Il 29/06/2012 09:45, Timo Sirainen ha scritto:
On 29.6.2012, at 10.39, Mailing List SVR wrote:
Il 29/06/2012 09:19, Timo Sirainen ha scritto:
On 29.6.2012, at 9.35, Mailing List SVR wrote:
I have some out of memory errors in my logs (file errors.txt attached)
How large is your auth process's VSZ when it starts up and has handled a couple
of logins? It's possible that it's not leaking at all, you're just not giving
enough memory for its normal operation. Some Linux distros nowadays build
binaries that eat up a lot of VSZ immediately when they start up.
ps aux report this:
dovecot 7454 0.0 0.0 85980 3776 ? S 09:36 0:00 dovecot/auth
before restarting dovecot the auth process was running since about 1 hour and
this is the output from ps aux
dovecot 25002 0.0 0.0 86112 3780 ? S 08:24 0:00 dovecot/auth
So you have 44 MB of VSZ available after startup. You also have 10 MB of auth
cache, which could in reality take somewhat more than 10 MB. It doesn't leave a
whole lot available for regular use. I'd increase the auth process's VSZ limit
and see if it still crashes.
I increased the limit to 192MB or should I set the limit to 256MB or
more? I'll wait some days to see if still crash
If you want to, you could also test with valgrind if there's a leak:
service auth {
executable = /usr/bin/valgrind --leak-check=full -q /usr/libexec/dovecot/auth
}
You'd then need to restart the auth process to make valgrind output the leaks.
for now I prefer to avoid valgrind on a production server if the crash
persist with the new limit I'll setup a test environment and I'll run
valgrind there,
thanks
Nicola
