public virtual void TestRecovery()
{
FilePath[] localLogDirs = GetLocalLogDirFiles(this.GetType().FullName, 2);
string localLogDirsString = localLogDirs[0].GetAbsolutePath() + "," + localLogDirs
[1].GetAbsolutePath();
conf.Set(YarnConfiguration.NmLogDirs, localLogDirsString);
conf.SetBoolean(YarnConfiguration.LogAggregationEnabled, false);
conf.SetLong(YarnConfiguration.NmLogRetainSeconds, YarnConfiguration.DefaultNmLogRetainSeconds
);
dirsHandler.Init(conf);
NMStateStoreService stateStore = new NMMemoryStateStoreService();
stateStore.Init(conf);
stateStore.Start();
TestNonAggregatingLogHandler.NonAggregatingLogHandlerWithMockExecutor logHandler =
new TestNonAggregatingLogHandler.NonAggregatingLogHandlerWithMockExecutor(this,
dispatcher, mockDelService, dirsHandler, stateStore);
logHandler.Init(conf);
logHandler.Start();
logHandler.Handle(new LogHandlerAppStartedEvent(appId, user, null, ContainerLogsRetentionPolicy
.AllContainers, null));
logHandler.Handle(new LogHandlerContainerFinishedEvent(container11, 0));
logHandler.Handle(new LogHandlerAppFinishedEvent(appId));
// simulate a restart and verify deletion is rescheduled
logHandler.Close();
logHandler = new TestNonAggregatingLogHandler.NonAggregatingLogHandlerWithMockExecutor
(this, dispatcher, mockDelService, dirsHandler, stateStore);
logHandler.Init(conf);
logHandler.Start();
ArgumentCaptor <Runnable> schedArg = ArgumentCaptor.ForClass <Runnable>();
Org.Mockito.Mockito.Verify(logHandler.mockSched).Schedule(schedArg.Capture(), Matchers.AnyLong
(), Matchers.Eq(TimeUnit.Milliseconds));
// execute the runnable and verify another restart has nothing scheduled
schedArg.GetValue().Run();
logHandler.Close();
logHandler = new TestNonAggregatingLogHandler.NonAggregatingLogHandlerWithMockExecutor
(this, dispatcher, mockDelService, dirsHandler, stateStore);
logHandler.Init(conf);
logHandler.Start();
Org.Mockito.Mockito.Verify(logHandler.mockSched, Org.Mockito.Mockito.Never()).Schedule
(Matchers.Any <Runnable>(), Matchers.AnyLong(), Matchers.Any <TimeUnit>());
logHandler.Close();
}
public virtual void TestRecovery()
{
YarnConfiguration conf = new YarnConfiguration();
conf.SetBoolean(YarnConfiguration.NmRecoveryEnabled, true);
NodeId nodeId = NodeId.NewInstance("somehost", 1234);
ContainerId cid1 = BuilderUtils.NewContainerId(1, 1, 1, 1);
ContainerId cid2 = BuilderUtils.NewContainerId(2, 2, 2, 2);
TestNMContainerTokenSecretManager.ContainerTokenKeyGeneratorForTest keygen = new
TestNMContainerTokenSecretManager.ContainerTokenKeyGeneratorForTest(conf);
NMMemoryStateStoreService stateStore = new NMMemoryStateStoreService();
stateStore.Init(conf);
stateStore.Start();
NMContainerTokenSecretManager secretMgr = new NMContainerTokenSecretManager(conf,
stateStore);
secretMgr.SetNodeId(nodeId);
MasterKey currentKey = keygen.GenerateKey();
secretMgr.SetMasterKey(currentKey);
ContainerTokenIdentifier tokenId1 = CreateContainerTokenId(cid1, nodeId, "user1",
secretMgr);
ContainerTokenIdentifier tokenId2 = CreateContainerTokenId(cid2, nodeId, "user2",
secretMgr);
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(tokenId1));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(tokenId2));
// restart and verify tokens still valid
secretMgr = new NMContainerTokenSecretManager(conf, stateStore);
secretMgr.SetNodeId(nodeId);
