[jira] Updated: (HBASE-1939) HLog group commit

["Nicolas Spiegelberg (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HBASE-1939?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Nicolas Spiegelberg updated HBASE-1939:
---------------------------------------

    Attachment: HBASE-1939-20.4.patch

Ported this patch back to version 0.20.  Note that this refactoring incidentally fixed a bug on the tip with HLog.sync().  To summarize, the code under sync() should read 
    sync(); syncFs() 
instead of 
    (syncFs) ? syncFs() : sync();  
sync() flushes the SequenceFile.Writer buffer to DFSOutputStream.  syncFs() flushes the output stream across the network to the datanodes.  So the current implementation isn't flushing the buffers properly and still gets stuck on local storage until the SequenceFile.Writer buffer is full. 

> HLog group commit
> -----------------
>
>                 Key: HBASE-1939
>                 URL: https://issues.apache.org/jira/browse/HBASE-1939
>             Project: Hadoop HBase
>          Issue Type: New Feature
>            Reporter: Jean-Daniel Cryans
>            Assignee: Jean-Daniel Cryans
>             Fix For: 0.21.0
>
>         Attachments: HBASE-1939-20.4.patch, HBASE-1939-v2.patch, HBASE-1939-v3.patch, HBASE-1939.patch
>
>
> Currently multiple clients writing to a RS are serialized when sync'ing their appends. Implementing group commit can help this by being aware of all the clients in that queue.

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