private void StartResourceManager(float utilizationThreshold)
{
conf.SetFloat(FairSchedulerConfiguration.PreemptionThreshold, utilizationThreshold
);
resourceManager = new MockRM(conf);
resourceManager.Start();
NUnit.Framework.Assert.IsTrue(resourceManager.GetResourceScheduler() is TestFairSchedulerPreemption.StubbedFairScheduler
);
scheduler = (FairScheduler)resourceManager.GetResourceScheduler();
scheduler.SetClock(clock);
scheduler.updateInterval = 60 * 1000;
}
public virtual void TestContainerTokenGeneratedOnPullRequest()
{
MockRM rm1 = new MockRM(conf);
rm1.Start();
MockNM nm1 = rm1.RegisterNode("127.0.0.1:1234", 8000);
RMApp app1 = rm1.SubmitApp(200);
MockAM am1 = MockRM.LaunchAndRegisterAM(app1, rm1, nm1);
// request a container.
am1.Allocate("127.0.0.1", 1024, 1, new AList <ContainerId>());
ContainerId containerId2 = ContainerId.NewContainerId(am1.GetApplicationAttemptId
(), 2);
rm1.WaitForState(nm1, containerId2, RMContainerState.Allocated);
RMContainer container = rm1.GetResourceScheduler().GetRMContainer(containerId2);
// no container token is generated.
NUnit.Framework.Assert.AreEqual(containerId2, container.GetContainerId());
NUnit.Framework.Assert.IsNull(container.GetContainer().GetContainerToken());
// acquire the container.
IList <Container> containers = am1.Allocate(new AList <ResourceRequest>(), new AList
<ContainerId>()).GetAllocatedContainers();
NUnit.Framework.Assert.AreEqual(containerId2, containers[0].GetId());
// container token is generated.
NUnit.Framework.Assert.IsNotNull(containers[0].GetContainerToken());
rm1.Stop();
}
public virtual void TestExistenceOfResourceRequestInRMContainer()
{
Configuration conf = new Configuration();
MockRM rm1 = new MockRM(conf);
rm1.Start();
MockNM nm1 = rm1.RegisterNode("unknownhost:1234", 8000);
RMApp app1 = rm1.SubmitApp(1024);
MockAM am1 = MockRM.LaunchAndRegisterAM(app1, rm1, nm1);
ResourceScheduler scheduler = rm1.GetResourceScheduler();
// request a container.
am1.Allocate("127.0.0.1", 1024, 1, new AList <ContainerId>());
ContainerId containerId2 = ContainerId.NewContainerId(am1.GetApplicationAttemptId
(), 2);
rm1.WaitForState(nm1, containerId2, RMContainerState.Allocated);
// Verify whether list of ResourceRequest is present in RMContainer
// while moving to ALLOCATED state
NUnit.Framework.Assert.IsNotNull(scheduler.GetRMContainer(containerId2).GetResourceRequests
());
// Allocate container
am1.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers
();
rm1.WaitForState(nm1, containerId2, RMContainerState.Acquired);
// After RMContainer moving to ACQUIRED state, list of ResourceRequest will
// be empty
NUnit.Framework.Assert.IsNull(scheduler.GetRMContainer(containerId2).GetResourceRequests
());
}
public virtual void TestUpdateMaxAllocationUsesTotal()
{
int configuredMaxVCores = 20;
int configuredMaxMemory = 10 * 1024;
Org.Apache.Hadoop.Yarn.Api.Records.Resource configuredMaximumResource = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(configuredMaxMemory, configuredMaxVCores);
ConfigureScheduler();
YarnConfiguration conf = GetConf();
conf.SetInt(YarnConfiguration.RmSchedulerMaximumAllocationVcores, configuredMaxVCores
);
conf.SetInt(YarnConfiguration.RmSchedulerMaximumAllocationMb, configuredMaxMemory
);
conf.SetLong(YarnConfiguration.RmWorkPreservingRecoverySchedulingWaitMs, 0);
MockRM rm = new MockRM(conf);
try
{
rm.Start();
AbstractYarnScheduler scheduler = (AbstractYarnScheduler)rm.GetResourceScheduler(
);
Org.Apache.Hadoop.Yarn.Api.Records.Resource emptyResource = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(0, 0);
Org.Apache.Hadoop.Yarn.Api.Records.Resource fullResource1 = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(1024, 5);
Org.Apache.Hadoop.Yarn.Api.Records.Resource fullResource2 = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(2048, 10);
SchedulerNode mockNode1 = Org.Mockito.Mockito.Mock <SchedulerNode>();
Org.Mockito.Mockito.When(mockNode1.GetNodeID()).ThenReturn(NodeId.NewInstance("foo"
, 8080));
Org.Mockito.Mockito.When(mockNode1.GetAvailableResource()).ThenReturn(emptyResource
);
Org.Mockito.Mockito.When(mockNode1.GetTotalResource()).ThenReturn(fullResource1);
SchedulerNode mockNode2 = Org.Mockito.Mockito.Mock <SchedulerNode>();
Org.Mockito.Mockito.When(mockNode1.GetNodeID()).ThenReturn(NodeId.NewInstance("bar"
, 8081));
Org.Mockito.Mockito.When(mockNode2.GetAvailableResource()).ThenReturn(emptyResource
);
Org.Mockito.Mockito.When(mockNode2.GetTotalResource()).ThenReturn(fullResource2);
VerifyMaximumResourceCapability(configuredMaximumResource, scheduler);
scheduler.nodes = new Dictionary <NodeId, SchedulerNode>();
scheduler.nodes[mockNode1.GetNodeID()] = mockNode1;
scheduler.UpdateMaximumAllocation(mockNode1, true);
VerifyMaximumResourceCapability(fullResource1, scheduler);
scheduler.nodes[mockNode2.GetNodeID()] = mockNode2;
scheduler.UpdateMaximumAllocation(mockNode2, true);
VerifyMaximumResourceCapability(fullResource2, scheduler);
Sharpen.Collections.Remove(scheduler.nodes, mockNode2.GetNodeID());
scheduler.UpdateMaximumAllocation(mockNode2, false);
VerifyMaximumResourceCapability(fullResource1, scheduler);
Sharpen.Collections.Remove(scheduler.nodes, mockNode1.GetNodeID());
scheduler.UpdateMaximumAllocation(mockNode1, false);
VerifyMaximumResourceCapability(configuredMaximumResource, scheduler);
}
finally
{
rm.Stop();
}
}
/// <exception cref="System.IO.IOException"/>
private void CreateClusterWithQueuesAndOneNode(int mem, int vCores, string policy
)
{
PrintWriter @out = new PrintWriter(new FileWriter(AllocFile));
@out.WriteLine("<?xml version=\"1.0\"?>");
@out.WriteLine("<allocations>");
@out.WriteLine("<queue name=\"root\" >");
@out.WriteLine(" <queue name=\"parentA\" >");
@out.WriteLine(" <weight>8</weight>");
@out.WriteLine(" <queue name=\"childA1\" />");
@out.WriteLine(" <queue name=\"childA2\" />");
@out.WriteLine(" <queue name=\"childA3\" />");
@out.WriteLine(" <queue name=\"childA4\" />");
@out.WriteLine(" </queue>");
@out.WriteLine(" <queue name=\"parentB\" >");
@out.WriteLine(" <weight>1</weight>");
@out.WriteLine(" <queue name=\"childB1\" />");
@out.WriteLine(" <queue name=\"childB2\" />");
@out.WriteLine(" </queue>");
@out.WriteLine("</queue>");
@out.WriteLine("<defaultQueueSchedulingPolicy>" + policy + "</defaultQueueSchedulingPolicy>"
);
@out.WriteLine("</allocations>");
@out.Close();
resourceManager = new MockRM(conf);
resourceManager.Start();
scheduler = (FairScheduler)resourceManager.GetResourceScheduler();
RMNode node1 = MockNodes.NewNodeInfo(1, Resources.CreateResource(mem, vCores), 1,
"127.0.0.1");
NodeAddedSchedulerEvent nodeEvent1 = new NodeAddedSchedulerEvent(node1);
scheduler.Handle(nodeEvent1);
}
// Test RM restarts after AM container is preempted, new RM should not count
// AM preemption failure towards the max-retry-account and should be able to
// re-launch the AM.
