/// <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 TestAppCleanupWhenRMRestartedBeforeAppFinished()
{
conf.SetInt(YarnConfiguration.RmAmMaxAttempts, 1);
MemoryRMStateStore memStore = new MemoryRMStateStore();
memStore.Init(conf);
// start RM
MockRM rm1 = new MockRM(conf, memStore);
rm1.Start();
MockNM nm1 = new MockNM("127.0.0.1:1234", 1024, rm1.GetResourceTrackerService());
nm1.RegisterNode();
MockNM nm2 = new MockNM("127.0.0.1:5678", 1024, rm1.GetResourceTrackerService());
nm2.RegisterNode();
// create app and launch the AM
RMApp app0 = rm1.SubmitApp(200);
MockAM am0 = LaunchAM(app0, rm1, nm1);
// alloc another container on nm2
AllocateResponse allocResponse = am0.Allocate(Arrays.AsList(ResourceRequest.NewInstance
(Priority.NewInstance(1), "*", Resource.NewInstance(1024, 0), 1)), null);
while (null == allocResponse.GetAllocatedContainers() || allocResponse.GetAllocatedContainers
().IsEmpty())
{
nm2.NodeHeartbeat(true);
allocResponse = am0.Allocate(null, null);
Sharpen.Thread.Sleep(1000);
}
// start new RM
MockRM rm2 = new MockRM(conf, memStore);
rm2.Start();
// nm1/nm2 register to rm2, and do a heartbeat
nm1.SetResourceTrackerService(rm2.GetResourceTrackerService());
nm1.RegisterNode(Arrays.AsList(NMContainerStatus.NewInstance(ContainerId.NewContainerId
(am0.GetApplicationAttemptId(), 1), ContainerState.Complete, Resource.NewInstance
(1024, 1), string.Empty, 0, Priority.NewInstance(0), 1234)), Arrays.AsList(app0.
GetApplicationId()));
nm2.SetResourceTrackerService(rm2.GetResourceTrackerService());
nm2.RegisterNode(Arrays.AsList(app0.GetApplicationId()));
// assert app state has been saved.
rm2.WaitForState(app0.GetApplicationId(), RMAppState.Failed);
// wait for application cleanup message received on NM1
WaitForAppCleanupMessageRecved(nm1, app0.GetApplicationId());
// wait for application cleanup message received on NM2
WaitForAppCleanupMessageRecved(nm2, app0.GetApplicationId());
rm1.Stop();
rm2.Stop();
}
/// <exception cref="System.Exception"/>
protected internal virtual void AllocateContainersAndValidateNMTokens(MockAM am,
AList <Container> containersReceived, int totalContainerRequested, Dictionary <string
, Token> nmTokens, MockNM nm)
{
AList <ContainerId> releaseContainerList = new AList <ContainerId>();
AllocateResponse response;
AList <ResourceRequest> resourceRequest = new AList <ResourceRequest>();
while (containersReceived.Count < totalContainerRequested)
{
nm.NodeHeartbeat(true);
Log.Info("requesting containers..");
response = am.Allocate(resourceRequest, releaseContainerList);
Sharpen.Collections.AddAll(containersReceived, response.GetAllocatedContainers());
if (!response.GetNMTokens().IsEmpty())
{
foreach (NMToken nmToken in response.GetNMTokens())
{
string nodeId = nmToken.GetNodeId().ToString();
if (nmTokens.Contains(nodeId))
{
NUnit.Framework.Assert.Fail("Duplicate NMToken received for : " + nodeId);
}
nmTokens[nodeId] = nmToken.GetToken();
}
}
Log.Info("Got " + containersReceived.Count + " containers. Waiting to get " + totalContainerRequested
);
Sharpen.Thread.Sleep(WaitSleepMs);
}
}
// 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);
}
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();
}
/// <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();
}
// The test verifies processing of NMContainerStatuses which are sent during
// NM registration.
// 1. Start the cluster-RM,NM,Submit app with 1024MB,Launch & register AM
// 2. AM sends ResourceRequest for 1 container with memory 2048MB.
// 3. Verify for number of container allocated by RM
// 4. Verify Memory Usage by cluster, it should be 3072. AM memory + requested
// memory. 1024 + 2048=3072
// 5. Re-register NM by sending completed container status
// 6. Verify for Memory Used, it should be 1024
// 7. Send AM heatbeat to RM. Allocated response should contain completed
// container.
/// <exception cref="System.Exception"/>
public virtual void TestProcessingNMContainerStatusesOnNMRestart()
{
conf.SetInt(YarnConfiguration.RmAmMaxAttempts, 1);
MemoryRMStateStore memStore = new MemoryRMStateStore();
memStore.Init(conf);
// 1. Start the cluster-RM,NM,Submit app with 1024MB,Launch & register AM
MockRM rm1 = new MockRM(conf, memStore);
rm1.Start();
int nmMemory = 8192;
int amMemory = 1024;
int containerMemory = 2048;
MockNM nm1 = new MockNM("127.0.0.1:1234", nmMemory, rm1.GetResourceTrackerService
());
nm1.RegisterNode();
RMApp app0 = rm1.SubmitApp(amMemory);
MockAM am0 = MockRM.LaunchAndRegisterAM(app0, rm1, nm1);
// 2. AM sends ResourceRequest for 1 container with memory 2048MB.
int noOfContainers = 1;
IList <Container> allocateContainers = am0.AllocateAndWaitForContainers(noOfContainers
, containerMemory, nm1);
// 3. Verify for number of container allocated by RM
NUnit.Framework.Assert.AreEqual(noOfContainers, allocateContainers.Count);
Container container = allocateContainers[0];
nm1.NodeHeartbeat(am0.GetApplicationAttemptId(), 1, ContainerState.Running);
nm1.NodeHeartbeat(am0.GetApplicationAttemptId(), container.GetId().GetContainerId
(), ContainerState.Running);
rm1.WaitForState(app0.GetApplicationId(), RMAppState.Running);
// 4. Verify Memory Usage by cluster, it should be 3072. AM memory +
// requested memory. 1024 + 2048=3072
ResourceScheduler rs = rm1.GetRMContext().GetScheduler();
int allocatedMB = rs.GetRootQueueMetrics().GetAllocatedMB();
NUnit.Framework.Assert.AreEqual(amMemory + containerMemory, allocatedMB);
// 5. Re-register NM by sending completed container status
IList <NMContainerStatus> nMContainerStatusForApp = CreateNMContainerStatusForApp(
am0);
nm1.RegisterNode(nMContainerStatusForApp, Arrays.AsList(app0.GetApplicationId()));
WaitForClusterMemory(nm1, rs, amMemory);
// 6. Verify for Memory Used, it should be 1024
NUnit.Framework.Assert.AreEqual(amMemory, rs.GetRootQueueMetrics().GetAllocatedMB
());
// 7. Send AM heatbeat to RM. Allocated response should contain completed
// container
AllocateRequest req = AllocateRequest.NewInstance(0, 0F, new AList <ResourceRequest
>(), new AList <ContainerId>(), null);
AllocateResponse allocate = am0.Allocate(req);
IList <ContainerStatus> completedContainersStatuses = allocate.GetCompletedContainersStatuses
();
NUnit.Framework.Assert.AreEqual(noOfContainers, completedContainersStatuses.Count
);
// Application clean up should happen Cluster memory used is 0
nm1.NodeHeartbeat(am0.GetApplicationAttemptId(), 1, ContainerState.Complete);
WaitForClusterMemory(nm1, rs, 0);
rm1.Stop();
}
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)
{
}
}
/// <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();
}
}
