public virtual void TestHAIDLookup()
{
//test implicitly lookup HA-ID
Configuration conf = new YarnConfiguration(configuration);
rm = new MockRM(conf);
rm.Init(conf);
NUnit.Framework.Assert.AreEqual(conf.Get(YarnConfiguration.RmHaId), Rm2NodeId);
//test explicitly lookup HA-ID
configuration.Set(YarnConfiguration.RmHaId, Rm1NodeId);
conf = new YarnConfiguration(configuration);
rm = new MockRM(conf);
rm.Init(conf);
NUnit.Framework.Assert.AreEqual(conf.Get(YarnConfiguration.RmHaId), Rm1NodeId);
//test if RM_HA_ID can not be found
configuration.Set(YarnConfiguration.RmHaIds, Rm1NodeId + "," + Rm3NodeId);
configuration.Unset(YarnConfiguration.RmHaId);
conf = new YarnConfiguration(configuration);
try
{
rm = new MockRM(conf);
rm.Init(conf);
NUnit.Framework.Assert.Fail("Should get an exception here.");
}
catch (Exception ex)
{
NUnit.Framework.Assert.IsTrue(ex.Message.Contains("Invalid configuration! Can not find valid RM_HA_ID."
));
}
}
/// <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();
}
public virtual void TestAdminRefreshQueuesWithFileSystemBasedConfigurationProvider
()
{
configuration.Set(YarnConfiguration.RmConfigurationProviderClass, "org.apache.hadoop.yarn.FileSystemBasedConfigurationProvider"
);
//upload default configurations
UploadDefaultConfiguration();
try
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Should not get any exceptions");
}
CapacityScheduler cs = (CapacityScheduler)rm.GetRMContext().GetScheduler();
int maxAppsBefore = cs.GetConfiguration().GetMaximumSystemApplications();
CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
csConf.Set("yarn.scheduler.capacity.maximum-applications", "5000");
UploadConfiguration(csConf, "capacity-scheduler.xml");
rm.adminService.RefreshQueues(RefreshQueuesRequest.NewInstance());
int maxAppsAfter = cs.GetConfiguration().GetMaximumSystemApplications();
NUnit.Framework.Assert.AreEqual(maxAppsAfter, 5000);
NUnit.Framework.Assert.IsTrue(maxAppsAfter != maxAppsBefore);
}
public virtual void TestAdminAclsWithFileSystemBasedConfigurationProvider()
{
configuration.Set(YarnConfiguration.RmConfigurationProviderClass, "org.apache.hadoop.yarn.FileSystemBasedConfigurationProvider"
);
//upload default configurations
UploadDefaultConfiguration();
try
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Should not get any exceptions");
}
string aclStringBefore = rm.adminService.GetAccessControlList().GetAclString().Trim
();
YarnConfiguration yarnConf = new YarnConfiguration();
yarnConf.Set(YarnConfiguration.YarnAdminAcl, "world:anyone:rwcda");
UploadConfiguration(yarnConf, "yarn-site.xml");
rm.adminService.RefreshAdminAcls(RefreshAdminAclsRequest.NewInstance());
string aclStringAfter = rm.adminService.GetAccessControlList().GetAclString().Trim
();
NUnit.Framework.Assert.IsTrue(!aclStringAfter.Equals(aclStringBefore));
NUnit.Framework.Assert.AreEqual(aclStringAfter, "world:anyone:rwcda," + UserGroupInformation
.GetCurrentUser().GetShortUserName());
}
// 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 TestRefreshUserToGroupsMappingsWithFileSystemBasedConfigurationProvider
()
{
configuration.Set(YarnConfiguration.RmConfigurationProviderClass, "org.apache.hadoop.yarn.FileSystemBasedConfigurationProvider"
);
string[] defaultTestUserGroups = new string[] { "dummy_group1", "dummy_group2" };
UserGroupInformation ugi = UserGroupInformation.CreateUserForTesting("dummyUser",
defaultTestUserGroups);
string user = ugi.GetUserName();
IList <string> groupWithInit = new AList <string>(2);
for (int i = 0; i < ugi.GetGroupNames().Length; i++)
{
groupWithInit.AddItem(ugi.GetGroupNames()[i]);
}
// upload default configurations
UploadDefaultConfiguration();
Configuration conf = new Configuration();
conf.SetClass(CommonConfigurationKeys.HadoopSecurityGroupMapping, typeof(TestRMAdminService.MockUnixGroupsMapping
