/// <exception cref="System.IO.IOException"/>
public virtual void TransitionToStandby(HAServiceProtocol.StateChangeRequestInfo
reqInfo)
{
lock (this)
{
// call refreshAdminAcls before HA state transition
// for the case that adminAcls have been updated in previous active RM
try
{
RefreshAdminAcls(false);
}
catch (YarnException ex)
{
throw new ServiceFailedException("Can not execute refreshAdminAcls", ex);
}
UserGroupInformation user = CheckAccess("transitionToStandby");
CheckHaStateChange(reqInfo);
try
{
rm.TransitionToStandby(true);
RMAuditLogger.LogSuccess(user.GetShortUserName(), "transitionToStandby", "RMHAProtocolService"
);
}
catch (Exception e)
{
RMAuditLogger.LogFailure(user.GetShortUserName(), "transitionToStandby", string.Empty
, "RMHAProtocolService", "Exception transitioning to standby");
throw new ServiceFailedException("Error when transitioning to Standby mode", e);
}
}
}
/// <exception cref="System.Exception"/>
public virtual void TestTransitionedToStandbyShouldNotHang()
{
configuration.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
Configuration conf = new YarnConfiguration(configuration);
MemoryRMStateStore memStore = new _MemoryRMStateStore_464();
memStore.Init(conf);
rm = new _MockRM_472(conf, memStore);
rm.Init(conf);
HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
NUnit.Framework.Assert.AreEqual(StateErr, HAServiceProtocol.HAServiceState.Initializing
, rm.adminService.GetServiceStatus().GetState());
NUnit.Framework.Assert.IsFalse("RM is ready to become active before being started"
, rm.adminService.GetServiceStatus().IsReadyToBecomeActive());
CheckMonitorHealth();
rm.Start();
CheckMonitorHealth();
CheckStandbyRMFunctionality();
// 2. Transition to Active.
rm.adminService.TransitionToActive(requestInfo);
// 3. Try Transition to standby
Sharpen.Thread t = new Sharpen.Thread(new _Runnable_498(this));
// TODO Auto-generated catch block
t.Start();
rm.GetRMContext().GetStateStore().UpdateApplicationState(null);
t.Join();
// wait for thread to finish
rm.adminService.TransitionToStandby(requestInfo);
CheckStandbyRMFunctionality();
rm.Stop();
}
private HAServiceProtocolProtos.HAStateChangeRequestInfoProto Convert(HAServiceProtocol.StateChangeRequestInfo
reqInfo)
{
HAServiceProtocolProtos.HARequestSource src;
switch (reqInfo.GetSource())
{
case HAServiceProtocol.RequestSource.RequestByUser:
{
src = HAServiceProtocolProtos.HARequestSource.RequestByUser;
break;
}
case HAServiceProtocol.RequestSource.RequestByUserForced:
{
src = HAServiceProtocolProtos.HARequestSource.RequestByUserForced;
break;
}
case HAServiceProtocol.RequestSource.RequestByZkfc:
{
src = HAServiceProtocolProtos.HARequestSource.RequestByZkfc;
break;
}
default:
{
throw new ArgumentException("Bad source: " + reqInfo.GetSource());
}
}
return((HAServiceProtocolProtos.HAStateChangeRequestInfoProto)HAServiceProtocolProtos.HAStateChangeRequestInfoProto
.NewBuilder().SetReqSource(src).Build());
}
// Other stuff is verified in the regular web-services related tests
/// <summary>Test to verify the following RM HA transitions to the following states.</summary>
/// <remarks>
/// Test to verify the following RM HA transitions to the following states.
/// 1. Standby: Should be a no-op
/// 2. Active: Active services should start
/// 3. Active: Should be a no-op.
