[ZOOKEEPER-3706] ZooKeeper.close() would leak SendThread when the network is broken

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.6.0, 3.4.14, 3.5.6
• Fix Version/s: 3.5.10, 3.6.3, 3.7.0, 3.8.0
• Component/s: java client
• Labels: pull-request-available
• External issue URL: https://github.com/apache/zookeeper/pull/1235
• Assignee: Pierre Yin
• Reporter: Pierre Yin
• Created: 25/Jan/20 08:28
• Updated: 17/May/22 09:02
• Resolved: 03/Mar/20 02:13

Description

The close method of ZooKeeper may cause the leak of SendThread when the network is broken.

When the network is broken, the SendThread of ZooKeeper client falls into the continuous reconnecting scenario. But there is an unsafe point which is just at the moment before startConnect() during the continuous reconnecting. If SendThread.close() in another thread hit the unsafe point, startConnect() would sleep some time and force to change state to States.CONNECTING although SendThread.close() already set state to States.CLOSED. In this case, the SendThread would be never be dead and nobody would change the state again.

In normal case, ZooKeeper.close() would be blocked forever to wait closeSession packet is finished until the network broken is recovered. But if user set the request timeout, ZooKeeper.close() would break the block waiting within timeout and invoke SendThread.close() to change state to CLOSED. That's why SendThread.close() can hit the unsafe point.

Set request timeout is a very common practice. 

I propose a patch and send it out later.

Maybe someone can help to review it.

 

Thanks

 

 

Issue Links

• is related to: ZOOKEEPER-4223 Backport ZOOKEEPER-3706 to branch-3.6
• links to: GitHub Pull Request #1235