curator-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tom Byrne (JIRA)" <>
Subject [jira] [Commented] (CURATOR-172) Deadlock when performing background operation
Date Wed, 24 Jun 2015 22:46:04 GMT


Tom Byrne commented on CURATOR-172:

OK, I think I might have a line on what's happening. 

Thread A is making a call. It gets a CuratorZKClient, which calls checkTimeouts (getting the
ConnectionState monitor). The connection is gone, and so reset() gets called, then HandleHolder
calls ZK.close, which then calls ZKClientCnxn.close. 

This calls submitRequest which queues a "disconnect" packet, then wait()'s on the packet object.

This wait() will only get notified when the packet is dequeued and notifyAll() is called in
finishPacket(). When this thread calls wait(), it releases the monitor lock on the Packet
object, but keeps the other two monitors that it has, one on the ConnectionState object, and
another on the Zookeeper object itself. 

During this wait, the SendThread (either already running or wakes up, whatever). Throws an
exception, which trickles through to cleanup().

inside cleanup(), the pending queue is drained, calling conLossPacket, then finishPacket().

Because they're pending aysnc packets, eventThread.queuePacket() is called on the pending
async packet. We call the callback, which in turn trickles back up to (in this case) the CreateBuilderImpl
AsyncCallback.StringCallback's processResult, which then tries to submit another event, which
will never get processed, because in order to get the ZKClient again, we call getZookeeperClient,
which will try to get the ConnectionState monitor lock that Thread A has.  

It will wait for that, never dequeuing, processing and notifying the "disconnect" packet that
Thread A is waiting on.

I think I can reproduce this with a small test, I'll see what I can come up with in the next
day or two.

> Deadlock when performing background operation
> ---------------------------------------------
>                 Key: CURATOR-172
>                 URL:
>             Project: Apache Curator
>          Issue Type: Bug
>          Components: Client
>    Affects Versions: 2.4.2
>         Environment: Linux HOSTNAME-REMOVED 2.6.32-279.19.1.el6.x86_64 #1 SMP Tue Dec
18 15:04:44 PST 2012 x86_64 x86_64 x86_64 GNU/Linux
> java version "1.7.0_60"
> Java(TM) SE Runtime Environment (build 1.7.0_60-b19)
> Java HotSpot(TM) 64-Bit Server VM (build 24.60-b09, mixed mode)
>            Reporter: Tom Byrne
> Had a box get into a state where our ZK connections were all deadlocked, waiting on an
object monitor. jstack shows that our background thread that was creating a node was waiting
on a lock that was held by the CuratorFramework thread, who was waiting on an object monitor
that looks like it couldn't be completed until our other write was finished (packet.finish
would never return true.) 
> We have seen this happen twice, but don't notice it until afterwards, and don't have
enough logging to know what's triggering it (possible ZK connections going away?) 
> Rest of the box is fine, network connections are not flapping, main IO threads continue
to accept and process connections, until we get backed up waiting for ZK. 
> Here are the two stack traces:
> "ZooChangeWatcher-BackgroundReader--2-1-SendThread()" daemon prio=10 tid=0x00007fcf64108000
nid=0x88d waiting for monitor entry [0x00007fcbf5d16000]
>    java.lang.Thread.State: BLOCKED (on object monitor)
> 	at org.apache.curator.ConnectionState.checkTimeouts(
> 	- waiting to lock <0x00000000d526bcc8> (a org.apache.curator.ConnectionState)
> 	at org.apache.curator.ConnectionState.getZooKeeper(
> 	at org.apache.curator.CuratorZookeeperClient.getZooKeeper(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl.performBackgroundOperation(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl.processBackgroundOperation(
> 	at org.apache.curator.framework.imps.CreateBuilderImpl.pathInBackground(
> 	at org.apache.curator.framework.imps.CreateBuilderImpl.forPath(
> 	at org.apache.curator.framework.imps.CreateBuilderImpl.forPath(
> 	at
> 	at$000(
> 	at$4.processResult(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl.sendToBackgroundCallback(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl.checkBackgroundRetry(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl.processBackgroundOperation(
> 	at org.apache.curator.framework.imps.CreateBuilderImpl.sendBackgroundResponse(
> 	at org.apache.curator.framework.imps.CreateBuilderImpl.access$600(
> 	at org.apache.curator.framework.imps.CreateBuilderImpl$6.processResult(
> 	at org.apache.zookeeper.ClientCnxn$EventThread.processEvent(
> 	at org.apache.zookeeper.ClientCnxn$EventThread.queuePacket(
> 	- locked <0x00000000fa8e16f8> (a java.util.concurrent.LinkedBlockingQueue)
> 	at org.apache.zookeeper.ClientCnxn.finishPacket(
> 	at org.apache.zookeeper.ClientCnxn.conLossPacket(
> 	at org.apache.zookeeper.ClientCnxn.access$2400(
> 	at org.apache.zookeeper.ClientCnxn$SendThread.cleanup(
> 	- locked <0x00000000fa8e1380> (a java.util.LinkedList)
> 	at org.apache.zookeeper.ClientCnxn$
> "CuratorFramework-0" daemon prio=10 tid=0x00007fd02cb57800 nid=0x4425 in Object.wait()
>    java.lang.Thread.State: WAITING (on object monitor)
> 	at java.lang.Object.wait(Native Method)
> 	at java.lang.Object.wait(
> 	at org.apache.zookeeper.ClientCnxn.submitRequest(
> 	- locked <0x00000000fa8e6750> (a org.apache.zookeeper.ClientCnxn$Packet)
> 	at org.apache.zookeeper.ClientCnxn.close(
> 	at org.apache.zookeeper.ZooKeeper.close(
> 	- locked <0x00000000fa8e0948> (a org.apache.zookeeper.ZooKeeper)
> 	at org.apache.curator.HandleHolder.internalClose(
> 	at org.apache.curator.HandleHolder.closeAndReset(
> 	at org.apache.curator.ConnectionState.reset(
> 	- locked <0x00000000d526bcc8> (a org.apache.curator.ConnectionState)
> 	at org.apache.curator.ConnectionState.checkTimeouts(
> 	- locked <0x00000000d526bcc8> (a org.apache.curator.ConnectionState)
> 	at org.apache.curator.ConnectionState.getZooKeeper(
> 	at org.apache.curator.CuratorZookeeperClient.getZooKeeper(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl.performBackgroundOperation(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl.backgroundOperationsLoop(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl.access$300(
> 	at org.apache.curator.framework.imps.CuratorFrameworkImpl$
> 	at java.util.concurrent.FutureTask$Sync.innerRun(
> 	at
> 	at java.util.concurrent.ThreadPoolExecutor.runWorker(
> 	at java.util.concurrent.ThreadPoolExecutor$
> 	at
> Help me Obi-Wan Kenobi, you're my only hope. 

This message was sent by Atlassian JIRA

View raw message