Mailbox list corruption

Andrew Morgan morgan at orst.edu
Wed Jun 27 13:15:08 EDT 2007


On Wed, 27 Jun 2007, Janne Peltonen wrote:

> On Tue, Jun 26, 2007 at 08:17:15PM +0300, Janne Peltonen wrote:
>> Hi!
>>
>> Sometimes I get the following while trying to restart a murder member:
>>
>> Jun 26 16:12:42 pcn4.mappi.helsinki.fi i10/master[21932]: exiting on SIGTERM/SIGINT
>> Jun 26 16:24:28 pcn4.mappi.helsinki.fi i10/master[19054]: process started
>> Jun 26 16:24:28 pcn4.mappi.helsinki.fi i10/ctl_cyrusdb[19056]: recovering cyrus databases
>> Jun 26 16:24:39 pcn4.mappi.helsinki.fi i10/ctl_cyrusdb[19056]: DBERROR: skiplist recovery /var/lib/imap/i
>> 10/mailboxes.db: 5BE0D50 should be ADD or DELETE
>> Jun 26 16:24:39 pcn4.mappi.helsinki.fi i10/ctl_cyrusdb[19056]: DBERROR: opening /var/lib/imap/i10/mailbox
>> es.db: cyrusdb error
>
> On Tue, Jun 26, 2007 at 10:10:11PM +0300, Janne Peltonen wrote:
>> Hi.
>>
>> I've got a line like the following in my cyrus.conf files:
>>
>>   mupdatepush   cmd="/usr/lib/cyrus-imapd/ctl_mboxlist -C /etc/imapd.conf.i01.master -m"
>>
>> It is meant to synchronize the backend mailbox list with the mupdate
>> master upon startup. Sometimes, it takes horribly long to complete: one
>> to three hours with 600000 records in the mailboxes list. (Sometimes it
>> takes a lot less time, something like 9 seconds).
>>
>> There are 24 nodes in my murder. Could this create locking problems on
>> the mupdate master?
>
> Apparently, these were both the result of ill-considered mount options
> on the config partitions. *sigh* There was a colleague that recommended
> setting the 'noatime, data=writeback' options on all cyrus filesystems
> during migration from our old system to the new one. Oh, well...

I'm running Cyrus 2.2.13 in a murder configuration.  My entire Cyrus 
directory (spool and config) in on an ext3 partition with the following 
mount options:

/dev/sdb1 on /private type ext3 (rw,noatime,data=ordered)

You shouldn't have any trouble with the "noatime" option (it gives a good 
performance boost).  data=writeback gives higher performance, but may 
allow corruption of files if you experience a system crash.  If you didn't 
have a system crash or unclean umount of the filesystem, then I am at a 
loss for why you would experience corruption.

 	Andy


More information about the Info-cyrus mailing list