RnD/infra

Files

znetsixe af354a4b9e feat(cloud): short function-based subdomains + harden keycloak with postgres

Subdomain rename (Versio side keeps original tool-named hostnames)
- nginx vhosts updated:
    grafana   -> dash.wbd-rd.nl
    gitea     -> git.wbd-rd.nl
    keycloak  -> auth.wbd-rd.nl
    node-red  -> flow.wbd-rd.nl
    mlflow    -> ml.wbd-rd.nl
    jupyter   -> hub.wbd-rd.nl
    portainer -> ops.wbd-rd.nl
    rabbitmq  -> mq.wbd-rd.nl
    jenkins   -> ci.wbd-rd.nl
    mqtt      -> mqtt.wbd-rd.nl   (no Versio conflict assumed)
- nginx-proxy README: bootstrap cert -d list + DNS A-record prereqs updated
- cloud/.env.example: GITEA_ROOT_URL, GRAFANA_ROOT_URL, KEYCLOAK_HOSTNAME

Function-based names are tool-agnostic (a Grafana -> Kibana swap leaves
dash.wbd-rd.nl meaningful) and avoid one-off "*2" suffixes.

Keycloak hardening
- Switch backend from bundled file storage to postgres (keycloak DB
  already provisioned by sql/config/init.d/01-databases.sh).
- KC_HOSTNAME=auth.wbd-rd.nl, KC_PROXY_HEADERS=xforwarded for nginx
  reverse-proxy posture; KC_HTTP_ENABLED=true since nginx terminates TLS.
- Added KC_HOSTNAME_STRICT, KC_HEALTH_ENABLED, KC_METRICS_ENABLED.
- Service joins app + mgmt + data networks (data needed for postgres).
- Mounted config/realms/ for realm-as-code (kc.sh import) — TODO to
  populate once realm + clients are designed.
- README documents the recommended realm structure (wbd realm, one
  client per app with redirect URIs) and the oauth2-proxy approach
  for apps without native OIDC (mlflow, portainer-CE).

cloud
- Uncomment keycloak include in cloud/compose.yml.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-21 13:54:57 +02:00

4.5 KiB

Raw Blame History

nginx-proxy

The single web ingress for cloud + edge. Reverse-proxies HTTPS UIs and stream-proxies MQTT-TLS to RabbitMQ. TLS certs managed by a certbot sidecar (Let's Encrypt).

Image: stock nginx:1.27-alpine (we don't use nginxproxy/nginx-proxy because we need the stream {} context for MQTT-TLS)
Sidecar: certbot/certbot:latest — renews every 12h via HTTP-01 webroot challenges
Networks: edge (the only port-publisher) + app (talks to upstream services)
Host ports: tcp/80, tcp/443, tcp/8883

Cert strategy

Interim (Versio DNS): HTTP-01 SAN cert covering all subdomains, issued via --webroot. Requires:

Public DNS A records for each subdomain pointing at the cloud host
tcp/80 reachable from the internet

After TransIP migration: switch to DNS-01 wildcard (*.wbd-rd.nl). Swap the certbot/certbot image for a build that includes certbot-dns-transip and reissue with --cert-name infra so the cert path stays stable — no vhost config changes needed.

Config layout

config/
├── nginx.conf                    # base — http + stream contexts
├── conf.d/
│   ├── 00-default.conf           # port 80: ACME challenge + HTTPS redirect
│   ├── grafana.conf              # grafana.wbd-rd.nl
│   ├── gitea.conf                # gitea.wbd-rd.nl
│   ├── keycloak.conf             # keycloak.wbd-rd.nl
│   ├── nodered.conf              # nodered.wbd-rd.nl
│   ├── mlflow.conf               # mlflow.wbd-rd.nl
│   ├── jupyter.conf              # jupyter.wbd-rd.nl
│   ├── portainer.conf            # portainer.wbd-rd.nl (HTTPS upstream)
│   ├── rabbitmq.conf             # rabbitmq.wbd-rd.nl (mgmt UI)
│   └── jenkins.conf              # jenkins.wbd-rd.nl
└── stream.d/
    └── mqtt.conf                 # mqtt.wbd-rd.nl:8883 → rabbitmq:1883

Volumes:

nginx-certs — Let's Encrypt cert chains at /etc/letsencrypt/; read-only into nginx, writable from certbot
nginx-acme-challenge — webroot for HTTP-01 challenges at /var/www/certbot/

All vhosts reference /etc/letsencrypt/live/infra/fullchain.pem and privkey.pem — a stable path independent of the issuance method.

First-run bootstrap

The HTTPS server blocks won't load without a cert at /etc/letsencrypt/live/infra/. Bootstrap procedure (one-time):

cd stacks/nginx-proxy

# 1. Self-signed fallback so nginx starts and serves /.well-known/acme-challenge/
docker compose run --rm --entrypoint=/bin/sh nginx -c \
  "mkdir -p /etc/letsencrypt/live/infra && \
   openssl req -x509 -nodes -days 1 -newkey rsa:2048 \
     -keyout /etc/letsencrypt/live/infra/privkey.pem \
     -out   /etc/letsencrypt/live/infra/fullchain.pem \
     -subj  '/CN=bootstrap-infra'"

# 2. Start nginx (HTTPS blocks load with the dummy cert)
docker compose up -d nginx

# 3. Issue the real cert via HTTP-01
docker compose run --rm certbot certonly \
  --webroot -w /var/www/certbot \
  --email "$LETSENCRYPT_EMAIL" --agree-tos --no-eff-email \
  --cert-name infra \
  -d git.wbd-rd.nl   -d auth.wbd-rd.nl  -d dash.wbd-rd.nl \
  -d flow.wbd-rd.nl  -d ml.wbd-rd.nl    -d hub.wbd-rd.nl \
  -d ops.wbd-rd.nl   -d mq.wbd-rd.nl    -d ci.wbd-rd.nl \
  -d mqtt.wbd-rd.nl

# 4. Reload nginx to pick up the real cert
docker compose exec nginx nginx -s reload

The certbot sidecar then renews every 12h automatically.

DNS prereqs (HTTP-01)

Before bootstrap, ensure A records exist in Versio for the 10 new short subdomains (the canonical tool-named ones — gitea.wbd-rd.nl, grafana.wbd-rd.nl, etc. — stay pointed at the existing Versio stack during the transition):

git.wbd-rd.nl    A  <cloud-public-ip>      # gitea (new)
auth.wbd-rd.nl   A  <cloud-public-ip>      # keycloak
dash.wbd-rd.nl   A  <cloud-public-ip>      # grafana (new)
flow.wbd-rd.nl   A  <cloud-public-ip>      # node-red (new)
ml.wbd-rd.nl     A  <cloud-public-ip>      # mlflow
hub.wbd-rd.nl    A  <cloud-public-ip>      # jupyterhub
ops.wbd-rd.nl    A  <cloud-public-ip>      # portainer
mq.wbd-rd.nl     A  <cloud-public-ip>      # rabbitmq mgmt UI
ci.wbd-rd.nl     A  <cloud-public-ip>      # jenkins
mqtt.wbd-rd.nl   A  <cloud-public-ip>      # MQTT-TLS broker

TODO

Wildcard cert via certbot-dns-transip (post TransIP migration)
OIDC auth_request to Keycloak in front of services without native OIDC (mlflow, portainer-CE)
Edge-side variant: bind to plant-LAN IP, internal CA for *.local hostnames
HSTS + security headers (add_header Strict-Transport-Security ...)

4.5 KiB Raw Blame History