.NewInstance(appId, 1);
UserGroupInformation ugi = UserGroupInformation.CreateRemoteUser(appAttemptId.ToString
());
ugi.AddTokenIdentifier(new NMTokenIdentifier(appAttemptId, this.context.GetNodeId
(), this._enclosing.user, this.context.GetNMTokenSecretManager().GetCurrentKey()
.GetKeyId()));
return(ugi);
}
/// <exception cref="System.Exception"/>
private AllocateResponse Allocate(ApplicationAttemptId attemptId, AllocateRequest
req)
{
UserGroupInformation ugi = UserGroupInformation.CreateRemoteUser(attemptId.ToString
());
Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> token = rm.GetRMContext
().GetRMApps()[attemptId.GetApplicationId()].GetRMAppAttempt(attemptId).GetAMRMToken
();
ugi.AddTokenIdentifier(token.DecodeIdentifier());
return(ugi.DoAs(new _PrivilegedExceptionAction_67(this, req)));
}
/// <exception cref="System.Exception"/>
public virtual void TestTokenIdentifiers()
{
UserGroupInformation ugi = UserGroupInformation.CreateUserForTesting("TheDoctor",
new string[] { "TheTARDIS" });
TokenIdentifier t1 = Org.Mockito.Mockito.Mock <TokenIdentifier>();
TokenIdentifier t2 = Org.Mockito.Mockito.Mock <TokenIdentifier>();
ugi.AddTokenIdentifier(t1);
ugi.AddTokenIdentifier(t2);
ICollection <TokenIdentifier> z = ugi.GetTokenIdentifiers();
Assert.True(z.Contains(t1));
Assert.True(z.Contains(t2));
Assert.Equal(2, z.Count);
// ensure that the token identifiers are passed through doAs
ICollection <TokenIdentifier> otherSet = ugi.DoAs(new _PrivilegedExceptionAction_639
());
Assert.True(otherSet.Contains(t1));
Assert.True(otherSet.Contains(t2));
Assert.Equal(2, otherSet.Count);
}
/// <exception cref="System.Exception"/>
private StartContainersResponse StartContainer(Context context, ContainerManagerImpl
cm, ContainerId cid, ContainerLaunchContext clc, LogAggregationContext logAggregationContext
)
{
UserGroupInformation user = UserGroupInformation.CreateRemoteUser(cid.GetApplicationAttemptId
().ToString());
StartContainerRequest scReq = StartContainerRequest.NewInstance(clc, TestContainerManager
.CreateContainerToken(cid, 0, context.GetNodeId(), user.GetShortUserName(), context
.GetContainerTokenSecretManager(), logAggregationContext));
IList <StartContainerRequest> scReqList = new AList <StartContainerRequest>();
scReqList.AddItem(scReq);
NMTokenIdentifier nmToken = new NMTokenIdentifier(cid.GetApplicationAttemptId(),
context.GetNodeId(), user.GetShortUserName(), context.GetNMTokenSecretManager().
GetCurrentKey().GetKeyId());
user.AddTokenIdentifier(nmToken);
return(user.DoAs(new _PrivilegedExceptionAction_264(cm, scReqList)));
}
/// <exception cref="System.Exception"/>
public virtual void UnregisterAppAttempt(FinishApplicationMasterRequest req, bool
waitForStateRunning)
{
if (waitForStateRunning)
{
WaitForState(RMAppAttemptState.Running);
}
if (ugi == null)
{
ugi = UserGroupInformation.CreateRemoteUser(attemptId.ToString());
Org.Apache.Hadoop.Security.Token.Token <AMRMTokenIdentifier> token = context.GetRMApps
()[attemptId.GetApplicationId()].GetRMAppAttempt(attemptId).GetAMRMToken();
ugi.AddTokenIdentifier(token.DecodeIdentifier());
}
try
{
ugi.DoAs(new _PrivilegedExceptionAction_276(this, req));
}
catch (UndeclaredThrowableException e)
{
throw (Exception)e.InnerException;
}
}
/// <exception cref="System.IO.IOException"/>
/// <exception cref="Org.Apache.Hadoop.Yarn.Exceptions.YarnException"/>
/// <exception cref="System.Exception"/>
public virtual void TestClearLocalDirWhenNodeReboot()
{
nm = new TestNodeManagerReboot.MyNodeManager(this);
nm.Start();
ContainerManagementProtocol containerManager = nm.GetContainerManager();
// create files under fileCache
CreateFiles(nmLocalDir.GetAbsolutePath(), ContainerLocalizer.Filecache, 100);
localResourceDir.Mkdirs();
ContainerLaunchContext containerLaunchContext = Records.NewRecord <ContainerLaunchContext
>();
// Construct the Container-id
ContainerId cId = CreateContainerId();
URL localResourceUri = ConverterUtils.GetYarnUrlFromPath(localFS.MakeQualified(new
Path(localResourceDir.GetAbsolutePath())));
LocalResource localResource = LocalResource.NewInstance(localResourceUri, LocalResourceType
.File, LocalResourceVisibility.Application, -1, localResourceDir.LastModified());
string destinationFile = "dest_file";
IDictionary <string, LocalResource> localResources = new Dictionary <string, LocalResource
>();
localResources[destinationFile] = localResource;
containerLaunchContext.SetLocalResources(localResources);
IList <string> commands = new AList <string>();
containerLaunchContext.SetCommands(commands);
NodeId nodeId = nm.GetNMContext().GetNodeId();
StartContainerRequest scRequest = StartContainerRequest.NewInstance(containerLaunchContext
, TestContainerManager.CreateContainerToken(cId, 0, nodeId, destinationFile, nm.
