hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sameet Agarwal (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HBASE-13316) Reduce the downtime on panned moves of regions
Date Mon, 23 Mar 2015 16:19:11 GMT

     [ https://issues.apache.org/jira/browse/HBASE-13316?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Sameet Agarwal updated HBASE-13316:
    Status: Patch Available  (was: Open)

This is a patch to do a very simple version of this change. We basically change the sequence
for a region move to be:
1. Warmup region (in the destination regionserver)
2. Close region in the source regionserver
3. Open region in the destination regionserver

Warmup region is the new step in the process. It basically opens the region and closes it
(without adding it to the list of open regions). The open of a region reads the most common
pages into the cache and then close doesn't throw them out. This means that after step #1,
the destination regionserver has the common pages in cache and reduces the time for step #3.

A more complete fix would involve tracking the most used pages and reading them in as part
of warmup. This can be added later but is not in this patch

> Reduce the downtime on panned moves of regions
> ----------------------------------------------
>                 Key: HBASE-13316
>                 URL: https://issues.apache.org/jira/browse/HBASE-13316
>             Project: HBase
>          Issue Type: Improvement
>          Components: Balancer
>    Affects Versions: 2.0.0, 1.0.1, 1.1.1
>            Reporter: Sameet Agarwal
>            Assignee: Sameet Agarwal
>            Priority: Minor
>             Fix For: 2.0.0
>   Original Estimate: 168h
>  Remaining Estimate: 168h
> The current behavior of a region move shuts down a region and then starts is up in another
regionserver. This causes increased latency and possibly timeouts till the new region's cache
is fully warmed up. We can make a region move less disruptive by warming the cache in the
destination region server before shutting dow the old region.

This message was sent by Atlassian JIRA

View raw message