Rule-based deletion of watched movies from Radarr (with files), driven by
Maintainerr. Raw manifests + directory-type Argo Application (no Helm).
Config on shared plex-data NFS PVC (subPath configs/maintainerr); Recreate
strategy since it uses SQLite on RWX NFS. ClusterIP only, no ingress —
access via kubectl/k9s port-forward.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Plan to move both Bitnami postgres instances (pgsql PG16 in default,
gitea-postgresql PG17 bundled in gitea) to CloudNativePG, since Bitnami
images are frozen (bitnamilegacy). Not executed -- planning doc only.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
0.16.1 chart defaults frr.enabled=false but frrk8s.enabled=true, which
deploys a heavy frr-k8s daemonset. With no BGP peers (pure L2/ARP), FRR is
unnecessary and its images caused DiskPressure on the Pi nodes, evicting a
speaker and stalling the rollout. Disable both frr and frrk8s for a single
-container L2 speaker.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Bumped gitea helm chart 12.4.0->12.6.0 (app 1.24.6->1.26.1). The chart
default RollingUpdate (maxSurge 100%/maxUnavailable 0) surges a second pod
that can't mount the single RWO NFS PVC, deadlocking 'helm upgrade --wait'.
Recreate avoids it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
AirVPN blocks outbound DNS-over-TLS (tcp/853), so gluetun's default DoT
resolver at 127.0.0.1 never gets answers. The startup healthcheck's
"lookup cloudflare.com" then times out and the VPN restarts every ~6s
in a permanent loop, leaving qbittorrent with no working DNS.
Verified inside the pod netns: tunnel egress works (ping 8.8.8.8 18ms),
AirVPN's pushed resolver 10.128.0.1 resolves fine, but tcp/853 to both
1.1.1.1 and 8.8.8.8 times out.
Set DOT=off and DNS_ADDRESS=10.128.0.1 so gluetun points resolv.conf at
AirVPN's pushed DNS, reached over the tunnel (no DNS leak, no port 853).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fixes VPN restart loop after :latest pulled a build with the Alpine 3.22
iptables parsing regression and the healthcheck race (#3123). v3.41.1
includes the k8s cluster-DNS auto-detection so DNS lookups in the
startup healthcheck no longer time out behind the killswitch.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Snapshot of /var/lib/rancher/k3s/server/manifests/traefik-config.yaml
on turing1 after dropping the v2 image pin during the Traefik v3
migration. Lives only on the node otherwise — track it here so it can
be restored on a node rebuild.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
k3s update bumped Traefik chart 37 → 39, dropping v2 support. Replace
the v2-only `whitelist.sourcerange` annotation on the gitea ingress
with an `ipAllowList` Middleware (resources/gitea-middleware.yaml),
referenced via `router.middlewares`. Switch the default-ns ingresses
(kube-plex, radarr, sonarr, lidarr) from the deprecated
`kubernetes.io/ingress.class` annotation to `spec.ingressClassName`.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
prowlarr was pinned to a stale digest (v2.0.5.5160) via
.argocd-source-prowlarr.yaml; remove the file so the live app's
helm.parameters (which already has the current :latest digest =
v2.3.5.5327) takes effect.
qbittorrent: bump 5.1.0 -> 5.2.0.
gilgamezh