/// <exception cref="System.Exception"/>
private void TestMinimumAllocation(YarnConfiguration conf, int testAlloc)
{
MockRM rm = new MockRM(conf);
rm.Start();
// Register node1
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb);
// Submit an application
RMApp app1 = rm.SubmitApp(testAlloc);
// kick the scheduling
nm1.NodeHeartbeat(true);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = rm.SendAMLaunched(attempt1.GetAppAttemptId());
am1.RegisterAppAttempt();
SchedulerNodeReport report_nm1 = rm.GetResourceScheduler().GetNodeReport(nm1.GetNodeId
());
int checkAlloc = conf.GetInt(YarnConfiguration.RmSchedulerMinimumAllocationMb, YarnConfiguration
.DefaultRmSchedulerMinimumAllocationMb);
NUnit.Framework.Assert.AreEqual(checkAlloc, report_nm1.GetUsedResource().GetMemory
());
rm.Stop();
}
/// <exception cref="System.Exception"/>
public virtual void TestAppOnMultiNode()
{
Logger rootLogger = LogManager.GetRootLogger();
rootLogger.SetLevel(Level.Debug);
conf.Set("yarn.scheduler.capacity.node-locality-delay", "-1");
MockRM rm = new MockRM(conf);
rm.Start();
MockNM nm1 = rm.RegisterNode("h1:1234", 5120);
MockNM nm2 = rm.RegisterNode("h2:5678", 10240);
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 = 13;
am.Allocate("h1", 1000, request, new AList <ContainerId>());
//kick the scheduler
IList <Container> conts = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId
>()).GetAllocatedContainers();
int contReceived = conts.Count;
while (contReceived < 3)
{
//only 3 containers are available on node1
nm1.NodeHeartbeat(true);
Sharpen.Collections.AddAll(conts, am.Allocate(new AList <ResourceRequest>(), new AList
<ContainerId>()).GetAllocatedContainers());
contReceived = conts.Count;
Log.Info("Got " + contReceived + " containers. Waiting to get " + 3);
Sharpen.Thread.Sleep(WaitSleepMs);
}
NUnit.Framework.Assert.AreEqual(3, conts.Count);
//send node2 heartbeat
conts = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>()).GetAllocatedContainers
();
contReceived = conts.Count;
while (contReceived < 10)
{
nm2.NodeHeartbeat(true);
Sharpen.Collections.AddAll(conts, am.Allocate(new AList <ResourceRequest>(), new AList
<ContainerId>()).GetAllocatedContainers());
contReceived = conts.Count;
Log.Info("Got " + contReceived + " containers. Waiting to get " + 10);
Sharpen.Thread.Sleep(WaitSleepMs);
}
NUnit.Framework.Assert.AreEqual(10, conts.Count);
am.UnregisterAppAttempt();
nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState.Complete);
am.WaitForState(RMAppAttemptState.Finished);
rm.Stop();
}
/// <exception cref="System.Exception"/>
public virtual void TestInvalidContainerReleaseRequest()
{
MockRM rm = new MockRM(conf);
try
{
rm.Start();
// Register node1
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb);
// Submit an application
RMApp app1 = rm.SubmitApp(1024);
// kick the scheduling
nm1.NodeHeartbeat(true);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = rm.SendAMLaunched(attempt1.GetAppAttemptId());
am1.RegisterAppAttempt();
am1.AddRequests(new string[] { "127.0.0.1" }, Gb, 1, 1);
AllocateResponse alloc1Response = am1.Schedule();
// send the request
// kick the scheduler
nm1.NodeHeartbeat(true);
while (alloc1Response.GetAllocatedContainers().Count < 1)
{
Log.Info("Waiting for containers to be created for app 1...");
Sharpen.Thread.Sleep(1000);
alloc1Response = am1.Schedule();
}
NUnit.Framework.Assert.IsTrue(alloc1Response.GetAllocatedContainers().Count > 0);
RMApp app2 = rm.SubmitApp(1024);
nm1.NodeHeartbeat(true);
RMAppAttempt attempt2 = app2.GetCurrentAppAttempt();
MockAM am2 = rm.SendAMLaunched(attempt2.GetAppAttemptId());
am2.RegisterAppAttempt();
// Now trying to release container allocated for app1 -> appAttempt1.
ContainerId cId = alloc1Response.GetAllocatedContainers()[0].GetId();
am2.AddContainerToBeReleased(cId);
try
{
am2.Schedule();
NUnit.Framework.Assert.Fail("Exception was expected!!");
}
catch (InvalidContainerReleaseException e)
{
StringBuilder sb = new StringBuilder("Cannot release container : ");
sb.Append(cId.ToString());
sb.Append(" not belonging to this application attempt : ");
sb.Append(attempt2.GetAppAttemptId().ToString());
NUnit.Framework.Assert.IsTrue(e.Message.Contains(sb.ToString()));
}
}
finally
{
if (rm != null)
{
rm.Stop();
}
}
}
/// <exception cref="System.Exception"/>
public static MockAM LaunchAndRegisterAM(RMApp app, Org.Apache.Hadoop.Yarn.Server.Resourcemanager.MockRM
rm, MockNM nm)
{
MockAM am = LaunchAM(app, rm, nm);
am.RegisterAppAttempt();
rm.WaitForState(app.GetApplicationId(), RMAppState.Running);
return(am);
}
public virtual void TestAMLaunchAndCleanup()
{
Logger rootLogger = LogManager.GetRootLogger();
rootLogger.SetLevel(Level.Debug);
TestApplicationMasterLauncher.MyContainerManagerImpl containerManager = new TestApplicationMasterLauncher.MyContainerManagerImpl
();
MockRMWithCustomAMLauncher rm = new MockRMWithCustomAMLauncher(containerManager);
rm.Start();
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 5120);
RMApp app = rm.SubmitApp(2000);
// kick the scheduling
nm1.NodeHeartbeat(true);
int waitCount = 0;
while (containerManager.launched == false && waitCount++ < 20)
{
Log.Info("Waiting for AM Launch to happen..");
Sharpen.Thread.Sleep(1000);
}
NUnit.Framework.Assert.IsTrue(containerManager.launched);
RMAppAttempt attempt = app.GetCurrentAppAttempt();
ApplicationAttemptId appAttemptId = attempt.GetAppAttemptId();
NUnit.Framework.Assert.AreEqual(appAttemptId.ToString(), containerManager.attemptIdAtContainerManager
);
NUnit.Framework.Assert.AreEqual(app.GetSubmitTime(), containerManager.submitTimeAtContainerManager
);
NUnit.Framework.Assert.AreEqual(app.GetRMAppAttempt(appAttemptId).GetMasterContainer
().GetId().ToString(), containerManager.containerIdAtContainerManager);
NUnit.Framework.Assert.AreEqual(nm1.GetNodeId().ToString(), containerManager.nmHostAtContainerManager
);
NUnit.Framework.Assert.AreEqual(YarnConfiguration.DefaultRmAmMaxAttempts, containerManager
.maxAppAttempts);
MockAM am = new MockAM(rm.GetRMContext(), rm.GetApplicationMasterService(), appAttemptId
);
am.RegisterAppAttempt();
am.UnregisterAppAttempt();
//complete the AM container to finish the app normally
nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState.Complete);
am.WaitForState(RMAppAttemptState.Finished);
waitCount = 0;
while (containerManager.cleanedup == false && waitCount++ < 20)
{
Log.Info("Waiting for AM Cleanup to happen..");
Sharpen.Thread.Sleep(1000);
}
NUnit.Framework.Assert.IsTrue(containerManager.cleanedup);
am.WaitForState(RMAppAttemptState.Finished);
rm.Stop();
}
/// <exception cref="System.Exception"/>
private MockAM LaunchAM(RMApp app, MockRM rm, MockNM nm)
{
RMAppAttempt attempt = app.GetCurrentAppAttempt();
nm.NodeHeartbeat(true);
MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId());
am.RegisterAppAttempt();
rm.WaitForState(app.GetApplicationId(), RMAppState.Running);
return(am);
}
// Test even if AM container is allocated with containerId not equal to 1, the
// following allocate requests from AM should be able to retrieve the
// corresponding NM Token.
