cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benedict (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (CASSANDRA-6689) Partially Off Heap Memtables
Date Tue, 25 Mar 2014 13:16:23 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-6689?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13946318#comment-13946318
] 

Benedict edited comment on CASSANDRA-6689 at 3/25/14 1:14 PM:
--------------------------------------------------------------

-Agreed-. Pushed updated version to branch, which also modifies the FileUtils.clean() code
to be the neater direct cast, and to check for the possibility by actually attempting a clean
of a DirectByteBuffer.

bq. and just fail over to the heap allocation

I've opted for dying with an error message, and prompting the user to change the config, but
while this is less in keeping with what is the normal approach, I think the danger of eating
up all your memory (and possibly crashing with OOM at random times) is bad enough to warrant
the break from the norm, and simply falling back to heap allocation doesn't allow you to fulfil
the config spec, as we cannot simply add the offheap space to the onheap space, as this would
inflate the ratio of heap space used, and ignoring the offheap space might be particularly
bad in e.g. later non-buffered implementations, as the heap space could have been tuned down
in the knowledge it is less necessary. 

I think it's better to say: cannot fulfil your request, please advise.


was (Author: benedict):
-Agreed-. Pushed updated version to branch, which also modifies the FileUtils.clean() code
to be the neater direct cast, and to check for the possibility by actually attempting a clean
of a DirectByteBuffer.

bq. and just fail over to the heap allocation

I've opted for dying with an error message, and prompting the user to change the config, but
while this is less in keeping with what is the normal approach, I think the danger of eating
up all your memory (and possibly crashing with OOM at random times) is bad enough to warrant
the break from the norm.

> Partially Off Heap Memtables
> ----------------------------
>
>                 Key: CASSANDRA-6689
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6689
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: Core
>            Reporter: Benedict
>            Assignee: Benedict
>              Labels: performance
>             Fix For: 2.1 beta2
>
>         Attachments: CASSANDRA-6689-final-changes.patch, CASSANDRA-6689-small-changes.patch
>
>
> Move the contents of ByteBuffers off-heap for records written to a memtable.
> (See comments for details)



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message