/// While active, submit a couple of jobs
/// 4. Standby: Active services should stop
/// 5. Active: Active services should start
/// 6. Stop the RM: All services should stop and RM should not be ready to
/// become Active
/// </remarks>
/// <exception cref="System.Exception"/>
public virtual void TestFailoverAndTransitions()
{
configuration.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
Configuration conf = new YarnConfiguration(configuration);
rm = new MockRM(conf);
rm.Init(conf);
HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
NUnit.Framework.Assert.AreEqual(StateErr, HAServiceProtocol.HAServiceState.Initializing
, rm.adminService.GetServiceStatus().GetState());
NUnit.Framework.Assert.IsFalse("RM is ready to become active before being started"
, rm.adminService.GetServiceStatus().IsReadyToBecomeActive());
CheckMonitorHealth();
rm.Start();
CheckMonitorHealth();
CheckStandbyRMFunctionality();
VerifyClusterMetrics(0, 0, 0, 0, 0, 0);
// 1. Transition to Standby - must be a no-op
rm.adminService.TransitionToStandby(requestInfo);
CheckMonitorHealth();
CheckStandbyRMFunctionality();
VerifyClusterMetrics(0, 0, 0, 0, 0, 0);
// 2. Transition to active
rm.adminService.TransitionToActive(requestInfo);
CheckMonitorHealth();
CheckActiveRMFunctionality();
VerifyClusterMetrics(1, 1, 1, 1, 2048, 1);
// 3. Transition to active - no-op
rm.adminService.TransitionToActive(requestInfo);
CheckMonitorHealth();
CheckActiveRMFunctionality();
VerifyClusterMetrics(1, 2, 2, 2, 2048, 2);
// 4. Transition to standby
rm.adminService.TransitionToStandby(requestInfo);
CheckMonitorHealth();
CheckStandbyRMFunctionality();
VerifyClusterMetrics(0, 0, 0, 0, 0, 0);
// 5. Transition to active to check Active->Standby->Active works
rm.adminService.TransitionToActive(requestInfo);
CheckMonitorHealth();
CheckActiveRMFunctionality();
VerifyClusterMetrics(1, 1, 1, 1, 2048, 1);
// 6. Stop the RM. All services should stop and RM should not be ready to
// become active
rm.Stop();
NUnit.Framework.Assert.AreEqual(StateErr, HAServiceProtocol.HAServiceState.Stopping
, rm.adminService.GetServiceStatus().GetState());
NUnit.Framework.Assert.IsFalse("RM is ready to become active even after it is stopped"
, rm.adminService.GetServiceStatus().IsReadyToBecomeActive());
NUnit.Framework.Assert.IsFalse("Active RM services are started", rm.AreActiveServicesRunning
());
CheckMonitorHealth();
}
public virtual void TestZKRootPathAcls()
{
HAServiceProtocol.StateChangeRequestInfo req = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
string rootPath = YarnConfiguration.DefaultZkRmStateStoreParentPath + "/" + ZKRMStateStore
.RootZnodeName;
// Start RM with HA enabled
Configuration conf = CreateHARMConf("rm1,rm2", "rm1", 1234);
conf.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
ResourceManager rm = new MockRM(conf);
rm.Start();
rm.GetRMContext().GetRMAdminService().TransitionToActive(req);
Stat stat = new Stat();
IList <ACL> acls = ((ZKRMStateStore)rm.GetRMContext().GetStateStore()).GetACLWithRetries
(rootPath, stat);
NUnit.Framework.Assert.AreEqual(acls.Count, 2);
// CREATE and DELETE permissions for root node based on RM ID
VerifyZKACL("digest", "localhost", ZooDefs.Perms.Create | ZooDefs.Perms.Delete, acls
);
VerifyZKACL("world", "anyone", ZooDefs.Perms.All ^ (ZooDefs.Perms.Create | ZooDefs.Perms
.Delete), acls);
rm.Close();
// Now start RM with HA disabled. NoAuth Exception should not be thrown.
