public virtual void Setup()
{
localFS.Delete(new Path(localDir.GetAbsolutePath()), true);
localFS.Delete(new Path(tmpDir.GetAbsolutePath()), true);
localFS.Delete(new Path(localLogDir.GetAbsolutePath()), true);
localFS.Delete(new Path(remoteLogDir.GetAbsolutePath()), true);
localDir.Mkdir();
tmpDir.Mkdir();
localLogDir.Mkdir();
remoteLogDir.Mkdir();
Log.Info("Created localDir in " + localDir.GetAbsolutePath());
Log.Info("Created tmpDir in " + tmpDir.GetAbsolutePath());
string bindAddress = "0.0.0.0:12345";
conf.Set(YarnConfiguration.NmAddress, bindAddress);
conf.Set(YarnConfiguration.NmLocalDirs, localDir.GetAbsolutePath());
conf.Set(YarnConfiguration.NmLogDirs, localLogDir.GetAbsolutePath());
conf.Set(YarnConfiguration.NmRemoteAppLogDir, remoteLogDir.GetAbsolutePath());
conf.SetLong(YarnConfiguration.NmLogRetainSeconds, 1);
// Default delSrvc
delSrvc = CreateDeletionService();
delSrvc.Init(conf);
exec = CreateContainerExecutor();
nodeHealthChecker = new NodeHealthCheckerService();
nodeHealthChecker.Init(conf);
dirsHandler = nodeHealthChecker.GetDiskHandler();
containerManager = CreateContainerManager(delSrvc);
((NodeManager.NMContext)context).SetContainerManager(containerManager);
nodeStatusUpdater.Init(conf);
containerManager.Init(conf);
nodeStatusUpdater.Start();
}
public _ContainersLauncher_148(DummyContainerManager _enclosing, Context baseArg1
, Dispatcher baseArg2, ContainerExecutor baseArg3, LocalDirsHandlerService baseArg4
, ContainerManagerImpl baseArg5)
: base(baseArg1, baseArg2, baseArg3, baseArg4, baseArg5)
{
this._enclosing = _enclosing;
}
/// <exception cref="System.Exception"/>
protected override void ServiceInit(Configuration conf)
{
conf.SetBoolean(Dispatcher.DispatcherExitOnErrorKey, true);
rmWorkPreservingRestartEnabled = conf.GetBoolean(YarnConfiguration.RmWorkPreservingRecoveryEnabled
, YarnConfiguration.DefaultRmWorkPreservingRecoveryEnabled);
InitAndStartRecoveryStore(conf);
NMContainerTokenSecretManager containerTokenSecretManager = new NMContainerTokenSecretManager
(conf, nmStore);
NMTokenSecretManagerInNM nmTokenSecretManager = new NMTokenSecretManagerInNM(nmStore
);
RecoverTokens(nmTokenSecretManager, containerTokenSecretManager);
this.aclsManager = new ApplicationACLsManager(conf);
ContainerExecutor exec = ReflectionUtils.NewInstance(conf.GetClass <ContainerExecutor
>(YarnConfiguration.NmContainerExecutor, typeof(DefaultContainerExecutor)), conf
);
try
{
exec.Init();
}
catch (IOException e)
{
throw new YarnRuntimeException("Failed to initialize container executor", e);
}
DeletionService del = CreateDeletionService(exec);
AddService(del);
// NodeManager level dispatcher
this.dispatcher = new AsyncDispatcher();
nodeHealthChecker = new NodeHealthCheckerService();
AddService(nodeHealthChecker);
dirsHandler = nodeHealthChecker.GetDiskHandler();
this.context = CreateNMContext(containerTokenSecretManager, nmTokenSecretManager,
nmStore);
nodeStatusUpdater = CreateNodeStatusUpdater(context, dispatcher, nodeHealthChecker
);
NodeResourceMonitor nodeResourceMonitor = CreateNodeResourceMonitor();
AddService(nodeResourceMonitor);
containerManager = CreateContainerManager(context, exec, del, nodeStatusUpdater,
this.aclsManager, dirsHandler);
AddService(containerManager);
((NodeManager.NMContext)context).SetContainerManager(containerManager);
WebServer webServer = CreateWebServer(context, containerManager.GetContainersMonitor
(), this.aclsManager, dirsHandler);
AddService(webServer);
((NodeManager.NMContext)context).SetWebServer(webServer);
dispatcher.Register(typeof(ContainerManagerEventType), containerManager);
dispatcher.Register(typeof(NodeManagerEventType), this);
AddService(dispatcher);
DefaultMetricsSystem.Initialize("NodeManager");
// StatusUpdater should be added last so that it get started last
// so that we make sure everything is up before registering with RM.