secretMgr.Recover();
NUnit.Framework.Assert.AreEqual(currentKey, secretMgr.GetCurrentKey());
NUnit.Framework.Assert.IsTrue(secretMgr.IsValidStartContainerRequest(tokenId1));
NUnit.Framework.Assert.IsTrue(secretMgr.IsValidStartContainerRequest(tokenId2));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(tokenId1));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(tokenId2));
// roll master key and start a container
secretMgr.StartContainerSuccessful(tokenId2);
currentKey = keygen.GenerateKey();
secretMgr.SetMasterKey(currentKey);
// restart and verify tokens still valid due to prev key persist
secretMgr = new NMContainerTokenSecretManager(conf, stateStore);
secretMgr.SetNodeId(nodeId);
secretMgr.Recover();
NUnit.Framework.Assert.AreEqual(currentKey, secretMgr.GetCurrentKey());
NUnit.Framework.Assert.IsTrue(secretMgr.IsValidStartContainerRequest(tokenId1));
NUnit.Framework.Assert.IsFalse(secretMgr.IsValidStartContainerRequest(tokenId2));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(tokenId1));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(tokenId2));
// roll master key again, restart, and verify keys no longer valid
currentKey = keygen.GenerateKey();
secretMgr.SetMasterKey(currentKey);
secretMgr = new NMContainerTokenSecretManager(conf, stateStore);
secretMgr.SetNodeId(nodeId);
secretMgr.Recover();
NUnit.Framework.Assert.AreEqual(currentKey, secretMgr.GetCurrentKey());
NUnit.Framework.Assert.IsTrue(secretMgr.IsValidStartContainerRequest(tokenId1));
NUnit.Framework.Assert.IsFalse(secretMgr.IsValidStartContainerRequest(tokenId2));
try
{
secretMgr.RetrievePassword(tokenId1);
NUnit.Framework.Assert.Fail("token should not be valid");
}
catch (SecretManager.InvalidToken)
{
}
// expected
try
{
secretMgr.RetrievePassword(tokenId2);
NUnit.Framework.Assert.Fail("token should not be valid");
}
catch (SecretManager.InvalidToken)
{
}
// expected
stateStore.Close();
}
public virtual void TestApplicationRecovery()
{
YarnConfiguration conf = new YarnConfiguration();
conf.SetBoolean(YarnConfiguration.NmRecoveryEnabled, true);
conf.Set(YarnConfiguration.NmAddress, "localhost:1234");
conf.SetBoolean(YarnConfiguration.YarnAclEnable, true);
conf.Set(YarnConfiguration.YarnAdminAcl, "yarn_admin_user");
NMStateStoreService stateStore = new NMMemoryStateStoreService();
stateStore.Init(conf);
stateStore.Start();
Context context = new NodeManager.NMContext(new NMContainerTokenSecretManager(conf
), new NMTokenSecretManagerInNM(), null, new ApplicationACLsManager(conf), stateStore
);
ContainerManagerImpl cm = CreateContainerManager(context);
cm.Init(conf);
cm.Start();
// simulate registration with RM
MasterKey masterKey = new MasterKeyPBImpl();
masterKey.SetKeyId(123);
masterKey.SetBytes(ByteBuffer.Wrap(new byte[] { 123 }));
context.GetContainerTokenSecretManager().SetMasterKey(masterKey);
context.GetNMTokenSecretManager().SetMasterKey(masterKey);
// add an application by starting a container
string appUser = "******";
string modUser = "******";
string viewUser = "******";
string enemyUser = "******";
ApplicationId appId = ApplicationId.NewInstance(0, 1);
ApplicationAttemptId attemptId = ApplicationAttemptId.NewInstance(appId, 1);
ContainerId cid = ContainerId.NewContainerId(attemptId, 1);
IDictionary <string, LocalResource> localResources = Collections.EmptyMap();
