You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "Jimmy Xiang (JIRA)" <ji...@apache.org> on 2012/10/19 21:58:15 UTC

[jira] [Commented] (HBASE-6977) Multithread processing ZK assignment events

    [ https://issues.apache.org/jira/browse/HBASE-6977?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13480325#comment-13480325 ] 

Jimmy Xiang commented on HBASE-6977:
------------------------------------

One more change I am going to do in this issue is to enhance the nodeChildrenChanged event handler.  We don't want to watch most of the children again since we are already watching them.  We just need to watch those new children.  In bulk assignment, there are so many children which should not be watched again.  So it will avoid some efforts wasted here.
                
> Multithread processing ZK assignment events
> -------------------------------------------
>
>                 Key: HBASE-6977
>                 URL: https://issues.apache.org/jira/browse/HBASE-6977
>             Project: HBase
>          Issue Type: Improvement
>          Components: Region Assignment
>            Reporter: Jimmy Xiang
>            Assignee: Jimmy Xiang
>
> Related to HBASE-6976 and HBASE-6611.  ZK events processing is a bottle neck for assignments, since there is only one ZK event thread.  If we can use multiple threads, it should be better.
> With multiple threads, the order of events could be messed up. However, if we pass all events related to one region always to the same worker thread, the order should be kept.
> We need to play with it and find out how much performance imrovement we can get.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira