Public: anyone can read this; nobody can change it.

germinate: Grafana SSO on daughters (roll onto the sealed OAuth client; Admin = this platform's operator) #134

Closed opened this view on origin
Closed without merging

trevato hit "application error" signing into a fresh daughter's Grafana with Forgejo SSO. Two defects, both in germination: Placeholder client forever.

Notes

stopped closed without shipping

opened this view on origin

trevato hit "application error" signing into a fresh daughter's Grafana with Forgejo SSO. Two defects, both in germination:

  1. Placeholder client forever. grafana-oauth is sealed with pending-registration first; Flux starts Grafana; fork_register_grafana_oauth seals the real client id/secret later — but envFromSecret is read ONCE at pod creation and nothing rolled the Deployment. Forgejo answered the authorize with 400 Client ID not registered (that page is the "application error"). Fix: the same finalize-time roll germinate already does for mcp, now for Grafana (best-effort).
  2. Viewer for everyone. role_attribute_path named the PARENT's operator (trevato), so on a daughter every SSO login was a Viewer and nobody could hold the Admin session /flow-api's forwardAuth needs. Fix: re-point it at the daughter's own admin at seal time (parent's RBAC shape, right name).

Proven live on the test daughter (hot-fixed the same way): full OAuth round trip → /api/user 200 as plat, /api/user/orgs role Admin, /api/org/users 200.

Separate, pre-existing (parent AND daughters): monitoring/grafana-folder-acl CronJob fails every 2 min with ConnectionRefused to the Grafana Service — Viewer folder-scoping isn't being enforced. Own issue.

commented

Merged as part of #135 (same finalize region; the front-door PR carried this commit). Closing.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs