Hi Timo,
I think I can still trigger this issue though, by opening Outlook 2003 (sorry :X), and holding down F9 for a bit (send/receive shortcut). When the send/receive query contains about 5+ actions, outlook stops responding.
But still a LOT better then before. At least when the problem occurs, other users can still login, which wasn't the case before.
Could it be because of Outlook? ;)
agreed++ :)
I tested it with 100 simultaneous virtual clients doing login+logout all the time and didn't notice any hangs. You could set auth_debug=yes and look at the logs what happens when the F9 is being hold. Does it log an ldap_bind() that just doesn't seem to get a reply?
Actually, it binds just 1 time, which succeeds just fine. If you now press F9 900 times, it will work just fine. It sends/reveives over one connection, no new bind()'s, no hangs.
Now if you also start browsing some folders (for which Outlook opens a new connection) while the 900 send/receive's are being handled, it is quite easy to make dovecot stop responding.
Hmm. Actually now that I think of it, the ldap_bind() must give a reply before the connection is used for anything else. So if the other users can still log in, I think there's a good chance that the bug isn't in Dovecot's side.
I see no indication of a ldap_bind or ldap_search (dn or user search) _not_ returning. All ldap code seems fine to me atm (yay).
I've attached the rawlogs from a co-working of mine, who just triggered it. I hope this is the relevant portion of the log.
Cheers, Marc