Bad index files.

[Earl R Shannon]

Hello,

Sadly we rotate the log files every 10 days, and any logs during the
time frame for this have rotated out. And we don't know how to replicate
the problem short of stopping the server. Which we don't want to
do just to troubleshoot this. :)  And I can't check to see if both
the index files you ask about exist since we have fixed all the affected
accounts. :(

Yes, we are talking about a stop/start cycle. Which is one reason
I asked a little while ago about the "proper" way to shut down
the server. All we currently do is send a SIGKILL to the master
process. I'm planning to update our init script to put a shutdown
message in place and wait five minutes before actually sending
the signal in hopes that will help reduce the occurences. Really its
not a large percentage. At most close to 100 on machines with over
8000 accounts on them. However, having to do it at all is somewhat
disturbing. Particularly when one doesn't know why.

Regards,
Earl Shannon
--
Systems Programmer, Information Technology Division
North Carolina State University
http://www.earl.ncsu.edu

Lawrence Greenfield wrote:
>    Date: Wed, 3 Mar 2004 15:15:58 -0500 (EST)
>    From: Rob Siemborski <rjs3 at andrew.cmu.edu>
> 
>    On Wed, 3 Mar 2004, Earl R Shannon wrote:
> 
>    > The stop/start is when this seems to happen. Is it not possible that
>    > imapd processes are being SIGKILLed and leaving the index files
>    > in an untenable state?
> 
>    The "running" index files are never updated directly -- the .NEW files are
>    written out, and then moved (atomicly via rename()) into place.
> 
> Err, that's not true. The index files for some things (like \Answer flag
> updates) are modified in place, since rewriting/renaming the entire
> file would be too slow.
> 
> However, if we're talking a stop/start cycle (and not even a reboot) I
> don't understand how these files are getting corrupted.
> 
> The existance of .NEW files seem to indicate that an EXPUNGE is going
> on. Is there both a cyrus.index.NEW _and_ cyrus.cache.NEW? (If one
> gets renamed and the other doesn't, that yields a corrupt mailbox.)
> 
> Are there any syslog'd messages when the user attempts to SELECT the
> corrupt mailbox?
> 
> Larry
> 

---
Home Page: http://asg.web.cmu.edu/cyrus
Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html