nan-vm-operator dev deploy never takes effect: CI bumps a values file the dev ArgoCD app does not read #205

Finding (2026-10-20)

The nan-vm-operator deploy workflow (deploy-dev job) commits the new dev-<sha> tag to helm/nan-vm-operator/values-dev.yaml in nan-devops.

The ArgoCD app prod-nan-vm-operator-dev renders with kubernetes/apps/prod/default/nan-system-dev/nan-vm-operator-dev/values.yaml, which pins:

image:
  tag: dev-42ed1632782093d657d20021af8557335090e3dc

The comment next to that pin says dev CI rewrites it, but nothing does. As a result, the dev operator still runs the image from nan-vm-operator#42, while prod runs v0.0.53-alpha (lease + leader-epoch fencing). Every dev deploy since then has been a no-op.

The app is also sync-manual. On 2026-10-20 it was synced, but only labels changed and the image stayed at dev-42ed163.

Fix options

1. Point the deploy-dev commit-back at kubernetes/apps/prod/default/nan-system-dev/nan-vm-operator-dev/values.yaml (image.tag).
2. Or make the dev app also read helm/nan-vm-operator/values-dev.yaml and drop the pin.

After the fix, bump dev and sync it: argocd --core app diff/sync prod-nan-vm-operator-dev. Then verify the pod image and that operator_leader_epoch in nan_platform_dev gets an epoch, since migration 124 is already applied there.

Context: the 2026-10-19 operator incident, postmortem in nan-devops#909.