conf.SetBoolean(YarnConfiguration.RmHaEnabled, false);
rm = new MockRM(conf);
rm.Start();
rm.GetRMContext().GetRMAdminService().TransitionToActive(req);
acls = ((ZKRMStateStore)rm.GetRMContext().GetStateStore()).GetACLWithRetries(rootPath
, stat);
NUnit.Framework.Assert.AreEqual(acls.Count, 1);
VerifyZKACL("world", "anyone", ZooDefs.Perms.All, acls);
rm.Close();
// Start RM with HA enabled.
conf.SetBoolean(YarnConfiguration.RmHaEnabled, true);
rm = new MockRM(conf);
rm.Start();
rm.GetRMContext().GetRMAdminService().TransitionToActive(req);
acls = ((ZKRMStateStore)rm.GetRMContext().GetStateStore()).GetACLWithRetries(rootPath
, stat);
NUnit.Framework.Assert.AreEqual(acls.Count, 2);
VerifyZKACL("digest", "localhost", ZooDefs.Perms.Create | ZooDefs.Perms.Delete, acls
);
VerifyZKACL("world", "anyone", ZooDefs.Perms.All ^ (ZooDefs.Perms.Create | ZooDefs.Perms
.Delete), acls);
rm.Close();
}
/// <exception cref="System.IO.IOException"/>
public virtual void TransitionToStandby(HAServiceProtocol.StateChangeRequestInfo
reqInfo)
{
try
{
HAServiceProtocolProtos.TransitionToStandbyRequestProto req = ((HAServiceProtocolProtos.TransitionToStandbyRequestProto
)HAServiceProtocolProtos.TransitionToStandbyRequestProto.NewBuilder().SetReqInfo
(Convert(reqInfo)).Build());
rpcProxy.TransitionToStandby(NullController, req);
}
catch (ServiceException e)
{
throw ProtobufHelper.GetRemoteException(e);
}
}
public virtual void TestRMDispatcherForHA()
{
string errorMessageForEventHandler = "Expect to get the same number of handlers";
string errorMessageForService = "Expect to get the same number of services";
configuration.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
Configuration conf = new YarnConfiguration(configuration);
rm = new _MockRM_313(conf);
rm.Init(conf);
int expectedEventHandlerCount = ((TestRMHA.MyCountingDispatcher)rm.GetRMContext()
.GetDispatcher()).GetEventHandlerCount();
int expectedServiceCount = rm.GetServices().Count;
NUnit.Framework.Assert.IsTrue(expectedEventHandlerCount != 0);
HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
NUnit.Framework.Assert.AreEqual(StateErr, HAServiceProtocol.HAServiceState.Initializing
, rm.adminService.GetServiceStatus().GetState());
NUnit.Framework.Assert.IsFalse("RM is ready to become active before being started"
, rm.adminService.GetServiceStatus().IsReadyToBecomeActive());
rm.Start();
//call transitions to standby and active a couple of times
rm.adminService.TransitionToStandby(requestInfo);
rm.adminService.TransitionToActive(requestInfo);
rm.adminService.TransitionToStandby(requestInfo);
rm.adminService.TransitionToActive(requestInfo);
rm.adminService.TransitionToStandby(requestInfo);
TestRMHA.MyCountingDispatcher dispatcher = (TestRMHA.MyCountingDispatcher)rm.GetRMContext
().GetDispatcher();
NUnit.Framework.Assert.IsTrue(!dispatcher.IsStopped());
rm.adminService.TransitionToActive(requestInfo);
NUnit.Framework.Assert.AreEqual(errorMessageForEventHandler, expectedEventHandlerCount
, ((TestRMHA.MyCountingDispatcher)rm.GetRMContext().GetDispatcher()).GetEventHandlerCount
());
NUnit.Framework.Assert.AreEqual(errorMessageForService, expectedServiceCount, rm.
