dovecot - Dec 2014 - Error: mremap_anon(###) failed: Cannot allocate memory

We're running dovecot 2.2.15 with pigeonhole 0.4.6, in a clustered 
environment, nfs with proxy and backend on all servers.

I've been seeing some odd errors from lmtp:

Error: mremap_anon(127930368) failed: Cannot allocate memory

It seems to affect specific users, but it doesn't seem to manifest in any 
particular way; no user complaints. Just the occasional log message.

I would guess this is a bug? I'm open to suggestions and I'd be happy to
post config if somebody has an idea.

Thanks,
Andy

---
Andy Dills
Xecunet, Inc.
www.xecu.net
301-682-9972
---

On 12/08/2014 02:54 AM, Andy Dills wrote:
> 
> We're running dovecot 2.2.15 with pigeonhole 0.4.6, in a clustered 
> environment, nfs with proxy and backend on all servers.
> 
> I've been seeing some odd errors from lmtp:
> 
> Error: mremap_anon(127930368) failed: Cannot allocate memory
> 
> It seems to affect specific users, but it doesn't seem to manifest in
any
> particular way; no user complaints. Just the occasional log message.
A config would always be useful, but I can venture a guess. Perhaps the affected
users have a dovecot.index.cache file
somehwere, e.g. under INBOX, that is larger than the memory limit for the lmtp
process. Try increasing "default_vsz_limit" or
the "service lmtp { vsz_limit }". Removing the overly large index
cache file should also, temporarily, help. In case you do not
get this error from the imap/pop3 processes, perhaps you have already set a
higher vsz_limit for those?

br,
Teemu Huovila

On 12/08/2014 03:07, Teemu Huovila wrote:
> A config would always be useful, but I can venture a guess. Perhaps the 
> affected users have a dovecot.index.cache file
> somehwere, e.g. under INBOX, that is larger than the memory limit for 
> the lmtp process. Try increasing "default_vsz_limit" or
> the "service lmtp { vsz_limit }". Removing the overly large index
cache
> file should also, temporarily, help. In case you do
> not get this error from the imap/pop3 processes, perhaps you have 
> already set a higher vsz_limit for those?
Teemu,

Thanks for your suggestion. I checked the output of doveconf, and by 
default it appears the vsz_limit is set to 18446744073709551615B for 
each of the services, and 256M for default_vsz_limit.

I checked a user in question, and their index.cache was indeed large, 
123M. Seemingly needlessly so, as I deleted the dovecot files and 
reindexed, and now it's 6K.

Thanks, I'll keep an eye on the users this affects and try to get their 
index.cache in order.

Thanks,
Andy