On Sat, 2006-02-25 at 17:22 +0100, Branko F. Gračnar wrote:
dovecot.log.1.gz:Feb 24 16:05:21 elfstone dovecot: imap(user3@domain.tld): Maildir /export/mail/domain.tld/user3 sync: UID inserted in the middle of mailbox (47681 > 47680, file = 1138960753.P9796Q0M642679.elfstone:2,T)
Well, this means that the file was lost, but then Dovecot saw it again later.
dovecot.log.1.gz:Feb 24 16:46:27 elfstone dovecot: imap(user1@domain.tld): Maildir /export/mail/domain.tld/user1 sync: UID < next_uid (1 < 199, file = 1122973973.V809Id001aafc.elfstone,S=8889:2,RS)
Same problem.
dovecot.log.1.gz:Feb 24 16:54:47 elfstone dovecot: imap(user5@domain.tld): Corrupted index file /export/data/dovecot/domain.tld/user5/.Sent/dovecot.index: keywords_count larger than header size dovecot.log.1.gz:Feb 24 17:01:00 elfstone dovecot: imap(user1@domain.tld): Corrupted transaction log file /export/data/dovecot/domain.tld/user1/.Sent/dovecot.index.log: Append with UID 1, but next_uid = 4850
These look almost like there's a problem with locking the indexes properly..
I have multi-domain setup with several hundred users, emails are delivered using dovecot-deliver LDA, dovecot indexes are outside mailboxes, indexes and mailboxes are on local filesystem (xfs). I'm using linux 2.6 on slackware platform.
Since the path contains /export, does it mean they're also exported to NFS?
I haven't seen any of these problems reported for a while, so I'd guess this has to be somehow specific to your system. One problem with Dovecot LDA is that it reads the configuration file by itself, so it might be possible that it reads for example the locking settings wrong.
Could you try running http://dovecot.org/tools/imaptest.c with some test account and see if it can make your Dovecot break? And if not, try sending mails to the test account with Dovecot LDA while running the imaptest?