Geographically distributed Confluence (Cluster): Is it possible?

[aseith]

Greetings to all sysadmins and operators out there,

the subject says it all. We are currently looking into possibilities to run multiple Confluence instances on geographically distributed locations, either with the offical Confluence Cluster edition or any other (more or less hacked) solution.

We are aware that Atlassian doesn't support any form of this clustering, but would like to know if anyone has done (or at least tried) to do this before and would be grateful for any experiences shared.

Kind Regards

Alex

[Andrew Frayling · Accepted solution]

Hi,

We tried, and failed, and that was with Confluence nodes in data centres just 1 mile apart. Support's suggested solution at the time was to host the servers in the same rack and connect them with a crossover cable. All the multicast tests were fine and reporting that the nodes could see each other, but we just couldn't get it to work reliably.

The problem is that the inderlying clustering technology (Oracle Coherence) seems to be incredibly flakey and the slightest blip means a node doesn't get updated with the correct ID in time and the nodes cluster panic. This is what we were seeing, and we resorted to running in active/passive with just one node receiving requests and the other in cold standby so that we could failover if there was a problem or if we needed to carry out planned maintenance.

Active / passive seems to be the multi-node option that most people seem to go for due to the reliability issues with clustering. This configuration will give you higher availability than a single node and you can vertically scale (to a point) by throwing more resources (RAM, CPU, faster disk, etc.) at the individual nodes.

Hope that's of some help?

Andrew.

[Chloe Sowers]

Yes we go

[Chloe Sowers]

Yes we have a cluster running in difference nearby cities. It is extremely slow however, and although we are not certain of the cause, I believe it may be due to the cluster. You may have to add the option

-Dconfluence.panic.disable=true

to the Java startup.

[Chloe Sowers]

Yes, it was because of the cluster. It crashed (locked up) often, with Coherence errors.

[Alexander Seith]

Thanks for your answers, Andrew and Beth. We also found that the most viable solution for HA-Clustering is an active/passive Cluster. We're already running this kind of setup under Linux with Pacemaker a couple of times.

Thanks again.

Alex

[Beth Schaefermann]

This is potentially somewhat related to the JIRA question/ conversation here:

https://answers.atlassian.com/questions/15984/can-we-run-jira-on-multiple-application-servers?page=1#42616