Henry Robinson updated ZOOKEEPER-783:

    Attachment: ZOOKEEPER-783.patch

Defensive copying added to getCommittedLog() and synchronization during 

No tests added; really not sure how best to test for this. It does fix my test 
case but it's very difficult to distill that into a test (plus it only fails 
once in about 100 runs). 

> committedLog in ZKDatabase is not properly synchronized
> -------------------------------------------------------
>                 Key: ZOOKEEPER-783
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-783
>             Project: Zookeeper
>          Issue Type: Bug
>          Components: server
>    Affects Versions: 3.3.1
>            Reporter: Henry Robinson
>            Priority: Critical
>         Attachments: ZOOKEEPER-783.patch
> ZKDatabase.getCommittedLog() returns a reference to the LinkedList<Proposal> 
> committedLog in ZKDatabase. This is then iterated over by at least one 
> caller. 
> I have seen a bug that causes a NPE in LinkedList.clear on committedLog, 
> which I am pretty sure is due to the lack of synchronization. This bug has 
> not been apparent in normal ZK operation, but in code that I have that starts 
> and stops a ZK server in process repeatedly (clear() is called from 
> ZooKeeperServerMain.shutdown()). 
> It's better style to defensively copy the list in getCommittedLog, and to 
> synchronize on the list in ZKDatabase.clear.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

Reply via email to