/// <exception cref="System.Exception"/>
public virtual void TestNMTokenSentForNormalContainer()
{
conf.Set(YarnConfiguration.RmScheduler, typeof(CapacityScheduler).GetCanonicalName
());
MockRM rm = new MockRM(conf);
rm.Start();
MockNM nm1 = rm.RegisterNode("h1:1234", 5120);
RMApp app = rm.SubmitApp(2000);
RMAppAttempt attempt = app.GetCurrentAppAttempt();
// Call getNewContainerId to increase container Id so that the AM container
// Id doesn't equal to one.
CapacityScheduler cs = (CapacityScheduler)rm.GetResourceScheduler();
cs.GetApplicationAttempt(attempt.GetAppAttemptId()).GetNewContainerId();
// kick the scheduling
nm1.NodeHeartbeat(true);
MockAM am = MockRM.LaunchAM(app, rm, nm1);
// am container Id not equal to 1.
NUnit.Framework.Assert.IsTrue(attempt.GetMasterContainer().GetId().GetContainerId
() != 1);
// NMSecretManager doesn't record the node on which the am is allocated.
NUnit.Framework.Assert.IsFalse(rm.GetRMContext().GetNMTokenSecretManager().IsApplicationAttemptNMTokenPresent
(attempt.GetAppAttemptId(), nm1.GetNodeId()));
am.RegisterAppAttempt();
rm.WaitForState(app.GetApplicationId(), RMAppState.Running);
int NumContainers = 1;
IList <Container> containers = new AList <Container>();
// nmTokens keeps track of all the nmTokens issued in the allocate call.
IList <NMToken> expectedNMTokens = new AList <NMToken>();
// am1 allocate 1 container on nm1.
while (true)
{
AllocateResponse response = am.Allocate("127.0.0.1", 2000, NumContainers, new AList
<ContainerId>());
nm1.NodeHeartbeat(true);
Sharpen.Collections.AddAll(containers, response.GetAllocatedContainers());
Sharpen.Collections.AddAll(expectedNMTokens, response.GetNMTokens());
if (containers.Count == NumContainers)
{
break;
}
Sharpen.Thread.Sleep(200);
System.Console.Out.WriteLine("Waiting for container to be allocated.");
}
NodeId nodeId = expectedNMTokens[0].GetNodeId();
// NMToken is sent for the allocated container.
NUnit.Framework.Assert.AreEqual(nm1.GetNodeId(), nodeId);
}
/// <exception cref="System.Exception"/>
public virtual void TestUsageWithOneAttemptAndOneContainer()
{
MockRM rm = new MockRM(conf);
rm.Start();
MockNM nm = new MockNM("127.0.0.1:1234", 15120, rm.GetResourceTrackerService());
nm.RegisterNode();
RMApp app0 = rm.SubmitApp(200);
RMAppMetrics rmAppMetrics = app0.GetRMAppMetrics();
NUnit.Framework.Assert.IsTrue("Before app submittion, memory seconds should have been 0 but was "
+ rmAppMetrics.GetMemorySeconds(), rmAppMetrics.GetMemorySeconds() == 0);
NUnit.Framework.Assert.IsTrue("Before app submission, vcore seconds should have been 0 but was "
+ rmAppMetrics.GetVcoreSeconds(), rmAppMetrics.GetVcoreSeconds() == 0);
RMAppAttempt attempt0 = app0.GetCurrentAppAttempt();
nm.NodeHeartbeat(true);
MockAM am0 = rm.SendAMLaunched(attempt0.GetAppAttemptId());
am0.RegisterAppAttempt();
RMContainer rmContainer = rm.GetResourceScheduler().GetRMContainer(attempt0.GetMasterContainer
().GetId());
// Allow metrics to accumulate.
int sleepInterval = 1000;
int cumulativeSleepTime = 0;
while (rmAppMetrics.GetMemorySeconds() <= 0 && cumulativeSleepTime < 5000)
{
Sharpen.Thread.Sleep(sleepInterval);
cumulativeSleepTime += sleepInterval;
}
rmAppMetrics = app0.GetRMAppMetrics();
NUnit.Framework.Assert.IsTrue("While app is running, memory seconds should be >0 but is "
+ rmAppMetrics.GetMemorySeconds(), rmAppMetrics.GetMemorySeconds() > 0);
NUnit.Framework.Assert.IsTrue("While app is running, vcore seconds should be >0 but is "
+ rmAppMetrics.GetVcoreSeconds(), rmAppMetrics.GetVcoreSeconds() > 0);
MockRM.FinishAMAndVerifyAppState(app0, rm, nm, am0);
AggregateAppResourceUsage ru = CalculateContainerResourceMetrics(rmContainer);
rmAppMetrics = app0.GetRMAppMetrics();
NUnit.Framework.Assert.AreEqual("Unexcpected MemorySeconds value", ru.GetMemorySeconds
(), rmAppMetrics.GetMemorySeconds());
NUnit.Framework.Assert.AreEqual("Unexpected VcoreSeconds value", ru.GetVcoreSeconds
(), rmAppMetrics.GetVcoreSeconds());
rm.Stop();
}
/// <exception cref="System.Exception"/>
public virtual void TestResourceTypes()
{
Dictionary <YarnConfiguration, EnumSet <YarnServiceProtos.SchedulerResourceTypes> >
driver = new Dictionary <YarnConfiguration, EnumSet <YarnServiceProtos.SchedulerResourceTypes
> >();
CapacitySchedulerConfiguration csconf = new CapacitySchedulerConfiguration();
csconf.SetResourceComparator(typeof(DominantResourceCalculator));
YarnConfiguration testCapacityDRConf = new YarnConfiguration(csconf);
testCapacityDRConf.SetClass(YarnConfiguration.RmScheduler, typeof(CapacityScheduler
), typeof(ResourceScheduler));
YarnConfiguration testCapacityDefConf = new YarnConfiguration();
testCapacityDefConf.SetClass(YarnConfiguration.RmScheduler, typeof(CapacityScheduler
), typeof(ResourceScheduler));
YarnConfiguration testFairDefConf = new YarnConfiguration();
testFairDefConf.SetClass(YarnConfiguration.RmScheduler, typeof(FairScheduler), typeof(
ResourceScheduler));
driver[conf] = EnumSet.Of(YarnServiceProtos.SchedulerResourceTypes.Memory);
driver[testCapacityDRConf] = EnumSet.Of(YarnServiceProtos.SchedulerResourceTypes.
