On Fri, 2009-12-18 at 14:56 +0000, Chris Wakelin wrote:
I guess the crash is caused by the "connection queue full". I'll see about fixing the crash, but you shouldn't let it get full in any case. So increase your login_max_connections value.
That's puzzling, I thought it was supposed to fire up another imap-login process (up to login_max_processes_count)?
It's supposed to, yes, but it's been buggy for a long time and I haven't figured out how to fix it. v2.0 should finally fix it..
We have something in the region of 1000 concurrent connections and login_max_processes_count * login_max_connections = 128 * 256 should be enough for 32000+ of them!
Would it be better to have login_max_connections = 1024 (or even higher)?
Yes. If you have only one CPU/core you shouldn't have more than 2 login processes.
I'm also confused as to why we never saw the problem on the Solaris box (with more connections as it was during term time).
It was probably doing a better job than Linux of distributing connections to processes.