GetNMContext().GetContainerTokenSecretManager()));
IList <StartContainerRequest> list = new AList <StartContainerRequest>();
list.AddItem(scRequest);
StartContainersRequest allRequests = StartContainersRequest.NewInstance(list);
UserGroupInformation currentUser = UserGroupInformation.CreateRemoteUser(cId.GetApplicationAttemptId
().ToString());
NMTokenIdentifier nmIdentifier = new NMTokenIdentifier(cId.GetApplicationAttemptId
(), nodeId, user, 123);
currentUser.AddTokenIdentifier(nmIdentifier);
currentUser.DoAs(new _PrivilegedExceptionAction_152(this, allRequests));
IList <ContainerId> containerIds = new AList <ContainerId>();
containerIds.AddItem(cId);
GetContainerStatusesRequest request = GetContainerStatusesRequest.NewInstance(containerIds
);
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Container.Container container
= nm.GetNMContext().GetContainers()[request.GetContainerIds()[0]];
int MaxTries = 20;
int numTries = 0;
while (!container.GetContainerState().Equals(ContainerState.Done) && numTries <=
MaxTries)
{
try
{
Sharpen.Thread.Sleep(500);
}
catch (Exception)
{
}
// Do nothing
numTries++;
}
NUnit.Framework.Assert.AreEqual(ContainerState.Done, container.GetContainerState(
));
NUnit.Framework.Assert.IsTrue("The container should create a subDir named currentUser: "******"under localDir/usercache", NumOfLocalDirs(nmLocalDir.GetAbsolutePath(
), ContainerLocalizer.Usercache) > 0);
NUnit.Framework.Assert.IsTrue("There should be files or Dirs under nm_private when "
+ "container is launched", NumOfLocalDirs(nmLocalDir.GetAbsolutePath(), ResourceLocalizationService
.NmPrivateDir) > 0);
// restart the NodeManager
RestartNM(MaxTries);
CheckNumOfLocalDirs();
Org.Mockito.Mockito.Verify(delService, Org.Mockito.Mockito.Times(1)).Delete((string
)Matchers.IsNull(), Matchers.ArgThat(new TestNodeManagerReboot.PathInclude(this,
ResourceLocalizationService.NmPrivateDir + "_DEL_")));
Org.Mockito.Mockito.Verify(delService, Org.Mockito.Mockito.Times(1)).Delete((string
)Matchers.IsNull(), Matchers.ArgThat(new TestNodeManagerReboot.PathInclude(this,
ContainerLocalizer.Filecache + "_DEL_")));
Org.Mockito.Mockito.Verify(delService, Org.Mockito.Mockito.Times(1)).ScheduleFileDeletionTask
(Matchers.ArgThat(new TestNodeManagerReboot.FileDeletionInclude(this, user, null
, new string[] { destinationFile })));
Org.Mockito.Mockito.Verify(delService, Org.Mockito.Mockito.Times(1)).ScheduleFileDeletionTask
(Matchers.ArgThat(new TestNodeManagerReboot.FileDeletionInclude(this, null, ContainerLocalizer
.Usercache + "_DEL_", new string[] { })));
// restart the NodeManager again
// this time usercache directory should be empty
RestartNM(MaxTries);
CheckNumOfLocalDirs();
}
// 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();
}
