You are not wrong.
We restart CAS once a week. We are currently running 3.6.x, it has a memory
leak with load over time. We do 400k logins per day. (CPU goes to 100%
doing nothing but garbage collection)
The weekly restart mitigates / solves (?) this issue,
But if in the interval we messed up
Yes, but one of the nice things about CAS was that we could drop new
service definitions into a running instance and have them picked up
automatically. Anything that requires us to actually restart our service
get us involved in a formal change management process and restricts us to
weekend early