hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jean-Marc Spaggiari (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-9189) IncreasingToUpperBoundRegionSplitPolicy.shouldSplit() should check all the stores before returning.
Date Mon, 12 Aug 2013 02:36:51 GMT

    [ https://issues.apache.org/jira/browse/HBASE-9189?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13736509#comment-13736509

Jean-Marc Spaggiari commented on HBASE-9189:

What was the implication of this? We'd get a split request, but then not split because there
are unsplittable stores?

Regions with references will be marked as "splittable". So when the memstore flush and check
for compactions, even if there is a reference file they might emit a split request.

However, in the SplitRequest, there is a check for "st.prepare" which will validate if region
is splittable which, in turn, will verify if there is references. And if any, will wait. So
impact seems to be just few more pending threads waiting for the references on the stores
to be removed but I'm not 100% sure if it's the only impact.

Also, DelimitedKeyPrefixRegionSplitPolicy and KeyPrefixRegionSplitPolicy are extending IncreasingToUpperBoundRegionSplitPolicy
so are impacted the same way.
> IncreasingToUpperBoundRegionSplitPolicy.shouldSplit() should check all the stores before
> ---------------------------------------------------------------------------------------------------
>                 Key: HBASE-9189
>                 URL: https://issues.apache.org/jira/browse/HBASE-9189
>             Project: HBase
>          Issue Type: Bug
>    Affects Versions: 0.95.1, 0.94.10, 0.96.0
>            Reporter: Jean-Marc Spaggiari
>            Assignee: Jean-Marc Spaggiari
>         Attachments: HBASE-9189-v0-0.94.patch, HBASE-9189-v0-trunk.patch, HBASE-9189-v1-0.94.patch
> IncreasingToUpperBoundRegionSplitPolicy.shouldSplit() return as soon as it found any
store big enough to be split. However, it should first go trough all all the stores to make
sure they are all splittable before returning.

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

View raw message