Comment 7 for bug 876663

I started digging into this a bit.

We seem to handle this reasonably gracefully once we're up and running. The attempted operation fails, but the service keeps running. However, if the connection fails during initialisation, we're in much more trouble, because the unhandled exception happens in the main thread which then dies.
If we see this error later on, it's (almost?) exclusively while handling incoming requests, which is done in separate threads.

There are a couple of different ways to attack this. I'll see what I can come up with.