/// <exception cref="System.Exception"/>
public virtual void TestPreemptedAMRestartOnRMRestart()
{
YarnConfiguration conf = new YarnConfiguration();
conf.SetClass(YarnConfiguration.RmScheduler, typeof(CapacityScheduler), typeof(ResourceScheduler
));
conf.SetBoolean(YarnConfiguration.RecoveryEnabled, true);
conf.SetBoolean(YarnConfiguration.RmWorkPreservingRecoveryEnabled, false);
conf.Set(YarnConfiguration.RmStore, typeof(MemoryRMStateStore).FullName);
// explicitly set max-am-retry count as 1.
conf.SetInt(YarnConfiguration.RmAmMaxAttempts, 1);
MemoryRMStateStore memStore = new MemoryRMStateStore();
memStore.Init(conf);
MockRM rm1 = new MockRM(conf, memStore);
rm1.Start();
MockNM nm1 = new MockNM("127.0.0.1:1234", 8000, rm1.GetResourceTrackerService());
nm1.RegisterNode();
RMApp app1 = rm1.SubmitApp(200);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = MockRM.LaunchAndRegisterAM(app1, rm1, nm1);
CapacityScheduler scheduler = (CapacityScheduler)rm1.GetResourceScheduler();
ContainerId amContainer = ContainerId.NewContainerId(am1.GetApplicationAttemptId(
), 1);
// Forcibly preempt the am container;
scheduler.KillContainer(scheduler.GetRMContainer(amContainer));
am1.WaitForState(RMAppAttemptState.Failed);
NUnit.Framework.Assert.IsTrue(!attempt1.ShouldCountTowardsMaxAttemptRetry());
rm1.WaitForState(app1.GetApplicationId(), RMAppState.Accepted);
// state store has 1 attempt stored.
ApplicationStateData appState = memStore.GetState().GetApplicationState()[app1.GetApplicationId
()];
NUnit.Framework.Assert.AreEqual(1, appState.GetAttemptCount());
// attempt stored has the preempted container exit status.
NUnit.Framework.Assert.AreEqual(ContainerExitStatus.Preempted, appState.GetAttempt
(am1.GetApplicationAttemptId()).GetAMContainerExitStatus());
// Restart rm.
MockRM rm2 = new MockRM(conf, memStore);
nm1.SetResourceTrackerService(rm2.GetResourceTrackerService());
nm1.RegisterNode();
rm2.Start();
// Restarted RM should re-launch the am.
MockAM am2 = rm2.WaitForNewAMToLaunchAndRegister(app1.GetApplicationId(), 2, nm1);
MockRM.FinishAMAndVerifyAppState(app1, rm2, nm1, am2);
RMAppAttempt attempt2 = rm2.GetRMContext().GetRMApps()[app1.GetApplicationId()].GetCurrentAppAttempt
();
NUnit.Framework.Assert.IsTrue(attempt2.ShouldCountTowardsMaxAttemptRetry());
NUnit.Framework.Assert.AreEqual(ContainerExitStatus.Invalid, appState.GetAttempt(
am2.GetApplicationAttemptId()).GetAMContainerExitStatus());
rm1.Stop();
rm2.Stop();
}
public virtual void Setup()
{
Configuration conf = CreateConfiguration();
resourceManager = new MockRM(conf);
resourceManager.Start();
scheduler = (FairScheduler)resourceManager.GetResourceScheduler();
}
private void CheckUsedResource(MockRM rm, string queueName, int memory, string label
)
{
CapacityScheduler scheduler = (CapacityScheduler)rm.GetResourceScheduler();
CSQueue queue = scheduler.GetQueue(queueName);
NUnit.Framework.Assert.AreEqual(memory, queue.GetQueueResourceUsage().GetUsed(label
).GetMemory());
}
public virtual void TestConcurrentAccess()
{
conf.Set(FairSchedulerConfiguration.AssignMultiple, "false");
resourceManager = new MockRM(conf);
resourceManager.Start();
scheduler = (FairScheduler)resourceManager.GetResourceScheduler();
string queueName = "root.queue1";
FSLeafQueue schedulable = scheduler.GetQueueManager().GetLeafQueue(queueName, true
);
ApplicationAttemptId applicationAttemptId = CreateAppAttemptId(1, 1);
RMContext rmContext = resourceManager.GetRMContext();
FSAppAttempt app = new FSAppAttempt(scheduler, applicationAttemptId, "user1", schedulable
, null, rmContext);
// this needs to be in sync with the number of runnables declared below
int testThreads = 2;
IList <Runnable> runnables = new AList <Runnable>();
// add applications to modify the list
runnables.AddItem(new _Runnable_257(schedulable, app));
// iterate over the list a couple of times in a different thread
runnables.AddItem(new _Runnable_267(schedulable));
IList <Exception> exceptions = Sharpen.Collections.SynchronizedList(new AList <Exception
>());
ExecutorService threadPool = Executors.NewFixedThreadPool(testThreads);
try
{
CountDownLatch allExecutorThreadsReady = new CountDownLatch(testThreads);
CountDownLatch startBlocker = new CountDownLatch(1);
CountDownLatch allDone = new CountDownLatch(testThreads);
foreach (Runnable submittedTestRunnable in runnables)
{
threadPool.Submit(new _Runnable_287(allExecutorThreadsReady, startBlocker, submittedTestRunnable
, exceptions, allDone));
}
// wait until all threads are ready
allExecutorThreadsReady.Await();
// start all test runners
startBlocker.CountDown();
int testTimeout = 2;
NUnit.Framework.Assert.IsTrue("Timeout waiting for more than " + testTimeout + " seconds"
, allDone.Await(testTimeout, TimeUnit.Seconds));
}
catch (Exception ie)
{
exceptions.AddItem(ie);
}
finally
{
threadPool.ShutdownNow();
}
NUnit.Framework.Assert.IsTrue("Test failed with exception(s)" + exceptions, exceptions
.IsEmpty());
}
public virtual void TestMaxAllocationAfterUpdateNodeResource()
{
int configuredMaxVCores = 20;
int configuredMaxMemory = 10 * 1024;