IDictionary <string, string> containerEnv = Sharpen.Collections.EmptyMap();
IList <string> containerCmds = Sharpen.Collections.EmptyList();
IDictionary <string, ByteBuffer> serviceData = Sharpen.Collections.EmptyMap();
Credentials containerCreds = new Credentials();
DataOutputBuffer dob = new DataOutputBuffer();
containerCreds.WriteTokenStorageToStream(dob);
ByteBuffer containerTokens = ByteBuffer.Wrap(dob.GetData(), 0, dob.GetLength());
IDictionary <ApplicationAccessType, string> acls = new Dictionary <ApplicationAccessType
, string>();
acls[ApplicationAccessType.ModifyApp] = modUser;
acls[ApplicationAccessType.ViewApp] = viewUser;
ContainerLaunchContext clc = ContainerLaunchContext.NewInstance(localResources, containerEnv
, containerCmds, serviceData, containerTokens, acls);
// create the logAggregationContext
LogAggregationContext logAggregationContext = LogAggregationContext.NewInstance("includePattern"
, "excludePattern", "includePatternInRollingAggregation", "excludePatternInRollingAggregation"
);
StartContainersResponse startResponse = StartContainer(context, cm, cid, clc, logAggregationContext
);
NUnit.Framework.Assert.IsTrue(startResponse.GetFailedRequests().IsEmpty());
NUnit.Framework.Assert.AreEqual(1, context.GetApplications().Count);
Org.Apache.Hadoop.Yarn.Server.Nodemanager.Containermanager.Application.Application
app = context.GetApplications()[appId];
NUnit.Framework.Assert.IsNotNull(app);
WaitForAppState(app, ApplicationState.Initing);
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(modUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ViewApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(enemyUser), ApplicationAccessType.ViewApp, appUser, appId));
// reset container manager and verify app recovered with proper acls
cm.Stop();
context = new NodeManager.NMContext(new NMContainerTokenSecretManager(conf), new
NMTokenSecretManagerInNM(), null, new ApplicationACLsManager(conf), stateStore);
cm = CreateContainerManager(context);
cm.Init(conf);
cm.Start();
NUnit.Framework.Assert.AreEqual(1, context.GetApplications().Count);
app = context.GetApplications()[appId];
NUnit.Framework.Assert.IsNotNull(app);
// check whether LogAggregationContext is recovered correctly
LogAggregationContext recovered = ((ApplicationImpl)app).GetLogAggregationContext
();
NUnit.Framework.Assert.IsNotNull(recovered);
NUnit.Framework.Assert.AreEqual(logAggregationContext.GetIncludePattern(), recovered
.GetIncludePattern());
NUnit.Framework.Assert.AreEqual(logAggregationContext.GetExcludePattern(), recovered
.GetExcludePattern());
NUnit.Framework.Assert.AreEqual(logAggregationContext.GetRolledLogsIncludePattern
(), recovered.GetRolledLogsIncludePattern());
NUnit.Framework.Assert.AreEqual(logAggregationContext.GetRolledLogsExcludePattern
(), recovered.GetRolledLogsExcludePattern());
WaitForAppState(app, ApplicationState.Initing);
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(modUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ViewApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(enemyUser), ApplicationAccessType.ViewApp, appUser, appId));
// simulate application completion
IList <ApplicationId> finishedApps = new AList <ApplicationId>();
finishedApps.AddItem(appId);
cm.Handle(new CMgrCompletedAppsEvent(finishedApps, CMgrCompletedAppsEvent.Reason.
