public virtual void TestDefaultSingleQueueMetrics()
{
string queueName = "single";
string user = "******";
QueueMetrics metrics = QueueMetrics.ForQueue(ms, queueName, null, false, conf);
MetricsSource queueSource = QueueSource(ms, queueName);
AppSchedulingInfo app = MockApp(user);
metrics.SubmitApp(user);
MetricsSource userSource = UserSource(ms, queueName, user);
CheckApps(queueSource, 1, 0, 0, 0, 0, 0, true);
metrics.SubmitAppAttempt(user);
CheckApps(queueSource, 1, 1, 0, 0, 0, 0, true);
metrics.SetAvailableResourcesToQueue(Resources.CreateResource(100 * Gb, 100));
metrics.IncrPendingResources(user, 5, Resources.CreateResource(3 * Gb, 3));
// Available resources is set externally, as it depends on dynamic
// configurable cluster/queue resources
CheckResources(queueSource, 0, 0, 0, 0, 0, 100 * Gb, 100, 15 * Gb, 15, 5, 0, 0, 0
);
metrics.RunAppAttempt(app.GetApplicationId(), user);
CheckApps(queueSource, 1, 0, 1, 0, 0, 0, true);
metrics.AllocateResources(user, 3, Resources.CreateResource(2 * Gb, 2), true);
CheckResources(queueSource, 6 * Gb, 6, 3, 3, 0, 100 * Gb, 100, 9 * Gb, 9, 2, 0, 0
, 0);
metrics.ReleaseResources(user, 1, Resources.CreateResource(2 * Gb, 2));
CheckResources(queueSource, 4 * Gb, 4, 2, 3, 1, 100 * Gb, 100, 9 * Gb, 9, 2, 0, 0
, 0);
metrics.FinishAppAttempt(app.GetApplicationId(), app.IsPending(), app.GetUser());
CheckApps(queueSource, 1, 0, 0, 0, 0, 0, true);
metrics.FinishApp(user, RMAppState.Finished);
CheckApps(queueSource, 1, 0, 0, 1, 0, 0, true);
NUnit.Framework.Assert.IsNull(userSource);
}
public virtual void Move(Queue newQueue)
{
lock (this)
{
QueueMetrics oldMetrics = queue.GetMetrics();
QueueMetrics newMetrics = newQueue.GetMetrics();
string user = GetUser();
foreach (RMContainer liveContainer in liveContainers.Values)
{
Org.Apache.Hadoop.Yarn.Api.Records.Resource resource = liveContainer.GetContainer
().GetResource();
oldMetrics.ReleaseResources(user, 1, resource);
newMetrics.AllocateResources(user, 1, resource, false);
}
foreach (IDictionary <NodeId, RMContainer> map in reservedContainers.Values)
{
foreach (RMContainer reservedContainer in map.Values)
{
Org.Apache.Hadoop.Yarn.Api.Records.Resource resource = reservedContainer.GetReservedResource
();
oldMetrics.UnreserveResource(user, resource);
newMetrics.ReserveResource(user, resource);
}
}
appSchedulingInfo.Move(newQueue);
this.queue = newQueue;
}
}
public virtual void TestMetricsCache()
{
MetricsSystem ms = new MetricsSystemImpl("cache");
ms.Start();
try
{
string p1 = "root1";
string leafQueueName = "root1.leaf";
QueueMetrics p1Metrics = QueueMetrics.ForQueue(ms, p1, null, true, conf);
Queue parentQueue1 = MockitoMaker.Make(MockitoMaker.Stub <Queue>().Returning(p1Metrics
).from.GetMetrics());
QueueMetrics metrics = QueueMetrics.ForQueue(ms, leafQueueName, parentQueue1, true
, conf);
NUnit.Framework.Assert.IsNotNull("QueueMetrics for A shoudn't be null", metrics);
// Re-register to check for cache hit, shouldn't blow up metrics-system...
