There are many pods (replicas) in k8s cluster that can restart, scaled up or down.
Each pod exposes an error counter that Prometheus scraps.
I want some how to
