cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Brandon Williams (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-6193) Move throughput-heavy activities (repair/compaction) into separate process
Date Sun, 13 Oct 2013 19:04:43 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-6193?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13793760#comment-13793760
] 

Brandon Williams commented on CASSANDRA-6193:
---------------------------------------------

It's complicated at the system level, too.  We can't just fork the process, because that will
require as much free memory to fork as the main process (which is why you need JNA or snapshotting
is extremely slow because the forking of 'ln' is so expensive.)  And then if one jvm OOMs
but the other does not...

> Move throughput-heavy activities (repair/compaction) into separate process
> --------------------------------------------------------------------------
>
>                 Key: CASSANDRA-6193
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6193
>             Project: Cassandra
>          Issue Type: Wish
>          Components: Core
>            Reporter: André Cruz
>            Priority: Minor
>              Labels: ponies
>
> Repairs and compactions are activities that I've seen cause Full GCs. It is difficult
to optimize the GC for pauseless behaviour when the jvm is performing such different functions
as serving client requests and processing large data files.
> Wouldn't it be possible to separate repairs/compactions into another process where Full
GCs would not be a problem?



--
This message was sent by Atlassian JIRA
(v6.1#6144)

Mime
View raw message