Find initial criteria for resource pruning about babylon

[sechmann]

My thoughts:

ReplicaSets where all pods have one of the the following states are considered safe to scale down:

• CreateContainerConfigError
• ImagePullBackOff
• ErrImagePull
• Init:CrashLoopBackOff

Pods that have a high (thousands) number of restarts and are in CrashLoopBackoff or Error should notify the owning team, with a possibility to mute the alert

[sondr3]

@sechmann Would be valid to also turn off pods that have been in a ContainerCreating for a very long time? We have a test deploy that never manages to mount a volume, so it just spins waiting for mounting forever.

[sechmann]

No, because the volume might become available at a later time, and the pods would start up (without any changes to the deployment).
The way i see it a container "stuck" in ContainerCreating waiting for a volume to become available is normal given the eventual consistency nature of k8s.