GetServices().Count);
// Keep the dispatcher reference before transitioning to standby
dispatcher = (TestRMHA.MyCountingDispatcher)rm.GetRMContext().GetDispatcher();
rm.adminService.TransitionToStandby(requestInfo);
NUnit.Framework.Assert.AreEqual(errorMessageForEventHandler, expectedEventHandlerCount
, ((TestRMHA.MyCountingDispatcher)rm.GetRMContext().GetDispatcher()).GetEventHandlerCount
());
NUnit.Framework.Assert.AreEqual(errorMessageForService, expectedServiceCount, rm.
GetServices().Count);
NUnit.Framework.Assert.IsTrue(dispatcher.IsStopped());
rm.Stop();
}
public virtual void TestFailoverClearsRMContext()
{
configuration.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
configuration.SetBoolean(YarnConfiguration.RecoveryEnabled, true);
Configuration conf = new YarnConfiguration(configuration);
MemoryRMStateStore memStore = new MemoryRMStateStore();
memStore.Init(conf);
// 1. start RM
rm = new MockRM(conf, memStore);
rm.Init(conf);
rm.Start();
HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
CheckMonitorHealth();
CheckStandbyRMFunctionality();
// 2. Transition to active
rm.adminService.TransitionToActive(requestInfo);
CheckMonitorHealth();
CheckActiveRMFunctionality();
VerifyClusterMetrics(1, 1, 1, 1, 2048, 1);
NUnit.Framework.Assert.AreEqual(1, rm.GetRMContext().GetRMNodes().Count);
NUnit.Framework.Assert.AreEqual(1, rm.GetRMContext().GetRMApps().Count);
// 3. Create new RM
rm = new _MockRM_550(conf, memStore);
rm.Init(conf);
rm.Start();
CheckMonitorHealth();
CheckStandbyRMFunctionality();
// 4. Try Transition to active, throw exception
try
{
rm.adminService.TransitionToActive(requestInfo);
NUnit.Framework.Assert.Fail("Transitioned to Active should throw exception.");
}
catch (Exception e)
{
NUnit.Framework.Assert.IsTrue("Error when transitioning to Active mode".Contains(
e.Message));
}
// 5. Clears the metrics
VerifyClusterMetrics(0, 0, 0, 0, 0, 0);
NUnit.Framework.Assert.AreEqual(0, rm.GetRMContext().GetRMNodes().Count);
NUnit.Framework.Assert.AreEqual(0, rm.GetRMContext().GetRMApps().Count);
}
/// <exception cref="System.IO.IOException"/>
public virtual void TransitionToActive(HAServiceProtocol.StateChangeRequestInfo reqInfo
)
{
lock (this)
{
// call refreshAdminAcls before HA state transition
// for the case that adminAcls have been updated in previous active RM
try
{
RefreshAdminAcls(false);
}
catch (YarnException ex)
{
throw new ServiceFailedException("Can not execute refreshAdminAcls", ex);
}
UserGroupInformation user = CheckAccess("transitionToActive");
CheckHaStateChange(reqInfo);
try
{
rm.TransitionToActive();
}
catch (Exception e)
{
RMAuditLogger.LogFailure(user.GetShortUserName(), "transitionToActive", string.Empty
, "RMHAProtocolService", "Exception transitioning to active");
throw new ServiceFailedException("Error when transitioning to Active mode", e);
}
try
{
// call all refresh*s for active RM to get the updated configurations.
