accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Keith Turner (JIRA)" <>
Subject [jira] [Commented] (ACCUMULO-4187) Rate limiting of major compactions
Date Wed, 13 Apr 2016 17:58:25 GMT


Keith Turner commented on ACCUMULO-4187:

Recording offline discussions I had w/ Shawn about this issue.

[~ShawnWalker] experimented with per table rate limits while working on this.   He identified
a serious issue with this, a table with a low rate limit could monopolize all of the compaction
threads causing problems for all other tables.  One possible solution to this we discussed
was Compaction Resource Groups.   Each CRG would have a thread pool and rate limit.  Multiple
named CRGs could be defined at the system level.   Tables and user initiated compactions could
be assigned to a CRG.  This solution allows a table to have low rate limit without impacting
tables with higher rate limits.

> Rate limiting of major compactions
> ----------------------------------
>                 Key: ACCUMULO-4187
>                 URL:
>             Project: Accumulo
>          Issue Type: Improvement
>          Components: core, tserver
>            Reporter: Shawn Walker
>            Assignee: Shawn Walker
>            Priority: Minor
>             Fix For: 1.8.0
> In discussing [ACCUMULO-4166] with Keith Turner, we decided that the underlying issue
is that major compactions can overwhelm a tablet server, rendering it nearly unresponsive.
> To address this, we should take a cue from Apache Cassandra and restrict how quickly
we perform major compactions.  Rate limiting reads and writes involved in major compactions
will directly affect the IO load caused by major compactions, and should also indirectly affect
the CPU load.

This message was sent by Atlassian JIRA

View raw message