Consequently... Things work on a 40 node stateful cluster just fine. But you're absolutely fucked if you ever want to run something ephemerally, "restart" anything, or even just sensibly provide something as a service for other people to run.
Why in the absolute fuck did we write ourselves into a corner as an industry where "oh sorry minimum specs are 187 vCPU, 4TB of ram, 20TB of disks" seen as something sensible instead of absolutely ludicrous beyond belief?