// also, verify parent-metrics
QueueMetrics alterMetrics = QueueMetrics.ForQueue(ms, leafQueueName, parentQueue1
, true, conf);
NUnit.Framework.Assert.IsNotNull("QueueMetrics for alterMetrics shoudn't be null"
, alterMetrics);
}
finally
{
ms.Shutdown();
}
}
public virtual void Move(Queue newQueue)
{
lock (this)
{
QueueMetrics oldMetrics = queue.GetMetrics();
QueueMetrics newMetrics = newQueue.GetMetrics();
foreach (IDictionary <string, ResourceRequest> asks in requests.Values)
{
ResourceRequest request = asks[ResourceRequest.Any];
if (request != null)
{
oldMetrics.DecrPendingResources(user, request.GetNumContainers(), request.GetCapability
());
newMetrics.IncrPendingResources(user, request.GetNumContainers(), request.GetCapability
());
}
}
oldMetrics.MoveAppFrom(this);
newMetrics.MoveAppTo(this);
activeUsersManager.DeactivateApplication(user, applicationId);
activeUsersManager = newQueue.GetActiveUsersManager();
activeUsersManager.ActivateApplication(user, applicationId);
this.queue = newQueue;
this.queueName = newQueue.GetQueueName();
}
}
public static MetricsSource UserSource(MetricsSystem ms, string queue, string user
)
{
MetricsSource s = ms.GetSource(QueueMetrics.SourceName(queue).Append(",user=").Append
(user).ToString());
return(s);
}
private Queue CreateQueue(string name, Queue parent)
{
QueueMetrics metrics = QueueMetrics.ForQueue(name, parent, false, conf);
ActiveUsersManager activeUsersManager = new ActiveUsersManager(metrics);
Queue queue = Org.Mockito.Mockito.Mock <Queue>();
Org.Mockito.Mockito.When(queue.GetMetrics()).ThenReturn(metrics);
Org.Mockito.Mockito.When(queue.GetActiveUsersManager()).ThenReturn(activeUsersManager
);
return(queue);
}
public virtual void TestMove()
{
string user = "******";
Queue parentQueue = CreateQueue("parent", null);
Queue oldQueue = CreateQueue("old", parentQueue);
Queue newQueue = CreateQueue("new", parentQueue);
QueueMetrics parentMetrics = parentQueue.GetMetrics();
QueueMetrics oldMetrics = oldQueue.GetMetrics();
QueueMetrics newMetrics = newQueue.GetMetrics();
ApplicationAttemptId appAttId = CreateAppAttemptId(0, 0);
RMContext rmContext = Org.Mockito.Mockito.Mock <RMContext>();
Org.Mockito.Mockito.When(rmContext.GetEpoch()).ThenReturn(3L);
SchedulerApplicationAttempt app = new SchedulerApplicationAttempt(appAttId, user,
oldQueue, oldQueue.GetActiveUsersManager(), rmContext);
oldMetrics.SubmitApp(user);
// confirm that containerId is calculated based on epoch.
NUnit.Framework.Assert.AreEqual(unchecked ((long)(0x30000000001L)), app.GetNewContainerId
());
// Resource request
Resource requestedResource = Resource.NewInstance(1536, 2);
Priority requestedPriority = Priority.NewInstance(2);
ResourceRequest request = ResourceRequest.NewInstance(requestedPriority, ResourceRequest
.Any, requestedResource, 3);
app.UpdateResourceRequests(Arrays.AsList(request));
// Allocated container
RMContainer container1 = CreateRMContainer(appAttId, 1, requestedResource);
app.liveContainers[container1.GetContainerId()] = container1;
SchedulerNode node = CreateNode();
app.appSchedulingInfo.Allocate(NodeType.OffSwitch, node, requestedPriority, request
, container1.GetContainer());
// Reserved container
Priority prio1 = Priority.NewInstance(1);
Resource reservedResource = Resource.NewInstance(2048, 3);
RMContainer container2 = CreateReservedRMContainer(appAttId, 1, reservedResource,
node.GetNodeID(), prio1);
IDictionary <NodeId, RMContainer> reservations = new Dictionary <NodeId, RMContainer
>();
reservations[node.GetNodeID()] = container2;
app.reservedContainers[prio1] = reservations;
oldMetrics.ReserveResource(user, reservedResource);
CheckQueueMetrics(oldMetrics, 1, 1, 1536, 2, 2048, 3, 3072, 4);
