You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "zhengchenyu (Jira)" <ji...@apache.org> on 2022/05/06 12:08:00 UTC
[jira] [Comment Edited] (YARN-11127) Potential deadlock in AsyncDispatcher caused by RMNodeImpl, SchedulerApplicationAttempt and RMAppImpl's lock contention.
[ https://issues.apache.org/jira/browse/YARN-11127?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17532824#comment-17532824 ]
zhengchenyu edited comment on YARN-11127 at 5/6/22 12:07 PM:
-------------------------------------------------------------
There are two change in this PR:
(a) aggregateLogReport introduce by YARN-1376 then two level writelock happened.
[~xgong], do you have some suggestion about update app aggeration report in asynchronous way.
(b) In YARN-3141, getResourceUsageReport will use write lock to replace synchronized.
[~wangda], I think getResourceUsageReport should use readlock, because only member variable ’lastMemoryAggregateAllocationUpdateTime' and 'lastResourceSecondsMap' are update, I think volatile is enough. do you have some suggestion?
was (Author: zhengchenyu):
aggregateLogReport introduce by YARN-1376 then two level writelock happened.
[~xgong], do you have some suggestion about update app aggeration report in asynchronous way.
In YARN-3141, getResourceUsageReport will use write lock to replace synchronized.
[~wangda], I think getResourceUsageReport should use readlock, because only member variable ’lastMemoryAggregateAllocationUpdateTime' and 'lastResourceSecondsMap' are update, I think volatile is enough. do you have some suggestion?
> Potential deadlock in AsyncDispatcher caused by RMNodeImpl, SchedulerApplicationAttempt and RMAppImpl's lock contention.
> ------------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-11127
> URL: https://issues.apache.org/jira/browse/YARN-11127
> Project: Hadoop YARN
> Issue Type: Bug
> Components: resourcemanager
> Affects Versions: 3.2.1
> Reporter: zhengchenyu
> Assignee: zhengchenyu
> Priority: Major
> Labels: pull-request-available
> Attachments: rm-dead-lock.png
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> I found rm deadlock in our cluster. It's a low probability event. some critical jstack information are below:
> {code:java}
> "RM Event dispatcher" #63 prio=5 os_prio=0 tid=0x00007f9a73aaa800 nid=0x221e7 waiting on condition [0x00007f85dd00b000]
> java.lang.Thread.State: WAITING (parking)
> at sun.misc.Unsafe.park(Native Method)
> - parking to wait for <0x00007f9389aab478> (a java.util.concurrent.locks.ReentrantReadWriteLock$NonfairSync)
> at java.util.concurrent.locks.LockSupport.park(LockSupport.java:175)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.parkAndCheckInterrupt(AbstractQueuedSynchronizer.java:836)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquireQueued(AbstractQueuedSynchronizer.java:870)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquire(AbstractQueuedSynchronizer.java:1199)
> at java.util.concurrent.locks.ReentrantReadWriteLock$WriteLock.lock(ReentrantReadWriteLock.java:943)
> at org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMAppLogAggregation.aggregateLogReport(RMAppLogAggregation.java:120)
> at org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMAppImpl.aggregateLogReport(RMAppImpl.java:1740)
> at org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNodeImpl.handleLogAggregationStatus(RMNodeImpl.java:1481)
> at org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNodeImpl.access$500(RMNodeImpl.java:104)
> at org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNodeImpl$StatusUpdateWhenHealthyTransition.transition(RMNodeImpl.java:1242)
> at org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNodeImpl$StatusUpdateWhenHealthyTransition.transition(RMNodeImpl.java:1198)
> at org.apache.hadoop.yarn.state.StateMachineFactory$MultipleInternalArc.doTransition(StateMachineFactory.java:385)
> at org.apache.hadoop.yarn.state.StateMachineFactory.doTransition(StateMachineFactory.java:302)
> at org.apache.hadoop.yarn.state.StateMachineFactory.access$500(StateMachineFactory.java:46)
> at org.apache.hadoop.yarn.state.StateMachineFactory$InternalStateMachine.doTransition(StateMachineFactory.java:487)
> - locked <0x00007f88db78c5c8> (a org.apache.hadoop.yarn.state.StateMachineFactory$InternalStateMachine)
> at org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNodeImpl.handle(RMNodeImpl.java:670)
> at org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNodeImpl.handle(RMNodeImpl.java:101)
> at org.apache.hadoop.yarn.server.resourcemanager.ResourceManager$NodeEventDispatcher.handle(ResourceManager.java:1116)
> at org.apache.hadoop.yarn.server.resourcemanager.ResourceManager$NodeEventDispatcher.handle(ResourceManager.java:1100)
> at org.apache.hadoop.yarn.event.AsyncDispatcher.dispatch(AsyncDispatcher.java:219)
> at org.apache.hadoop.yarn.event.AsyncDispatcher$1.run(AsyncDispatcher.java:133)
> at java.lang.Thread.run(Thread.java:748)
> "IPC Server handler 264 on default port 8032" #1717 daemon prio=5 os_prio=0 tid=0x000055b69acc2800 nid=0x229a5 waiting on condition [0x00007f8574ba2000]
> java.lang.Thread.State: WAITING (parking)