Org.Apache.Hadoop.Yarn.Api.Records.Resource configuredMaximumResource = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(configuredMaxMemory, configuredMaxVCores);
ConfigureScheduler();
YarnConfiguration conf = GetConf();
conf.SetInt(YarnConfiguration.RmSchedulerMaximumAllocationVcores, configuredMaxVCores
);
conf.SetInt(YarnConfiguration.RmSchedulerMaximumAllocationMb, configuredMaxMemory
);
conf.SetLong(YarnConfiguration.RmWorkPreservingRecoverySchedulingWaitMs, 0);
MockRM rm = new MockRM(conf);
try
{
rm.Start();
AbstractYarnScheduler scheduler = (AbstractYarnScheduler)rm.GetResourceScheduler(
);
VerifyMaximumResourceCapability(configuredMaximumResource, scheduler);
Org.Apache.Hadoop.Yarn.Api.Records.Resource resource1 = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(2048, 5);
Org.Apache.Hadoop.Yarn.Api.Records.Resource resource2 = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(4096, 10);
Org.Apache.Hadoop.Yarn.Api.Records.Resource resource3 = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(512, 1);
Org.Apache.Hadoop.Yarn.Api.Records.Resource resource4 = Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(1024, 2);
RMNode node1 = MockNodes.NewNodeInfo(0, resource1, 1, "127.0.0.2");
scheduler.Handle(new NodeAddedSchedulerEvent(node1));
RMNode node2 = MockNodes.NewNodeInfo(0, resource3, 2, "127.0.0.3");
scheduler.Handle(new NodeAddedSchedulerEvent(node2));
VerifyMaximumResourceCapability(resource1, scheduler);
// increase node1 resource
scheduler.UpdateNodeResource(node1, ResourceOption.NewInstance(resource2, 0));
VerifyMaximumResourceCapability(resource2, scheduler);
// decrease node1 resource
scheduler.UpdateNodeResource(node1, ResourceOption.NewInstance(resource1, 0));
VerifyMaximumResourceCapability(resource1, scheduler);
// increase node2 resource
scheduler.UpdateNodeResource(node2, ResourceOption.NewInstance(resource4, 0));
VerifyMaximumResourceCapability(resource1, scheduler);
// decrease node2 resource
scheduler.UpdateNodeResource(node2, ResourceOption.NewInstance(resource3, 0));
VerifyMaximumResourceCapability(resource1, scheduler);
}
finally
{
rm.Stop();
}
}
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);
}
public virtual void Setup()
{
mockClock = new FairSchedulerTestBase.MockClock();
conf = CreateConfiguration();
resourceManager = new MockRM(conf);
resourceManager.Start();
scheduler = (FairScheduler)resourceManager.GetResourceScheduler();
scheduler.SetClock(mockClock);
NUnit.Framework.Assert.IsTrue(scheduler.IsContinuousSchedulingEnabled());
NUnit.Framework.Assert.AreEqual(FairSchedulerConfiguration.DefaultContinuousSchedulingSleepMs
, scheduler.GetContinuousSchedulingSleepMs());
NUnit.Framework.Assert.AreEqual(mockClock, scheduler.GetClock());
}
public virtual void TestRefreshQueuesWithReservations()
{
CapacityScheduler cs = (CapacityScheduler)rm.GetResourceScheduler();
// Test add one reservation dynamically and manually modify capacity
ReservationQueue a1 = new ReservationQueue(cs, "a1", (PlanQueue)cs.GetQueue("a"));
cs.AddQueue(a1);
a1.SetEntitlement(new QueueEntitlement(A1Capacity / 100, 1f));
// Test add another reservation queue and use setEntitlement to modify
// capacity
ReservationQueue a2 = new ReservationQueue(cs, "a2", (PlanQueue)cs.GetQueue("a"));
cs.AddQueue(a2);
cs.SetEntitlement("a2", new QueueEntitlement(A2Capacity / 100, 1.0f));
// Verify all allocations match
tcs.CheckQueueCapacities(cs, ACapacity, BCapacity);
// Reinitialize and verify all dynamic queued survived
CapacitySchedulerConfiguration conf = cs.GetConfiguration();
conf.SetCapacity(A, 80f);
conf.SetCapacity(B, 20f);
cs.Reinitialize(conf, rm.GetRMContext());
tcs.CheckQueueCapacities(cs, 80f, 20f);
}
/// <exception cref="System.Exception"/>
public virtual void Test()
{
conf.Set(FairSchedulerConfiguration.AllocationFile, AllocFile);
PrintWriter @out = new PrintWriter(new FileWriter(AllocFile));
@out.WriteLine("<?xml version=\"1.0\"?>");
@out.WriteLine("<allocations>");
@out.WriteLine("<queue name=\"queueA\">");
@out.WriteLine("<minResources>2048mb,0vcores</minResources>");
@out.WriteLine("</queue>");
@out.WriteLine("<queue name=\"queueB\">");
@out.WriteLine("<minResources>2048mb,0vcores</minResources>");
@out.WriteLine("</queue>");
@out.WriteLine("</allocations>");
@out.Close();
resourceManager = new MockRM(conf);
resourceManager.Start();
scheduler = (FairScheduler)resourceManager.GetResourceScheduler();
// Add one big node (only care about aggregate capacity)
RMNode node1 = MockNodes.NewNodeInfo(1, Resources.CreateResource(4 * 1024, 4), 1,
"127.0.0.1");
NodeAddedSchedulerEvent nodeEvent1 = new NodeAddedSchedulerEvent(node1);
scheduler.Handle(nodeEvent1);
scheduler.Update();
// Queue A wants 3 * 1024. Node update gives this all to A
CreateSchedulingRequest(3 * 1024, "queueA", "user1");
scheduler.Update();
NodeUpdateSchedulerEvent nodeEvent2 = new NodeUpdateSchedulerEvent(node1);
scheduler.Handle(nodeEvent2);
// Queue B arrives and wants 1 * 1024
CreateSchedulingRequest(1 * 1024, "queueB", "user1");
scheduler.Update();
ICollection <FSLeafQueue> queues = scheduler.GetQueueManager().GetLeafQueues();
NUnit.Framework.Assert.AreEqual(3, queues.Count);
// Queue A should be above min share, B below.