AddService(nodeStatusUpdater);
base.ServiceInit(conf);
}
public ContainersLauncher(Context context, Dispatcher dispatcher, ContainerExecutor
exec, LocalDirsHandlerService dirsHandler, ContainerManagerImpl containerManager
)
: base("containers-launcher")
{
this.exec = exec;
this.context = context;
this.dispatcher = dispatcher;
this.dirsHandler = dirsHandler;
this.containerManager = containerManager;
}
/// <exception cref="System.Exception"/>
protected internal virtual void WaitForNMToReceiveNMTokenKey(NMTokenSecretManagerInNM
nmTokenSecretManagerNM, NodeManager nm)
{
int attempt = 60;
ContainerManagerImpl cm = ((ContainerManagerImpl)nm.GetNMContext().GetContainerManager
());
while ((cm.GetBlockNewContainerRequestsStatus() || nmTokenSecretManagerNM.GetNodeId
() == null) && attempt-- > 0)
{
Sharpen.Thread.Sleep(2000);
}
}
/// <exception cref="System.Exception"/>
internal static void WaitForApplicationState(ContainerManagerImpl containerManager
, ApplicationId appID, ApplicationState finalState)
{
// Wait for app-finish
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app = containerManager.GetContext().GetApplications()[appID];
int timeout = 0;
while (!(app.GetApplicationState().Equals(finalState)) && timeout++ < 15)
{
Log.Info("Waiting for app to reach " + finalState + ".. Current state is " + app.
GetApplicationState());
Sharpen.Thread.Sleep(1000);
}
NUnit.Framework.Assert.IsTrue("App is not in " + finalState + " yet!! Timedout!!"
, app.GetApplicationState().Equals(finalState));
}
public ContainerLaunch(Context context, Configuration configuration, Dispatcher dispatcher
, ContainerExecutor exec, Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app, Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Container.Container
container, LocalDirsHandlerService dirsHandler, ContainerManagerImpl containerManager
)
{
this.context = context;
this.conf = configuration;
this.app = app;
this.exec = exec;
this.container = container;
this.dispatcher = dispatcher;
this.dirsHandler = dirsHandler;
this.containerManager = containerManager;
this.sleepDelayBeforeSigKill = conf.GetLong(YarnConfiguration.NmSleepDelayBeforeSigkillMs
, YarnConfiguration.DefaultNmSleepDelayBeforeSigkillMs);
this.maxKillWaitTime = conf.GetLong(YarnConfiguration.NmProcessKillWaitMs, YarnConfiguration
.DefaultNmProcessKillWaitMs);
}
/// <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)));
}
public virtual void TestApplicationRecovery()
{
YarnConfiguration conf = new YarnConfiguration();
conf.SetBoolean(YarnConfiguration.NmRecoveryEnabled, true);
conf.Set(YarnConfiguration.NmAddress, "localhost:1234");
conf.SetBoolean(YarnConfiguration.YarnAclEnable, true);
conf.Set(YarnConfiguration.YarnAdminAcl, "yarn_admin_user");
NMStateStoreService stateStore = new NMMemoryStateStoreService();
stateStore.Init(conf);
stateStore.Start();
Context context = new NodeManager.NMContext(new NMContainerTokenSecretManager(conf
), new NMTokenSecretManagerInNM(), null, new ApplicationACLsManager(conf), stateStore
);
ContainerManagerImpl cm = CreateContainerManager(context);
cm.Init(conf);
cm.Start();
// simulate registration with RM
MasterKey masterKey = new MasterKeyPBImpl();
masterKey.SetKeyId(123);
masterKey.SetBytes(ByteBuffer.Wrap(new byte[] { 123 }));
context.GetContainerTokenSecretManager().SetMasterKey(masterKey);
context.GetNMTokenSecretManager().SetMasterKey(masterKey);
// add an application by starting a container
string appUser = "******";
string modUser = "******";
string viewUser = "******";
string enemyUser = "******";
ApplicationId appId = ApplicationId.NewInstance(0, 1);
ApplicationAttemptId attemptId = ApplicationAttemptId.NewInstance(appId, 1);
ContainerId cid = ContainerId.NewContainerId(attemptId, 1);
IDictionary <string, LocalResource> localResources = Collections.EmptyMap();
IDictionary <string, string> containerEnv = Sharpen.Collections.EmptyMap();
IList <string> containerCmds = Sharpen.Collections.EmptyList();
IDictionary <string, ByteBuffer> serviceData = Sharpen.Collections.EmptyMap();
Credentials containerCreds = new Credentials();
DataOutputBuffer dob = new DataOutputBuffer();
containerCreds.WriteTokenStorageToStream(dob);
ByteBuffer containerTokens = ByteBuffer.Wrap(dob.GetData(), 0, dob.GetLength());
IDictionary <ApplicationAccessType, string> acls = new Dictionary <ApplicationAccessType
, string>();
acls[ApplicationAccessType.ModifyApp] = modUser;
acls[ApplicationAccessType.ViewApp] = viewUser;
ContainerLaunchContext clc = ContainerLaunchContext.NewInstance(localResources, containerEnv
, containerCmds, serviceData, containerTokens, acls);
// create the logAggregationContext
LogAggregationContext logAggregationContext = LogAggregationContext.NewInstance("includePattern"
