Here's the (short) silly cut I made of it last summer: https://www.youtube.com/watch?v=UuSZ37vMIks
Here's another (longer) remix: https://www.youtube.com/watch?v=rRbY3TMUcgQ
Not Erlang but same concept and I'm not sure what the software development environment was, but the thing was loaded to the gills and almost all ops save major software upgrade only required warm restarts which preserved existing connections (perhaps with some delays of digit processing).
Two major problems I remember in those days was a bug in operator position handling where a certain operation created some sort of race condition that brought it all the way down. That was fixed but the other was not so easy... mainly because it was doing triple-duty as class 3/4/5, what functioned as a stack/event queue occupied an insufficient pool of memory to handle extraordinary events. And in that early state of development overflow of the event queue was a fatal cold start error ... from a cold start it took ~15 minutes to come up again.
Murphy's Law supplied TWO great examples to illustrate the problem. One was a severe cable cut, an auger down the street from the telco wrapped itself around ~3,000 pairs and triggered many events. The next was an earthquake... and everyone picked up their phones at once.