hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Blue Arrow (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-1869) Access to zkAcl should be synchronized in ZKRMStateStore#addStoreOrUpdateOps()
Date Fri, 28 Mar 2014 05:23:15 GMT

    [ https://issues.apache.org/jira/browse/YARN-1869?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13950384#comment-13950384
] 

Blue Arrow commented on YARN-1869:
----------------------------------

Shorter duration locks are good as a general principle for performance but do we know if the
difference is worth worrying about in this case? One nice thing about the method being synchronized
is that it makes the synchronization clear and visible when future code changes are made,
which is always a plus when working with concurrency. (It does leave the possibility of deadlocks
via external references versus synchronizing using a private). If we want to implement a shorter
duration lock, we should make that consistent by modifying the other methods that are accessing
zkAcl? Do we know if access/manipulation of zkAcl is the only critical section in these methods?
Thoughts? 

> Access to zkAcl should be synchronized in ZKRMStateStore#addStoreOrUpdateOps()
> ------------------------------------------------------------------------------
>
>                 Key: YARN-1869
>                 URL: https://issues.apache.org/jira/browse/YARN-1869
>             Project: Hadoop YARN
>          Issue Type: Bug
>            Reporter: Ted Yu
>            Priority: Minor
>         Attachments: yarn-1869.patch
>
>
> Here is related code:
> {code}
>       } else {
>         opList.add(Op.create(nodeCreatePath, tokenOs.toByteArray(), zkAcl,
>             CreateMode.PERSISTENT));
>       }
> {code}
> The other methods accessing zkAcl are synchronized.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message