CheckQueueMetrics(newMetrics, 0, 0, 0, 0, 0, 0, 0, 0);
CheckQueueMetrics(parentMetrics, 1, 1, 1536, 2, 2048, 3, 3072, 4);
app.Move(newQueue);
CheckQueueMetrics(oldMetrics, 0, 0, 0, 0, 0, 0, 0, 0);
CheckQueueMetrics(newMetrics, 1, 1, 1536, 2, 2048, 3, 3072, 4);
CheckQueueMetrics(parentMetrics, 1, 1, 1536, 2, 2048, 3, 3072, 4);
}
public virtual void TestMetricsInitializedOnRMInit()
{
YarnConfiguration conf = new YarnConfiguration();
conf.SetClass(YarnConfiguration.RmScheduler, typeof(FifoScheduler), typeof(ResourceScheduler
));
MockRM rm = new MockRM(conf);
QueueMetrics metrics = rm.GetResourceScheduler().GetRootQueueMetrics();
CheckApps(metrics, 0, 0, 0, 0, 0, 0, true);
MetricsAsserts.AssertGauge("ReservedContainers", 0, metrics);
}
private void CheckQueueMetrics(QueueMetrics metrics, int activeApps, int runningApps
, int allocMb, int allocVcores, int reservedMb, int reservedVcores, int pendingMb
, int pendingVcores)
{
NUnit.Framework.Assert.AreEqual(activeApps, metrics.GetActiveApps());
NUnit.Framework.Assert.AreEqual(runningApps, metrics.GetAppsRunning());
NUnit.Framework.Assert.AreEqual(allocMb, metrics.GetAllocatedMB());
NUnit.Framework.Assert.AreEqual(allocVcores, metrics.GetAllocatedVirtualCores());
NUnit.Framework.Assert.AreEqual(reservedMb, metrics.GetReservedMB());
NUnit.Framework.Assert.AreEqual(reservedVcores, metrics.GetReservedVirtualCores()
);
NUnit.Framework.Assert.AreEqual(pendingMb, metrics.GetPendingMB());
NUnit.Framework.Assert.AreEqual(pendingVcores, metrics.GetPendingVirtualCores());
}
public virtual void Stop(RMAppAttemptState rmAppAttemptFinalState)
{
lock (this)
{
// clear pending resources metrics for the application
QueueMetrics metrics = queue.GetMetrics();
foreach (IDictionary <string, ResourceRequest> asks in requests.Values)
{
ResourceRequest request = asks[ResourceRequest.Any];
if (request != null)
{
metrics.DecrPendingResources(user, request.GetNumContainers(), request.GetCapability
());
}
}
metrics.FinishAppAttempt(applicationId, pending, user);
// Clear requests themselves
ClearRequests();
}
}
public virtual void RecoverContainer(RMContainer rmContainer)
{
lock (this)
{
QueueMetrics metrics = queue.GetMetrics();
if (pending)
{
// If there was any container to recover, the application was
// running from scheduler's POV.
pending = false;
metrics.RunAppAttempt(applicationId, user);
}
// Container is completed. Skip recovering resources.
if (rmContainer.GetState().Equals(RMContainerState.Completed))
{
return;
}
metrics.AllocateResources(user, 1, rmContainer.GetAllocatedResource(), false);
}
}
public virtual void TestQueueAppMetricsForMultipleFailures()
{
string queueName = "single";
string user = "******";
QueueMetrics metrics = QueueMetrics.ForQueue(ms, queueName, null, false, new Configuration
());
MetricsSource queueSource = QueueSource(ms, queueName);
AppSchedulingInfo app = MockApp(user);
metrics.SubmitApp(user);
MetricsSource userSource = UserSource(ms, queueName, user);
CheckApps(queueSource, 1, 0, 0, 0, 0, 0, true);
metrics.SubmitAppAttempt(user);
CheckApps(queueSource, 1, 1, 0, 0, 0, 0, true);
metrics.RunAppAttempt(app.GetApplicationId(), user);
CheckApps(queueSource, 1, 0, 1, 0, 0, 0, true);
metrics.FinishAppAttempt(app.GetApplicationId(), app.IsPending(), app.GetUser());
CheckApps(queueSource, 1, 0, 0, 0, 0, 0, true);
// As the application has failed, framework retries the same application
// based on configuration
metrics.SubmitAppAttempt(user);
CheckApps(queueSource, 1, 1, 0, 0, 0, 0, true);
metrics.RunAppAttempt(app.GetApplicationId(), user);
CheckApps(queueSource, 1, 0, 1, 0, 0, 0, true);
// Suppose say application has failed this time as well.