RefreshAll();
}
catch (Exception e)
{
Log.Error("RefreshAll failed so firing fatal event", e);
rmContext.GetDispatcher().GetEventHandler().Handle(new RMFatalEvent(RMFatalEventType
.TransitionToActiveFailed, e));
throw new ServiceFailedException("Error on refreshAll during transistion to Active"
, e);
}
RMAuditLogger.LogSuccess(user.GetShortUserName(), "transitionToActive", "RMHAProtocolService"
);
}
}
public virtual void TestFencing()
{
HAServiceProtocol.StateChangeRequestInfo req = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
Configuration conf1 = CreateHARMConf("rm1,rm2", "rm1", 1234);
conf1.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
ResourceManager rm1 = new ResourceManager();
rm1.Init(conf1);
rm1.Start();
rm1.GetRMContext().GetRMAdminService().TransitionToActive(req);
NUnit.Framework.Assert.AreEqual("RM with ZKStore didn't start", Service.STATE.Started
, rm1.GetServiceState());
NUnit.Framework.Assert.AreEqual("RM should be Active", HAServiceProtocol.HAServiceState
.Active, rm1.GetRMContext().GetRMAdminService().GetServiceStatus().GetState());
Configuration conf2 = CreateHARMConf("rm1,rm2", "rm2", 5678);
conf2.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
ResourceManager rm2 = new ResourceManager();
rm2.Init(conf2);
rm2.Start();
rm2.GetRMContext().GetRMAdminService().TransitionToActive(req);
NUnit.Framework.Assert.AreEqual("RM with ZKStore didn't start", Service.STATE.Started
, rm2.GetServiceState());
NUnit.Framework.Assert.AreEqual("RM should be Active", HAServiceProtocol.HAServiceState
.Active, rm2.GetRMContext().GetRMAdminService().GetServiceStatus().GetState());
for (int i = 0; i < ZkTimeoutMs / 50; i++)
{
if (HAServiceProtocol.HAServiceState.Active == rm1.GetRMContext().GetRMAdminService
().GetServiceStatus().GetState())
{
Sharpen.Thread.Sleep(100);
}
}
NUnit.Framework.Assert.AreEqual("RM should have been fenced", HAServiceProtocol.HAServiceState
.Standby, rm1.GetRMContext().GetRMAdminService().GetServiceStatus().GetState());
NUnit.Framework.Assert.AreEqual("RM should be Active", HAServiceProtocol.HAServiceState
.Active, rm2.GetRMContext().GetRMAdminService().GetServiceStatus().GetState());
}
/// <exception cref="System.Exception"/>
public virtual void TestTransitionedToActiveRefreshFail()
{
configuration.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
YarnConfiguration conf = new YarnConfiguration(configuration);
configuration = new CapacitySchedulerConfiguration(conf);
rm = new _MockRM_588(this, configuration);
rm.Init(configuration);
rm.Start();
HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
configuration.Set("yarn.scheduler.capacity.root.default.capacity", "100");
rm.adminService.TransitionToStandby(requestInfo);
NUnit.Framework.Assert.AreEqual(HAServiceProtocol.HAServiceState.Standby, rm.GetRMContext
().GetHAServiceState());
configuration.Set("yarn.scheduler.capacity.root.default.capacity", "200");
try
{
rm.adminService.TransitionToActive(requestInfo);
}
catch (Exception e)
{
NUnit.Framework.Assert.IsTrue("Error on refreshAll during transistion to Active".
Contains(e.Message));
}
TestRMHA.FailFastDispatcher dispatcher = ((TestRMHA.FailFastDispatcher)rm.rmContext
.GetDispatcher());
dispatcher.Await();
NUnit.Framework.Assert.AreEqual(1, dispatcher.GetEventCount());
// Making correct conf and check the state
configuration.Set("yarn.scheduler.capacity.root.default.capacity", "100");
rm.adminService.TransitionToActive(requestInfo);
NUnit.Framework.Assert.AreEqual(HAServiceProtocol.HAServiceState.Active, rm.GetRMContext
().GetHAServiceState());
rm.adminService.TransitionToStandby(requestInfo);
NUnit.Framework.Assert.AreEqual(HAServiceProtocol.HAServiceState.Standby, rm.GetRMContext
().GetHAServiceState());
}
/// <exception cref="System.Exception"/>
public virtual void TestFailoverWhenTransitionToActiveThrowException()
{
configuration.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
Configuration conf = new YarnConfiguration(configuration);
MemoryRMStateStore memStore = new _MemoryRMStateStore_414();
// first time throw exception
// start RM
memStore.Init(conf);
rm = new MockRM(conf, memStore);
rm.Init(conf);
HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
NUnit.Framework.Assert.AreEqual(StateErr, HAServiceProtocol.HAServiceState.Initializing
, rm.adminService.GetServiceStatus().GetState());
NUnit.Framework.Assert.IsFalse("RM is ready to become active before being started"
, rm.adminService.GetServiceStatus().IsReadyToBecomeActive());
CheckMonitorHealth();
rm.Start();
CheckMonitorHealth();
CheckStandbyRMFunctionality();
// 2. Try Transition to active, throw exception
try
{
rm.adminService.TransitionToActive(requestInfo);
NUnit.Framework.Assert.Fail("Transitioned to Active should throw exception.");
}
catch (Exception e)
{
NUnit.Framework.Assert.IsTrue("Error when transitioning to Active mode".Contains(
e.Message));
}
// 3. Transition to active, success
rm.adminService.TransitionToActive(requestInfo);
CheckMonitorHealth();
CheckActiveRMFunctionality();
}
/// <summary>Check that a request to change this node's HA state is valid.</summary>
/// <remarks>
/// Check that a request to change this node's HA state is valid.
