I recreated the Keycloak account from LDAP, and then manually patched the databases for all OIDC-based services to the new account UUID, so the existing accounts are linked to the new Keycloak account.

I have two Keycloak accounts, one in the master realm for administrative purposes, and one in the apps realm for all my services, so I didn’t break access to Keycloak

I already had Keycloak set up, but a few services don’t support OIDC or SAML (Jellyfin, Reposilite), so I’ve deployed lldap and connected those services and Keycloak to it. Now I really have a single user across all services

I’m using istio (in ambient mode, so it doesn’t put its sidecar into every pod I create), and Nextcloud behind it, works flawlessly. It does consume a bit of time to set up properly though.

TLS certificates are handled via cert-manager, which works with both Ingress and Gateway API, independent of which implementation you choose.

I’ve also tried envoy gateway (had issues with uploading larger files via Nextcloud Web UI and Gradle publish to maven repo).

Cilium (a CNI) also has an integrated Gateway and Ingress API implementation, works fine, but lacks some of the most recent stable features, and quite some experimental features. But if you just need a HTTPS gateway without much fuss, it’s rock-solid. Setup time is also minimal for the Ingress/Gateway controller part. Cilium CNI on the other hand takes a lot of time to understand and configure, if you’re unfamiliar with networking in general.

Traefik’s featureset is comparable with Cilium’s Gateway API implementation, also works, simple to setup, and might be better suited, if you’re already familiar with it using docker, as it’s the same concepts, just slightly different config (docker labels vs. Ingress/Gateway API with Traefik specific annotations).

Traefik also supports running on K8s, yes.

On that note, if you’re moving to K8s, I recommend looking into K8s Gateway API, it’s the successor to the old Ingress API. There are other, more complete, implementations of it than Traefik. See https://gateway-api.sigs.k8s.io/

Caddy or Traefik, depending on whether you want quick and simple (Caddy), or integration with docker, so you don’t have to write config files by hand (Traefik)

Mayastor or Linstor, Ceph requires too much CPU for these nodes

You may need to configure nginx to pass through some additional headers, I haven’t used it in a while. It could also be that memos refuses IP addresses as SITE_URL, and needs a proper domain name.

I’m using traefik (on kubernetes) as reverse proxy, and I don’t even set SITE_URL, but it still works.

There’s no reason not to expose those services to the Internet, they have authentication, and noone can access them without logging in first. There are actually reasons for exposing them, you can share a memo or a file to other people. You should enable HTTPS though to prevent passwords being transferred in clear text.

If you’re exposing memos through nginx, the SITE_URL needs to be the public url where nginx exposed memos (so exactly the same as you enter in your browser), not the Public-IP and the internal port of memos.