metrics.FinishAppAttempt(app.GetApplicationId(), app.IsPending(), app.GetUser());
CheckApps(queueSource, 1, 0, 0, 0, 0, 0, true);
// As the application has failed, framework retries the same application
// based on configuration
metrics.SubmitAppAttempt(user);
CheckApps(queueSource, 1, 1, 0, 0, 0, 0, true);
metrics.RunAppAttempt(app.GetApplicationId(), user);
CheckApps(queueSource, 1, 0, 1, 0, 0, 0, true);
// Suppose say application has failed, and there's no more retries.
metrics.FinishAppAttempt(app.GetApplicationId(), app.IsPending(), app.GetUser());
CheckApps(queueSource, 1, 0, 0, 0, 0, 0, true);
metrics.FinishApp(user, RMAppState.Failed);
CheckApps(queueSource, 1, 0, 0, 0, 1, 0, true);
NUnit.Framework.Assert.IsNull(userSource);
}
public virtual void TestCollectAllMetrics()
{
string queueName = "single";
QueueMetrics.ForQueue(ms, queueName, null, false, conf);
MetricsSource queueSource = QueueSource(ms, queueName);
CheckApps(queueSource, 0, 0, 0, 0, 0, 0, true);
try
{
// do not collect all metrics
CheckApps(queueSource, 0, 0, 0, 0, 0, 0, false);
NUnit.Framework.Assert.Fail();
}
catch (Exception e)
{
NUnit.Framework.Assert.IsTrue(e.Message.Contains("Expected exactly one metric for name "
));
}
// collect all metrics
CheckApps(queueSource, 0, 0, 0, 0, 0, 0, true);
}
/// <summary>
/// Resources have been allocated to this application by the resource
/// scheduler.
/// </summary>
/// <remarks>
/// Resources have been allocated to this application by the resource
/// scheduler. Track them.
/// </remarks>
/// <param name="type">the type of the node</param>
/// <param name="node">the nodeinfo of the node</param>
/// <param name="priority">the priority of the request.</param>
/// <param name="request">the request</param>
/// <param name="container">the containers allocated.</param>
public virtual IList <ResourceRequest> Allocate(NodeType type, SchedulerNode node,
Priority priority, ResourceRequest request, Container container)
{
lock (this)
{
IList <ResourceRequest> resourceRequests = new AList <ResourceRequest>();
if (type == NodeType.NodeLocal)
{
AllocateNodeLocal(node, priority, request, container, resourceRequests);
}
else
{
if (type == NodeType.RackLocal)
{
AllocateRackLocal(node, priority, request, container, resourceRequests);
}
else
{
AllocateOffSwitch(node, priority, request, container, resourceRequests);
}
}
QueueMetrics metrics = queue.GetMetrics();
if (pending)
{
// once an allocation is done we assume the application is
// running from scheduler's POV.
pending = false;
metrics.RunAppAttempt(applicationId, user);
}
if (Log.IsDebugEnabled())
{
Log.Debug("allocate: applicationId=" + applicationId + " container=" + container.
