[server-admins] [Fwd: Logwatch for www.crosswire.org (Linux)]

[Jonathan Marsden]

Karl Kleinpaste wrote:

> At least as likely is that not-yet-timed-out dovecot connections
> exist, credited against the total, when the devices had their own
> network failure or reboot, and opened a new connection.

That would get us up to two connections active per device, i.e. six 
simultaneous connections total from one IP, in Chris's case.  To get to 
more than ten connections total that way would mean the devices were all 
restarting three times within a two minute window (TIME_WAIT period), 
which seems unlikely.  For them to do this three times in one day, as 
the log indicates... that seems even more unlikely, to me.

> For that matter, perhaps a plain dovecot bug in which processes
> either do not exit as intended, or the per-connection count is being
> mishandled.

Could be, although I'd expect to see more log entries like this in that 
case.  The closest thing I have to compare with is a production RHEL 5 
server running an older dovecot, dovecot-1.0.7-7.el5, and I am not 
seeing these messages logged on it, with same the default max 
connections per IP per user.  Maybe none of our users are as 
well-connected as Chris is (!), but if there were a bug like this, it 
"ought" to affect more than just one user, I would think.  Or maybe this 
setting and warning message were only added in dovecot 1.1, and noone 
else on crosswire.org is affected.

In any event, turning on dovecot debug logging on crosswire.org for a 
while should help gain a better understanding of what is going on with 
this (assuming it recurs at all).  If that is too much work, the max can 
be set higher, as I suggested in an earlier email.

Jonathan