Cpu, YarnServiceProtos.SchedulerResourceTypes.Memory);
driver[testCapacityDefConf] = EnumSet.Of(YarnServiceProtos.SchedulerResourceTypes
.Memory);
driver[testFairDefConf] = EnumSet.Of(YarnServiceProtos.SchedulerResourceTypes.Memory
, YarnServiceProtos.SchedulerResourceTypes.Cpu);
foreach (KeyValuePair <YarnConfiguration, EnumSet <YarnServiceProtos.SchedulerResourceTypes
> > entry in driver)
{
EnumSet <YarnServiceProtos.SchedulerResourceTypes> expectedValue = entry.Value;
MockRM rm = new MockRM(entry.Key);
rm.Start();
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb);
RMApp app1 = rm.SubmitApp(2048);
nm1.NodeHeartbeat(true);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = rm.SendAMLaunched(attempt1.GetAppAttemptId());
RegisterApplicationMasterResponse resp = am1.RegisterAppAttempt();
EnumSet <YarnServiceProtos.SchedulerResourceTypes> types = resp.GetSchedulerResourceTypes
();
Log.Info("types = " + types.ToString());
NUnit.Framework.Assert.AreEqual(expectedValue, types);
rm.Stop();
}
}
/// <exception cref="System.Exception"/>
public virtual void TestFinishApplicationMasterBeforeRegistering()
{
MockRM rm = new MockRM(conf);
try
{
rm.Start();
// Register node1
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb);
// Submit an application
RMApp app1 = rm.SubmitApp(2048);
MockAM am1 = MockRM.LaunchAM(app1, rm, nm1);
FinishApplicationMasterRequest req = FinishApplicationMasterRequest.NewInstance(FinalApplicationStatus
.Failed, string.Empty, string.Empty);
try
{
am1.UnregisterAppAttempt(req, false);
NUnit.Framework.Assert.Fail("ApplicationMasterNotRegisteredException should be thrown"
);
}
catch (ApplicationMasterNotRegisteredException e)
{
NUnit.Framework.Assert.IsNotNull(e);
NUnit.Framework.Assert.IsNotNull(e.Message);
NUnit.Framework.Assert.IsTrue(e.Message.Contains("Application Master is trying to unregister before registering for:"
));
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("ApplicationMasterNotRegisteredException should be thrown"
);
}
am1.RegisterAppAttempt();
am1.UnregisterAppAttempt(req, false);
am1.WaitForState(RMAppAttemptState.Finishing);
}
finally
{
if (rm != null)
{
rm.Stop();
}
}
}
/// <summary>Validate killing an application when it is at accepted state.</summary>
/// <exception cref="System.Exception">exception</exception>
public virtual void TestApplicationKillAtAcceptedState()
{
Dispatcher dispatcher = new _AsyncDispatcher_573();
MockRM rm = new _MockRM_596(dispatcher, conf);
// test metrics
QueueMetrics metrics = rm.GetResourceScheduler().GetRootQueueMetrics();
int appsKilled = metrics.GetAppsKilled();
int appsSubmitted = metrics.GetAppsSubmitted();
rm.Start();
MockNM nm1 = new MockNM("127.0.0.1:1234", 15120, rm.GetResourceTrackerService());
nm1.RegisterNode();
// a failed app
RMApp application = rm.SubmitApp(200);
MockAM am = MockRM.LaunchAM(application, rm, nm1);
am.WaitForState(RMAppAttemptState.Launched);
nm1.NodeHeartbeat(am.GetApplicationAttemptId(), 1, ContainerState.Running);
rm.WaitForState(application.GetApplicationId(), RMAppState.Accepted);
// Now kill the application before new attempt is launched, the app report
// returns the invalid AM host and port.
KillApplicationRequest request = KillApplicationRequest.NewInstance(application.GetApplicationId
());
rm.GetClientRMService().ForceKillApplication(request);
// Specific test for YARN-1689 follows
// Now let's say a race causes AM to register now. This should not crash RM.
am.RegisterAppAttempt(false);
// We explicitly intercepted the kill-event to RMAppAttempt, so app should
// still be in KILLING state.
rm.WaitForState(application.GetApplicationId(), RMAppState.Killing);
// AM should now be in running
rm.WaitForState(am.GetApplicationAttemptId(), RMAppAttemptState.Running);
// Simulate that appAttempt is killed.
