[Dovecot] UID inserted in the middle of mailbox errors
Hi,
We are still getting, on a daily basis, users who cannot move messages to Trash or expunge Trash due to these errors:
Mar 28 10:43:57 tm2 dovecot: [ID 107833 mail.error] IMAP(kss021): mbox sync: UID inserted in the middle of mailbox /home/kss021/Mail/Trash (3137 > 536, seq=2, idx_msgs=76)
Our users have no other access to their mailbox other than Dovecot, and we are using mbox format and 1.1rc3.
Deleting the indexes fixes the problem, so it seems to be a problem with index corruption and not something inherently wrong with the mailbox. That is, once I delete the index files, the problem does not come back immediately.
I've seen others on the list with the same problem, but I wasn't able to find a workaround. It's happening more and more frequently.
I guess what I don't understand is why Dovecot cant just mark the index invalid and purge it instead of failing over and over. I thought the whole point behind Dovecot's indexing is that it was able to detect index inconsistencies and mark the index invalid.
Any help is much appreciated. Even a quick stopgap fix would be a lifesaver, short of a cronjob that just purges everyone's index files every 2 hours (which sort of defeats the point of using Dovecot).
-- Dean Brooks dean@iglou.com
On Fri, 28 Mar 2008 10:53:49 -0400 Dean Brooks wrote:
With maildir format too
Mar 27 10:19:26 mailbox dovecot: imap(exim4000): Maildir /var/Mail/Boxes/e/exim4000/.Trash sync: UID inserted in the middle of mailbox (105265 > 105264, file = 1206604287.H478110P28054.mailbox.tu-berlin.de,S=6185:2,b)
On Fri, Mar 28, 2008 at 10:53:49AM -0400, Dean Brooks wrote:
For what it's worth, I set "mbox_lazy_writes = no" in dovecot.conf and the problem hasn't happened all weekend. Previously, it was happening hourly.
It seems that there may be some lingering bugs in the lazy_write logic. If anyone is running into this problem on a regular basis, I would try disabling lazy writes and see if that helps.
I'm not sure why everyone isn't having this problem given how such a stock vanilla configuration we're using, but maybe it's just related to server volume.
-- Dean Brooks dean@iglou.com
On Fri, 2008-03-28 at 10:53 -0400, Dean Brooks wrote:
It wasn't actually failing at all from user's point of view. It just kept logging the same error over and over again because it always delayed writing fixes to the mbox file.
I changed the code to now always write pending changes to mbox if syncing had errors. This still causes errors to be written twice, but that can't be fixed easily. http://hg.dovecot.org/dovecot-1.1/rev/591145eaaf87
As for why these happen in the first place when copying, I've really no idea.. I tried to reproduce this with imaptest, but couldn't. Does imaptest show any errors with you? http://imapwiki.org/ImapTest
participants (3)
-
Dean Brooks
-
Frank Elsner
-
Timo Sirainen