/// In particular, verifies that, if auto failover is enabled, non-forced
/// requests from the HAAdmin CLI are rejected, and vice versa.
/// </remarks>
/// <param name="req">the request to check</param>
/// <exception cref="Org.Apache.Hadoop.Security.AccessControlException">if the request is disallowed
/// </exception>
private void CheckHaStateChange(HAServiceProtocol.StateChangeRequestInfo req)
{
switch (req.GetSource())
{
case HAServiceProtocol.RequestSource.RequestByUser:
{
if (autoFailoverEnabled)
{
throw new AccessControlException("Manual failover for this ResourceManager is disallowed, "
+ "because automatic failover is enabled.");
}
break;
}
case HAServiceProtocol.RequestSource.RequestByUserForced:
{
if (autoFailoverEnabled)
{
Log.Warn("Allowing manual failover from " + Org.Apache.Hadoop.Ipc.Server.GetRemoteAddress
() + " even though automatic failover is enabled, because the user " + "specified the force flag"
);
}
break;
}
case HAServiceProtocol.RequestSource.RequestByZkfc:
{
if (!autoFailoverEnabled)
{
throw new AccessControlException("Request from ZK failover controller at " + Org.Apache.Hadoop.Ipc.Server
.GetRemoteAddress() + " denied " + "since automatic failover is not enabled");
}
break;
}
}
}
public virtual void TestTransitionsWhenAutomaticFailoverEnabled()
{
string ErrUnforcedRequest = "User request succeeded even when " + "automatic failover is enabled";
Configuration conf = new YarnConfiguration(configuration);
rm = new MockRM(conf);
rm.Init(conf);
rm.Start();
HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
// Transition to standby
try
{
rm.adminService.TransitionToStandby(requestInfo);
NUnit.Framework.Assert.Fail(ErrUnforcedRequest);
}
catch (AccessControlException)
{
}
// expected
CheckMonitorHealth();
CheckStandbyRMFunctionality();
// Transition to active
try
{
rm.adminService.TransitionToActive(requestInfo);
NUnit.Framework.Assert.Fail(ErrUnforcedRequest);
}
catch (AccessControlException)
{
}
// expected
CheckMonitorHealth();
CheckStandbyRMFunctionality();
string ErrForcedRequest = "Forced request by user should work " + "even if automatic failover is enabled";
requestInfo = new HAServiceProtocol.StateChangeRequestInfo(HAServiceProtocol.RequestSource
.RequestByUserForced);
// Transition to standby
try
{
rm.adminService.TransitionToStandby(requestInfo);
}
catch (AccessControlException)
{
NUnit.Framework.Assert.Fail(ErrForcedRequest);
}
CheckMonitorHealth();
CheckStandbyRMFunctionality();
// Transition to active
try
{
rm.adminService.TransitionToActive(requestInfo);
}
catch (AccessControlException)
{
NUnit.Framework.Assert.Fail(ErrForcedRequest);
}
CheckMonitorHealth();
CheckActiveRMFunctionality();
}
public virtual void TestRMHAWithFileSystemBasedConfiguration()
{
HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo
