Complete identity smoke path: id_token claims, registration entry, cutover docs

Prefer verified KeyCape id_token claims when /userinfo returns 401; soft-fail
userinfo. Add CSOC-WP-0003 registration entry (disabled until NetKingdom URL),
AAL step-up hooks, smoke/cutover evidence for tegwick OIDC without MFA.
This commit is contained in:
tegwick 2026-08-09 22:42:51 +02:00
parent 3bc16b581b
commit 29a9ff735e
14 changed files with 513 additions and 41 deletions

View file

@ -1,11 +1,30 @@
# Deploy notes (stub — CSOC-WP-0002-T08)
# Deploy notes
## Shape
Standalone service: commit-SHA images → registry
`gitea.coulomb.social/coulomb/coulomb-social` → `railiance-apps` values →
`forgejo.coulomb.social/coulomb/coulomb-social` → `railiance-apps` Helm values →
railiance01 (same lane as `vergabe-teilnahme`).
Chart/values/ingress live in **`railiance-apps`**
(`helm/coulomb-social-values.yaml`, `docs/coulomb-social.md`).
## Current cluster status (2026-08-09)
| Item | State |
|------|--------|
| Namespace | `coulomb-social` Active |
| Deployment | 1/1 Ready, image `:7067145` |
| Service | ClusterIP :80 |
| Ingress | `coulomb.social` → Traefik, cert-manager annotation |
| Env secret | `coulomb-social-env` (`SECRET_KEY`, `DATABASE_URL`, `USER_ENGINE_PROXY_SECRET`) |
| OIDC | enabled; issuer `https://kc.coulomb.social`; public client |
| Public DNS | **still Cloudflare / Bubble** |
| TLS secret | **pending** HTTP-01 until DNS points at the cluster |
In-cluster smoke (with `Host: coulomb.social`): `/healthz` ok, landing 200,
`/auth/login/` → KeyCape authorize. Full browser session needs cutover.
## Runtime secrets (names only)
K8s Secret `coulomb-social-env` in namespace `coulomb-social` (chart `envFrom`):
@ -26,24 +45,82 @@ make coulomb-social-env-secret
./scripts/create-env-secret.sh
```
Script: `railiance-apps/tools/create-coulomb-social-env-secret.sh`
OIDC is a **public** client — no client secret.
Non-secret env (OIDC, ALLOWED_HOSTS, user-engine URL) lives in Helm values.
## Health
- `GET /healthz``{"status":"ok"}`
- `GET /healthz``{"status":"ok","service":"coulomb-social"}`
- Probes use `Host: coulomb.social` (`probes.hostHeader`)
## Build
## Build / deploy
```bash
SHA=$(git rev-parse --short HEAD)
docker build -t forgejo.coulomb.social/coulomb/coulomb-social:$SHA .
# push, then:
# COULOMB_SOCIAL_IMAGE_TAG=$SHA make coulomb-social-deploy # in railiance-apps
```
Runtime env (no secrets in image): `SECRET_KEY`, `DATABASE_URL`,
`OIDC_*`, `USER_ENGINE_*`, `DEFAULT_TENANT_ID`, `ALLOWED_HOSTS`.
## Cutover checklist (DNS → live Railiance)
## Status
**Goal:** `https://coulomb.social` serves this app (identity shell), not Bubble.
- Dockerfile present (gunicorn, non-root, `/healthz` check).
- railiance-apps Helm values / cluster Service **not** yet landed.
- Local `make run` + `make test` remain the default verification path.
### Preconditions
1. [x] Image + Helm release healthy
2. [x] KeyCape client `coulomb-social` with prod redirect `https://coulomb.social/auth/callback/`
3. [x] In-cluster OIDC start redirect works
4. [ ] Local browser OIDC + MFA completed once (proves IdP + user-engine path)
5. [ ] Operator accepts brief public outage / Bubble freeze during DNS switch
6. [ ] Optional: export Bubble data if still needed (CSOC-WP-0001) — not required for identity-only cutover
### DNS switch
1. In Cloudflare (or DNS host): lower TTL on `coulomb.social` if possible (e.g. 300s) ahead of time.
2. Point apex (and `www` if used) **A** to **`92.205.62.239`** (railiance01 ingress).
- Prefer DNS-only (grey cloud) first so LE HTTP-01 and Traefik see real traffic; re-enable proxy only if you understand TLS termination path.
3. Wait for propagation: `dig +short coulomb.social A``92.205.62.239`.
4. cert-manager should finish HTTP-01; confirm:
```bash
kubectl -n coulomb-social get certificate coulomb-social-tls
# READY=True
```
5. Smoke public HTTPS:
```bash
curl -fsS https://coulomb.social/healthz
curl -sI https://coulomb.social/auth/login/ | grep -i location
# Location: https://kc.coulomb.social/authorize?...
```
6. **Browser:** Sign in → Authelia MFA → land on `/app/` with principal.
7. Sign out; confirm `/app/` requires login.
8. Second login: same member row / user-engine user_id.
### Rollback
- Point DNS A (or Cloudflare origin) back to Bubble/Cloudflare target.
- Cluster release can stay; it only receives traffic when DNS aims at the node.
### After cutover residuals
| Item | Note |
|------|------|
| Bubble freeze | Stop editing live Bubble as source of truth |
| Content/UI | CSOC-WP-0001 + design extract — not required for identity shell |
| flex-auth Service | leave `FLEX_AUTH_BASE_URL` unset (local vocabulary) until PDP exists |
| apps-pg backup/HA | business-app contract |
| OpenBao CCR | replace kubectl-sourced env secret when ready |
| Image CI | Forgejo/Gitea pipeline for SHA tags |
## Local verification (no cutover)
```bash
make test
make run # offline identity
# or OIDC vars from docs/identity/oidc-client.md
```