ByResourcemanager));
WaitForAppState(app, ApplicationState.ApplicationResourcesCleaningup);
// restart and verify app is marked for finishing
cm.Stop();
context = new NodeManager.NMContext(new NMContainerTokenSecretManager(conf), new
NMTokenSecretManagerInNM(), null, new ApplicationACLsManager(conf), stateStore);
cm = CreateContainerManager(context);
cm.Init(conf);
cm.Start();
NUnit.Framework.Assert.AreEqual(1, context.GetApplications().Count);
app = context.GetApplications()[appId];
NUnit.Framework.Assert.IsNotNull(app);
WaitForAppState(app, ApplicationState.ApplicationResourcesCleaningup);
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(modUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ModifyApp, appUser, appId));
NUnit.Framework.Assert.IsTrue(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(viewUser), ApplicationAccessType.ViewApp, appUser, appId));
NUnit.Framework.Assert.IsFalse(context.GetApplicationACLsManager().CheckAccess(UserGroupInformation
.CreateRemoteUser(enemyUser), ApplicationAccessType.ViewApp, appUser, appId));
// simulate log aggregation completion
app.Handle(new ApplicationEvent(app.GetAppId(), ApplicationEventType.ApplicationResourcesCleanedup
));
NUnit.Framework.Assert.AreEqual(app.GetApplicationState(), ApplicationState.Finished
);
app.Handle(new ApplicationEvent(app.GetAppId(), ApplicationEventType.ApplicationLogHandlingFinished
));
// restart and verify app is no longer present after recovery
cm.Stop();
context = new NodeManager.NMContext(new NMContainerTokenSecretManager(conf), new
NMTokenSecretManagerInNM(), null, new ApplicationACLsManager(conf), stateStore);
cm = CreateContainerManager(context);
cm.Init(conf);
cm.Start();
NUnit.Framework.Assert.IsTrue(context.GetApplications().IsEmpty());
cm.Stop();
}
public virtual void TestRecovery()
{
YarnConfiguration conf = new YarnConfiguration();
conf.SetBoolean(YarnConfiguration.NmRecoveryEnabled, true);
NodeId nodeId = NodeId.NewInstance("somehost", 1234);
ApplicationAttemptId attempt1 = ApplicationAttemptId.NewInstance(ApplicationId.NewInstance
(1, 1), 1);
ApplicationAttemptId attempt2 = ApplicationAttemptId.NewInstance(ApplicationId.NewInstance
(2, 2), 2);
TestNMTokenSecretManagerInNM.NMTokenKeyGeneratorForTest keygen = new TestNMTokenSecretManagerInNM.NMTokenKeyGeneratorForTest
();
NMMemoryStateStoreService stateStore = new NMMemoryStateStoreService();
stateStore.Init(conf);
stateStore.Start();
NMTokenSecretManagerInNM secretMgr = new NMTokenSecretManagerInNM(stateStore);
secretMgr.SetNodeId(nodeId);
MasterKey currentKey = keygen.GenerateKey();
secretMgr.SetMasterKey(currentKey);
NMTokenIdentifier attemptToken1 = GetNMTokenId(secretMgr.CreateNMToken(attempt1,
nodeId, "user1"));
NMTokenIdentifier attemptToken2 = GetNMTokenId(secretMgr.CreateNMToken(attempt2,
nodeId, "user2"));
secretMgr.AppAttemptStartContainer(attemptToken1);
secretMgr.AppAttemptStartContainer(attemptToken2);
NUnit.Framework.Assert.IsTrue(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt1));
NUnit.Framework.Assert.IsTrue(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt2));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(attemptToken1));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(attemptToken2));
// restart and verify key is still there and token still valid
secretMgr = new NMTokenSecretManagerInNM(stateStore);
secretMgr.Recover();
secretMgr.SetNodeId(nodeId);
NUnit.Framework.Assert.AreEqual(currentKey, secretMgr.GetCurrentKey());
NUnit.Framework.Assert.IsTrue(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt1));
NUnit.Framework.Assert.IsTrue(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt2));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(attemptToken1));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(attemptToken2));
// roll master key and remove an app
