fix(storage): add longhorn-wffc SC + pin portainer/forgejo-runner to az-a — fixes PVC attach

k8s/applications/portainer/portainer-values.yaml

@@ -33,7 +33,7 @@ tls:
 # RWO block volume. 10Gi is generous for config data but cheap on Longhorn.
 persistence:
   enabled: true
-  storageClass: "longhorn"
+  storageClass: "longhorn-wffc"
   size: 10Gi
 
 resources:
@@ -51,3 +51,8 @@ tolerations:
   - key: node-role.kubernetes.io/control-plane
     operator: Exists
     effect: NoSchedule
+
+# Pin to az-a (talos-cp-1) — sole Longhorn storage node. Its RWO PVC can only
+# attach there; without this the pod may land on cp-2/cp-3 and fail to mount.
+nodeSelector:
+  topology.kubernetes.io/zone: az-a

k8s/infrastructure/longhorn/kustomization.yaml

@@ -1,5 +1,8 @@
 apiVersion: kustomize.config.k8s.io/v1beta1
 kind: Kustomization
 namespace: longhorn-system
-resources: []
-# Longhorn deployed via Helm chart (values in parent layer)
+resources:
+  - longhorn-wffc-storageclass.yaml
+# Longhorn deployed via Helm chart (values in parent layer). The WFFC
+# StorageClass above is the new default; the chart's Immediate `longhorn` SC is
+# demoted (default-class annotation removed on the live object at bootstrap).

k8s/infrastructure/longhorn/longhorn-wffc-storageclass.yaml

@@ -0,0 +1,29 @@
+# longhorn-wffc — Longhorn StorageClass with WaitForFirstConsumer binding.
+#
+# The chart's default `longhorn` SC uses Immediate binding: the PV binds before
+# the pod is scheduled, so on this single-storage-node topology (only talos-cp-1
+# runs Longhorn) the scheduler often places the pod on cp-2/cp-3 where the volume
+# can't attach ("CSINode does not contain driver driver.longhorn.io").
+#
+# WaitForFirstConsumer defers binding until the pod is scheduled, so the volume
+# is provisioned on the node the pod lands on — and with a single Longhorn node
+# that co-locates pod + volume on cp-1 automatically. This is the new default;
+# the chart's `longhorn` SC is demoted (see longhorn-values default-class=false).
+#
+# volumeBindingMode is immutable, so this is a distinct SC (not an edit of the
+# chart's). Existing volumes stay on `longhorn`; new PVCs use this.
+apiVersion: storage.k8s.io/v1
+kind: StorageClass
+metadata:
+  name: longhorn-wffc
+  annotations:
+    storageclass.kubernetes.io/is-default-class: "true"
+provisioner: driver.longhorn.io
+allowVolumeExpansion: true
+reclaimPolicy: Delete
+volumeBindingMode: WaitForFirstConsumer
+parameters:
+  numberOfReplicas: "1"
+  staleReplicaTimeout: "30"
+  fromBackup: ""
+  dataLocality: "best-effort"

k8s/security/ci-cd/charts/forgejo-runner/templates/deployment.yaml

@@ -19,6 +19,10 @@ spec:
     spec:
       tolerations:
         {{- toYaml .Values.tolerations | nindent 8 }}
+      {{- with .Values.nodeSelector }}
+      nodeSelector:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
 
       initContainers:
         - name: register

k8s/security/ci-cd/charts/forgejo-runner/values.yaml

@@ -30,13 +30,18 @@ dind:
 
 persistence:
   reg:
-    storageClass: longhorn
+    storageClass: longhorn-wffc
     size: 1Gi         # .runner registration file + config — survives pod restarts
   dind:
-    storageClass: longhorn
+    storageClass: longhorn-wffc
     size: 30Gi        # docker layer cache — keeps rebuilds fast across restarts
 
 tolerations:
   - key: node-role.kubernetes.io/control-plane
     operator: Exists
     effect: NoSchedule
+
+# Pin to az-a (talos-cp-1) — sole Longhorn node; RWO PVCs (reg/dind cache) only
+# attach there.
+nodeSelector:
+  topology.kubernetes.io/zone: az-a