hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Adela Maznikar (JIRA)" <j...@apache.org>
Subject [jira] [Resolved] (HBASE-10940) [0.89-fb] Rebalance primaries in-non disruptive way
Date Wed, 09 Apr 2014 00:31:24 GMT

     [ https://issues.apache.org/jira/browse/HBASE-10940?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Adela Maznikar resolved HBASE-10940.

    Resolution: Fixed

code in RegionPlacement.balancePrimaries

> [0.89-fb] Rebalance primaries in-non disruptive way
> ---------------------------------------------------
>                 Key: HBASE-10940
>                 URL: https://issues.apache.org/jira/browse/HBASE-10940
>             Project: HBase
>          Issue Type: New Feature
>          Components: Region Assignment
>    Affects Versions: 0.89-fb
>            Reporter: Adela Maznikar
>            Assignee: Adela Maznikar
>             Fix For: 0.89-fb
> Added an option RegionPlacement -bp (balance primaries).
> Sometimes in production we hit a case where some of the regionservers are serving less
number of regions than the rest and just updating the assignment plan doesn't help since locality
is highest on those locations and locality is the highest parameter considered when we produce
assignment plan for primaries. 
> We want to switch primary and tertiary in a case where the regionserver has
> more than average primaries and more primaries assigned than the
> regionserver where teritary is (and tertiary has less than average). We do
> bfs traversal from each node (node being a regionserver, and there is an edge from machine
serving primary to machine serving tertiary for a particular region) until we find the node
to which there is a
> path and that node has less primaries assigned. Then we switch primary and
>  tertiary for all nodes on the path. We do that for each node. (node ==
> machine).
> The motivation of the approach is - we want to move the region to a location where locality
is high enough.

This message was sent by Atlassian JIRA

View raw message