I'm looking at this now, and have read the forum discussions, and I haven't really seen anything convincing regarding clustered moodledata.
Is everyone who has a load-balanced/clustered Moodle setup on Linux using NFS? How do you deal with the single point of failure?
At present we have one applicaton server, a "management node" that runs the cron job, and a database server. As part of the cron job, the management node rsyncs moodledata from the app server to its own copy, then pushes back the changes (e.g. RSS feeds, course backups) to the app server.
We have reached the point where the app server is the bottleneck.
I've looked at presenting an OCFS2 clustered filesystem to multiple app servers, but it hasn't proven reliable.
Has anyone looked at Lustre, or GPFS, or GFS2? Or found a way to remove the SPOF from an NFS setup?
Not sure if this is useful to you, but as I understand it our setup uses two standard Linux NFS servers (each of which are connected directly to the institutional SAN).
In this configuration the NFS server remains a single point of failure as only one is used at a time - but there are a pair in fail-over configuration. Not sure if this is automatic or manual, but even if it's manual, it means we shouldn't be down for long (er for that reason).
Thanks, this is useful. A fellow sysadmin in another department recently said "every time I think about Moodle, I wonder what the OU does". So now I can tell him.
In your NFS configuration, do both servers point to the same filestore, i.e. LUN or similar? Or are they separate and rsynced?