rm.GetRMContext().GetDispatcher().GetEventHandler().Handle(new RMAppEvent(application
.GetApplicationId(), RMAppEventType.AttemptKilled));
rm.WaitForState(application.GetApplicationId(), RMAppState.Killed);
// test metrics
metrics = rm.GetResourceScheduler().GetRootQueueMetrics();
NUnit.Framework.Assert.AreEqual(appsKilled + 1, metrics.GetAppsKilled());
NUnit.Framework.Assert.AreEqual(appsSubmitted + 1, metrics.GetAppsSubmitted());
}
/// <exception cref="System.Exception"/>
public virtual void TestAppWithNoContainers()
{
Logger rootLogger = LogManager.GetRootLogger();
rootLogger.SetLevel(Level.Debug);
MockRM rm = new MockRM(conf);
rm.Start();
MockNM nm1 = rm.RegisterNode("h1:1234", 5120);
RMApp app = rm.SubmitApp(2000);
//kick the scheduling
nm1.NodeHeartbeat(true);
RMAppAttempt attempt = app.GetCurrentAppAttempt();
MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId());
am.RegisterAppAttempt();
am.UnregisterAppAttempt();
nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState.Complete);
am.WaitForState(RMAppAttemptState.Finished);
rm.Stop();
}
/// <exception cref="System.Exception"/>
public virtual void TestRMIdentifierOnContainerAllocation()
{
MockRM rm = new MockRM(conf);
rm.Start();
// Register node1
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb);
// Submit an application
RMApp app1 = rm.SubmitApp(2048);
// kick the scheduling
nm1.NodeHeartbeat(true);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = rm.SendAMLaunched(attempt1.GetAppAttemptId());
am1.RegisterAppAttempt();
am1.AddRequests(new string[] { "127.0.0.1" }, Gb, 1, 1);
AllocateResponse alloc1Response = am1.Schedule();
// send the request
// kick the scheduler
nm1.NodeHeartbeat(true);
while (alloc1Response.GetAllocatedContainers().Count < 1)
{
Log.Info("Waiting for containers to be created for app 1...");
Sharpen.Thread.Sleep(1000);
alloc1Response = am1.Schedule();
}
// assert RMIdentifer is set properly in allocated containers
Container allocatedContainer = alloc1Response.GetAllocatedContainers()[0];
ContainerTokenIdentifier tokenId = BuilderUtils.NewContainerTokenIdentifier(allocatedContainer
.GetContainerToken());
NUnit.Framework.Assert.AreEqual(MockRM.GetClusterTimeStamp(), tokenId.GetRMIdentifier
());
rm.Stop();
}
public virtual void TestAllocateContainerOnNodeWithoutOffSwitchSpecified()
{
Logger rootLogger = LogManager.GetRootLogger();
rootLogger.SetLevel(Level.Debug);
MockRM rm = new MockRM(conf);
rm.Start();
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb);
RMApp app1 = rm.SubmitApp(2048);
// kick the scheduling, 2 GB given to AM1, remaining 4GB on nm1
nm1.NodeHeartbeat(true);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = rm.SendAMLaunched(attempt1.GetAppAttemptId());
am1.RegisterAppAttempt();
// add request for containers
IList <ResourceRequest> requests = new AList <ResourceRequest>();
requests.AddItem(am1.CreateResourceReq("127.0.0.1", 1 * Gb, 1, 1));
requests.AddItem(am1.CreateResourceReq("/default-rack", 1 * Gb, 1, 1));
am1.Allocate(requests, null);
// send the request
try
{
// kick the schedule
nm1.NodeHeartbeat(true);
}
catch (ArgumentNullException)
{
NUnit.Framework.Assert.Fail("NPE when allocating container on node but " + "forget to set off-switch request should be handled"
);
}
rm.Stop();
}
public virtual void Test()
{
Logger rootLogger = LogManager.GetRootLogger();
rootLogger.SetLevel(Level.Debug);
MockRM rm = new MockRM(conf);
rm.Start();
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb);
MockNM nm2 = rm.RegisterNode("127.0.0.2:5678", 4 * Gb);
RMApp app1 = rm.SubmitApp(2048);
// kick the scheduling, 2 GB given to AM1, remaining 4GB on nm1
nm1.NodeHeartbeat(true);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = rm.SendAMLaunched(attempt1.GetAppAttemptId());
am1.RegisterAppAttempt();
SchedulerNodeReport report_nm1 = rm.GetResourceScheduler().GetNodeReport(nm1.GetNodeId
());
NUnit.Framework.Assert.AreEqual(2 * Gb, report_nm1.GetUsedResource().GetMemory());
RMApp app2 = rm.SubmitApp(2048);
// kick the scheduling, 2GB given to AM, remaining 2 GB on nm2
nm2.NodeHeartbeat(true);
RMAppAttempt attempt2 = app2.GetCurrentAppAttempt();
MockAM am2 = rm.SendAMLaunched(attempt2.GetAppAttemptId());
am2.RegisterAppAttempt();
SchedulerNodeReport report_nm2 = rm.GetResourceScheduler().GetNodeReport(nm2.GetNodeId
());
NUnit.Framework.Assert.AreEqual(2 * Gb, report_nm2.GetUsedResource().GetMemory());
// add request for containers
am1.AddRequests(new string[] { "127.0.0.1", "127.0.0.2" }, Gb, 1, 1);
AllocateResponse alloc1Response = am1.Schedule();
// send the request
// add request for containers
am2.AddRequests(new string[] { "127.0.0.1", "127.0.0.2" }, 3 * Gb, 0, 1);
AllocateResponse alloc2Response = am2.Schedule();
// send the request
// kick the scheduler, 1 GB and 3 GB given to AM1 and AM2, remaining 0
nm1.NodeHeartbeat(true);
while (alloc1Response.GetAllocatedContainers().Count < 1)
{
Log.Info("Waiting for containers to be created for app 1...");
Sharpen.Thread.Sleep(1000);
alloc1Response = am1.Schedule();
}
while (alloc2Response.GetAllocatedContainers().Count < 1)
{
Log.Info("Waiting for containers to be created for app 2...");
Sharpen.Thread.Sleep(1000);
alloc2Response = am2.Schedule();
}
// kick the scheduler, nothing given remaining 2 GB.
nm2.NodeHeartbeat(true);
IList <Container> allocated1 = alloc1Response.GetAllocatedContainers();
NUnit.Framework.Assert.AreEqual(1, allocated1.Count);
NUnit.Framework.Assert.AreEqual(1 * Gb, allocated1[0].GetResource().GetMemory());
NUnit.Framework.Assert.AreEqual(nm1.GetNodeId(), allocated1[0].GetNodeId());
IList <Container> allocated2 = alloc2Response.GetAllocatedContainers();
NUnit.Framework.Assert.AreEqual(1, allocated2.Count);
NUnit.Framework.Assert.AreEqual(3 * Gb, allocated2[0].GetResource().GetMemory());
NUnit.Framework.Assert.AreEqual(nm1.GetNodeId(), allocated2[0].GetNodeId());
report_nm1 = rm.GetResourceScheduler().GetNodeReport(nm1.GetNodeId());
report_nm2 = rm.GetResourceScheduler().GetNodeReport(nm2.GetNodeId());
NUnit.Framework.Assert.AreEqual(0, report_nm1.GetAvailableResource().GetMemory());
NUnit.Framework.Assert.AreEqual(2 * Gb, report_nm2.GetAvailableResource().GetMemory
());
NUnit.Framework.Assert.AreEqual(6 * Gb, report_nm1.GetUsedResource().GetMemory());
NUnit.Framework.Assert.AreEqual(2 * Gb, report_nm2.GetUsedResource().GetMemory());
Container c1 = allocated1[0];
NUnit.Framework.Assert.AreEqual(Gb, c1.GetResource().GetMemory());
ContainerStatus containerStatus = BuilderUtils.NewContainerStatus(c1.GetId(), ContainerState
.Complete, string.Empty, 0);
nm1.ContainerStatus(containerStatus);
int waitCount = 0;
while (attempt1.GetJustFinishedContainers().Count < 1 && waitCount++ != 20)
{
Log.Info("Waiting for containers to be finished for app 1... Tried " + waitCount
+ " times already..");
Sharpen.Thread.Sleep(1000);
}
NUnit.Framework.Assert.AreEqual(1, attempt1.GetJustFinishedContainers().Count);
NUnit.Framework.Assert.AreEqual(1, am1.Schedule().GetCompletedContainersStatuses(
).Count);
report_nm1 = rm.GetResourceScheduler().GetNodeReport(nm1.GetNodeId());
NUnit.Framework.Assert.AreEqual(5 * Gb, report_nm1.GetUsedResource().GetMemory());
rm.Stop();
}
/// <exception cref="System.Exception"/>
private void AmRestartTests(bool keepRunningContainers)
{
MockRM rm = new MockRM(conf);
rm.Start();
RMApp app = rm.SubmitApp(200, "name", "user", new Dictionary <ApplicationAccessType
, string>(), false, "default", -1, null, "MAPREDUCE", false, keepRunningContainers
);
MockNM nm = new MockNM("127.0.0.1:1234", 10240, rm.GetResourceTrackerService());
nm.RegisterNode();
MockAM am0 = MockRM.LaunchAndRegisterAM(app, rm, nm);
int NumContainers = 1;
// allocate NUM_CONTAINERS containers
am0.Allocate("127.0.0.1", 1024, NumContainers, new AList <ContainerId>());
nm.NodeHeartbeat(true);
// wait for containers to be allocated.
