TargetDown alert fired while it shouldn't.
Prometheus endpoints are not always properly unregistered and the alert will therefore think that some Kube service endpoints are down

The problem as always been there.

Not reproducible.
Most of the time Prometheus endpoints are properly unregistered.
Aim here is to get the TargetDown Prometheus expression be more resilient; this can be tested on past metrics data in which the unregistration issue was encountered.

N/A

TargetDown alert triggered while Kube service endpoints are all up & running.

TargetDown alert should not have been trigerred.