, "excludePattern", "includePatternInRollingAggregation", "excludePatternInRollingAggregation"
);
StartContainersResponse startResponse = StartContainer(context, cm, cid, clc, logAggregationContext
);
NUnit.Framework.Assert.IsTrue(startResponse.GetFailedRequests().IsEmpty());
NUnit.Framework.Assert.AreEqual(1, context.GetApplications().Count);
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app = context.GetApplications()[appId];
NUnit.Framework.Assert.IsNotNull(app);
WaitForAppState(app, ApplicationState.Initing);
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(modUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ViewApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(enemyUser), ApplicationAccessType.ViewApp, appUser, appId));
// reset container manager and verify app recovered with proper acls
cm.Stop();
context = new NodeManager.NMContext(new NMContainerTokenSecretManager(conf), new
NMTokenSecretManagerInNM(), null, new ApplicationACLsManager(conf), stateStore);
cm = CreateContainerManager(context);
cm.Init(conf);
cm.Start();
NUnit.Framework.Assert.AreEqual(1, context.GetApplications().Count);
app = context.GetApplications()[appId];
NUnit.Framework.Assert.IsNotNull(app);
// check whether LogAggregationContext is recovered correctly
LogAggregationContext recovered = ((ApplicationImpl)app).GetLogAggregationContext
();
NUnit.Framework.Assert.IsNotNull(recovered);
NUnit.Framework.Assert.AreEqual(logAggregationContext.GetIncludePattern(), recovered
.GetIncludePattern());
NUnit.Framework.Assert.AreEqual(logAggregationContext.GetExcludePattern(), recovered
.GetExcludePattern());
NUnit.Framework.Assert.AreEqual(logAggregationContext.GetRolledLogsIncludePattern
(), recovered.GetRolledLogsIncludePattern());
NUnit.Framework.Assert.AreEqual(logAggregationContext.GetRolledLogsExcludePattern
(), recovered.GetRolledLogsExcludePattern());
WaitForAppState(app, ApplicationState.Initing);
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(modUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ViewApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(enemyUser), ApplicationAccessType.ViewApp, appUser, appId));
// simulate application completion
IList <ApplicationId> finishedApps = new AList <ApplicationId>();
finishedApps.AddItem(appId);
cm.Handle(new CMgrCompletedAppsEvent(finishedApps, CMgrCompletedAppsEvent.Reason.
ByResourcemanager));
WaitForAppState(app, ApplicationState.ApplicationResourcesCleaningup);
// restart and verify app is marked for finishing
cm.Stop();
context = new NodeManager.NMContext(new NMContainerTokenSecretManager(conf), new
NMTokenSecretManagerInNM(), null, new ApplicationACLsManager(conf), stateStore);
cm = CreateContainerManager(context);
cm.Init(conf);
cm.Start();
NUnit.Framework.Assert.AreEqual(1, context.GetApplications().Count);
app = context.GetApplications()[appId];
NUnit.Framework.Assert.IsNotNull(app);
WaitForAppState(app, ApplicationState.ApplicationResourcesCleaningup);
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(modUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ViewApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(enemyUser), ApplicationAccessType.ViewApp, appUser, appId));
// simulate log aggregation completion
app.Handle(new ApplicationEvent(app.GetAppId(), ApplicationEventType.ApplicationResourcesCleanedup
));
NUnit.Framework.Assert.AreEqual(app.GetApplicationState(), ApplicationState.Finished
);
app.Handle(new ApplicationEvent(app.GetAppId(), ApplicationEventType.ApplicationLogHandlingFinished
));
// restart and verify app is no longer present after recovery
cm.Stop();
context = new NodeManager.NMContext(new NMContainerTokenSecretManager(conf), new
NMTokenSecretManagerInNM(), null, new ApplicationACLsManager(conf), stateStore);
cm = CreateContainerManager(context);
cm.Init(conf);
cm.Start();
NUnit.Framework.Assert.IsTrue(context.GetApplications().IsEmpty());
cm.Stop();
}
public _ContainersLauncher_309(Context baseArg1, Dispatcher baseArg2, ContainerExecutor
baseArg3, LocalDirsHandlerService baseArg4, ContainerManagerImpl baseArg5)
: base(baseArg1, baseArg2, baseArg3, baseArg4, baseArg5)
{
}
public _PrivilegedExceptionAction_264(ContainerManagerImpl cm, IList <StartContainerRequest
> scReqList)
{
this.cm = cm;
this.scReqList = scReqList;
}
public RecoveredContainerLaunch(Context context, Configuration configuration, Dispatcher
dispatcher, ContainerExecutor exec, Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app, Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Container.Container
container, LocalDirsHandlerService dirsHandler, ContainerManagerImpl containerManager
)
: base(context, configuration, dispatcher, exec, app, container, dirsHandler, containerManager
)
{
this.shouldLaunchContainer.Set(true);
}