IList <Container> containers = am0.Allocate(new AList <ResourceRequest>(), new AList
<ContainerId>()).GetAllocatedContainers();
while (containers.Count != NumContainers)
{
nm.NodeHeartbeat(true);
Sharpen.Collections.AddAll(containers, am0.Allocate(new AList <ResourceRequest>(),
new AList <ContainerId>()).GetAllocatedContainers());
Sharpen.Thread.Sleep(200);
}
// launch the 2nd container.
ContainerId containerId2 = ContainerId.NewContainerId(am0.GetApplicationAttemptId
(), 2);
nm.NodeHeartbeat(am0.GetApplicationAttemptId(), containerId2.GetContainerId(), ContainerState
.Running);
rm.WaitForState(nm, containerId2, RMContainerState.Running);
// Capture the containers here so the metrics can be calculated after the
// app has completed.
ICollection <RMContainer> rmContainers = rm.scheduler.GetSchedulerAppInfo(am0.GetApplicationAttemptId
()).GetLiveContainers();
// fail the first app attempt by sending CONTAINER_FINISHED event without
// registering.
ContainerId amContainerId = app.GetCurrentAppAttempt().GetMasterContainer().GetId
();
nm.NodeHeartbeat(am0.GetApplicationAttemptId(), amContainerId.GetContainerId(), ContainerState
.Complete);
am0.WaitForState(RMAppAttemptState.Failed);
long memorySeconds = 0;
long vcoreSeconds = 0;
// Calculate container usage metrics for first attempt.
if (keepRunningContainers)
{
// Only calculate the usage for the one container that has completed.
foreach (RMContainer c in rmContainers)
{
if (c.GetContainerId().Equals(amContainerId))
{
AggregateAppResourceUsage ru = CalculateContainerResourceMetrics(c);
memorySeconds += ru.GetMemorySeconds();
vcoreSeconds += ru.GetVcoreSeconds();
}
else
{
// The remaining container should be RUNNING.
NUnit.Framework.Assert.IsTrue("After first attempt failed, remaining container "
+ "should still be running. ", c.GetContainerState().Equals(ContainerState.Running
));
}
}
}
else
{
// If keepRunningContainers is false, all live containers should now
// be completed. Calculate the resource usage metrics for all of them.
foreach (RMContainer c in rmContainers)
{
AggregateAppResourceUsage ru = CalculateContainerResourceMetrics(c);
memorySeconds += ru.GetMemorySeconds();
vcoreSeconds += ru.GetVcoreSeconds();
}
}
// wait for app to start a new attempt.
rm.WaitForState(app.GetApplicationId(), RMAppState.Accepted);
// assert this is a new AM.
RMAppAttempt attempt2 = app.GetCurrentAppAttempt();
NUnit.Framework.Assert.IsFalse(attempt2.GetAppAttemptId().Equals(am0.GetApplicationAttemptId
()));
// launch the new AM
nm.NodeHeartbeat(true);
MockAM am1 = rm.SendAMLaunched(attempt2.GetAppAttemptId());
am1.RegisterAppAttempt();
// allocate NUM_CONTAINERS containers
am1.Allocate("127.0.0.1", 1024, NumContainers, new AList <ContainerId>());
nm.NodeHeartbeat(true);
// wait for containers to be allocated.
containers = am1.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>())
.GetAllocatedContainers();
while (containers.Count != NumContainers)
{
nm.NodeHeartbeat(true);
Sharpen.Collections.AddAll(containers, am1.Allocate(new AList <ResourceRequest>(),
new AList <ContainerId>()).GetAllocatedContainers());
Sharpen.Thread.Sleep(200);
}
rm.WaitForState(app.GetApplicationId(), RMAppState.Running);
// Capture running containers for later use by metrics calculations.
rmContainers = rm.scheduler.GetSchedulerAppInfo(attempt2.GetAppAttemptId()).GetLiveContainers
();
// complete container by sending the container complete event which has
// earlier attempt's attemptId
amContainerId = app.GetCurrentAppAttempt().GetMasterContainer().GetId();
nm.NodeHeartbeat(am0.GetApplicationAttemptId(), amContainerId.GetContainerId(), ContainerState
.Complete);
MockRM.FinishAMAndVerifyAppState(app, rm, nm, am1);
// Calculate container usage metrics for second attempt.
foreach (RMContainer c_1 in rmContainers)
{
AggregateAppResourceUsage ru = CalculateContainerResourceMetrics(c_1);
memorySeconds += ru.GetMemorySeconds();
vcoreSeconds += ru.GetVcoreSeconds();
}
RMAppMetrics rmAppMetrics = app.GetRMAppMetrics();
NUnit.Framework.Assert.AreEqual("Unexcpected MemorySeconds value", memorySeconds,
rmAppMetrics.GetMemorySeconds());
NUnit.Framework.Assert.AreEqual("Unexpected VcoreSeconds value", vcoreSeconds, rmAppMetrics
.GetVcoreSeconds());
rm.Stop();
return;
}
/// <exception cref="System.Exception"/>
public virtual void TestUsageWithMultipleContainersAndRMRestart()
{
// Set max attempts to 1 so that when the first attempt fails, the app
// won't try to start a new one.
conf.SetInt(YarnConfiguration.RmAmMaxAttempts, 1);
conf.SetBoolean(YarnConfiguration.RecoveryEnabled, true);
conf.SetBoolean(YarnConfiguration.RmWorkPreservingRecoveryEnabled, false);
MemoryRMStateStore memStore = new MemoryRMStateStore();
memStore.Init(conf);
MockRM rm0 = new MockRM(conf, memStore);
rm0.Start();
MockNM nm = new MockNM("127.0.0.1:1234", 65536, rm0.GetResourceTrackerService());
nm.RegisterNode();
RMApp app0 = rm0.SubmitApp(200);
rm0.WaitForState(app0.GetApplicationId(), RMAppState.Accepted);
RMAppAttempt attempt0 = app0.GetCurrentAppAttempt();
ApplicationAttemptId attemptId0 = attempt0.GetAppAttemptId();
rm0.WaitForState(attemptId0, RMAppAttemptState.Scheduled);
nm.NodeHeartbeat(true);
rm0.WaitForState(attemptId0, RMAppAttemptState.Allocated);
MockAM am0 = rm0.SendAMLaunched(attempt0.GetAppAttemptId());
am0.RegisterAppAttempt();
int NumContainers = 2;
am0.Allocate("127.0.0.1", 1000, NumContainers, new AList <ContainerId>());
nm.NodeHeartbeat(true);
IList <Container> conts = am0.Allocate(new AList <ResourceRequest>(), new AList <ContainerId
>()).GetAllocatedContainers();
while (conts.Count != NumContainers)
{
nm.NodeHeartbeat(true);
Sharpen.Collections.AddAll(conts, am0.Allocate(new AList <ResourceRequest>(), new
AList <ContainerId>()).GetAllocatedContainers());
Sharpen.Thread.Sleep(500);
}
// launch the 2nd and 3rd containers.
foreach (Container c in conts)
{
nm.NodeHeartbeat(attempt0.GetAppAttemptId(), c.GetId().GetContainerId(), ContainerState
.Running);
rm0.WaitForState(nm, c.GetId(), RMContainerState.Running);
}
// Get the RMContainers for all of the live containers, to be used later
// for metrics calculations and comparisons.
