[
https://issues.apache.org/jira/browse/HBASE-8769?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13688137#comment-13688137
]
Nicolas Liochon commented on HBASE-8769:
----------------------------------------
It would be a good thing to do, but it seems non trivial to me:
With a scenario like: 3 tables of 50 regions each. I read from the first two
and write in the third, using the default config everywhere. HBase cannot know
that the first two write buffers will remain empty, so it may lowers the memory
available for the 3rd table while it's not necessary.
Or, if we want to have a global view in real time we will have to add some
synchronisation at the HConnection level (i.e: AtomicLong or so), likely for
each inserts.
Today, we're pushing this to the user... Agreed it's not perfect.
> Calculates the client write buffer size from the number of region
> -----------------------------------------------------------------
>
> Key: HBASE-8769
> URL: https://issues.apache.org/jira/browse/HBASE-8769
> Project: HBase
> Issue Type: Improvement
> Components: Client
> Affects Versions: 0.98.0, 0.95.1
> Reporter: Nicolas Liochon
> Assignee: Nicolas Liochon
>
> The default size is ~2Mb. There are two issues:
> 1) It has not changed for a few years, while the memory available has
> increased
> 2) It's a fixed size, that does not take into account the number of region.
> I think we should have something like 100Kb per region, with a max at 6Mb,
> all this configurable of course.
> I plan to do this after HBASE-6295
--
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