GetId() + " host=" + container.GetNodeId().ToString() + " user="******" resource="
+ request.GetCapability());
}
metrics.AllocateResources(user, 1, request.GetCapability(), true);
return(resourceRequests);
}
}
public ActiveUsersManager(QueueMetrics metrics)
{
this.metrics = metrics;
}
public virtual void TestTwoLevelWithUserMetrics()
{
string parentQueueName = "root";
string leafQueueName = "root.leaf";
string user = "******";
QueueMetrics parentMetrics = QueueMetrics.ForQueue(ms, parentQueueName, null, true
, conf);
Queue parentQueue = MockitoMaker.Make(MockitoMaker.Stub <Queue>().Returning(parentMetrics
).from.GetMetrics());
QueueMetrics metrics = QueueMetrics.ForQueue(ms, leafQueueName, parentQueue, true
, conf);
MetricsSource parentQueueSource = QueueSource(ms, parentQueueName);
MetricsSource queueSource = QueueSource(ms, leafQueueName);
AppSchedulingInfo app = MockApp(user);
metrics.SubmitApp(user);
MetricsSource userSource = UserSource(ms, leafQueueName, user);
MetricsSource parentUserSource = UserSource(ms, parentQueueName, user);
CheckApps(queueSource, 1, 0, 0, 0, 0, 0, true);
CheckApps(parentQueueSource, 1, 0, 0, 0, 0, 0, true);
CheckApps(userSource, 1, 0, 0, 0, 0, 0, true);
CheckApps(parentUserSource, 1, 0, 0, 0, 0, 0, true);
metrics.SubmitAppAttempt(user);
CheckApps(queueSource, 1, 1, 0, 0, 0, 0, true);
CheckApps(parentQueueSource, 1, 1, 0, 0, 0, 0, true);
CheckApps(userSource, 1, 1, 0, 0, 0, 0, true);
CheckApps(parentUserSource, 1, 1, 0, 0, 0, 0, true);
parentMetrics.SetAvailableResourcesToQueue(Resources.CreateResource(100 * Gb, 100
));
metrics.SetAvailableResourcesToQueue(Resources.CreateResource(100 * Gb, 100));
parentMetrics.SetAvailableResourcesToUser(user, Resources.CreateResource(10 * Gb,
10));
metrics.SetAvailableResourcesToUser(user, Resources.CreateResource(10 * Gb, 10));
metrics.IncrPendingResources(user, 5, Resources.CreateResource(3 * Gb, 3));
CheckResources(queueSource, 0, 0, 0, 0, 0, 100 * Gb, 100, 15 * Gb, 15, 5, 0, 0, 0
);
CheckResources(parentQueueSource, 0, 0, 0, 0, 0, 100 * Gb, 100, 15 * Gb, 15, 5, 0
, 0, 0);
CheckResources(userSource, 0, 0, 0, 0, 0, 10 * Gb, 10, 15 * Gb, 15, 5, 0, 0, 0);
CheckResources(parentUserSource, 0, 0, 0, 0, 0, 10 * Gb, 10, 15 * Gb, 15, 5, 0, 0
, 0);
metrics.RunAppAttempt(app.GetApplicationId(), user);
CheckApps(queueSource, 1, 0, 1, 0, 0, 0, true);
CheckApps(userSource, 1, 0, 1, 0, 0, 0, true);
metrics.AllocateResources(user, 3, Resources.CreateResource(2 * Gb, 2), true);
metrics.ReserveResource(user, Resources.CreateResource(3 * Gb, 3));
// Available resources is set externally, as it depends on dynamic
// configurable cluster/queue resources
CheckResources(queueSource, 6 * Gb, 6, 3, 3, 0, 100 * Gb, 100, 9 * Gb, 9, 2, 3 *
Gb, 3, 1);
CheckResources(parentQueueSource, 6 * Gb, 6, 3, 3, 0, 100 * Gb, 100, 9 * Gb, 9, 2
, 3 * Gb, 3, 1);
CheckResources(userSource, 6 * Gb, 6, 3, 3, 0, 10 * Gb, 10, 9 * Gb, 9, 2, 3 * Gb,
3, 1);
CheckResources(parentUserSource, 6 * Gb, 6, 3, 3, 0, 10 * Gb, 10, 9 * Gb, 9, 2, 3
* Gb, 3, 1);
metrics.ReleaseResources(user, 1, Resources.CreateResource(2 * Gb, 2));
metrics.UnreserveResource(user, Resources.CreateResource(3 * Gb, 3));
CheckResources(queueSource, 4 * Gb, 4, 2, 3, 1, 100 * Gb, 100, 9 * Gb, 9, 2, 0, 0
, 0);
CheckResources(parentQueueSource, 4 * Gb, 4, 2, 3, 1, 100 * Gb, 100, 9 * Gb, 9, 2
, 0, 0, 0);
CheckResources(userSource, 4 * Gb, 4, 2, 3, 1, 10 * Gb, 10, 9 * Gb, 9, 2, 0, 0, 0
);
