Problem with cyrreconstruct

Ciprian Marius Vizitiu cvizitiu at gbif.org
Thu Nov 6 08:13:23 EST 2008


> I am trying to migrate cyrus to a new server, but i got a big problem.
> 
> I made a backup of /var/lib/cyrus and /var/spool/cyrus. If i copy the 
> /var/lib/cyrus backup to the new server the "/usr/bin/cyrreconstruct -rf 
> user/*" does not works. It opens a process that uses 100% proc and does 
> nothing.  There is no useful info in syslog.

Unless you're moving between significantly *close* versions of Cyrus I 
don't think blindly copying /var/lib/cyrus would do. Which brings up the 
next problem question: what version(s) of Cyrus/OS are you using in the 
migration?

> If i do not use /var/lib/cyrus backup the "/usr/bin/cyrreconstruct -rf 
> user/*" command works fine. But i have some other problems like:
> 
> * No quota
> * All messages appears as not read.
> 
> My permissions are ok, and i am using debian 4.0.
> 
> Any suggestions?

Broadly speaking Cyrus uses some metadata for bookeeping. :-) Problem is 
there is a wide array of options for storing this metadata and the 
biggest culprit is typically Berkeley. This metadata includes at least 
the mailbox list, quota and the "seen" state. So, instead of blindly 
copying the metadata folder (most likely /var/lib/cyrus on Debian, 
/var/lib/imap on Red Hat) I suggest you investigate the format of the 
files and/or versions in order to see which will match what; e.g. a 
Berkeley 3.x will have problems with Berkeley 4.x and so on. The 
approach which worked for me was to export and import the metadata via 
plain text.



More information about the Info-cyrus mailing list