(HAServiceProtocol.RequestSource.RequestByUser);
configuration.Set(YarnConfiguration.RmConfigurationProviderClass, "org.apache.hadoop.yarn.FileSystemBasedConfigurationProvider"
);
configuration.SetBoolean(YarnConfiguration.RmHaEnabled, true);
configuration.SetBoolean(YarnConfiguration.AutoFailoverEnabled, false);
configuration.Set(YarnConfiguration.RmHaIds, "rm1,rm2");
int @base = 100;
foreach (string confKey in YarnConfiguration.GetServiceAddressConfKeys(configuration
))
{
configuration.Set(HAUtil.AddSuffix(confKey, "rm1"), "0.0.0.0:" + (@base + 20));
configuration.Set(HAUtil.AddSuffix(confKey, "rm2"), "0.0.0.0:" + (@base + 40));
@base = @base * 2;
}
Configuration conf1 = new Configuration(configuration);
conf1.Set(YarnConfiguration.RmHaId, "rm1");
Configuration conf2 = new Configuration(configuration);
conf2.Set(YarnConfiguration.RmHaId, "rm2");
// upload default configurations
UploadDefaultConfiguration();
MockRM rm1 = null;
MockRM rm2 = null;
try
{
rm1 = new MockRM(conf1);
rm1.Init(conf1);
rm1.Start();
NUnit.Framework.Assert.IsTrue(rm1.GetRMContext().GetHAServiceState() == HAServiceProtocol.HAServiceState
.Standby);
rm2 = new MockRM(conf2);
rm2.Init(conf1);
rm2.Start();
NUnit.Framework.Assert.IsTrue(rm2.GetRMContext().GetHAServiceState() == HAServiceProtocol.HAServiceState
.Standby);
rm1.adminService.TransitionToActive(requestInfo);
NUnit.Framework.Assert.IsTrue(rm1.GetRMContext().GetHAServiceState() == HAServiceProtocol.HAServiceState
.Active);
CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
csConf.Set("yarn.scheduler.capacity.maximum-applications", "5000");
UploadConfiguration(csConf, "capacity-scheduler.xml");
rm1.adminService.RefreshQueues(RefreshQueuesRequest.NewInstance());
int maxApps = ((CapacityScheduler)rm1.GetRMContext().GetScheduler()).GetConfiguration
().GetMaximumSystemApplications();
NUnit.Framework.Assert.AreEqual(maxApps, 5000);
// Before failover happens, the maxApps is
// still the default value on the standby rm : rm2
int maxAppsBeforeFailOver = ((CapacityScheduler)rm2.GetRMContext().GetScheduler()
).GetConfiguration().GetMaximumSystemApplications();
NUnit.Framework.Assert.AreEqual(maxAppsBeforeFailOver, 10000);
// Do the failover
rm1.adminService.TransitionToStandby(requestInfo);
rm2.adminService.TransitionToActive(requestInfo);
NUnit.Framework.Assert.IsTrue(rm1.GetRMContext().GetHAServiceState() == HAServiceProtocol.HAServiceState
.Standby);
NUnit.Framework.Assert.IsTrue(rm2.GetRMContext().GetHAServiceState() == HAServiceProtocol.HAServiceState
.Active);
int maxAppsAfter = ((CapacityScheduler)rm2.GetRMContext().GetScheduler()).GetConfiguration
().GetMaximumSystemApplications();
NUnit.Framework.Assert.AreEqual(maxAppsAfter, 5000);
}
finally
{
if (rm1 != null)
{
rm1.Stop();
}
if (rm2 != null)
{
rm2.Stop();
}
}
}