CheckResources(parentUserSource, 4 * Gb, 4, 2, 3, 1, 10 * Gb, 10, 9 * Gb, 9, 2, 0
, 0, 0);
metrics.FinishAppAttempt(app.GetApplicationId(), app.IsPending(), app.GetUser());
CheckApps(queueSource, 1, 0, 0, 0, 0, 0, true);
CheckApps(parentQueueSource, 1, 0, 0, 0, 0, 0, true);
CheckApps(userSource, 1, 0, 0, 0, 0, 0, true);
CheckApps(parentUserSource, 1, 0, 0, 0, 0, 0, true);
metrics.FinishApp(user, RMAppState.Finished);
CheckApps(queueSource, 1, 0, 0, 1, 0, 0, true);
CheckApps(parentQueueSource, 1, 0, 0, 1, 0, 0, true);
CheckApps(userSource, 1, 0, 0, 1, 0, 0, true);
CheckApps(parentUserSource, 1, 0, 0, 1, 0, 0, true);
}
public virtual void TearDown()
{
QueueMetrics.ClearQueueMetrics();
DefaultMetricsSystem.Shutdown();
}
public virtual void SetUp()
{
ms = new MetricsSystemImpl();
QueueMetrics.ClearQueueMetrics();
}
public static MetricsSource QueueSource(MetricsSystem ms, string queue)
{
MetricsSource s = ms.GetSource(QueueMetrics.SourceName(queue).ToString());
return(s);
}
/// <summary>
/// The ApplicationMaster is updating resource requirements for the
/// application, by asking for more resources and releasing resources acquired
/// by the application.
/// </summary>
/// <param name="requests">resources to be acquired</param>
/// <param name="recoverPreemptedRequest">recover Resource Request on preemption</param>
public virtual void UpdateResourceRequests(IList <ResourceRequest> requests, bool
recoverPreemptedRequest)
{
lock (this)
{
QueueMetrics metrics = queue.GetMetrics();
// Update resource requests
foreach (ResourceRequest request in requests)
{
Priority priority = request.GetPriority();
string resourceName = request.GetResourceName();
bool updatePendingResources = false;
ResourceRequest lastRequest = null;
if (resourceName.Equals(ResourceRequest.Any))
{
if (Log.IsDebugEnabled())
{
Log.Debug("update:" + " application=" + applicationId + " request=" + request);
}
updatePendingResources = true;
// Premature optimization?
// Assumes that we won't see more than one priority request updated
// in one call, reasonable assumption... however, it's totally safe
// to activate same application more than once.
// Thus we don't need another loop ala the one in decrementOutstanding()
// which is needed during deactivate.
if (request.GetNumContainers() > 0)
{
activeUsersManager.ActivateApplication(user, applicationId);
}
}
IDictionary <string, ResourceRequest> asks = this.requests[priority];
if (asks == null)
{
asks = new ConcurrentHashMap <string, ResourceRequest>();
this.requests[priority] = asks;
this.priorities.AddItem(priority);
}
lastRequest = asks[resourceName];
if (recoverPreemptedRequest && lastRequest != null)
{
// Increment the number of containers to 1, as it is recovering a
// single container.
request.SetNumContainers(lastRequest.GetNumContainers() + 1);
}
asks[resourceName] = request;
if (updatePendingResources)
{
// Similarly, deactivate application?
if (request.GetNumContainers() <= 0)
{
Log.Info("checking for deactivate of application :" + this.applicationId);
CheckForDeactivation();
}
int lastRequestContainers = lastRequest != null?lastRequest.GetNumContainers() :
0;
Org.Apache.Hadoop.Yarn.Api.Records.Resource lastRequestCapability = lastRequest !=
null?lastRequest.GetCapability() : Resources.None();
metrics.IncrPendingResources(user, request.GetNumContainers(), request.GetCapability
());
metrics.DecrPendingResources(user, lastRequestContainers, lastRequestCapability);
}
}
}
}
