hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jeffrey Zhong (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-11314) Remove MultiVersionConsistencyControl#writeQueue
Date Tue, 10 Jun 2014 17:55:01 GMT

    [ https://issues.apache.org/jira/browse/HBASE-11314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14026762#comment-14026762
] 

Jeffrey Zhong commented on HBASE-11314:
---------------------------------------

So far I don't have other better ways. As you suggested before, using disruptor might be another
option as writeQueue is a queue implementation. :-). 

> Remove MultiVersionConsistencyControl#writeQueue
> ------------------------------------------------
>
>                 Key: HBASE-11314
>                 URL: https://issues.apache.org/jira/browse/HBASE-11314
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: Jeffrey Zhong
>             Fix For: 0.99.0
>
>
> This is a follow up improvement of HBase-8763. After we combined MVCC & LogSeqId,
the memstore read point advances in the logSeqId increasing order. The MultiVersionConsistencyControl#writeQueue
can be possibly removed.
> In order to remove writeQueue, we need to return the disruptor sequence of the last WALEdit
with durability >= SYNC_WAL inside disruptor consumer code before assigning log sequence
Id. If current WALEdit has durability >= SYNC_WAL, we return the disruptor sequence of
current WALEdit.
> During syncOrDefer, we always waits for the sequence of the last WALEdit has durability
>= SYNC_WAL.
> The tricky part is how to return the disruptor sequence of the last WALEdit with durability
>= SYNC_WAL cleanly from disruptor consumer. 
> We could put those info in HLogKey because we always create a new HLogKey object for
each change otherwise we have to create a new object for this purpose.



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

Mime
View raw message