hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew Purtell (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-14404) Backport HBASE-14098 (Allow dropping caches behind compactions) to 0.98
Date Sat, 19 Sep 2015 01:17:05 GMT

    [ https://issues.apache.org/jira/browse/HBASE-14404?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14876786#comment-14876786
] 

Andrew Purtell commented on HBASE-14404:
----------------------------------------

As a backport issue we are carrying back a change from later branches that alter reader and
writer behavior unconditionally. I can make it conditional upon HBase settings changes but
can't use the backport. It has to be a different approach that looks at configuration where
we set up the readers and writers, rather than once in CacheConfig. Would no longer be a backport,
but instead something unique to 0.98. I don't think that's what we want. Since I have a +1
to commit the earlier patch I'm going to drop the half-assed thing I put up a few minutes
ago and commit the original backport patch shortly.

> Backport HBASE-14098 (Allow dropping caches behind compactions) to 0.98
> -----------------------------------------------------------------------
>
>                 Key: HBASE-14404
>                 URL: https://issues.apache.org/jira/browse/HBASE-14404
>             Project: HBase
>          Issue Type: Task
>            Reporter: Andrew Purtell
>            Assignee: Andrew Purtell
>             Fix For: 0.98.15
>
>         Attachments: HBASE-14404-0.98.patch
>
>
> HBASE-14098 adds a new configuration toggle - "hbase.hfile.drop.behind.compaction" -
which if set to "true" tells compactions to drop pages from the OS blockcache after write.
 It's on by default where committed so far but a backport to 0.98 would default it to off.
(The backport would also retain compat methods to LimitedPrivate interface StoreFileScanner.)
What could make it a controversial change in 0.98 is it changes the default setting of 'hbase.regionserver.compaction.private.readers'
from "false" to "true".  I think it's fine, we use private readers in production. They're
stable and do not present perf issues.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message