[Dovecot] No buffer space available

[Rick Romero]

I forgot to add that I also upgraded from 1.1.1 to 1.1.3 yesterday in  
an attempt to fix the problem.


On Sep 9, 2008, at 8:58 AM, Rick Romero <rick at havokmon.com> wrote:

> Hi,
>
> I'm running Dovecot 1.1.3 on FreeBSD 7.0.    In the past couple  
> days, I've been seeing errors like this:
>
> Sep  9 08:43:51 sysvol dovecot: pop3-login: socketpair() failed: No  
> buffer space available
> Sep  9 08:43:51 sysvol dovecot: pop3-login: socket(default) failed:  
> No buffer space available
> Sep  9 08:43:51 sysvol dovecot: pop3-login: Can't connect to auth  
> server at default: No buffer space available
> Sep  9 08:25:15 sysvol dovecot: pop3-login: socketpair() failed: No  
> buffer space available
> pop3-login: Can't connect to auth server at default: No buffer space  
> available
>
> What ends up happening is that the port doesn't accept any new  
> connections until there is enough buffer space.
>
> Yesterday, I saw it a LOT with POP3 and changed my config login  
> processes:
> login_process_per_connection: no
> login_process_size: 256
> login_processes_count: 6
> This seemed to clear up the issues, but now it's occurring again  
> this morning.
>
> I haven't had any of these since I made that change either.. So  
> there's some progress... :
> Sep  8 12:53:40 sysvol dovecot: Temporary failure in creating login  
> processes, slowing down for now
>
>
> TOP shows:
> Mem: 267M Active, 5558M Inact, 1637M Wired, 217M Cache, 214M Buf,  
> 87M Free
> Swap: 4096M Total, 4096M Free
>
> At this time, there are 120 imap processes, 10 POP and 1 dovecot and  
> 1 dovecot-auth
> pop and imap both have 6 login processes.  It doesn't seem like a  
> load issue, but obviously something is set too low or unavailable...
> It's fixed for a period of time when I restart dovecot.  I did a  
> netstat -m prior to restart, and after restart, and the only main  
> difference was this line:
> Before:
> 11/414/425/12800 4k (page size) jumbo clusters in use (current/cache/ 
> total/max)
> After:
> 0/425/425/12800 4k (page size) jumbo clusters in use (current/cache/ 
> total/max)
> (After about 5 minutes 'current' went to 3)
>
> This machine also runs NFSd and MySQL.
> Any ideas?
>
> Thanks,
> Rick