Apr 14 23:48:12 mercury dovecot: child 25983 (imap) killed with signal 11 Apr 14 23:48:12 mercury dovecot: child 25992 (imap) killed with signal 11 Apr 14 23:48:14 mercury dovecot: child 26034 (imap) killed with signal 11
Could these be the FETCH MIME crashes that Evolution caused? What clients do you mostly have anyway?
It could be. I'll try the test4 and see what it reports. I generally stage the installation through a test machine and my rather low-volume tests to that machine do not show these failures. So there probably is some difference between my tests and other user accesses.
In fact it appears that all examples of "Corrupted index" are pop3 instances, and all examples of "signal 11" are imap..
Are you sure you rebuilt/reinstalled the pop3 binary? I haven't actually tried pop3 at all for a long time, but it shares all the backend code with imap.
Very sure..
Another difference that occured to me: I built the recent -test versions with the libiconv library, and the previous ones without. This would seem (from my naive perspective) to also be a difference between pop3 and imap code paths.
I tried setting
mail_drop_priv_before_exec = yes
as you suggested, to generate a core dump, but if it's dumping core I can't find it anywhere. dovecot was started from /var/tmp and I have verified that the running dovecot and dovecot-auth processes have /var/tmp as a working directory.
Does everyone have write access to that directory then? It's the imap process that needs it.. Or maybe you have just disabled core dumping? Dovecot uses the ulimit -c that was active when dovecot was started.
I started it from /var/tmp to help with the core dumping (because that is indeed a directory that everybody can write to). I didn't mention it but I did indeed look at the dovecot process limits (via "rlimit") to verify that it did not have a restricted coredump limit.
I also continue to see sporadic "(imap) returned error 89" with this version and with 0.99.8.1
Without more specific error above it?
Not that I have been able to find.. i.e. no correlation with the pid given.
mm