FSLeafQueue queueA = scheduler.GetQueueManager().GetLeafQueue("queueA", false);
FSLeafQueue queueB = scheduler.GetQueueManager().GetLeafQueue("queueB", false);
NUnit.Framework.Assert.IsFalse(queueA.IsStarvedForMinShare());
NUnit.Framework.Assert.IsTrue(queueB.IsStarvedForMinShare());
// Node checks in again, should allocate for B
scheduler.Handle(nodeEvent2);
// Now B should have min share ( = demand here)
NUnit.Framework.Assert.IsFalse(queueB.IsStarvedForMinShare());
}
/// <exception cref="Org.Codehaus.Jettison.Json.JSONException"/>
/// <exception cref="System.Exception"/>
public virtual void VerifyClusterMetrics(int submittedApps, int completedApps, int
reservedMB, int availableMB, int allocMB, int reservedVirtualCores, int availableVirtualCores
, int allocVirtualCores, int totalVirtualCores, int containersAlloc, int totalMB
, int totalNodes, int lostNodes, int unhealthyNodes, int decommissionedNodes, int
rebootedNodes, int activeNodes)
{
ResourceScheduler rs = rm.GetResourceScheduler();
QueueMetrics metrics = rs.GetRootQueueMetrics();
ClusterMetrics clusterMetrics = ClusterMetrics.GetMetrics();
long totalMBExpect = metrics.GetAvailableMB() + metrics.GetAllocatedMB();
long totalVirtualCoresExpect = metrics.GetAvailableVirtualCores() + metrics.GetAllocatedVirtualCores
();
NUnit.Framework.Assert.AreEqual("appsSubmitted doesn't match", metrics.GetAppsSubmitted
(), submittedApps);
NUnit.Framework.Assert.AreEqual("appsCompleted doesn't match", metrics.GetAppsCompleted
(), completedApps);
NUnit.Framework.Assert.AreEqual("reservedMB doesn't match", metrics.GetReservedMB
(), reservedMB);
NUnit.Framework.Assert.AreEqual("availableMB doesn't match", metrics.GetAvailableMB
(), availableMB);
NUnit.Framework.Assert.AreEqual("allocatedMB doesn't match", metrics.GetAllocatedMB
(), allocMB);
NUnit.Framework.Assert.AreEqual("reservedVirtualCores doesn't match", metrics.GetReservedVirtualCores
(), reservedVirtualCores);
NUnit.Framework.Assert.AreEqual("availableVirtualCores doesn't match", metrics.GetAvailableVirtualCores
(), availableVirtualCores);
NUnit.Framework.Assert.AreEqual("allocatedVirtualCores doesn't match", totalVirtualCoresExpect
, allocVirtualCores);
NUnit.Framework.Assert.AreEqual("containersAllocated doesn't match", 0, containersAlloc
);
NUnit.Framework.Assert.AreEqual("totalMB doesn't match", totalMBExpect, totalMB);
NUnit.Framework.Assert.AreEqual("totalNodes doesn't match", clusterMetrics.GetNumActiveNMs
() + clusterMetrics.GetNumLostNMs() + clusterMetrics.GetNumDecommisionedNMs() +
clusterMetrics.GetNumRebootedNMs() + clusterMetrics.GetUnhealthyNMs(), totalNodes
);
NUnit.Framework.Assert.AreEqual("lostNodes doesn't match", clusterMetrics.GetNumLostNMs
(), lostNodes);
NUnit.Framework.Assert.AreEqual("unhealthyNodes doesn't match", clusterMetrics.GetUnhealthyNMs
(), unhealthyNodes);
NUnit.Framework.Assert.AreEqual("decommissionedNodes doesn't match", clusterMetrics
.GetNumDecommisionedNMs(), decommissionedNodes);
NUnit.Framework.Assert.AreEqual("rebootedNodes doesn't match", clusterMetrics.GetNumRebootedNMs
(), rebootedNodes);
NUnit.Framework.Assert.AreEqual("activeNodes doesn't match", clusterMetrics.GetNumActiveNMs
(), activeNodes);
}
public virtual void TestMaximimumAllocationMemory()
{
int node1MaxMemory = 15 * 1024;
int node2MaxMemory = 5 * 1024;
int node3MaxMemory = 6 * 1024;
int configuredMaxMemory = 10 * 1024;
ConfigureScheduler();
YarnConfiguration conf = GetConf();
conf.SetInt(YarnConfiguration.RmSchedulerMaximumAllocationMb, configuredMaxMemory
);
conf.SetLong(YarnConfiguration.RmWorkPreservingRecoverySchedulingWaitMs, 1000 * 1000
);
MockRM rm = new MockRM(conf);
try
{
rm.Start();
TestMaximumAllocationMemoryHelper((AbstractYarnScheduler)rm.GetResourceScheduler(
), node1MaxMemory, node2MaxMemory, node3MaxMemory, configuredMaxMemory, configuredMaxMemory
, configuredMaxMemory, configuredMaxMemory, configuredMaxMemory, configuredMaxMemory
);
}
finally
{
rm.Stop();
}
conf.SetLong(YarnConfiguration.RmWorkPreservingRecoverySchedulingWaitMs, 0);
rm = new MockRM(conf);
try
{
rm.Start();
TestMaximumAllocationMemoryHelper((AbstractYarnScheduler)rm.GetResourceScheduler(
), node1MaxMemory, node2MaxMemory, node3MaxMemory, configuredMaxMemory, configuredMaxMemory
, configuredMaxMemory, node2MaxMemory, node3MaxMemory, node2MaxMemory);
}
finally
{
rm.Stop();
}
}
public virtual void TestMaximimumAllocationVCores()
{
int node1MaxVCores = 15;
int node2MaxVCores = 5;
int node3MaxVCores = 6;
int configuredMaxVCores = 10;
ConfigureScheduler();
YarnConfiguration conf = GetConf();
conf.SetInt(YarnConfiguration.RmSchedulerMaximumAllocationVcores, configuredMaxVCores
);
conf.SetLong(YarnConfiguration.RmWorkPreservingRecoverySchedulingWaitMs, 1000 * 1000
);
MockRM rm = new MockRM(conf);
try
{
rm.Start();
TestMaximumAllocationVCoresHelper((AbstractYarnScheduler)rm.GetResourceScheduler(
), node1MaxVCores, node2MaxVCores, node3MaxVCores, configuredMaxVCores, configuredMaxVCores
, configuredMaxVCores, configuredMaxVCores, configuredMaxVCores, configuredMaxVCores
);
}
finally
{
rm.Stop();
}
conf.SetLong(YarnConfiguration.RmWorkPreservingRecoverySchedulingWaitMs, 0);
rm = new MockRM(conf);
try
{
rm.Start();
TestMaximumAllocationVCoresHelper((AbstractYarnScheduler)rm.GetResourceScheduler(
), node1MaxVCores, node2MaxVCores, node3MaxVCores, configuredMaxVCores, configuredMaxVCores
, configuredMaxVCores, node2MaxVCores, node3MaxVCores, node2MaxVCores);
}
finally
{
rm.Stop();
}
}
/// <exception cref="Org.Codehaus.Jettison.Json.JSONException"/>
/// <exception cref="System.Exception"/>
public virtual void VerifyNodeInfoGeneric(MockNM nm, string state, string rack, string
id, string nodeHostName, string nodeHTTPAddress, long lastHealthUpdate, string
healthReport, int numContainers, long usedMemoryMB, long availMemoryMB, long usedVirtualCores
, long availVirtualCores, string version)
{
RMNode node = rm.GetRMContext().GetRMNodes()[nm.GetNodeId()];
ResourceScheduler sched = rm.GetResourceScheduler();
SchedulerNodeReport report = sched.GetNodeReport(nm.GetNodeId());
WebServicesTestUtils.CheckStringMatch("state", node.GetState().ToString(), state);
WebServicesTestUtils.CheckStringMatch("rack", node.GetRackName(), rack);
WebServicesTestUtils.CheckStringMatch("id", nm.GetNodeId().ToString(), id);