ICollection <RMContainer> rmContainers = rm0.scheduler.GetSchedulerAppInfo(attempt0
.GetAppAttemptId()).GetLiveContainers();
// Allow metrics to accumulate.
int sleepInterval = 1000;
int cumulativeSleepTime = 0;
while (app0.GetRMAppMetrics().GetMemorySeconds() <= 0 && cumulativeSleepTime < 5000
)
{
Sharpen.Thread.Sleep(sleepInterval);
cumulativeSleepTime += sleepInterval;
}
// Stop all non-AM containers
foreach (Container c_1 in conts)
{
if (c_1.GetId().GetContainerId() == 1)
{
continue;
}
nm.NodeHeartbeat(attempt0.GetAppAttemptId(), c_1.GetId().GetContainerId(), ContainerState
.Complete);
rm0.WaitForState(nm, c_1.GetId(), RMContainerState.Completed);
}
// After all other containers have completed, manually complete the master
// container in order to trigger a save to the state store of the resource
// usage metrics. This will cause the attempt to fail, and, since the max
// attempt retries is 1, the app will also fail. This is intentional so
// that all containers will complete prior to saving.
ContainerId cId = ContainerId.NewContainerId(attempt0.GetAppAttemptId(), 1);
nm.NodeHeartbeat(attempt0.GetAppAttemptId(), cId.GetContainerId(), ContainerState
.Complete);
rm0.WaitForState(nm, cId, RMContainerState.Completed);
// Check that the container metrics match those from the app usage report.
long memorySeconds = 0;
long vcoreSeconds = 0;
foreach (RMContainer c_2 in rmContainers)
{
AggregateAppResourceUsage ru = CalculateContainerResourceMetrics(c_2);
memorySeconds += ru.GetMemorySeconds();
vcoreSeconds += ru.GetVcoreSeconds();
}
RMAppMetrics metricsBefore = app0.GetRMAppMetrics();
NUnit.Framework.Assert.AreEqual("Unexcpected MemorySeconds value", memorySeconds,
metricsBefore.GetMemorySeconds());
NUnit.Framework.Assert.AreEqual("Unexpected VcoreSeconds value", vcoreSeconds, metricsBefore
.GetVcoreSeconds());
// create new RM to represent RM restart. Load up the state store.
MockRM rm1 = new MockRM(conf, memStore);
rm1.Start();
RMApp app0After = rm1.GetRMContext().GetRMApps()[app0.GetApplicationId()];
// Compare container resource usage metrics from before and after restart.
RMAppMetrics metricsAfter = app0After.GetRMAppMetrics();
NUnit.Framework.Assert.AreEqual("Vcore seconds were not the same after RM Restart"
, metricsBefore.GetVcoreSeconds(), metricsAfter.GetVcoreSeconds());
NUnit.Framework.Assert.AreEqual("Memory seconds were not the same after RM Restart"
, metricsBefore.GetMemorySeconds(), metricsAfter.GetMemorySeconds());
rm0.Stop();
rm0.Close();
rm1.Stop();
rm1.Close();
}
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 virtual void TestContainerCleanup()
{
Logger rootLogger = LogManager.GetRootLogger();
rootLogger.SetLevel(Level.Debug);
DrainDispatcher dispatcher = new DrainDispatcher();
MockRM rm = new _MockRM_167(this, dispatcher);
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>());
dispatcher.Await();
//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
();
dispatcher.Await();
contReceived += conts.Count;
nm1.NodeHeartbeat(true);
}
NUnit.Framework.Assert.AreEqual(request, contReceived);
// Release a container.
AList <ContainerId> release = new AList <ContainerId>();
release.AddItem(conts[0].GetId());
am.Allocate(new AList <ResourceRequest>(), release);
dispatcher.Await();
// Send one more heartbeat with a fake running container. This is to
// simulate the situation that can happen if the NM reports that container
// is running in the same heartbeat when the RM asks it to clean it up.
IDictionary <ApplicationId, IList <ContainerStatus> > containerStatuses = new Dictionary
<ApplicationId, IList <ContainerStatus> >();
AList <ContainerStatus> containerStatusList = new AList <ContainerStatus>();
containerStatusList.AddItem(BuilderUtils.NewContainerStatus(conts[0].GetId(), ContainerState
.Running, "nothing", 0));
containerStatuses[app.GetApplicationId()] = containerStatusList;
NodeHeartbeatResponse resp = nm1.NodeHeartbeat(containerStatuses, true);
WaitForContainerCleanup(dispatcher, nm1, resp);
// Now to test the case when RM already gave cleanup, and NM suddenly
// realizes that the container is running.
Log.Info("Testing container launch much after release and " + "NM getting cleanup"
);
containerStatuses.Clear();
containerStatusList.Clear();
containerStatusList.AddItem(BuilderUtils.NewContainerStatus(conts[0].GetId(), ContainerState
.Running, "nothing", 0));
containerStatuses[app.GetApplicationId()] = containerStatusList;
resp = nm1.NodeHeartbeat(containerStatuses, true);
// The cleanup list won't be instantaneous as it is given out by scheduler
// and not RMNodeImpl.
