[
https://issues.apache.org/jira/browse/YARN-3798?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14593221#comment-14593221
]
Rakesh R commented on YARN-3798:
--------------------------------
[~varun_saxena], sorry I didn't see your ping. I have gone through few
ZooKeeper related comments in the jira. Based on my understanding I will try to
help you guys:)
bq. So I guess even ZKRMStateStore can get inconsistent data if it disconnects
with one zookeeper server and connects to other. Not a 100% sure on this
though. Maybe a zookeeper guy can chime in on this. Rakesh R, will sync work in
this scenario ?
Generally sync() is recommended only when accessing ZooKeeper service from
multiple clients. If only one ZooKeeper client is performing the update
operation it is not required to call syncup after the connection
re-establishment. For example, zkclient connected to ZK1 and successfully
created a znode. Now, assume zkclient got disconnected and successfully
reconnected to ZK2 the client zkclient will see the same znode in this server
also. I agree with [~ozawa]'s comments [about
sync()|https://issues.apache.org/jira/browse/YARN-3798?focusedCommentId=14584141&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14584141]
FYI: sync() call is a costly operation, internally this will force the
connected server to sync up the data from the Leader ZK server. Probably in
your logic after creating the new ZooKeeper connection it can do a sync() call
before performing any operation.
bq. It looks exhaust, but it's not: reconnection to other ZooKeeper servers are
done at ClientCnxn#startConnect in a main thread of ZooKeeper's client. Please
note that a session is not equal to a connection in ZooKeeper. What we can do
it to retry with current zookeeper client. I also noticed that we shouldn't
create new session when SESSIONMOVED occurs.
IMHO it is fine not to recreate a connection on SESSIONMOVED. If someone uses
saved sessionid and uses the constructor {{new ZooKeeper(connectString,
sessionTimeout, watcher, sessionId, sessionPasswd)}}, it may give unexpected
result. I hope you are not using this one in YARN.
bq. Coming to the patch: By definition, CONNECTIONLOSS also means that we
should recreate the connection?
Not required to create a new zkclient connection on errors except
SESSIONEXPIRED. Because zkclient internally does connection retries to all the
servers that are passed in the ZooKeeper constructor. I have one suggestion,
after creating new ZooKeeper connection it can do a {{sync()}} call before
performing any operation. This way will ensure consistency of data I feel.
> ZKRMStateStore shouldn't create new session without occurrance of
> SESSIONEXPIED
> -------------------------------------------------------------------------------
>
> Key: YARN-3798
> URL: https://issues.apache.org/jira/browse/YARN-3798
> Project: Hadoop YARN
> Issue Type: Bug
> Components: resourcemanager
> Affects Versions: 2.7.0
> Environment: Suse 11 Sp3
> Reporter: Bibin A Chundatt
> Assignee: Varun Saxena
> Priority: Blocker
> Attachments: RM.log, YARN-3798-branch-2.7.002.patch,
> YARN-3798-branch-2.7.patch
>
>
> RM going down with NoNode exception during create of znode for appattempt
> *Please find the exception logs*
> {code}
> 2015-06-09 10:09:44,732 INFO
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore:
> ZKRMStateStore Session connected
> 2015-06-09 10:09:44,732 INFO
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore:
> ZKRMStateStore Session restored
> 2015-06-09 10:09:44,886 INFO
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore:
> Exception while executing a ZK operation.
> org.apache.zookeeper.KeeperException$NoNodeException: KeeperErrorCode = NoNode
> at org.apache.zookeeper.KeeperException.create(KeeperException.java:115)
> at org.apache.zookeeper.ZooKeeper.multiInternal(ZooKeeper.java:1405)
> at org.apache.zookeeper.ZooKeeper.multi(ZooKeeper.java:1310)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$4.run(ZKRMStateStore.java:926)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$4.run(ZKRMStateStore.java:923)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$ZKAction.runWithCheck(ZKRMStateStore.java:1101)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$ZKAction.runWithRetries(ZKRMStateStore.java:1122)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.doStoreMultiWithRetries(ZKRMStateStore.java:923)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.doStoreMultiWithRetries(ZKRMStateStore.java:937)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.createWithRetries(ZKRMStateStore.java:970)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.updateApplicationAttemptStateInternal(ZKRMStateStore.java:671)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$UpdateAppAttemptTransition.transition(RMStateStore.java:275)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$UpdateAppAttemptTransition.transition(RMStateStore.java:260)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory$SingleInternalArc.doTransition(StateMachineFactory.java:362)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory.doTransition(StateMachineFactory.java:302)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory.access$300(StateMachineFactory.java:46)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory$InternalStateMachine.doTransition(StateMachineFactory.java:448)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore.handleStoreEvent(RMStateStore.java:837)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$ForwardingEventHandler.handle(RMStateStore.java:900)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$ForwardingEventHandler.handle(RMStateStore.java:895)
> at
> org.apache.hadoop.yarn.event.AsyncDispatcher.dispatch(AsyncDispatcher.java:175)
> at
> org.apache.hadoop.yarn.event.AsyncDispatcher$1.run(AsyncDispatcher.java:108)
> at java.lang.Thread.run(Thread.java:745)
> 2015-06-09 10:09:44,887 INFO
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore: Maxed
> out ZK retries. Giving up!
