/// <exception cref="System.IO.IOException"/>
private Dictionary <string, string> AddAppContainers(Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app)
{
Dispatcher dispatcher = new AsyncDispatcher();
ApplicationAttemptId appAttemptId = BuilderUtils.NewApplicationAttemptId(app.GetAppId
(), 1);
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Container.Container container1
= new MockContainer(appAttemptId, dispatcher, conf, app.GetUser(), app.GetAppId
(), 1);
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Container.Container container2
= new MockContainer(appAttemptId, dispatcher, conf, app.GetUser(), app.GetAppId
(), 2);
nmContext.GetContainers()[container1.GetContainerId()] = container1;
nmContext.GetContainers()[container2.GetContainerId()] = container2;
app.GetContainers()[container1.GetContainerId()] = container1;
app.GetContainers()[container2.GetContainerId()] = container2;
Dictionary <string, string> hash = new Dictionary <string, string>();
hash[container1.GetContainerId().ToString()] = container1.GetContainerId().ToString
();
hash[container2.GetContainerId().ToString()] = container2.GetContainerId().ToString
();
return(hash);
}
/// <exception cref="Org.Codehaus.Jettison.Json.JSONException"/>
/// <exception cref="System.Exception"/>
public virtual void VerifyNodeAppInfoGeneric(Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app, string id, string state, string user)
{
WebServicesTestUtils.CheckStringMatch("id", app.GetAppId().ToString(), id);
WebServicesTestUtils.CheckStringMatch("state", app.GetApplicationState().ToString
(), state);
WebServicesTestUtils.CheckStringMatch("user", app.GetUser().ToString(), user);
}
/// <exception cref="System.Exception"/>
private void WaitForAppState(Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app, ApplicationState state)
{
int msecPerSleep = 10;
int msecLeft = 5000;
while (app.GetApplicationState() != state && msecLeft > 0)
{
Sharpen.Thread.Sleep(msecPerSleep);
msecLeft -= msecPerSleep;
}
NUnit.Framework.Assert.AreEqual(state, app.GetApplicationState());
}
/// <exception cref="Org.Codehaus.Jettison.Json.JSONException"/>
/// <exception cref="System.Exception"/>
public virtual void VerifyNodeAppInfo(JSONObject info, Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app, Dictionary <string, string> hash)
{
NUnit.Framework.Assert.AreEqual("incorrect number of elements", 4, info.Length());
VerifyNodeAppInfoGeneric(app, info.GetString("id"), info.GetString("state"), info
.GetString("user"));
JSONArray containerids = info.GetJSONArray("containerids");
for (int i = 0; i < containerids.Length(); i++)
{
string id = containerids.GetString(i);
NUnit.Framework.Assert.AreEqual("extra containerid: " + id, id, Sharpen.Collections.Remove
(hash, id));
}
NUnit.Framework.Assert.IsTrue("missing containerids", hash.IsEmpty());
}
/// <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 virtual AppInfo GetNodeApp(string appId)
{
Init();
ApplicationId id = ConverterUtils.ToApplicationId(recordFactory, appId);
if (id == null)
{
throw new NotFoundException("app with id " + appId + " not found");
}
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app = this.nmContext.GetApplications()[id];
if (app == null)
{
throw new NotFoundException("app with id " + appId + " not found");
}
return(new AppInfo(app));
}
/// <exception cref="Org.Codehaus.Jettison.Json.JSONException"/>
/// <exception cref="System.Exception"/>
public virtual void VerifyNodeAppInfoXML(NodeList nodes, Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app, Dictionary <string, string> hash)
{
for (int i = 0; i < nodes.GetLength(); i++)
{
Element element = (Element)nodes.Item(i);
VerifyNodeAppInfoGeneric(app, WebServicesTestUtils.GetXmlString(element, "id"), WebServicesTestUtils
.GetXmlString(element, "state"), WebServicesTestUtils.GetXmlString(element, "user"
));
NodeList ids = element.GetElementsByTagName("containerids");
for (int j = 0; j < ids.GetLength(); j++)
{
Element line = (Element)ids.Item(j);
Node first = line.GetFirstChild();
string val = first.GetNodeValue();
NUnit.Framework.Assert.AreEqual("extra containerid: " + val, val, Sharpen.Collections.Remove
(hash, val));
}
NUnit.Framework.Assert.IsTrue("missing containerids", hash.IsEmpty());
}
}
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 override void Handle(LocalizationEvent @event)
{
switch (@event.GetType())
{
case LocalizationEventType.InitApplicationResources:
{
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app = ((ApplicationLocalizationEvent)@event).GetApplication();
// Simulate event from ApplicationLocalization.
this.dispatcher.GetEventHandler().Handle(new ApplicationInitedEvent(app.GetAppId(
)));
break;
}
case LocalizationEventType.InitContainerResources:
{
ContainerLocalizationRequestEvent rsrcReqs = (ContainerLocalizationRequestEvent)@event;
// simulate localization of all requested resources
foreach (ICollection <LocalResourceRequest> rc in rsrcReqs.GetRequestedResources()
.Values)
{
foreach (LocalResourceRequest req in rc)
{
Org.Apache.Hadoop.Yarn.Server.Nodemanager.DummyContainerManager.Log.Info("DEBUG: "
+ req + ":" + rsrcReqs.GetContainer().GetContainerId());
this.dispatcher.GetEventHandler().Handle(new ContainerResourceLocalizedEvent(rsrcReqs
.GetContainer().GetContainerId(), req, new Path("file:///local" + req.GetPath().
ToUri().GetPath())));
}
}
break;
}
case LocalizationEventType.CleanupContainerResources:
{
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Container.Container container
= ((ContainerLocalizationEvent)@event).GetContainer();
// TODO: delete the container dir
this.dispatcher.GetEventHandler().Handle(new ContainerEvent(container.GetContainerId
(), ContainerEventType.ContainerResourcesCleanedup));
break;
}
case LocalizationEventType.DestroyApplicationResources:
{
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
application = ((ApplicationLocalizationEvent)@event).GetApplication();
// decrement reference counts of all resources associated with this
// app
this.dispatcher.GetEventHandler().Handle(new ApplicationEvent(application.GetAppId
(), ApplicationEventType.ApplicationResourcesCleanedup));
break;
}
default:
{
NUnit.Framework.Assert.Fail("Unexpected event: " + @event.GetType());
break;
}
}
}