RollingUpdate maxSurge cannot schedule a second 50m pod on a node
packed to 99% CPU requests. Recreate replaces in place; the Service
already has no ready endpoints.
Image sha256:ec15f63d… is commit bc3d80f. Local release check passed
against disposable Postgres. No schema migration. Live Ready still
depends on a fresh runtime lease from ESO/OpenBao.
Assistant: grok
Assistant-Session: 01a0a182-bab7-7f11-b32b-d06f3af52082
