Timo Sirainen wrote:
On 14.1.2013, at 17.16, Daniel Parthey daniel.parthey@informatik.tu-chemnitz.de wrote:
Last week we upgraded to 2.1.13 and since then we encountered the same issues as Morten Stevens. We got a lot of error messages with corrupt index cache and FETCH errors: So we had to roll back to dovecot 2.1.10. Now the errors are gone.
Jan 12 16:48:35 10.129.3.233 dovecot: mailbox: mail: imap(user1@example.org): <TaljVxnT4gAKgQPp>: Error: Cached message size larger than expected (5194 > 4399)
Does reverting this change also fix it for you? http://hg.dovecot.org/dovecot-2.1/rev/6c1b4b9f527c
Thanks, did not find this hint on the mailinglist yet. Will try (sooner or later) and keep you informed.
I guess you're using only LMTP to deliver mails?
Yes, LMTP via director port 20024 only.
How many CPU cores do you have?
4 physical machines with Dual Core.
Also how much is "a lot", especially compared to the number of users you have / emails you get? I think it didn't happen very often for Morten (something like 1 error per several tens of thousands of mails?)
I must admit, "a lot" was way too exaggerated ;)
Around 10000 successful LMTP deliveries a day. Around 1 to 3 errors a day.
But data corruption is something we can neither accept, nor explain to our customers, since indexes are very important with mdbox.