> at sun.misc.Unsafe.park(Native Method)
> - parking to wait for <0x00007f938976e818> (a java.util.concurrent.locks.ReentrantReadWriteLock$NonfairSync)
> at java.util.concurrent.locks.LockSupport.park(LockSupport.java:175)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.parkAndCheckInterrupt(AbstractQueuedSynchronizer.java:836)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquireQueued(AbstractQueuedSynchronizer.java:870)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquire(AbstractQueuedSynchronizer.java:1199)
> at java.util.concurrent.locks.ReentrantReadWriteLock$WriteLock.lock(ReentrantReadWriteLock.java:943)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp.getResourceUsageReport(FiCaSchedulerApp.java:1115)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.AbstractYarnScheduler.getAppResourceUsageReport(AbstractYarnScheduler.java:433)
> at org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttemptMetrics.getAggregateAppResourceUsage(RMAppAttemptMetrics.java:143)
> at org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMAppImpl.getRMAppMetrics(RMAppImpl.java:1693)
> at org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMAppImpl.createAndGetApplicationReport(RMAppImpl.java:742)
> at org.apache.hadoop.yarn.server.resourcemanager.ClientRMService.getApplicationReport(ClientRMService.java:428)
> at org.apache.hadoop.yarn.api.impl.pb.service.ApplicationClientProtocolPBServiceImpl.getApplicationReport(ApplicationClientProtocolPBServiceImpl.java:247)
> at org.apache.hadoop.yarn.proto.ApplicationClientProtocol$ApplicationClientProtocolService$2.callBlockingMethod(ApplicationClientProtocol.java:609)
> at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:528)
> at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:1070)
> at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:1040)
> at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:958)
> at java.security.AccessController.doPrivileged(Native Method)
> at javax.security.auth.Subject.doAs(Subject.java:422)
> at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1730)
> at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2972)
> "SchedulerEventDispatcher:Event Processor" #38 prio=5 os_prio=0 tid=0x000055b699c20000 nid=0x223be waiting on condition [0x00007f85d3686000]
> java.lang.Thread.State: WAITING (parking)
> at sun.misc.Unsafe.park(Native Method)
> - parking to wait for <0x00007f88d6708f68> (a java.util.concurrent.locks.ReentrantReadWriteLock$NonfairSync)
> at java.util.concurrent.locks.LockSupport.park(LockSupport.java:175)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.parkAndCheckInterrupt(AbstractQueuedSynchronizer.java:836)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.doAcquireShared(AbstractQueuedSynchronizer.java:967)
> at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquireShared(AbstractQueuedSynchronizer.java:1283)
> at java.util.concurrent.locks.ReentrantReadWriteLock$ReadLock.lock(ReentrantReadWriteLock.java:727)
> at org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNodeImpl.getState(RMNodeImpl.java:563)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp.accept(FiCaSchedulerApp.java:464)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.tryCommit(CapacityScheduler.java:3035)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.submitResourceCommitRequest(CapacityScheduler.java:2816)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.allocateOrReserveNewContainers(CapacityScheduler.java:1637)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.allocateContainerOnSingleNode(CapacityScheduler.java:1624)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.allocateContainersToNode(CapacityScheduler.java:1727)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.allocateContainersToNode(CapacityScheduler.java:1476)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.nodeUpdate(CapacityScheduler.java:1312)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.handle(CapacityScheduler.java:1785)
> at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.handle(CapacityScheduler.java:171)
> at org.apache.hadoop.yarn.event.EventDispatcher$EventProcessor.run(EventDispatcher.java:66)
> at java.lang.Thread.run(Thread.java:748){code}
> Here we found dead lock, because of three thread's lock contention on RMNodeImpl, SchedulerApplicationAttempt and RMAppImpl rwlock.
> For easy to understand the lock contention, you can see below image.
>
> !rm-dead-lock.png|width=578,height=214!
> Mulit-level lock with write lock is very danger.
> I reproduce this bug in my uni-test, only one nodemnager and one app with specical parameter, and keep nm have LogAggregationStatusForApps, the probability of this bug increase.
> How to fix?
> * For thread "RM Event dispatcher", we have lock writelock twice. We could update aggregateLogReport in asynchronous way, then only lock writelock one time.
> * For thread "IPC Server handler 264 on default port 8032", I think getResourceUsageReport could use readLock.
> Note: Any of the two proposal could solve this problem. But I think we need repair all.
>
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org