public virtual void TestUpdateHeartbeatResponseForCleanup()
{
RMNodeImpl node = GetRunningNode();
NodeId nodeId = node.GetNodeID();
// Expire a container
ContainerId completedContainerId = BuilderUtils.NewContainerId(BuilderUtils.NewApplicationAttemptId
(BuilderUtils.NewApplicationId(0, 0), 0), 0);
node.Handle(new RMNodeCleanContainerEvent(nodeId, completedContainerId));
NUnit.Framework.Assert.AreEqual(1, node.GetContainersToCleanUp().Count);
// Finish an application
ApplicationId finishedAppId = BuilderUtils.NewApplicationId(0, 1);
node.Handle(new RMNodeCleanAppEvent(nodeId, finishedAppId));
NUnit.Framework.Assert.AreEqual(1, node.GetAppsToCleanup().Count);
// Verify status update does not clear containers/apps to cleanup
// but updating heartbeat response for cleanup does
RMNodeStatusEvent statusEvent = GetMockRMNodeStatusEvent();
node.Handle(statusEvent);
NUnit.Framework.Assert.AreEqual(1, node.GetContainersToCleanUp().Count);
NUnit.Framework.Assert.AreEqual(1, node.GetAppsToCleanup().Count);
NodeHeartbeatResponse hbrsp = Org.Apache.Hadoop.Yarn.Util.Records.NewRecord <NodeHeartbeatResponse
>();
node.UpdateNodeHeartbeatResponseForCleanup(hbrsp);
NUnit.Framework.Assert.AreEqual(0, node.GetContainersToCleanUp().Count);
NUnit.Framework.Assert.AreEqual(0, node.GetAppsToCleanup().Count);
NUnit.Framework.Assert.AreEqual(1, hbrsp.GetContainersToCleanup().Count);
NUnit.Framework.Assert.AreEqual(completedContainerId, hbrsp.GetContainersToCleanup
()[0]);
NUnit.Framework.Assert.AreEqual(1, hbrsp.GetApplicationsToCleanup().Count);
NUnit.Framework.Assert.AreEqual(finishedAppId, hbrsp.GetApplicationsToCleanup()[0
]);
}
/// <exception cref="System.Exception"/>
private void WaitForAppCleanupMessageRecved(MockNM nm, ApplicationId appId)
{
while (true)
{
NodeHeartbeatResponse response = nm.NodeHeartbeat(true);
if (response.GetApplicationsToCleanup() != null && response.GetApplicationsToCleanup
().Count == 1 && appId.Equals(response.GetApplicationsToCleanup()[0]))
{
return;
}
Log.Info("Haven't got application=" + appId.ToString() + " in cleanup list from node heartbeat response, "
+ "sleep for a while before next heartbeat");
Sharpen.Thread.Sleep(1000);
}
}
public virtual void TestAppCleanup()
{
Logger rootLogger = LogManager.GetRootLogger();
rootLogger.SetLevel(Level.Debug);
MockRM rm = new MockRM();
rm.Start();
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 5000);
RMApp app = rm.SubmitApp(2000);
//kick the scheduling
nm1.NodeHeartbeat(true);
RMAppAttempt attempt = app.GetCurrentAppAttempt();
MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId());
am.RegisterAppAttempt();
//request for containers
int request = 2;
am.Allocate("127.0.0.1", 1000, request, new AList <ContainerId>());
//kick the scheduler
nm1.NodeHeartbeat(true);
IList <Container> conts = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId
>()).GetAllocatedContainers();
int contReceived = conts.Count;
int waitCount = 0;
while (contReceived < request && waitCount++ < 200)
{
Log.Info("Got " + contReceived + " containers. Waiting to get " + request);
Sharpen.Thread.Sleep(100);
conts = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers
();
contReceived += conts.Count;
nm1.NodeHeartbeat(true);
}
NUnit.Framework.Assert.AreEqual(request, contReceived);
am.UnregisterAppAttempt();
NodeHeartbeatResponse resp = nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState
.Complete);
am.WaitForState(RMAppAttemptState.Finished);
//currently only containers are cleaned via this
//AM container is cleaned via container launcher
resp = nm1.NodeHeartbeat(true);
IList <ContainerId> containersToCleanup = resp.GetContainersToCleanup();
IList <ApplicationId> appsToCleanup = resp.GetApplicationsToCleanup();
int numCleanedContainers = containersToCleanup.Count;
int numCleanedApps = appsToCleanup.Count;
waitCount = 0;
while ((numCleanedContainers < 2 || numCleanedApps < 1) && waitCount++ < 200)
{
Log.Info("Waiting to get cleanup events.. cleanedConts: " + numCleanedContainers
+ " cleanedApps: " + numCleanedApps);
Sharpen.Thread.Sleep(100);
resp = nm1.NodeHeartbeat(true);
IList <ContainerId> deltaContainersToCleanup = resp.GetContainersToCleanup();
IList <ApplicationId> deltaAppsToCleanup = resp.GetApplicationsToCleanup();
// Add the deltas to the global list
Sharpen.Collections.AddAll(containersToCleanup, deltaContainersToCleanup);
Sharpen.Collections.AddAll(appsToCleanup, deltaAppsToCleanup);
// Update counts now
numCleanedContainers = containersToCleanup.Count;
numCleanedApps = appsToCleanup.Count;
}
NUnit.Framework.Assert.AreEqual(1, appsToCleanup.Count);
NUnit.Framework.Assert.AreEqual(app.GetApplicationId(), appsToCleanup[0]);
NUnit.Framework.Assert.AreEqual(1, numCleanedApps);
NUnit.Framework.Assert.AreEqual(2, numCleanedContainers);
rm.Stop();
}
public void Run()
{
int lastHeartBeatID = 0;
while (!this._enclosing.isStopped)
{
try
{
NodeHeartbeatResponse response = null;
NodeStatus nodeStatus = this._enclosing.GetNodeStatus(lastHeartBeatID);
NodeHeartbeatRequest request = NodeHeartbeatRequest.NewInstance(nodeStatus, this.
