Skip to content

Add resources and priorityClass to unprotected addon workloads#356

Open
nogueiraanderson wants to merge 1 commit into
mainfrom
fix/resilience-values
Open

Add resources and priorityClass to unprotected addon workloads#356
nogueiraanderson wants to merge 1 commit into
mainfrom
fix/resilience-values

Conversation

@nogueiraanderson

Copy link
Copy Markdown
Collaborator

Bug

  • Grafana and authentik ran BestEffort (no requests/limits) on the single obs-state node they share with the mtr postgres, making the most user-visible services the first OOM-kill victims
  • Five workloads set no priorityClassName and sat at priority 0, fully preemptible

Fix

  • Resources for grafana and authentik server/worker/postgresql (Burstable: cpu+memory requests, memory-only limits per house style)
  • Priority classes assigned: platform-default for cloudnative-pg, headlamp, jenkins-endpoint-reconciler and snapscheduler, platform-stateful-high for the mtr CNPG cluster
  • Every chart re-rendered and the new fields asserted in the parsed pod specs, not just a passing render

Tickets

  • Repo review finding (MED, reliability): BestEffort SSO/dashboards on the shared obs-state node

- grafana: add resources (requests cpu 250m/mem 512Mi, limit mem 1Gi) so the
  container is Burstable, not BestEffort, on the shared obs-state node
- authentik: add resources for server (250m/512Mi req, 1Gi mem limit), worker
  (100m/256Mi, 512Mi), and the bundled postgresql (100m/256Mi, 512Mi, overrides
  the chart nano preset)
- authentik: no redis resources, the pinned goauthentik 2026.2.3 chart bundles
  no redis subchart (the existing redis block was already inert)
- cloudnative-pg, headlamp, snapscheduler: set priorityClassName platform-default
  (were unset, landing at implicit priority 0)
- jenkins-endpoint-reconciler: set priorityClassName platform-default on the
  CronJob pod template
- mtr: set priorityClassName platform-stateful-high on the CNPG Cluster spec (the
  stateful test-history postgres)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant