Hey all,

Date: Tue, 31 Jan 2017 12:56:41 +0200 From: Aki Tuomi <aki.tuomi@dovecot.fi> To: dovecot@dovecot.org Subject: Re: dovecot mdbox never fix broken indexes Message-ID: <e6881175-8a48-ee24-6681-14427256d3a1@dovecot.fi> Content-Type: text/plain; charset=utf-8

...

...

Also, if the file has indeed corrupted due to power loss, it's possible it's beyond repair, and you will have to delete the index file. This has the down side of losing flags, but it should repair this. Why dovecot not repair it by itself?

It really should, as it's saying... Have to see why it's not successful in this.

Aki

is there anything new known about this?

We are running dovecot 2.2.27 on solaris 10 and have several hundred accounts with this map inconsistency: Apr 27 12:51:06 <servername> dovecot: [ID 583609 local0.warning] lda(<userid>): Warning: mdbox /<storage-path>: Inconsistency in map index (10,29152 != 10,63560)

The users of these accounts can't delete or move mails and often have trouble logging in or lags, because dovecot will try to fix the error and possibly disconnect them at any action. The warning does not occur on every action (imap/pop3 or lda).

We did not have a hard reboot or any hardware trouble out of the ordinary (the machine is old and intermittently under higher load).

After a variable length of time with these warnings the user accounts develop the following error: Apr 26 22:50:09 <servername> dovecot: [ID 583609 local0.error] imap(<userid>@): Error: Log synchronization error at seq=28,offset=117000 for /<storage-path>/dovecot.map.index: Extension record inc drops number below zero (uid=<uid>, diff=-1, orig=0)

force-resync did not help at any stage of the problem and only got the index to continually count up. The only solution we found through which we could save the users flags is to copy the mdbox-files, delete the originals, and then re-import the copy into the user's account via doveadm import.

Best regards, Catherina Koehl