dovecot - Dec 2008 - 1.1.6 -> 1.1.7 regression: dovecot: pipe() failed: Too many open files

Hi all,

We updated to 1.1.7 a week or two back from 1.1.6 (both standard atrpms
builds). Since doing so, twice now I've seen the imap services accept
connections but they hang before the banner is printed (pop seems to be
unaffected). I see this in the logs:

Dec 23 14:02:15 mail9 dovecot: pipe() failed: Too many open files
Dec 23 14:02:15 mail9 dovecot: Temporary failure in creating login
processes, slowing down for now
Dec 23 14:02:15 mail9 dovecot: Created login processes successfully,
unstalling
Dec 23 14:02:16 mail9 dovecot: pipe() failed: Too many open files
Dec 23 14:02:16 mail9 dovecot: Temporary failure in creating login
processes, slowing down for now
Dec 23 14:02:16 mail9 dovecot: Created login processes successfully,
unstalling

The ulimit for the number of open files on these servers has remained
unchanged at 8192 and there are no warnings about too few fd's on
dovecot startup. There weren't amazing numbers of imap sessions open at
the time (only something like 400-500; we've had 1.5k in the past), so I
suspect this is a regression between 1.1.6 and 1.1.7 where you've got a
leak of fd's ? If it happens again I'll try and get a snapshot of the
fd's open by the master dovecot process.

Thanks,

Mark

--
Mark Zealey -- Shared Hosting Team Leader
Product Development * Webfusion
123-reg.co.uk, webfusion.co.uk, donhost.co.uk, supanames.co.uk

This mail is subject to http://www.gxn.net/disclaimer

On Tue, 2008-12-23 at 14:23 +0000, Mark Zealey wrote:
> Dec 23 14:02:15 mail9 dovecot: pipe() failed: Too many open files
dovecot master process has opened too many files.
> The ulimit for the number of open files on these servers has remained
> unchanged at 8192
Should be enough..
> and there are no warnings about too few fd's on
> dovecot startup.
I'm not completely sure that the warning check is correct. I think it
should be though.
>  There weren't amazing numbers of imap sessions open at
> the time (only something like 400-500; we've had 1.5k in the past), so
I
> suspect this is a regression between 1.1.6 and 1.1.7 where you've got a
> leak of fd's ? 
Sounds like a leak, but I doubt it was a regression. Probably you just
hit a rare condition.
> If it happens again I'll try and get a snapshot of the
> fd's open by the master dovecot process.
Yes, that would help figuring it out.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
URL:
<http://dovecot.org/pipermail/dovecot/attachments/20090108/bf109025/attachment-0002.bin>

> >  There weren't amazing numbers of imap sessions open at
> > the time (only something like 400-500; we've had 1.5k in 
> the past), so I
> > suspect this is a regression between 1.1.6 and 1.1.7 where 
> you've got a
> > leak of fd's ? 
> 
> Sounds like a leak, but I doubt it was a regression. Probably you just
> hit a rare condition.
Yeah I've not seen it happen again since 23rd Dec...

Mark