WaitForContainerCleanup(dispatcher, nm1, resp);
rm.Stop();
}
/// <exception cref="System.Exception"/>
public virtual void TestallocateBeforeAMRegistration()
{
Logger rootLogger = LogManager.GetRootLogger();
bool thrown = false;
rootLogger.SetLevel(Level.Debug);
MockRM rm = new MockRM();
rm.Start();
MockNM nm1 = rm.RegisterNode("h1:1234", 5000);
RMApp app = rm.SubmitApp(2000);
// kick the scheduling
nm1.NodeHeartbeat(true);
RMAppAttempt attempt = app.GetCurrentAppAttempt();
MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId());
// request for containers
int request = 2;
AllocateResponse ar = null;
try
{
ar = am.Allocate("h1", 1000, request, new AList <ContainerId>());
NUnit.Framework.Assert.Fail();
}
catch (ApplicationMasterNotRegisteredException)
{
}
// kick the scheduler
nm1.NodeHeartbeat(true);
AllocateResponse amrs = null;
try
{
amrs = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>());
NUnit.Framework.Assert.Fail();
}
catch (ApplicationMasterNotRegisteredException)
{
}
am.RegisterAppAttempt();
try
{
am.RegisterAppAttempt(false);
NUnit.Framework.Assert.Fail();
}
catch (Exception e)
{
NUnit.Framework.Assert.AreEqual("Application Master is already registered : " + attempt
.GetAppAttemptId().GetApplicationId(), e.Message);
}
// Simulate an AM that was disconnected and app attempt was removed
// (responseMap does not contain attemptid)
am.UnregisterAppAttempt();
nm1.NodeHeartbeat(attempt.GetAppAttemptId(), 1, ContainerState.Complete);
am.WaitForState(RMAppAttemptState.Finished);
try
{
amrs = am.Allocate(new AList <ResourceRequest>(), new AList <ContainerId>());
NUnit.Framework.Assert.Fail();
}
catch (ApplicationAttemptNotFoundException)
{
}
}
public virtual void TestResourceOverCommit()
{
MockRM rm = new MockRM(conf);
rm.Start();
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 4 * Gb);
RMApp app1 = rm.SubmitApp(2048);
// kick the scheduling, 2 GB given to AM1, remaining 2GB on nm1
nm1.NodeHeartbeat(true);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = rm.SendAMLaunched(attempt1.GetAppAttemptId());
am1.RegisterAppAttempt();
SchedulerNodeReport report_nm1 = rm.GetResourceScheduler().GetNodeReport(nm1.GetNodeId
());
// check node report, 2 GB used and 2 GB available
NUnit.Framework.Assert.AreEqual(2 * Gb, report_nm1.GetUsedResource().GetMemory());
NUnit.Framework.Assert.AreEqual(2 * Gb, report_nm1.GetAvailableResource().GetMemory
());
// add request for containers
am1.AddRequests(new string[] { "127.0.0.1", "127.0.0.2" }, 2 * Gb, 1, 1);
AllocateResponse alloc1Response = am1.Schedule();
// send the request
// kick the scheduler, 2 GB given to AM1, resource remaining 0
nm1.NodeHeartbeat(true);
while (alloc1Response.GetAllocatedContainers().Count < 1)
{
Log.Info("Waiting for containers to be created for app 1...");
Sharpen.Thread.Sleep(1000);
alloc1Response = am1.Schedule();
}
IList <Container> allocated1 = alloc1Response.GetAllocatedContainers();
NUnit.Framework.Assert.AreEqual(1, allocated1.Count);
NUnit.Framework.Assert.AreEqual(2 * Gb, allocated1[0].GetResource().GetMemory());
NUnit.Framework.Assert.AreEqual(nm1.GetNodeId(), allocated1[0].GetNodeId());
report_nm1 = rm.GetResourceScheduler().GetNodeReport(nm1.GetNodeId());
// check node report, 4 GB used and 0 GB available
NUnit.Framework.Assert.AreEqual(0, report_nm1.GetAvailableResource().GetMemory());
NUnit.Framework.Assert.AreEqual(4 * Gb, report_nm1.GetUsedResource().GetMemory());
// check container is assigned with 2 GB.
Container c1 = allocated1[0];
NUnit.Framework.Assert.AreEqual(2 * Gb, c1.GetResource().GetMemory());
// update node resource to 2 GB, so resource is over-consumed.
IDictionary <NodeId, ResourceOption> nodeResourceMap = new Dictionary <NodeId, ResourceOption
>();
nodeResourceMap[nm1.GetNodeId()] = ResourceOption.NewInstance(Org.Apache.Hadoop.Yarn.Api.Records.Resource
.NewInstance(2 * Gb, 1), -1);
UpdateNodeResourceRequest request = UpdateNodeResourceRequest.NewInstance(nodeResourceMap
);
AdminService @as = rm.adminService;
@as.UpdateNodeResource(request);
// Now, the used resource is still 4 GB, and available resource is minus value.
report_nm1 = rm.GetResourceScheduler().GetNodeReport(nm1.GetNodeId());
NUnit.Framework.Assert.AreEqual(4 * Gb, report_nm1.GetUsedResource().GetMemory());
NUnit.Framework.Assert.AreEqual(-2 * Gb, report_nm1.GetAvailableResource().GetMemory
());
// Check container can complete successfully in case of resource over-commitment.
ContainerStatus containerStatus = BuilderUtils.NewContainerStatus(c1.GetId(), ContainerState
.Complete, string.Empty, 0);
nm1.ContainerStatus(containerStatus);
int waitCount = 0;
while (attempt1.GetJustFinishedContainers().Count < 1 && waitCount++ != 20)
{
Log.Info("Waiting for containers to be finished for app 1... Tried " + waitCount
+ " times already..");
Sharpen.Thread.Sleep(100);
}
NUnit.Framework.Assert.AreEqual(1, attempt1.GetJustFinishedContainers().Count);
NUnit.Framework.Assert.AreEqual(1, am1.Schedule().GetCompletedContainersStatuses(
).Count);
report_nm1 = rm.GetResourceScheduler().GetNodeReport(nm1.GetNodeId());
NUnit.Framework.Assert.AreEqual(2 * Gb, report_nm1.GetUsedResource().GetMemory());
// As container return 2 GB back, the available resource becomes 0 again.
NUnit.Framework.Assert.AreEqual(0 * Gb, report_nm1.GetAvailableResource().GetMemory
());
rm.Stop();
}
/// <exception cref="System.Exception"/>
public virtual void TestProgressFilter()
{
MockRM rm = new MockRM(conf);
rm.Start();
// Register node1
MockNM nm1 = rm.RegisterNode("127.0.0.1:1234", 6 * Gb);
// Submit an application
RMApp app1 = rm.SubmitApp(2048);
nm1.NodeHeartbeat(true);
RMAppAttempt attempt1 = app1.GetCurrentAppAttempt();
MockAM am1 = rm.SendAMLaunched(attempt1.GetAppAttemptId());
am1.RegisterAppAttempt();
AllocateRequestPBImpl allocateRequest = new AllocateRequestPBImpl();
IList <ContainerId> release = new AList <ContainerId>();
IList <ResourceRequest> ask = new AList <ResourceRequest>();
allocateRequest.SetReleaseList(release);
allocateRequest.SetAskList(ask);
allocateRequest.SetProgress(float.PositiveInfinity);
am1.Allocate(allocateRequest);
while (attempt1.GetProgress() != 1)
{
Log.Info("Waiting for allocate event to be handled ...");
Sharpen.Thread.Sleep(100);
}
allocateRequest.SetProgress(float.NaN);
am1.Allocate(allocateRequest);
while (attempt1.GetProgress() != 0)
{
Log.Info("Waiting for allocate event to be handled ...");
Sharpen.Thread.Sleep(100);
}
allocateRequest.SetProgress((float)9);
am1.Allocate(allocateRequest);
while (attempt1.GetProgress() != 1)
{
Log.Info("Waiting for allocate event to be handled ...");
Sharpen.Thread.Sleep(100);
}
allocateRequest.SetProgress(float.NegativeInfinity);
am1.Allocate(allocateRequest);
while (attempt1.GetProgress() != 0)
{
Log.Info("Waiting for allocate event to be handled ...");
Sharpen.Thread.Sleep(100);
}
allocateRequest.SetProgress((float)0.5);
am1.Allocate(allocateRequest);
while (attempt1.GetProgress() != 0.5)
{
Log.Info("Waiting for allocate event to be handled ...");
Sharpen.Thread.Sleep(100);
}
allocateRequest.SetProgress((float)-1);
am1.Allocate(allocateRequest);
while (attempt1.GetProgress() != 0)
{
Log.Info("Waiting for allocate event to be handled ...");
Sharpen.Thread.Sleep(100);
}
}
/// <exception cref="System.Exception"/>
public virtual void TestNMToken()
{
MockRM rm = new MockRM(conf);
try
{
rm.Start();
MockNM nm1 = rm.RegisterNode("h1:1234", 10000);
NMTokenSecretManagerInRM nmTokenSecretManager = rm.GetRMContext().GetNMTokenSecretManager
();
// submitting new application
RMApp app = rm.SubmitApp(1000);
// start scheduling.
