[
https://issues.apache.org/jira/browse/CASSANDRA-3206?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13104632#comment-13104632
]
Jonathan Ellis commented on CASSANDRA-3206:
-------------------------------------------
bq. 5? per sstable
This is incorrect -- when using mmap'd I/O, we don't consume FD-per-sstable (we
use one temporarily when setting up the mmap, then release it). And even in
buffered I/O mode we use one FD per sstable being read, per reading thread.
(Which will be much less than one FD per sstable in most cases, although range
scans under LevelDB do not yet do a very good job of cutting back the number of
SSTables consulted.)
So it sounds like we mostly need to make sure we have it high enough that we
tolerate high numbers of unpooled connections (common in PHP environments, I'm
told).
> increase file descriptor limit in deb, rpm packages
> ---------------------------------------------------
>
> Key: CASSANDRA-3206
> URL: https://issues.apache.org/jira/browse/CASSANDRA-3206
> Project: Cassandra
> Issue Type: Bug
> Components: Packaging
> Reporter: Jonathan Ellis
> Assignee: paul cannon
> Priority: Minor
> Fix For: 0.8.6
>
>
> We can use a lot of file descriptors (one per socket, 5? per sstable).
> People hit this regularly on the user list and it will get worse with Leveled
> compaction, which limits sstable size to a relatively low size (currently
> 5MB).
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
