Nick Ustinov wrote:
> Hi!
>
> I use cyrus 2.0.12 + pam_mysql, sendmail 8.11.0 on a dual p3/550 with 1G ram
> (linux redhat7, kernel 2.4.2).
> I've faced a very strange problem -- when the system is under heavy load
> (however, there is ~300Mb of free memory and free ~50% cpu) master process
> is not forking new processes or doing it in a very strange way. If i telnet
> to port 143 it just opens the connection but no cyrus' prompt... it might
> delay for 40-50 seconds. Meanwhile sendmail tries to deliver local mail and
> opens many connections to localhost that eat memory, cpu, swap and then my
> box dies. If I kill sendmail when there are 900 of them and then start it
> again after 5-10 minutes it's ok... So i suppose there are very many
> simultanous requests to master which it can't handle.
>
> Has anyone experienced something similar? Maybe it's a Linux problem, but I
> have no clue what to do.
>
> Nick Ustinov
>
> [EMAIL PROTECTED]
> http://www.videinfra.com
Nick,
just to confirm, we are seeing exactly the same with cyrus 2.0.9 & pam_ldap.
We also tried 2.2 and 2.4 kernels but the problem doesn't go away. We have
about
70 k users and at rush hours cyrus slows down, there are many sendmail
processes
in "localhost user open" state and restarting cyrus & sendmail temporary solves
the problem.
There is no problem when there is only few concurent connections to cyrus. We
are struggling
with this problem quite for some time. Just for a tiny lead that we plan to
investigate - there
was a discussion on linux-kernel list about some big delays on /dev/random in
some cases
http://boudicca.tux.org/hypermail/linux-kernel/2001week15/0064.html
Maybe somebody more familiar with cyrus internals could tell more about this.
BTW. When this happens load is not high ( 2-3 ) only cyrus is completely stuck.
Regards,
Dori Seliskar
