Post

Seerr and Tasterr on k3s: Wiring a Media Request Front End Through Cloudflare and a VPN Pod

Seerr and Tasterr on k3s: Wiring a Media Request Front End Through Cloudflare and a VPN Pod

Purpose

Seerr is the request front end of the media stack. Someone asks for a show or a movie, and Seerr hands the request to Sonarr or Radarr. It is the merged Overseerr/Jellyseerr project. Tasterr is a “browse and taste” UI that sits on top of Seerr and signs in through Seerr’s own local accounts.

The interesting part is the network path. Seerr has to reach Jellyfin over the LAN, because the public name goes through Cloudflare and Cloudflare answers its API calls with 403. It reaches Sonarr and Radarr through one Service in front of a VPN pod. The Jellyfin side is covered in the Jellyfin and HandBrake post.

flowchart LR
    B["Browser"] --> T["Traefik"]
    T --> S["Seerr"]
    TA["Tasterr"] --> S
    S --> J["Jellyfin<br/>(LAN name)"]
    S -.->|"403"| CF["Cloudflare<br/>(public name)"]
    CF -.-> J
    S --> G["gluetun pod Service"]
    G --> SR["Sonarr and Radarr"]
    S --> P["PVC (SQLite)"]
    P --> O["Nightly offsite pull"]

A second, smaller picture shows how changes reach the cluster.

flowchart LR
    GIT["Git (k3s branch)"] --> A["ArgoCD"]
    A --> C["Cluster"]

Deployment

Seerr runs in the default namespace as one replica of ghcr.io/seerr-team/seerr:v3.5.2. It was first deployed on 2026-10-03 on v3.1.0 and is now on v3.5.2. Tasterr runs next to it as ghcr.io/zacharyarthur/tasterr:2.4.0, also one replica.

The Seerr Deployment uses the Recreate strategy. Seerr keeps its state in SQLite on one ReadWriteOnce volume, so a rolling update would put two pods on the same database. The container listens on port 5055 behind a ClusterIP Service, and Tasterr reaches Seerr through that Service’s cluster-internal name.

Configuration

The pod runs as a non-root user and group 1000, with fsGroup 1000, privilege escalation disallowed and all capabilities dropped:

1
2
3
4
5
6
7
securityContext:
  runAsUser: 1000
  runAsGroup: 1000
  fsGroup: 1000
  allowPrivilegeEscalation: false
  capabilities:
    drop: ["ALL"]

The health probe path is /api/v1/settings/public. Resources are requests of 100m CPU and 256 MiB, with limits of 1 CPU and 768 MiB.

Jellyfin: LAN name, not the public one

The setup wizard has to use Jellyfin’s LAN name (the Docker host directly). The public hostname goes through Cloudflare, which answers Seerr’s API calls with 403. Seerr then shows “Something went wrong while syncing libraries”. Use the LAN name for the API and the public name only for links.

Sonarr and Radarr in a VPN pod

Sonarr and Radarr run inside the gluetun VPN pod. Seerr points at that pod’s Service by its cluster-internal name: gluetun.default.svc.cluster.local on port 8989 for Sonarr and 7878 for Radarr. One Service fronts several apps because they share the pod’s network namespace, so the same Service also exposes ports for the other apps in that pod.

Tasterr

Tasterr signs in through Seerr’s own local accounts and reaches Seerr over the cluster Service.

Storage

Seerr’s PVC is seerr-config-zfs: 2 GiB on local-zfs, ReadWriteOnce, mounted at /app/config. It holds the settings and the SQLite database. The nightly k3s PV pull backs it up offsite (see the offsite ZFS pull post). Tasterr has its own 1 GiB local-zfs ReadWriteOnce PVC.

Access

The web UI is reached through a Traefik IngressRoute with tls: {}, which uses the cluster’s default wildcard certificate (see the wildcard TLS post). The ArgoCD app seerr tracks the k3s branch of the deploy repo with automated sync, prune and self-heal, and shows Synced and Healthy.

Notes

  • Renovate never automerges on the k3s branch, because a merge there is a deploy. See the Renovate post for the general setup.
  • Both pods’ logs reach Graylog through the cluster’s Vector DaemonSet, and both have a one-minute Gatus check through Traefik.
This post is licensed under CC BY 4.0 by the author.