[Dovecot] Very High Load on Dovecot 2 and Errors in mail.err.

[Jesper Dahl Nyerup]

On May 20  16:29, Urban Loesch wrote:
> I checked my kernel and the patch mentioned in
> https://bugzilla.redhat.com/show_bug.cgi?id=681578
> 
> (comment 31) is not applied. It comes in version 3.0.30 and 3.2.17.
> 
> I will see what tomorrow happens under more load.
> If I have the problem again, I give 3.2.17 a chance.

We've seen similar behavior on a similar system with a similar workload.

We've tried a 3.0.31 - after the epoll patch was applied upstream -
without seeing a difference. Right now we're running a 3.3.7 with
vs2.3.3.4, and this has reduced the problem quite a bit, but not
eliminated it completely.

Stracing the processes in D state from before they hang has just
revealed something interesting, however, pointing to an issue with
inotify rather than epoll.

[snip]
[...]
15414 23:27:36 inotify_init()           = 12 <0.000024>
[...]
15414 23:27:36 close(12 <unfinished ...>
15414 23:28:51 <... close resumed> )    = 0 <74.593917>
15414 23:28:51 close(9 <unfinished ...>
15414 23:28:51 <... close resumed> )    = 0 <0.000080>
15414 23:28:51 exit_group(0)            = ?
[/snip]

In short, as far as we can tell, all the processes in D state appear to
be waiting to close the file handle they got from their inotify_init(),
and eventually all these close()s go through almost simultaneously.

Right now we're trawling for locking issues related to inotify, with our
focus mainly at the VServer patch set. I would very much appreciate
updates on your - or anyone else's - findings and progress.

Yours,

Jesper Nyerup.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://dovecot.org/pipermail/dovecot/attachments/20120611/930695fb/attachment.bin>