Yes this is my default --- let it crash, and figure out why, and fix the problem. It also helps identify services that nobody really uses. If a service is down, and nobody complains, maybe it's not needed at all.

Sometimes there are external demands that require the system to always be up or to minimize any downtime. In an ideal world, that requirement should be designed into the system architecture from the beginning, but in the real world that doesn't always happen.