This blog post points out that automatically restarting processes can hide problems, which is certainly true. Doesn’t have to be systemd. Something like supervisor trees in #Erlang/#OTP might do the same.
If you’re restarting something, measure restarts and plot them on a graph. If it’s happening, understand why it’s happening. If it’s designed to fail, fail it on purpose at a regular cadence to make sure that failure is being compensated for correctly.
