Just some odd comments

mfoss — Fri, 20 May 2005 22:28:48 +0000

Warning: some of those comments could be erroneous.

The communication between ejabberd nodes in a cluster is done by an unified database, either with replicated or distributed tables. Additional services like MUC does not have any kind of distribution since some of the information they use is not stored in a table.

A user registered in the Jabber server can login to his account on any of the cluster nodes. He specifies the username, server name, password and additionally the server to connect to (psi, tkabber, exodus allow this).

I guess that didn't help in any way :( I experienced with ejabberd clustering last summer and I was able to login on any node. I forgot most of that, and I will not have spare time until this summer.

cluster related, I'm assuming

bbrazell — Wed, 11 May 2005 21:34:26 +0000

I've been battling with clustering ejabberd for about 2-3 weeks (part time) and am still at a loss for understanding how it's supposed to work. Perhaps I'm a retard. I hope that someone (perhaps even Sir Badlop himself) can assist here.

I thought I was getting close when I was able to register accounts with the pooled hostname. I've read that others have used DNS round-robin to achieve load balancing across two or more nodes. Like I said, that was looking good until it came to service discovery. Another article said something about giving the nodes the same name. I can't imagine how that would work.

Right now, I've settled with having server1 be the server that users will register on, and server2 is doing only database replication. Sure, it's an interconnected node, but if server1 goes down, the whole subsystem is down. Is the clustering offered by ejabberd supposed to provide fault tolerance? If so, how?

ejabberd - Comments for "vjud not available"

Just some odd comments

cluster related, I'm assuming