Any tips on managing rollback in an ejabberd cluster?

Question

Context:
I have a system that will have to support 20,000 connected chat users spreading over more than 100 chats. During performance testing, I found that I could get up to 6,000 connected users on one box before I get a crash dump, so in production I will probably go with four servers in the cluster.

My question is:
I understand that the chat is connected to the node server, so if the node dies, the chat disappears with it and the users no longer belong to the room. Is there a way to "replicate" the chat to another node so that the remaining users are moved to the replicated room? If not, what do you do to maintain continuity for users?

Answer 1

What equipment do you use? 6,000 connected users seem a bit low. In addition, ejabberd should not fall under load. This may slow down, but not crash.

Something is wrong with your installation.

About node chat replication, it's not easy. Better handle smooth client-side reconnection.

But then again, ejabberd should not fall under such a load, unless something is wrong.