Files

kubernetes/apps/media/plex/ks.yaml

1 ---

# yaml-language-server: $schema=https://k8s-schemas.home-operations.com/kustomize.toolkit.fluxcd.io/kustomization_v1.json

3 apiVersion: kustomize.toolkit.fluxcd.io/v1

4 kind: Kustomization

5 metadata:

6 name: plex

7 spec:

8 components:

9 - ../../../../components/zeroscaler

10 - ../../../../components/gpu

11 - ../../../../components/kopiur/backup

12 dependsOn:

13 - name: k8s-gpu-dra-driver

14 namespace: kube-system

15 - name: miroir-config

16 namespace: miroir-system

17 interval: 1h

18 path: ./kubernetes/apps/media/plex/app

19 postBuild:

20 substitute:

21 APP: plex

22 KOPIUR_CAPACITY: 50Gi

23 KOPIUR_COPYMETHOD: Direct

24 KOPIUR_STORAGECLASS: miroir-fast-local

25 prune: true

26 sourceRef:

27 kind: GitRepository

28 name: flux-system

29 namespace: flux-system

30 targetNamespace: media

31 wait: false

wai: GPU + zeroscaler + kopiur backup

why ·ok L12

Plex is the only media app that needs all three of these components together: GPU for HW transcoding, zeroscaler because it's a personal-use server with predictable off-hours, and kopiur/backup for the `/config` claim that holds the libraries DB and metadata. Sonarr/Radarr don't need GPU; kavita doesn't need GPU either. Both `dependsOn` entries gate on infra being ready (DRA driver must exist before the ResourceClaim is admitted; miroir-config must exist before kopiur can snapshot).

Show the rest

**Why:** dropping any of these three components breaks a different feature; bundling them as the default Plex kustomization is intentional. **How to apply:** if adding another media app that wants GPU + backup, follow this ks as the template; the `KOPIUR_CAPACITY: 50Gi` in postBuild matches Plex's metadata DB size and should be retuned per app.

2026-08-0701KZDY1P5JQ4TXFTFAXPXEVT3N

KOPIUR_CAPACITY: 50Gi for Plex

why ·ok L19

Plex's `/config` holds watched-state, library metadata, the SQLite DB, and a sizable transcode temporary tree — far larger than the 5Gi used by Sonarr or 20Gi used by Jellyfin/Kavita. 50Gi is the round number that survived a real kopiur restore after the openebs → miroir migration when 20Gi filled within a week.

**Why:** kopiur restore fails terminally when capacity is too small; sizing was chosen post-incident, not from docs. **How to apply:** if Plex starts failing kopiur Restores with "no space", bump this in 10Gi increments — kopiur does not auto-grow.

2026-08-0701KZDY1P6H0SV68JA1PRWXPPFF