currentKey = keygen.GenerateKey();
secretMgr.SetMasterKey(currentKey);
secretMgr.AppFinished(attempt1.GetApplicationId());
// restart and verify attempt1 key is still valid due to prev key persist
secretMgr = new NMTokenSecretManagerInNM(stateStore);
secretMgr.Recover();
secretMgr.SetNodeId(nodeId);
NUnit.Framework.Assert.AreEqual(currentKey, secretMgr.GetCurrentKey());
NUnit.Framework.Assert.IsFalse(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt1));
NUnit.Framework.Assert.IsTrue(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt2));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(attemptToken1));
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(attemptToken2));
// roll master key again, restart, and verify attempt1 key is bad but
// attempt2 is still good due to app key persist
currentKey = keygen.GenerateKey();
secretMgr.SetMasterKey(currentKey);
secretMgr = new NMTokenSecretManagerInNM(stateStore);
secretMgr.Recover();
secretMgr.SetNodeId(nodeId);
NUnit.Framework.Assert.AreEqual(currentKey, secretMgr.GetCurrentKey());
NUnit.Framework.Assert.IsFalse(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt1));
NUnit.Framework.Assert.IsTrue(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt2));
try
{
secretMgr.RetrievePassword(attemptToken1);
NUnit.Framework.Assert.Fail("attempt token should not still be valid");
}
catch (SecretManager.InvalidToken)
{
}
// expected
NUnit.Framework.Assert.IsNotNull(secretMgr.RetrievePassword(attemptToken2));
// remove last attempt, restart, verify both tokens are now bad
secretMgr.AppFinished(attempt2.GetApplicationId());
secretMgr = new NMTokenSecretManagerInNM(stateStore);
secretMgr.Recover();
secretMgr.SetNodeId(nodeId);
NUnit.Framework.Assert.AreEqual(currentKey, secretMgr.GetCurrentKey());
NUnit.Framework.Assert.IsFalse(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt1));
NUnit.Framework.Assert.IsFalse(secretMgr.IsAppAttemptNMTokenKeyPresent(attempt2));
try
{
secretMgr.RetrievePassword(attemptToken1);
NUnit.Framework.Assert.Fail("attempt token should not still be valid");
}
catch (SecretManager.InvalidToken)
{
}
// expected
try
{
secretMgr.RetrievePassword(attemptToken2);
NUnit.Framework.Assert.Fail("attempt token should not still be valid");
}
catch (SecretManager.InvalidToken)
{
}
// expected
stateStore.Close();
}
public virtual void TestRecovery()
{
Random r = new Random();
long seed = r.NextLong();
r.SetSeed(seed);
System.Console.Out.WriteLine("SEED: " + seed);
IList <Path> baseDirs = BuildDirs(r, @base, 4);
CreateDirs(new Path("."), baseDirs);
IList <Path> content = BuildDirs(r, new Path("."), 10);
foreach (Path b in baseDirs)
{
CreateDirs(b, content);
}
Configuration conf = new YarnConfiguration();
conf.SetBoolean(YarnConfiguration.NmRecoveryEnabled, true);
conf.SetInt(YarnConfiguration.DebugNmDeleteDelaySec, 1);
NMMemoryStateStoreService stateStore = new NMMemoryStateStoreService();
stateStore.Init(conf);
stateStore.Start();
DeletionService del = new DeletionService(new TestDeletionService.FakeDefaultContainerExecutor
(), stateStore);
try
{
del.Init(conf);
del.Start();
foreach (Path p in content)
{
NUnit.Framework.Assert.IsTrue(lfs.Util().Exists(new Path(baseDirs[0], p)));
del.Delete((long.Parse(p.GetName()) % 2) == 0 ? null : "dingo", p, Sharpen.Collections.ToArray
(baseDirs, new Path[4]));
}
// restart the deletion service
del.Stop();
del = new DeletionService(new TestDeletionService.FakeDefaultContainerExecutor(),
stateStore);
del.Init(conf);
del.Start();
// verify paths are still eventually deleted
int msecToWait = 10 * 1000;
foreach (Path p_1 in baseDirs)
{
foreach (Path q in content)
{
Path fp = new Path(p_1, q);
while (msecToWait > 0 && lfs.Util().Exists(fp))
{
Sharpen.Thread.Sleep(100);
msecToWait -= 100;
}
NUnit.Framework.Assert.IsFalse(lfs.Util().Exists(fp));
}
}
}
finally
{
del.Close();
stateStore.Close();
}
}