08 January 2009, 1:43

Hi,

I wonder if anyone can provide an aswer to the following... if we run Mahara in a cluster and one of the servers crashed or had to be taken out, will it automatically relocate users the remaining working servers? Would the users then be required to re-authenticate against the server? Or will all this happen transparently?

Cheers Andrew

08 January 2009, 2:30

That depends entirely on how you set up your cluster, but I don't think you could do anything other than make it work, to be honest. Mainly because you're going to have to have dataroot on a server that all the webservers can talk to, and sessions are in the dataroot.

08 January 2009, 19:18

Thanks Nigel.

Regards... Andrew

16 February 2009, 1:53

Hi we have same questions regarding to the multiple front ends of Mahara.

We have implemented multiple front ends with the moodle.

Is anybody there who implemented same sort of thing with Mahara ? And uses one database ?

16 February 2009, 4:49

Hi Yaju,

please describe what you mean with "multiple front ends", you have done with moodle.

Sounds very interesting.

Greeetings from the black forest

Heinz

16 February 2009, 6:06

Heinz - I think he means running Mahara across more than one server - e.g. having one web server and a separate database server.

16 February 2009, 6:05

Hi - yes, mahara.org itself is running on multiple webservers with a database server behind that. Mahara is designed to work with this model of cluster, and it does so just fine. The devil is in the details, as with all of these things - how you set up your cluster will have a big impact on whether it works for Mahara or not.

But, if you have set up a Moodle cluster, the same principles apply. You should be able to roll out Mahara on the same servers as your Moodle without any fuss.

16 February 2009, 18:28

Hi,  Yes that is right.  The University is looking at multiple webservers with a database behind them on a separate system.  We are also looking at whether we can "cluster" the database server for more HA.

Howvever, the primary concern at the moment is the configuration with multiple webservers.  I believe we will have load balancers out in front of the application layer.  I guess we are after any pointers on what not to do and any best practice or procedures for configuring multiple webservers.  Any advice would be greatly appreciated.

Please note, yes we know about sharing a common dataroot between the webservers so as to retain session info etc.  I also guess that Mahara must play some part in any locking required for sharing of data in this common area.

An item of interest was how the apache server configuration was done on each webserver.  Rather, what were the differences please?

 

16 February 2009, 20:34

Well all the webservers should be nearly identical, except hostname. This includes apache conf - it can get very confusing when one server is configured to do something slightly different, and then once every two or three requests something different happens!

If you're doing layer 4 load balancing, e.g. with ipvs, you don't need to worry about server affinity for requests. Sessions in dataroot handles this fine.

Mahara doesn't handle locking for dataroot - NFS can handle this by itself. Configuring NFS to perform can be a bit of a pain, but is possible.

17 February 2009, 1:28

OK, yes we are in the process of configuring the load balancers right now.  The Network Team tell me that we will be using layer 4 balancing.  I think we will be using GFS instead of NFS for sharing the dataroot  Don't believe there are additonal issues over NFS at this time.

Our mahara application layer servers have names like mahapp[1|n].  The outside, as seen by customers, would be eportfolio.us.edu for example.    The question then is the Apache config file must surely ensure that entries put into the database look like "eportfolio.us.edu" even though the system is actually called makapp1.  Is it possible to obtain extract sof relevant entries  in a couple examples of httpd.conf so I can work out our changes please.