[sr-dev] kamailio gets frozen when rsyslog-mysql cannot send the SQL query

[Jon Bonilla (Manwe)]

El Wed, 28 Jul 2010 22:27:26 +0200
"Olle E. Johansson" <oej at edvina.net> escribió:

> 
> 28 jul 2010 kl. 22.18 skrev Iñaki Baz Castillo:
> 
> > 2010/7/28 Olle E. Johansson <oej at edvina.net>:
> >>> Being a multiprocess blocking design it would be a bit difficult such
> >>> solution. Perhaps the "ping syslog" mechanism I suggested in my
> >>> initial mail :)
> >> 
> >> I thought it wasn't easy... But still, we need some sort of solution
> >> since this can affect a lot of systems.
> > 
> > Let me suggest this proposal:
> > 
> > - Having a dedicated kamailio process to act as syslogger.
> > 
> > - Other kamailio processes send logging messages to such syslogger
> > process by using a posix queue fifo which is non blocking. This means
> > that kamailio workers would never wait for the "syslog()" call  to
> > return.
> > 
> > - Posix mesage queues implement a configurable queue. But also, the
> > syslogger process could manage its own queue (by using an array as a
> > FIFO system) so messages received by the posix queue  would be
> > inserted in the FIFO and extracted from there when they are logged to
> > syslog.
> > 
> > - In this way, if syslog daemon gets frozen kamailio workers wouldn't
> > realize of it. Some logs would be lost in case the FIFO gets full but
> > this is not so important as getting Kamailio non processing SIP
> > messages :)
> > 
> > Opinnions? Do you consider it feasible?
> 
> Other people are better judges of the architecture, but I do agree with the
> results. We can accept syslog messages getting lost, as long as Kamailio keeps
> running the SIP handling processes.
>

I do not know from the programmer point of view if this approach is feasible.
But kamailio is never the only process running in a system. Other processes as
cron, a mta, ssh server (at least) run in the same machine. If syslog is
blocked, you might get a frozen system anyways. Almost every service relies on
syslog, which makes this solution incomplete.

I think that the solution to this problem is in syslog's side. It's a critical
service in any system and this solution would just mask the problem.