WebServicesTestUtils.CheckStringMatch("nodeHostName", nm.GetNodeId().GetHost(), nodeHostName
);
WebServicesTestUtils.CheckStringMatch("healthReport", node.GetHealthReport().ToString
(), healthReport);
string expectedHttpAddress = nm.GetNodeId().GetHost() + ":" + nm.GetHttpPort();
WebServicesTestUtils.CheckStringMatch("nodeHTTPAddress", expectedHttpAddress, nodeHTTPAddress
);
WebServicesTestUtils.CheckStringMatch("version", node.GetNodeManagerVersion(), version
);
long expectedHealthUpdate = node.GetLastHealthReportTime();
NUnit.Framework.Assert.AreEqual("lastHealthUpdate doesn't match, got: " + lastHealthUpdate
+ " expected: " + expectedHealthUpdate, expectedHealthUpdate, lastHealthUpdate);
if (report != null)
{
NUnit.Framework.Assert.AreEqual("numContainers doesn't match: " + numContainers,
report.GetNumContainers(), numContainers);
NUnit.Framework.Assert.AreEqual("usedMemoryMB doesn't match: " + usedMemoryMB, report
.GetUsedResource().GetMemory(), usedMemoryMB);
NUnit.Framework.Assert.AreEqual("availMemoryMB doesn't match: " + availMemoryMB,
report.GetAvailableResource().GetMemory(), availMemoryMB);
NUnit.Framework.Assert.AreEqual("usedVirtualCores doesn't match: " + usedVirtualCores
, report.GetUsedResource().GetVirtualCores(), usedVirtualCores);
NUnit.Framework.Assert.AreEqual("availVirtualCores doesn't match: " + availVirtualCores
, report.GetAvailableResource().GetVirtualCores(), availVirtualCores);
}
}
/// <exception cref="System.Exception"/>
private void CheckAppQueue(MockRM resourceManager, string user, string submissionQueue
, string expected)
{
RMApp app = resourceManager.SubmitApp(200, "name", user, new Dictionary <ApplicationAccessType
, string>(), false, submissionQueue, -1, null, "MAPREDUCE", false);
RMAppState expectedState = expected.IsEmpty() ? RMAppState.Failed : RMAppState.Accepted;
resourceManager.WaitForState(app.GetApplicationId(), expectedState);
// get scheduler app
CapacityScheduler cs = (CapacityScheduler)resourceManager.GetResourceScheduler();
SchedulerApplication schedulerApp = cs.GetSchedulerApplications()[app.GetApplicationId
()];
string queue = string.Empty;
if (schedulerApp != null)
{
queue = schedulerApp.GetQueue().GetQueueName();
}
NUnit.Framework.Assert.IsTrue("expected " + expected + " actual " + queue, expected
.Equals(queue));
NUnit.Framework.Assert.AreEqual(expected, app.GetQueue());
}
public virtual void TestUpdateDemand()
{
conf.Set(FairSchedulerConfiguration.AssignMultiple, "false");
resourceManager = new MockRM(conf);
resourceManager.Start();
scheduler = (FairScheduler)resourceManager.GetResourceScheduler();
scheduler.allocConf = Org.Mockito.Mockito.Mock <AllocationConfiguration>();
string queueName = "root.queue1";
Org.Mockito.Mockito.When(scheduler.allocConf.GetMaxResources(queueName)).ThenReturn
(maxResource);
Org.Mockito.Mockito.When(scheduler.allocConf.GetMinResources(queueName)).ThenReturn
(Resources.None());
FSLeafQueue schedulable = new FSLeafQueue(queueName, scheduler, null);
FSAppAttempt app = Org.Mockito.Mockito.Mock <FSAppAttempt>();
Org.Mockito.Mockito.When(app.GetDemand()).ThenReturn(maxResource);
schedulable.AddAppSchedulable(app);
schedulable.AddAppSchedulable(app);
schedulable.UpdateDemand();
NUnit.Framework.Assert.IsTrue("Demand is greater than max allowed ", Resources.Equals
(schedulable.GetDemand(), maxResource));
}
public virtual void TestBlackListNodes()
{
Configuration conf = new Configuration();
conf.SetClass(YarnConfiguration.RmScheduler, typeof(FifoScheduler), typeof(ResourceScheduler
));
MockRM rm = new MockRM(conf);
rm.Start();
FifoScheduler fs = (FifoScheduler)rm.GetResourceScheduler();
string host = "127.0.0.1";
RMNode node = MockNodes.NewNodeInfo(0, MockNodes.NewResource(4 * Gb), 1, host);
fs.Handle(new NodeAddedSchedulerEvent(node));
ApplicationId appId = BuilderUtils.NewApplicationId(100, 1);
ApplicationAttemptId appAttemptId = BuilderUtils.NewApplicationAttemptId(appId, 1
);
CreateMockRMApp(appAttemptId, rm.GetRMContext());
SchedulerEvent appEvent = new AppAddedSchedulerEvent(appId, "default", "user");
fs.Handle(appEvent);
SchedulerEvent attemptEvent = new AppAttemptAddedSchedulerEvent(appAttemptId, false
);
fs.Handle(attemptEvent);
// Verify the blacklist can be updated independent of requesting containers
fs.Allocate(appAttemptId, Sharpen.Collections.EmptyList <ResourceRequest>(), Sharpen.Collections
.EmptyList <ContainerId>(), Sharpen.Collections.SingletonList(host), null);
NUnit.Framework.Assert.IsTrue(fs.GetApplicationAttempt(appAttemptId).IsBlacklisted
(host));
fs.Allocate(appAttemptId, Sharpen.Collections.EmptyList <ResourceRequest>(), Sharpen.Collections
.EmptyList <ContainerId>(), null, Sharpen.Collections.SingletonList(host));
NUnit.Framework.Assert.IsFalse(fs.GetApplicationAttempt(appAttemptId).IsBlacklisted
(host));
rm.Stop();
}
// AM container preempted, nm disk failure
// should not be counted towards AM max retry count.
/// <exception cref="System.Exception"/>
public virtual void TestShouldNotCountFailureToMaxAttemptRetry()
{
YarnConfiguration conf = new YarnConfiguration();
conf.SetClass(YarnConfiguration.RmScheduler, typeof(CapacityScheduler), typeof(ResourceScheduler
));
// explicitly set max-am-retry count as 1.
conf.SetInt(YarnConfiguration.RmAmMaxAttempts, 1);
conf.SetBoolean(YarnConfiguration.RecoveryEnabled, true);
conf.Set(YarnConfiguration.RmStore, typeof(MemoryRMStateStore).FullName);
MemoryRMStateStore memStore = new MemoryRMStateStore();
memStore.Init(conf);
MockRM rm1 = new MockRM(conf, memStore);
rm1.Start();
MockNM nm1 = new MockNM("127.0.0.1:1234", 8000, rm1.GetResourceTrackerService());
nm1.RegisterNode();
RMApp app1 = rm1.SubmitApp(200);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = MockRM.LaunchAndRegisterAM(app1, rm1, nm1);
CapacityScheduler scheduler = (CapacityScheduler)rm1.GetResourceScheduler();
ContainerId amContainer = ContainerId.NewContainerId(am1.GetApplicationAttemptId(
), 1);
// Preempt the first attempt;
scheduler.KillContainer(scheduler.GetRMContainer(amContainer));
am1.WaitForState(RMAppAttemptState.Failed);
NUnit.Framework.Assert.IsTrue(!attempt1.ShouldCountTowardsMaxAttemptRetry());
rm1.WaitForState(app1.GetApplicationId(), RMAppState.Accepted);
ApplicationStateData appState = memStore.GetState().GetApplicationState()[app1.GetApplicationId
()];
// AM should be restarted even though max-am-attempt is 1.
