How does your server redundancy work?
Unfortunately Murmur (the back-end server for Mumble) doesn't support any sort of Master/Slave or multiple server setups. Because of this we had to design our own solution.
How It Works
We have two servers which maintain constant communication with each other. In the event that one of them goes down, the other knows immediately. Then it loads up the server, and takes over the IP address which you use to connect to the server.
What this means is that when a server crashes, you will get disconnected for a matter of 5-10 seconds while the new server takes over. Since the Mumble client auto-reconnects, you will get a disconnection notice in Mumble, and then should get connected automatically.
At the same time, we are notified that the switchover occurred, and we will investigate to ensure its not a major issue.
This also means that we can upgrade servers/hardware without causing significant downtime. We can simply upgrade the slave, swap the IP to the slave, and then upgrade the master. It makes no difference which server you are connected to, they are identical.