Check whether replicas are all on the same node

[marians]

Is your feature request related to a problem? Please describe.

One of the common use cases for running a container orchestrator is the possibility to increase the availability of workloads by spreading replicas over several nodes. However, by default, the Kubernetes scheduler does not take distribution of risk into account. So it can happen that all replicas of a Deployment or StatefulSet run on the same node. A node failure in this case would lead to complete unavailability of the workload/service.

Users can achieve distribution across nodes by setting appropriate anti-affinity settings for pods

Describe the solution you'd like

Popeye should issue a warning if there is a StatefulSet or Deployment with all replicas on the same node. The warning text should mention configuring anti-affinity as a way to mitigate. Example:

All pods on the same node. Consider setting anti-affinity.

Describe alternatives you've considered

The solution above looks at the symptom and may or may not catch the problem, depending on the distribution of pods in the moment of checking.

Parsing the pods's .spec.affinity and trying to deduct whether it would lead to distribution over nodes would solve this in theory. However, that appears way too complex.

Additional context

n/a

[marians]

PR #149 attempts to close this issue and is ready for review.