[
https://issues.apache.org/jira/browse/CASSANDRA-2521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13056239#comment-13056239
]
Terje Marthinussen commented on CASSANDRA-2521:
-----------------------------------------------
First impression is good.
I merged this with 2816 last night and tested overnight with a dataset that
normally uses 32-40% disk space when compacted.
Calling repair on a handful of nodes, I would normally have a good chance of
seeing a node or two with 90%+ full disk and regular full GC's.
With these patches, disk use seem to peak in the 55-65% disk area which is a
very significant improvement.
However, unfortunately, I started repairs on 5 of 12 nodes and one of them has
gone crazy however and filled the disk 100%. :(
I did happen to start repair twice on this node by accident.
I somewhat do not think that is the real problem however, but I am not sure.
Will make a ticket anyway on adding functionality to prevent repairs from
getting started twice on the same CF and I will test more to see if this
happens again.
However I noticed this in the log:
INFO [Thread-185] 2011-06-28 05:01:15,390 StorageService.java (line 2083)
requesting GC to free disk space
I guess we can get rid of that?
Should be nothing good from calling full GC there anymore?
If anything, maybe we should consider a variation here where cassandra instead
aborts all repairs and compactions and clean up everything before trying again
with smaller amounts of data in each compaction/repair attempt?
Another ticket on that maybe?
> Move away from Phantom References for Compaction/Memtable
> ---------------------------------------------------------
>
> Key: CASSANDRA-2521
> URL: https://issues.apache.org/jira/browse/CASSANDRA-2521
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Chris Goffinet
> Assignee: Sylvain Lebresne
> Fix For: 1.0
>
> Attachments:
> 0001-Use-reference-counting-to-decide-when-a-sstable-can-.patch,
> 0001-Use-reference-counting-to-decide-when-a-sstable-can-v2.patch,
> 0002-Force-unmapping-files-before-deletion-v2.patch, 2521-v3.txt, 2521-v4.txt
>
>
> http://wiki.apache.org/cassandra/MemtableSSTable
> Let's move to using reference counting instead of relying on GC to be called
> in StorageService.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
