[Gluster-users] Database clustering in Gluster

[Keith Freedman]

At 05:37 AM 8/4/2008, Bobirkhon Ismailov wrote:
>Hello!
>
>Right now I'd like to try AFR feature of Gluster (ver. 1.3.10) by 
>using it with database files (Postgresql).
>
>In examples/afr.vol file the following feature is on:
>
>volume brick1
>   type features/posix-locks
>   option mandatory on          # enables mandatory locking on all files
>   subvolumes posix1
>end-volume
>
>My question is, how critical is locking for database clustering+AFR 
>and is there information about
>using/configuring Gluster for database clastering in general?

my guess is, if you're only going to be writing to the database at 
one location, locking probably isn't as important.   you should 
enable locks on that machine I'm guessing, but they may not need to 
be mandatory.

The devels can shed more light on the exact process, but I'm thinking 
you may run into some troubles with AFR for such large files.

It's not clear to me how the auto-heal works exactl.
IF you get disconnected, and one of the servers tries to auto-heal 
the database file,  Does it just move over the diffs, which is a fast 
operation, or does it copy over the entire file.  If it's the latter, 
this could become quite a problem.

If you're going to be writing the db files at multiple locations YOU 
MUST enable locking and it MUST be mandatory.

I've not personally experienced this, but I've heard of it working ok 
with mysql -- which had to be explicitly set to use external locks 
and it worked ok, but I think the ramifications of 
split-brain/auto-heal are pretty severe in a multi-write database environment.

those are my thoughts.


>Thank you very much,
>
>Bobir
>_______________________________________________
>Gluster-users mailing list
>Gluster-users at gluster.org
>http://zresearch.com/cgi-bin/mailman/listinfo/gluster-users