The flipside of uptime
Long uptime is not a virtue metric. Reboot on your schedule, or discover broken boot paths during an outage.
A change that “worked” for weeks can still leave a host that will not boot cleanly. Monitoring stays green until the next reboot — then the latent break surfaces at the worst time.
Pride metrics
Multi-year uptime used to signal careful ops. Security patches (and kernels that need a reboot) invert that: time since last controlled restart often matters more than the vanity counter.
Controlled restarts
Reboot on a cadence you choose. The interval depends on the system; the point does not: eventually something will restart the box. Prefer finding broken init, mounts, and service order while you are watching.
That implies either scheduled maintenance windows or an architecture where any single host can leave the pool without interrupting service. “Enterprise” labels do not grant that property by themselves.