), typeof(GroupMappingServiceProvider));
UploadConfiguration(conf, "core-site.xml");
try
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Should not get any exceptions");
}
// Make sure RM will use the updated GroupMappingServiceProvider
IList <string> groupBefore = new AList <string>(Groups.GetUserToGroupsMappingService
(configuration).GetGroups(user));
NUnit.Framework.Assert.IsTrue(groupBefore.Contains("test_group_A") && groupBefore
.Contains("test_group_B") && groupBefore.Contains("test_group_C") && groupBefore
.Count == 3);
NUnit.Framework.Assert.IsTrue(groupWithInit.Count != groupBefore.Count);
NUnit.Framework.Assert.IsFalse(groupWithInit.Contains("test_group_A") || groupWithInit
.Contains("test_group_B") || groupWithInit.Contains("test_group_C"));
// update the groups
TestRMAdminService.MockUnixGroupsMapping.UpdateGroups();
rm.adminService.RefreshUserToGroupsMappings(RefreshUserToGroupsMappingsRequest.NewInstance
());
IList <string> groupAfter = Groups.GetUserToGroupsMappingService(configuration).GetGroups
(user);
// should get the updated groups
NUnit.Framework.Assert.IsTrue(groupAfter.Contains("test_group_D") && groupAfter.Contains
("test_group_E") && groupAfter.Contains("test_group_F") && groupAfter.Count == 3
);
}
public virtual void TestRefreshSuperUserGroupsWithFileSystemBasedConfigurationProvider
()
{
configuration.Set(YarnConfiguration.RmConfigurationProviderClass, "org.apache.hadoop.yarn.FileSystemBasedConfigurationProvider"
);
//upload default configurations
UploadDefaultConfiguration();
try
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Should not get any exceptions");
}
Configuration coreConf = new Configuration(false);
coreConf.Set("hadoop.proxyuser.test.groups", "test_groups");
coreConf.Set("hadoop.proxyuser.test.hosts", "test_hosts");
UploadConfiguration(coreConf, "core-site.xml");
rm.adminService.RefreshSuperUserGroupsConfiguration(RefreshSuperUserGroupsConfigurationRequest
.NewInstance());
NUnit.Framework.Assert.IsTrue(ProxyUsers.GetDefaultImpersonationProvider().GetProxyGroups
()["hadoop.proxyuser.test.groups"].Count == 1);
NUnit.Framework.Assert.IsTrue(ProxyUsers.GetDefaultImpersonationProvider().GetProxyGroups
()["hadoop.proxyuser.test.groups"].Contains("test_groups"));
NUnit.Framework.Assert.IsTrue(ProxyUsers.GetDefaultImpersonationProvider().GetProxyHosts
()["hadoop.proxyuser.test.hosts"].Count == 1);
NUnit.Framework.Assert.IsTrue(ProxyUsers.GetDefaultImpersonationProvider().GetProxyHosts
()["hadoop.proxyuser.test.hosts"].Contains("test_hosts"));
Configuration yarnConf = new Configuration(false);
yarnConf.Set("yarn.resourcemanager.proxyuser.test.groups", "test_groups_1");
yarnConf.Set("yarn.resourcemanager.proxyuser.test.hosts", "test_hosts_1");
UploadConfiguration(yarnConf, "yarn-site.xml");
// RM specific configs will overwrite the common ones
rm.adminService.RefreshSuperUserGroupsConfiguration(RefreshSuperUserGroupsConfigurationRequest
.NewInstance());
NUnit.Framework.Assert.IsTrue(ProxyUsers.GetDefaultImpersonationProvider().GetProxyGroups
()["hadoop.proxyuser.test.groups"].Count == 1);
NUnit.Framework.Assert.IsTrue(ProxyUsers.GetDefaultImpersonationProvider().GetProxyGroups
()["hadoop.proxyuser.test.groups"].Contains("test_groups_1"));
NUnit.Framework.Assert.IsTrue(ProxyUsers.GetDefaultImpersonationProvider().GetProxyHosts
()["hadoop.proxyuser.test.hosts"].Count == 1);
NUnit.Framework.Assert.IsTrue(ProxyUsers.GetDefaultImpersonationProvider().GetProxyHosts
()["hadoop.proxyuser.test.hosts"].Contains("test_hosts_1"));
}
public virtual void TestRefreshNodesWithLocalConfigurationProvider()
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
try
{
rm.adminService.RefreshNodes(RefreshNodesRequest.NewInstance());
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Using localConfigurationProvider. Should not get any exception."