MockAM am2 = rm1.WaitForNewAMToLaunchAndRegister(app1.GetApplicationId(), 2, nm1);
RMAppAttempt attempt2 = app1.GetCurrentAppAttempt();
NUnit.Framework.Assert.IsTrue(((RMAppAttemptImpl)attempt2).MayBeLastAttempt());
// Preempt the second attempt.
ContainerId amContainer2 = ContainerId.NewContainerId(am2.GetApplicationAttemptId
(), 1);
scheduler.KillContainer(scheduler.GetRMContainer(amContainer2));
am2.WaitForState(RMAppAttemptState.Failed);
NUnit.Framework.Assert.IsTrue(!attempt2.ShouldCountTowardsMaxAttemptRetry());
rm1.WaitForState(app1.GetApplicationId(), RMAppState.Accepted);
MockAM am3 = rm1.WaitForNewAMToLaunchAndRegister(app1.GetApplicationId(), 3, nm1);
RMAppAttempt attempt3 = app1.GetCurrentAppAttempt();
NUnit.Framework.Assert.IsTrue(((RMAppAttemptImpl)attempt3).MayBeLastAttempt());
// mimic NM disk_failure
ContainerStatus containerStatus = Org.Apache.Hadoop.Yarn.Util.Records.NewRecord <ContainerStatus
>();
containerStatus.SetContainerId(attempt3.GetMasterContainer().GetId());
containerStatus.SetDiagnostics("mimic NM disk_failure");
containerStatus.SetState(ContainerState.Complete);
containerStatus.SetExitStatus(ContainerExitStatus.DisksFailed);
IDictionary <ApplicationId, IList <ContainerStatus> > conts = new Dictionary <ApplicationId
, IList <ContainerStatus> >();
conts[app1.GetApplicationId()] = Sharpen.Collections.SingletonList(containerStatus
);
nm1.NodeHeartbeat(conts, true);
am3.WaitForState(RMAppAttemptState.Failed);
NUnit.Framework.Assert.IsTrue(!attempt3.ShouldCountTowardsMaxAttemptRetry());
NUnit.Framework.Assert.AreEqual(ContainerExitStatus.DisksFailed, appState.GetAttempt
(am3.GetApplicationAttemptId()).GetAMContainerExitStatus());
rm1.WaitForState(app1.GetApplicationId(), RMAppState.Accepted);
MockAM am4 = rm1.WaitForNewAMToLaunchAndRegister(app1.GetApplicationId(), 4, nm1);
RMAppAttempt attempt4 = app1.GetCurrentAppAttempt();
NUnit.Framework.Assert.IsTrue(((RMAppAttemptImpl)attempt4).MayBeLastAttempt());
// create second NM, and register to rm1
MockNM nm2 = new MockNM("127.0.0.1:2234", 8000, rm1.GetResourceTrackerService());
nm2.RegisterNode();
// nm1 heartbeats to report unhealthy
// This will mimic ContainerExitStatus.ABORT
nm1.NodeHeartbeat(false);
am4.WaitForState(RMAppAttemptState.Failed);
NUnit.Framework.Assert.IsTrue(!attempt4.ShouldCountTowardsMaxAttemptRetry());
NUnit.Framework.Assert.AreEqual(ContainerExitStatus.Aborted, appState.GetAttempt(
am4.GetApplicationAttemptId()).GetAMContainerExitStatus());
// launch next AM in nm2
nm2.NodeHeartbeat(true);
MockAM am5 = rm1.WaitForNewAMToLaunchAndRegister(app1.GetApplicationId(), 5, nm2);
RMAppAttempt attempt5 = app1.GetCurrentAppAttempt();
NUnit.Framework.Assert.IsTrue(((RMAppAttemptImpl)attempt5).MayBeLastAttempt());
// fail the AM normally
nm2.NodeHeartbeat(am5.GetApplicationAttemptId(), 1, ContainerState.Complete);
am5.WaitForState(RMAppAttemptState.Failed);
NUnit.Framework.Assert.IsTrue(attempt5.ShouldCountTowardsMaxAttemptRetry());
// AM should not be restarted.
rm1.WaitForState(app1.GetApplicationId(), RMAppState.Failed);
NUnit.Framework.Assert.AreEqual(5, app1.GetAppAttempts().Count);
rm1.Stop();
}
/// <exception cref="System.Exception"/>
public virtual void TestAMRestartWithExistingContainers()
{
YarnConfiguration conf = new YarnConfiguration();
conf.SetInt(YarnConfiguration.RmAmMaxAttempts, 2);
MockRM rm1 = new MockRM(conf);
rm1.Start();
RMApp app1 = rm1.SubmitApp(200, "name", "user", new Dictionary <ApplicationAccessType
, string>(), false, "default", -1, null, "MAPREDUCE", false, true);
MockNM nm1 = new MockNM("127.0.0.1:1234", 10240, rm1.GetResourceTrackerService());
nm1.RegisterNode();
MockNM nm2 = new MockNM("127.0.0.1:2351", 4089, rm1.GetResourceTrackerService());
nm2.RegisterNode();
MockAM am1 = MockRM.LaunchAndRegisterAM(app1, rm1, nm1);
int NumContainers = 3;
// allocate NUM_CONTAINERS containers
am1.Allocate("127.0.0.1", 1024, NumContainers, new AList <ContainerId>());
nm1.NodeHeartbeat(true);
// wait for containers to be allocated.
IList <Container> containers = am1.Allocate(new AList <ResourceRequest>(), new AList
<ContainerId>()).GetAllocatedContainers();
while (containers.Count != NumContainers)
{
nm1.NodeHeartbeat(true);
Sharpen.Collections.AddAll(containers, am1.Allocate(new AList <ResourceRequest>(),
new AList <ContainerId>()).GetAllocatedContainers());
Sharpen.Thread.Sleep(200);
}
// launch the 2nd container, for testing running container transferred.
nm1.NodeHeartbeat(am1.GetApplicationAttemptId(), 2, ContainerState.Running);
ContainerId containerId2 = ContainerId.NewContainerId(am1.GetApplicationAttemptId
(), 2);
rm1.WaitForState(nm1, containerId2, RMContainerState.Running);
// launch the 3rd container, for testing container allocated by previous
// attempt is completed by the next new attempt/
nm1.NodeHeartbeat(am1.GetApplicationAttemptId(), 3, ContainerState.Running);
ContainerId containerId3 = ContainerId.NewContainerId(am1.GetApplicationAttemptId
(), 3);
rm1.WaitForState(nm1, containerId3, RMContainerState.Running);
// 4th container still in AQUIRED state. for testing Acquired container is
// always killed.
