Upgrading cyrus

[Jeroen van Meeuwen (Kolab Systems)]

Patrick Goetz wrote:
> On 09/14/2010 04:16 PM, Jeroen van Meeuwen (Kolab Systems) wrote:
> > Indeed it [cyrus-imapd.cvt_cyrusdb_all] exports the databases on service
> > shutdown, and then imports them back on service startup; noted that a
> > condrestart happens on the RPM update, this about covers every thinkable
> > scenario (or at least those I've had in the past 7 years).
> 
> This seems like it would considerably slow down startup/shutdown on
> systems with large databases....
> 

A spool with 179GB (du -sh, ext3 4k), takes 30 seconds to export and 12 
seconds to import, noted that this is not the only thing the machine is doing. 
No raid whatsoever.

Yes, there is trade-off. There's also trade-up. Anyone may take their pick. 
Maybe this export/import thing is something we can stuff into 
/etc/default/cyrus-imapd or /etc/sysconfig/cyrus-imapd?

I'm not sure whether "large databases" having to reboot a partition server can 
ever not justify a hmm-times-42 second delay if they choose to transition 
database formats or upgrade -mostly because I'm almost certain it takes longer 
to check memory / initialize/load/check the BIOS, spin up disks and initialize 
RAID on such servers. Also, these type of servers shouldn't have to be updated 
that frequently, and are commonly located with organizations that have proper 
definitions of a service window -usually longer then hmm-times-42 seconds.

> > If there is no information in /etc/imapd.conf explicitly, the defaults
> > are being used -just like the service itself would. The defaults for RPM
> > (for the purpose of the script) is to use skiplist for all databases.
> 
> This is another question that I have:  is the set/type of database files
> determined at run time (i.e. in imapd.conf) or at compile time? 

At compile time, you set the defaults (from lib/imapoptions, which you may or 
may not have patched), defaults which can thus also be used in a script.

The script can thus also use these defaults, because any other format then the 
default is in /etc/imapd.conf.

> In
> particular, the set of databases I have on 2.1.16 only has a very small
> intersection with the set described in the forwarded email below:

I'll respond to the thread directly on the list.

Kind regards,

-- 
Jeroen van Meeuwen
Senior Engineer, Kolab Systems AG

e: vanmeeuwen at kolabsys.com
t: +316 42 801 403
w: http://www.kolabsys.com

pgp: 9342 BF08