_enclosing.context.GetContainerTokenSecretManager().GetCurrentKey(), this._enclosing
.context.GetNMTokenSecretManager().GetCurrentKey());
response = this._enclosing.resourceTracker.NodeHeartbeat(request);
this._enclosing.nextHeartBeatInterval = response.GetNextHeartBeatInterval();
this.UpdateMasterKeys(response);
if (response.GetNodeAction() == NodeAction.Shutdown)
{
Org.Apache.Hadoop.Yarn.Server.Nodemanager.NodeStatusUpdaterImpl.Log.Warn("Recieved SHUTDOWN signal from Resourcemanager as part of heartbeat,"
+ " hence shutting down.");
Org.Apache.Hadoop.Yarn.Server.Nodemanager.NodeStatusUpdaterImpl.Log.Warn("Message from ResourceManager: "
+ response.GetDiagnosticsMessage());
this._enclosing.context.SetDecommissioned(true);
this._enclosing.dispatcher.GetEventHandler().Handle(new NodeManagerEvent(NodeManagerEventType
.Shutdown));
break;
}
if (response.GetNodeAction() == NodeAction.Resync)
{
Org.Apache.Hadoop.Yarn.Server.Nodemanager.NodeStatusUpdaterImpl.Log.Warn("Node is out of sync with ResourceManager,"
+ " hence resyncing.");
Org.Apache.Hadoop.Yarn.Server.Nodemanager.NodeStatusUpdaterImpl.Log.Warn("Message from ResourceManager: "
+ response.GetDiagnosticsMessage());
this._enclosing.rmIdentifier = ResourceManagerConstants.RmInvalidIdentifier;
this._enclosing.dispatcher.GetEventHandler().Handle(new NodeManagerEvent(NodeManagerEventType
.Resync));
this._enclosing.pendingCompletedContainers.Clear();
break;
}
this._enclosing.RemoveOrTrackCompletedContainersFromContext(response.GetContainersToBeRemovedFromNM
());
lastHeartBeatID = response.GetResponseId();
IList <ContainerId> containersToCleanup = response.GetContainersToCleanup();
if (!containersToCleanup.IsEmpty())
{
this._enclosing.dispatcher.GetEventHandler().Handle(new CMgrCompletedContainersEvent
(containersToCleanup, CMgrCompletedContainersEvent.Reason.ByResourcemanager));
}
IList <ApplicationId> appsToCleanup = response.GetApplicationsToCleanup();
this._enclosing.TrackAppsForKeepAlive(appsToCleanup);
if (!appsToCleanup.IsEmpty())
{
this._enclosing.dispatcher.GetEventHandler().Handle(new CMgrCompletedAppsEvent(appsToCleanup
, CMgrCompletedAppsEvent.Reason.ByResourcemanager));
}
IDictionary <ApplicationId, ByteBuffer> systemCredentials = response.GetSystemCredentialsForApps
();
if (systemCredentials != null && !systemCredentials.IsEmpty())
{
((NodeManager.NMContext) this._enclosing.context).SetSystemCrendentialsForApps(Org.Apache.Hadoop.Yarn.Server.Nodemanager.NodeStatusUpdaterImpl
.ParseCredentials(systemCredentials));
}
}
catch (ConnectException e)
{
this._enclosing.dispatcher.GetEventHandler().Handle(new NodeManagerEvent(NodeManagerEventType
.Shutdown));
throw new YarnRuntimeException(e);
}
catch (Exception e)
{
Org.Apache.Hadoop.Yarn.Server.Nodemanager.NodeStatusUpdaterImpl.Log.Error("Caught exception in status-updater"
, e);
}
finally
{
lock (this._enclosing.heartbeatMonitor)
{
this._enclosing.nextHeartBeatInterval = this._enclosing.nextHeartBeatInterval <=
0 ? YarnConfiguration.DefaultRmNmHeartbeatIntervalMs : this._enclosing.nextHeartBeatInterval;
try
{
Sharpen.Runtime.Wait(this._enclosing.heartbeatMonitor, this._enclosing.nextHeartBeatInterval
);
}
catch (Exception)
{
}
}
}
}
}