ContainerId containerId4 = ContainerId.NewContainerId(am1.GetApplicationAttemptId
(), 4);
rm1.WaitForState(nm1, containerId4, RMContainerState.Acquired);
// 5th container is in Allocated state. for testing allocated container is
// always killed.
am1.Allocate("127.0.0.1", 1024, 1, new AList <ContainerId>());
nm1.NodeHeartbeat(true);
ContainerId containerId5 = ContainerId.NewContainerId(am1.GetApplicationAttemptId
(), 5);
rm1.WaitForContainerAllocated(nm1, containerId5);
rm1.WaitForState(nm1, containerId5, RMContainerState.Allocated);
// 6th container is in Reserved state.
am1.Allocate("127.0.0.1", 6000, 1, new AList <ContainerId>());
ContainerId containerId6 = ContainerId.NewContainerId(am1.GetApplicationAttemptId
(), 6);
nm1.NodeHeartbeat(true);
SchedulerApplicationAttempt schedulerAttempt = ((AbstractYarnScheduler)rm1.GetResourceScheduler
()).GetCurrentAttemptForContainer(containerId6);
while (schedulerAttempt.GetReservedContainers().IsEmpty())
{
System.Console.Out.WriteLine("Waiting for container " + containerId6 + " to be reserved."
);
nm1.NodeHeartbeat(true);
Sharpen.Thread.Sleep(200);
}
// assert containerId6 is reserved.
NUnit.Framework.Assert.AreEqual(containerId6, schedulerAttempt.GetReservedContainers
()[0].GetContainerId());
// fail the AM by sending CONTAINER_FINISHED event without registering.
nm1.NodeHeartbeat(am1.GetApplicationAttemptId(), 1, ContainerState.Complete);
am1.WaitForState(RMAppAttemptState.Failed);
// wait for some time. previous AM's running containers should still remain
// in scheduler even though am failed
Sharpen.Thread.Sleep(3000);
rm1.WaitForState(nm1, containerId2, RMContainerState.Running);
// acquired/allocated containers are cleaned up.
NUnit.Framework.Assert.IsNull(rm1.GetResourceScheduler().GetRMContainer(containerId4
));
NUnit.Framework.Assert.IsNull(rm1.GetResourceScheduler().GetRMContainer(containerId5
));
// wait for app to start a new attempt.
rm1.WaitForState(app1.GetApplicationId(), RMAppState.Accepted);
// assert this is a new AM.
ApplicationAttemptId newAttemptId = app1.GetCurrentAppAttempt().GetAppAttemptId();
NUnit.Framework.Assert.IsFalse(newAttemptId.Equals(am1.GetApplicationAttemptId())
);
// launch the new AM
RMAppAttempt attempt2 = app1.GetCurrentAppAttempt();
nm1.NodeHeartbeat(true);
MockAM am2 = rm1.SendAMLaunched(attempt2.GetAppAttemptId());
RegisterApplicationMasterResponse registerResponse = am2.RegisterAppAttempt();
// Assert two containers are running: container2 and container3;
NUnit.Framework.Assert.AreEqual(2, registerResponse.GetContainersFromPreviousAttempts
().Count);
bool containerId2Exists = false;
bool containerId3Exists = false;
foreach (Container container in registerResponse.GetContainersFromPreviousAttempts
())
{
if (container.GetId().Equals(containerId2))
{
containerId2Exists = true;
}
if (container.GetId().Equals(containerId3))
{
containerId3Exists = true;
}
}
NUnit.Framework.Assert.IsTrue(containerId2Exists && containerId3Exists);
rm1.WaitForState(app1.GetApplicationId(), RMAppState.Running);
// complete container by sending the container complete event which has earlier
// attempt's attemptId
nm1.NodeHeartbeat(am1.GetApplicationAttemptId(), 3, ContainerState.Complete);
// Even though the completed container containerId3 event was sent to the
// earlier failed attempt, new RMAppAttempt can also capture this container
// info.
// completed containerId4 is also transferred to the new attempt.
RMAppAttempt newAttempt = app1.GetRMAppAttempt(am2.GetApplicationAttemptId());
// 4 containers finished, acquired/allocated/reserved/completed.
WaitForContainersToFinish(4, newAttempt);
bool container3Exists = false;
bool container4Exists = false;
bool container5Exists = false;
bool container6Exists = false;
foreach (ContainerStatus status in newAttempt.GetJustFinishedContainers())
{
if (status.GetContainerId().Equals(containerId3))
{
// containerId3 is the container ran by previous attempt but finished by the
// new attempt.
container3Exists = true;
}
if (status.GetContainerId().Equals(containerId4))
{
// containerId4 is the Acquired Container killed by the previous attempt,
// it's now inside new attempt's finished container list.
container4Exists = true;
}
if (status.GetContainerId().Equals(containerId5))
{
// containerId5 is the Allocated container killed by previous failed attempt.
container5Exists = true;
}
if (status.GetContainerId().Equals(containerId6))
{
// containerId6 is the reserved container killed by previous failed attempt.
container6Exists = true;
}
}
NUnit.Framework.Assert.IsTrue(container3Exists && container4Exists && container5Exists &&
container6Exists);
// New SchedulerApplicationAttempt also has the containers info.
rm1.WaitForState(nm1, containerId2, RMContainerState.Running);
// record the scheduler attempt for testing.
SchedulerApplicationAttempt schedulerNewAttempt = ((AbstractYarnScheduler)rm1.GetResourceScheduler
()).GetCurrentAttemptForContainer(containerId2);
// finish this application
MockRM.FinishAMAndVerifyAppState(app1, rm1, nm1, am2);
// the 2nd attempt released the 1st attempt's running container, when the
// 2nd attempt finishes.
NUnit.Framework.Assert.IsFalse(schedulerNewAttempt.GetLiveContainers().Contains(containerId2
));
// all 4 normal containers finished.