);
}
}
/// <exception cref="System.IO.IOException"/>
protected internal virtual void StartRMs(MockRM rm1, Configuration confForRM1, MockRM
rm2, Configuration confForRM2)
{
rm1.Init(confForRM1);
rm1.Start();
NUnit.Framework.Assert.IsTrue(rm1.GetRMContext().GetHAServiceState() == HAServiceProtocol.HAServiceState
.Standby);
rm2.Init(confForRM2);
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);
}
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);
}
public virtual void TestRefreshNodesWithFileSystemBasedConfigurationProvider()
{
configuration.Set(YarnConfiguration.RmConfigurationProviderClass, "org.apache.hadoop.yarn.FileSystemBasedConfigurationProvider"
);
// upload default configurations
UploadDefaultConfiguration();
try
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Should not get any exceptions");
}
FilePath excludeHostsFile = new FilePath(tmpDir.ToString(), "excludeHosts");
if (excludeHostsFile.Exists())
{
excludeHostsFile.Delete();
}
if (!excludeHostsFile.CreateNewFile())
{
NUnit.Framework.Assert.Fail("Can not create " + "excludeHosts");
}
PrintWriter fileWriter = new PrintWriter(excludeHostsFile);
fileWriter.Write("0.0.0.0:123");
fileWriter.Close();
UploadToRemoteFileSystem(new Path(excludeHostsFile.GetAbsolutePath()));
Configuration yarnConf = new YarnConfiguration();
yarnConf.Set(YarnConfiguration.RmNodesExcludeFilePath, this.workingPath + "/excludeHosts"
);
UploadConfiguration(yarnConf, YarnConfiguration.YarnSiteConfigurationFile);
rm.adminService.RefreshNodes(RefreshNodesRequest.NewInstance());
ICollection <string> excludeHosts = rm.GetNodesListManager().GetHostsReader().GetExcludedHosts
();
NUnit.Framework.Assert.IsTrue(excludeHosts.Count == 1);
NUnit.Framework.Assert.IsTrue(excludeHosts.Contains("0.0.0.0:123"));
}
public virtual void TestRefreshAclWithDaemonUser()
{
string daemonUser = UserGroupInformation.GetCurrentUser().GetShortUserName();
configuration.Set(YarnConfiguration.RmConfigurationProviderClass, "org.apache.hadoop.yarn.FileSystemBasedConfigurationProvider"
);
UploadDefaultConfiguration();
YarnConfiguration yarnConf = new YarnConfiguration();
yarnConf.Set(YarnConfiguration.YarnAdminAcl, daemonUser + "xyz");
UploadConfiguration(yarnConf, "yarn-site.xml");
try
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Should not get any exceptions");
}
NUnit.Framework.Assert.AreEqual(daemonUser + "xyz," + daemonUser, rm.adminService
.GetAccessControlList().GetAclString().Trim());
yarnConf = new YarnConfiguration();
yarnConf.Set(YarnConfiguration.YarnAdminAcl, daemonUser + "abc");
UploadConfiguration(yarnConf, "yarn-site.xml");
try
{
rm.adminService.RefreshAdminAcls(RefreshAdminAclsRequest.NewInstance());
}
catch (YarnException e)
{
if (e.InnerException != null && e.InnerException is AccessControlException)
{
NUnit.Framework.Assert.Fail("Refresh should not have failed due to incorrect ACL"
);
}
throw;
}
NUnit.Framework.Assert.AreEqual(daemonUser + "abc," + daemonUser, rm.adminService
.GetAccessControlList().GetAclString().Trim());
}
public virtual void TestAdminRefreshQueuesWithLocalConfigurationProvider()
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
CapacityScheduler cs = (CapacityScheduler)rm.GetRMContext().GetScheduler();
int maxAppsBefore = cs.GetConfiguration().GetMaximumSystemApplications();
try
{
rm.adminService.RefreshQueues(RefreshQueuesRequest.NewInstance());
NUnit.Framework.Assert.AreEqual(maxAppsBefore, cs.GetConfiguration().GetMaximumSystemApplications
());
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Using localConfigurationProvider. Should not get any exception."
