From: "John C. Amodeo" <[EMAIL PROTECTED]>
> My current config is 2.0.16 / RH 7.1 / 2.4.3-12 kernel
Yes, I'm currently running 2.0.15 on a custom-built 2.4.3 kernel (RH 7.1 as
well).
> Once in a while, when a user goes to copy a message from their Inbox to
> a subfolder, it locks their IMAP session, After closing the mail client
> (Webmail or Netscape Messenger), when they log back in, now they cannot
> delete any messages in any of their folders, and, in addition, they
> cannot write to any folders under their account. Furthermore, Postfix
> starts deferring mail because it cannot write to the users Inbox either.
I configed cyrus to prefork 2 lmtp processes and that is what I see during
normal operation. But sometimes I periodically notice a bunch of lmtp
processes hanging about. Restarting cyrus became problematic because my users
have pine as their shell and they get scared when the connection drops.
I use the following recipe to remedy the problem without restarting the whole
thing:
1) Check to make sure the lmtp (or deliver) process is stuck because it is
waiting for an exclusive lock. This will also give you the file descriptor of
the stuck process.
# strace -p <PID of stuck LMTP process>
2) Find the file it is stuck on.
# lsof -p <PID of stuck LMTP process> | grep <FD number>u
3) Once you have the file check which processes have that file open.
# lsof | grep <file> | awk '{print $2}' | sort -n | uniq | xargs ps -e
4) That will give you a list of processes. I usally find the oldest one
(look at the date field reported by ps) which has always been an imapd process
and kill that. After doing this, the rest of the processes will be able to
grab the exclusive lock and finish what they were doing.
The good thing about doing that is that you're not disconnecting *everyone*.
The bad is that I heard complains that the same message will appear more than
once (my guess is that sendmail runs the queue and spawns another lmtp).
I still don't know why this is happening in the first place though. I recently
compiled 2.0.16 with -g so I can attach gdb to a stuck process and figure out
whats freezing it. Also I configured the new 2.0.16 with fcntl instead of
flock which I hope is the reason -- no idea though.
If anyone has any other thoughts, I'd love to hear.
--
/// max kalika
// [EMAIL PROTECTED]
"and after a while you'll stop making computers out of stuff and start making
stuff out of computers" -- douglas adams 4/5/01