System.Console.Out.WriteLine("New attempt's just finished containers: " + newAttempt
.GetJustFinishedContainers());
WaitForContainersToFinish(5, newAttempt);
rm1.Stop();
}
/// <exception cref="System.Exception"/>
public virtual void TestIsStarvedForFairShare()
{
conf.Set(FairSchedulerConfiguration.AllocationFile, AllocFile);
PrintWriter @out = new PrintWriter(new FileWriter(AllocFile));
@out.WriteLine("<?xml version=\"1.0\"?>");
@out.WriteLine("<allocations>");
@out.WriteLine("<queue name=\"queueA\">");
@out.WriteLine("<weight>.2</weight>");
@out.WriteLine("</queue>");
@out.WriteLine("<queue name=\"queueB\">");
@out.WriteLine("<weight>.8</weight>");
@out.WriteLine("<fairSharePreemptionThreshold>.4</fairSharePreemptionThreshold>");
@out.WriteLine("<queue name=\"queueB1\">");
@out.WriteLine("</queue>");
@out.WriteLine("<queue name=\"queueB2\">");
@out.WriteLine("<fairSharePreemptionThreshold>.6</fairSharePreemptionThreshold>");
@out.WriteLine("</queue>");
@out.WriteLine("</queue>");
@out.WriteLine("<defaultFairSharePreemptionThreshold>.5</defaultFairSharePreemptionThreshold>"
);
@out.WriteLine("</allocations>");
@out.Close();
resourceManager = new MockRM(conf);
resourceManager.Start();
scheduler = (FairScheduler)resourceManager.GetResourceScheduler();
// Add one big node (only care about aggregate capacity)
RMNode node1 = MockNodes.NewNodeInfo(1, Resources.CreateResource(10 * 1024, 10),
1, "127.0.0.1");
NodeAddedSchedulerEvent nodeEvent1 = new NodeAddedSchedulerEvent(node1);
scheduler.Handle(nodeEvent1);
scheduler.Update();
// Queue A wants 4 * 1024. Node update gives this all to A
CreateSchedulingRequest(1 * 1024, "queueA", "user1", 4);
scheduler.Update();
NodeUpdateSchedulerEvent nodeEvent2 = new NodeUpdateSchedulerEvent(node1);
for (int i = 0; i < 4; i++)
{
scheduler.Handle(nodeEvent2);
}
QueueManager queueMgr = scheduler.GetQueueManager();
FSLeafQueue queueA = queueMgr.GetLeafQueue("queueA", false);
NUnit.Framework.Assert.AreEqual(4 * 1024, queueA.GetResourceUsage().GetMemory());
// Both queue B1 and queue B2 want 3 * 1024
CreateSchedulingRequest(1 * 1024, "queueB.queueB1", "user1", 3);
CreateSchedulingRequest(1 * 1024, "queueB.queueB2", "user1", 3);
scheduler.Update();
for (int i_1 = 0; i_1 < 4; i_1++)
{
scheduler.Handle(nodeEvent2);
}
FSLeafQueue queueB1 = queueMgr.GetLeafQueue("queueB.queueB1", false);
FSLeafQueue queueB2 = queueMgr.GetLeafQueue("queueB.queueB2", false);
NUnit.Framework.Assert.AreEqual(2 * 1024, queueB1.GetResourceUsage().GetMemory());
NUnit.Framework.Assert.AreEqual(2 * 1024, queueB2.GetResourceUsage().GetMemory());
// For queue B1, the fairSharePreemptionThreshold is 0.4, and the fair share
// threshold is 1.6 * 1024
NUnit.Framework.Assert.IsFalse(queueB1.IsStarvedForFairShare());
// For queue B2, the fairSharePreemptionThreshold is 0.6, and the fair share
// threshold is 2.4 * 1024
NUnit.Framework.Assert.IsTrue(queueB2.IsStarvedForFairShare());
// Node checks in again
scheduler.Handle(nodeEvent2);
scheduler.Handle(nodeEvent2);
NUnit.Framework.Assert.AreEqual(3 * 1024, queueB1.GetResourceUsage().GetMemory());
NUnit.Framework.Assert.AreEqual(3 * 1024, queueB2.GetResourceUsage().GetMemory());
// Both queue B1 and queue B2 usages go to 3 * 1024
NUnit.Framework.Assert.IsFalse(queueB1.IsStarvedForFairShare());
NUnit.Framework.Assert.IsFalse(queueB2.IsStarvedForFairShare());
}
/// <exception cref="System.Exception"/>
public virtual void TestQueueMapping()
{
CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
SetupQueueConfiguration(csConf);
YarnConfiguration conf = new YarnConfiguration(csConf);
CapacityScheduler cs = new CapacityScheduler();
RMContext rmContext = TestUtils.GetMockRMContext();
cs.SetConf(conf);
cs.SetRMContext(rmContext);
cs.Init(conf);
cs.Start();
conf.SetClass(CommonConfigurationKeys.HadoopSecurityGroupMapping, typeof(SimpleGroupsMapping
), typeof(GroupMappingServiceProvider));
conf.Set(CapacitySchedulerConfiguration.EnableQueueMappingOverride, "true");
// configuration parsing tests - negative test cases
CheckInvalidQMapping(conf, cs, "x:a:b", "invalid specifier");
CheckInvalidQMapping(conf, cs, "u:a", "no queue specified");
CheckInvalidQMapping(conf, cs, "g:a", "no queue specified");
CheckInvalidQMapping(conf, cs, "u:a:b,g:a", "multiple mappings with invalid mapping"
);
CheckInvalidQMapping(conf, cs, "u:a:b,g:a:d:e", "too many path segments");
CheckInvalidQMapping(conf, cs, "u::", "empty source and queue");
CheckInvalidQMapping(conf, cs, "u:", "missing source missing queue");
CheckInvalidQMapping(conf, cs, "u:a:", "empty source missing q");
// simple base case for mapping user to queue
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "u:a:" + Q1);
cs.Reinitialize(conf, null);
CheckQMapping("a", Q1, cs);
// group mapping test
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "g:agroup:" + Q1);
cs.Reinitialize(conf, null);
CheckQMapping("a", Q1, cs);
// %user tests
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "u:%user:"******"a", Q2, cs);
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "u:%user:%user");
cs.Reinitialize(conf, null);
CheckQMapping("a", "a", cs);
// %primary_group tests
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "u:%user:%primary_group");
cs.Reinitialize(conf, null);
CheckQMapping("a", "agroup", cs);
// non-primary group mapping
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "g:asubgroup1:" + Q1);
cs.Reinitialize(conf, null);
CheckQMapping("a", Q1, cs);
// space trimming
conf.Set(CapacitySchedulerConfiguration.QueueMapping, " u : a : " + Q1);
cs.Reinitialize(conf, null);
CheckQMapping("a", Q1, cs);
csConf = new CapacitySchedulerConfiguration();
csConf.Set(YarnConfiguration.RmScheduler, typeof(CapacityScheduler).FullName);
SetupQueueConfiguration(csConf);
conf = new YarnConfiguration(csConf);
resourceManager = new MockRM(csConf);
resourceManager.Start();
conf.SetClass(CommonConfigurationKeys.HadoopSecurityGroupMapping, typeof(SimpleGroupsMapping
), typeof(GroupMappingServiceProvider));
conf.Set(CapacitySchedulerConfiguration.EnableQueueMappingOverride, "true");
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "u:user:"******"user", Q2, Q1);
// toggle admin override and retry
conf.SetBoolean(CapacitySchedulerConfiguration.EnableQueueMappingOverride, false);
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "u:user:"******"user", Q2, Q2);
// ensure that if a user does not specify a Q, the user mapping is used
CheckAppQueue(resourceManager, "user", null, Q1);
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "g:usergroup:" + Q2);
SetupQueueConfiguration(csConf);
resourceManager.GetResourceScheduler().Reinitialize(conf, null);
// ensure that if a user does not specify a Q, the group mapping is used
CheckAppQueue(resourceManager, "user", null, Q2);
// if the mapping specifies a queue that does not exist, the job is rejected
conf.Set(CapacitySchedulerConfiguration.QueueMapping, "u:user:non_existent_queue"
);
SetupQueueConfiguration(csConf);
bool fail = false;
try
{
resourceManager.GetResourceScheduler().Reinitialize(conf, null);
}
catch (IOException)
{
fail = true;
}
NUnit.Framework.Assert.IsTrue("queue initialization failed for non-existent q", fail
);
resourceManager.Stop();
}
public virtual void TestAddAndRemoveAppFromFiFoScheduler()
{
Configuration conf = new Configuration();
conf.SetClass(YarnConfiguration.RmScheduler, typeof(FifoScheduler), typeof(ResourceScheduler
));
MockRM rm = new MockRM(conf);
AbstractYarnScheduler <SchedulerApplicationAttempt, SchedulerNode> fs = (AbstractYarnScheduler
<SchedulerApplicationAttempt, SchedulerNode>)rm.GetResourceScheduler();
TestSchedulerUtils.VerifyAppAddedAndRemovedFromScheduler(fs.GetSchedulerApplications
(), fs, "queue");
}