> 2015-06-09 10:09:44,887 ERROR
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore: Error
> updating appAttempt: appattempt_1433764310492_7152_000001
> org.apache.zookeeper.KeeperException$NoNodeException: KeeperErrorCode = NoNode
> at org.apache.zookeeper.KeeperException.create(KeeperException.java:115)
> at org.apache.zookeeper.ZooKeeper.multiInternal(ZooKeeper.java:1405)
> at org.apache.zookeeper.ZooKeeper.multi(ZooKeeper.java:1310)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$4.run(ZKRMStateStore.java:926)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$4.run(ZKRMStateStore.java:923)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$ZKAction.runWithCheck(ZKRMStateStore.java:1101)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$ZKAction.runWithRetries(ZKRMStateStore.java:1122)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.doStoreMultiWithRetries(ZKRMStateStore.java:923)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.doStoreMultiWithRetries(ZKRMStateStore.java:937)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.createWithRetries(ZKRMStateStore.java:970)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.updateApplicationAttemptStateInternal(ZKRMStateStore.java:671)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$UpdateAppAttemptTransition.transition(RMStateStore.java:275)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$UpdateAppAttemptTransition.transition(RMStateStore.java:260)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory$SingleInternalArc.doTransition(StateMachineFactory.java:362)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory.doTransition(StateMachineFactory.java:302)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory.access$300(StateMachineFactory.java:46)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory$InternalStateMachine.doTransition(StateMachineFactory.java:448)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore.handleStoreEvent(RMStateStore.java:837)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$ForwardingEventHandler.handle(RMStateStore.java:900)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$ForwardingEventHandler.handle(RMStateStore.java:895)
> at
> org.apache.hadoop.yarn.event.AsyncDispatcher.dispatch(AsyncDispatcher.java:175)
> at
> org.apache.hadoop.yarn.event.AsyncDispatcher$1.run(AsyncDispatcher.java:108)
> at java.lang.Thread.run(Thread.java:745)
> 2015-06-09 10:09:44,898 INFO
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore: Updating
> info for app: application_1433764310492_7152
> 2015-06-09 10:09:44,898 FATAL
> org.apache.hadoop.yarn.server.resourcemanager.ResourceManager: Received a
> org.apache.hadoop.yarn.server.resourcemanager.RMFatalEvent of type
> STATE_STORE_OP_FAILED. Cause:
> org.apache.zookeeper.KeeperException$NoNodeException: KeeperErrorCode = NoNode
> at org.apache.zookeeper.KeeperException.create(KeeperException.java:115)
> at org.apache.zookeeper.ZooKeeper.multiInternal(ZooKeeper.java:1405)
> at org.apache.zookeeper.ZooKeeper.multi(ZooKeeper.java:1310)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$4.run(ZKRMStateStore.java:926)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$4.run(ZKRMStateStore.java:923)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$ZKAction.runWithCheck(ZKRMStateStore.java:1101)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore$ZKAction.runWithRetries(ZKRMStateStore.java:1122)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.doStoreMultiWithRetries(ZKRMStateStore.java:923)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.doStoreMultiWithRetries(ZKRMStateStore.java:937)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.createWithRetries(ZKRMStateStore.java:970)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore.updateApplicationAttemptStateInternal(ZKRMStateStore.java:671)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$UpdateAppAttemptTransition.transition(RMStateStore.java:275)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$UpdateAppAttemptTransition.transition(RMStateStore.java:260)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory$SingleInternalArc.doTransition(StateMachineFactory.java:362)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory.doTransition(StateMachineFactory.java:302)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory.access$300(StateMachineFactory.java:46)
> at
> org.apache.hadoop.yarn.state.StateMachineFactory$InternalStateMachine.doTransition(StateMachineFactory.java:448)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore.handleStoreEvent(RMStateStore.java:837)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$ForwardingEventHandler.handle(RMStateStore.java:900)
> at
> org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$ForwardingEventHandler.handle(RMStateStore.java:895)
> at
> org.apache.hadoop.yarn.event.AsyncDispatcher.dispatch(AsyncDispatcher.java:175)
> at
> org.apache.hadoop.yarn.event.AsyncDispatcher$1.run(AsyncDispatcher.java:108)
> at java.lang.Thread.run(Thread.java:745)
> 2015-06-09 10:09:44,920 INFO org.apache.hadoop.util.ExitUtil: Exiting with
> status 1
> {code}
> Zk leader process down has happened almost at the same time
> On startup of zk process znode for application was available
> *Current*
> RM going down and Job failure
> *Expected*
> Submitted Job can fail but RM shutdown i not required
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
