safeguard against pod eviction and OOM kill

[paulomach]

Guaranteed QoS Class requirements (docs):

• Every Container in the Pod must have a memory limit and a memory request.
• For every Container in the Pod, the memory limit must equal the memory request.
• Every Container in the Pod must have a CPU limit and a CPU request.
• For every Container in the Pod, the CPU limit must equal the CPU request.

BestEffort (current) is the first class to be evicted, followed by Burstable and Guaranteed

Method:

1. Patch statefulset at the charm initialization
   1. allow rolling update of pods before any initialization
   2. allow set of other desirable/unrelated statefulset values, e.g:
      1. terminationGracePeriodSeconds to 24hs
   3. Allow testing k8s cluster access ASAP (trust enabled or not)

Problem:

• setting request may prevent scheduling of pods due resource unavailability

Proposal:

• only set when both cpu and memory constraints are explicitly configured by the user on deployment, i.e. juju deploy mysql-k8s --constraints 'mem=4G cpu-power=2000'
  • cpu-power is set in millis
  • we have no knowledge of concurrent workloads or cluster topology
• patch QoS requirements on startup otherwise to match Burstable criteria (doc)
  • always patch request for initContainer is enough
  • still use method to patch some unrelated SS values / validate cluster access

Catches:

• juju will rewrite statefulset on charm refresh, which will require the patching
• Extra controls to ensure rollingupdate is not interpreted as and juju refresh

[github-actions]

https://warthogs.atlassian.net/browse/DPE-2918