nm1.NodeHeartbeat(true);
// Starting application attempt and launching
// It should get registered with NMTokenSecretManager.
RMAppAttempt attempt = app.GetCurrentAppAttempt();
MockAM am = rm.SendAMLaunched(attempt.GetAppAttemptId());
NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptRegistered
(attempt.GetAppAttemptId()));
// This will register application master.
am.RegisterAppAttempt();
AList <Container> containersReceivedForNM1 = new AList <Container>();
IList <ContainerId> releaseContainerList = new AList <ContainerId>();
Dictionary <string, Token> nmTokens = new Dictionary <string, Token>();
// initially requesting 2 containers.
AllocateResponse response = am.Allocate("h1", 1000, 2, releaseContainerList);
NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count);
AllocateContainersAndValidateNMTokens(am, containersReceivedForNM1, 2, nmTokens,
nm1);
NUnit.Framework.Assert.AreEqual(1, nmTokens.Count);
// requesting 2 more containers.
response = am.Allocate("h1", 1000, 2, releaseContainerList);
NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count);
AllocateContainersAndValidateNMTokens(am, containersReceivedForNM1, 4, nmTokens,
nm1);
NUnit.Framework.Assert.AreEqual(1, nmTokens.Count);
// We will be simulating NM restart so restarting newly added h2:1234
// NM 2 now registers.
MockNM nm2 = rm.RegisterNode("h2:1234", 10000);
nm2.NodeHeartbeat(true);
AList <Container> containersReceivedForNM2 = new AList <Container>();
response = am.Allocate("h2", 1000, 2, releaseContainerList);
NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count);
AllocateContainersAndValidateNMTokens(am, containersReceivedForNM2, 2, nmTokens,
nm2);
NUnit.Framework.Assert.AreEqual(2, nmTokens.Count);
// Simulating NM-2 restart.
nm2 = rm.RegisterNode("h2:1234", 10000);
// Wait for reconnect to make it through the RM and create a new RMNode
IDictionary <NodeId, RMNode> nodes = rm.GetRMContext().GetRMNodes();
while (nodes[nm2.GetNodeId()].GetLastNodeHeartBeatResponse().GetResponseId() > 0)
{
Sharpen.Thread.Sleep(WaitSleepMs);
}
int interval = 40;
// Wait for nm Token to be cleared.
while (nmTokenSecretManager.IsApplicationAttemptNMTokenPresent(attempt.GetAppAttemptId
(), nm2.GetNodeId()) && interval-- > 0)
{
Log.Info("waiting for nmToken to be cleared for : " + nm2.GetNodeId());
Sharpen.Thread.Sleep(WaitSleepMs);
}
NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptRegistered
(attempt.GetAppAttemptId()));
// removing NMToken for h2:1234
Sharpen.Collections.Remove(nmTokens, nm2.GetNodeId().ToString());
NUnit.Framework.Assert.AreEqual(1, nmTokens.Count);
// We should again receive the NMToken.
response = am.Allocate("h2", 1000, 2, releaseContainerList);
NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count);
AllocateContainersAndValidateNMTokens(am, containersReceivedForNM2, 4, nmTokens,
nm2);
NUnit.Framework.Assert.AreEqual(2, nmTokens.Count);
// Now rolling over NMToken masterKey. it should resend the NMToken in
// next allocate call.
NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent
(attempt.GetAppAttemptId(), nm1.GetNodeId()));
NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent
(attempt.GetAppAttemptId(), nm2.GetNodeId()));
nmTokenSecretManager.RollMasterKey();
nmTokenSecretManager.ActivateNextMasterKey();
NUnit.Framework.Assert.IsFalse(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent
(attempt.GetAppAttemptId(), nm1.GetNodeId()));
NUnit.Framework.Assert.IsFalse(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent
(attempt.GetAppAttemptId(), nm2.GetNodeId()));
// It should not remove application attempt entry.
NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptRegistered
(attempt.GetAppAttemptId()));
nmTokens.Clear();
NUnit.Framework.Assert.AreEqual(0, nmTokens.Count);
// We should again receive the NMToken.
response = am.Allocate("h2", 1000, 1, releaseContainerList);
NUnit.Framework.Assert.AreEqual(0, response.GetAllocatedContainers().Count);
AllocateContainersAndValidateNMTokens(am, containersReceivedForNM2, 5, nmTokens,
nm2);
NUnit.Framework.Assert.AreEqual(1, nmTokens.Count);
NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptNMTokenPresent
(attempt.GetAppAttemptId(), nm2.GetNodeId()));
// After AM is finished making sure that nmtoken entry for app
NUnit.Framework.Assert.IsTrue(nmTokenSecretManager.IsApplicationAttemptRegistered
(attempt.GetAppAttemptId()));
am.UnregisterAppAttempt();
// marking all the containers as finished.
foreach (Container container in containersReceivedForNM1)
{
nm1.NodeHeartbeat(attempt.GetAppAttemptId(), container.GetId().GetContainerId(),
ContainerState.Complete);
}
foreach (Container container_1 in containersReceivedForNM2)
{
nm2.NodeHeartbeat(attempt.GetAppAttemptId(), container_1.GetId().GetContainerId()
, ContainerState.Complete);
}
nm1.NodeHeartbeat(am.GetApplicationAttemptId(), 1, ContainerState.Complete);
am.WaitForState(RMAppAttemptState.Finished);
NUnit.Framework.Assert.IsFalse(nmTokenSecretManager.IsApplicationAttemptRegistered
(attempt.GetAppAttemptId()));
}
finally
{
rm.Stop();
}
}
