Cyrus, clusters, GFS - HA yet again

[Paul Dekkers]

Hi,

Janne Peltonen wrote:
> People /are/ running Cyrus setups on different *nixen with different
> clustered filesystems. But those cases don't seem to be documented on
> the Web, either. That's why I'm asking here. ;)
>   

If I read Ken's (historic) post correctly then the memory-mapping on the
clustering filesystem is the biggest problem, right? Since the
replicated murder from 2.3 solves the issue for the meta-data/mailboxes
file.

Isn't there a way to (stress?) test this mmap-functionality? Apparently
it should fail on (traditional) NFS, should work with a couple of
clustering file-systems and -maybe- even NFSv4 (haven't seen any posts
on that after questions regarding NFSv4, but it would be interesting for
running this on for instance NetApp filters). IF a clustering-fs works
is then just a matter of testing...

Paul

P.S. Regarding tests: I have been playing (a while back) with the simple
imaptest-thing from dovecot, to stress Cyrus and replication, actually.
Interesting thing. I was able to break the replication a couple of times
(so I could test my restart-script for the sync_client ;-)) and I
noticed how expunge settings influenced this. I also noticed a
difference between the RPM-ified version of Cyrus and the version from
the port on FreeBSD (which is closer to the actual distribution, I
believe): the RPM-ified version popped up some errors (that did not
appear with the 2.2 version of the rpm) and the 2.3-baked version on
FreeBSD did not. It's still my intention to find some time to play with
this some more, and maybe spin a thread on these experiences.

Oh well, not really related to this clustering-filesystem-thing, but I
thought of it because of this testing. Interesting enough to share so
maybe others can play with it and share experiences.