);
}
}
public virtual void TestRMStartsWithoutConfigurationFilesProvided()
{
// enable FileSystemBasedConfigurationProvider without uploading
// any configuration files into Remote File System.
configuration.Set(YarnConfiguration.RmConfigurationProviderClass, "org.apache.hadoop.yarn.FileSystemBasedConfigurationProvider"
);
// The configurationProvider will return NULL instead of
// throwing out Exceptions, if there are no configuration files provided.
// RM will not load the remote Configuration files,
// and should start successfully.
try
{
rm = new MockRM(configuration);
rm.Init(configuration);
rm.Start();
}
catch (Exception)
{
NUnit.Framework.Assert.Fail("Should not get any exceptions");
}
}
/// <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();
}
public virtual void InnerTestHAWithRMHostName(bool includeBindHost)
{
//this is run two times, with and without a bind host configured
if (includeBindHost)
{
configuration.Set(YarnConfiguration.RmBindHost, "9.9.9.9");
}
//test if both RM_HOSTBANE_{rm_id} and RM_RPCADDRESS_{rm_id} are set
//We should only read rpc addresses from RM_RPCADDRESS_{rm_id} configuration
configuration.Set(HAUtil.AddSuffix(YarnConfiguration.RmHostname, Rm1NodeId), "1.1.1.1"
);
configuration.Set(HAUtil.AddSuffix(YarnConfiguration.RmHostname, Rm2NodeId), "0.0.0.0"
);
configuration.Set(HAUtil.AddSuffix(YarnConfiguration.RmHostname, Rm3NodeId), "2.2.2.2"
);
try
{
Configuration conf = new YarnConfiguration(configuration);
rm = new MockRM(conf);
rm.Init(conf);
foreach (string confKey in YarnConfiguration.GetServiceAddressConfKeys(conf))
{
NUnit.Framework.Assert.AreEqual("RPC address not set for " + confKey, Rm1Address,
conf.Get(HAUtil.AddSuffix(confKey, Rm1NodeId)));
NUnit.Framework.Assert.AreEqual("RPC address not set for " + confKey, Rm2Address,
conf.Get(HAUtil.AddSuffix(confKey, Rm2NodeId)));
NUnit.Framework.Assert.AreEqual("RPC address not set for " + confKey, Rm3Address,
conf.Get(HAUtil.AddSuffix(confKey, Rm3NodeId)));
if (includeBindHost)
{
NUnit.Framework.Assert.AreEqual("Web address misconfigured WITH bind-host", Sharpen.Runtime.Substring
(rm.webAppAddress, 0, 7), "9.9.9.9");
}
else
{
//YarnConfiguration tries to figure out which rm host it's on by binding to it,
//which doesn't happen for any of these fake addresses, so we end up with 0.0.0.0
NUnit.Framework.Assert.AreEqual("Web address misconfigured WITHOUT bind-host", Sharpen.Runtime.Substring
(rm.webAppAddress, 0, 7), "0.0.0.0");
}
}
}
catch (YarnRuntimeException)
{
NUnit.Framework.Assert.Fail("Should not throw any exceptions.");
}
//test if only RM_HOSTBANE_{rm_id} is set
configuration.Clear();
configuration.SetBoolean(YarnConfiguration.RmHaEnabled, true);
configuration.Set(YarnConfiguration.RmHaIds, Rm1NodeId + "," + Rm2NodeId);
configuration.Set(HAUtil.AddSuffix(YarnConfiguration.RmHostname, Rm1NodeId), "1.1.1.1"
);
configuration.Set(HAUtil.AddSuffix(YarnConfiguration.RmHostname, Rm2NodeId), "0.0.0.0"
);
try
{
Configuration conf = new YarnConfiguration(configuration);
rm = new MockRM(conf);
rm.Init(conf);
NUnit.Framework.Assert.AreEqual("RPC address not set for " + YarnConfiguration.RmAddress
, "1.1.1.1:8032", conf.Get(HAUtil.AddSuffix(YarnConfiguration.RmAddress, Rm1NodeId
)));
NUnit.Framework.Assert.AreEqual("RPC address not set for " + YarnConfiguration.RmAddress
, "0.0.0.0:8032", conf.Get(HAUtil.AddSuffix(YarnConfiguration.RmAddress, Rm2NodeId
)));
}
catch (YarnRuntimeException)
{
NUnit.Framework.Assert.Fail("Should not throw any exceptions.");
}
}
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();
}
}
}
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();
}