core: repair node role reconciler

infrastructure/core/node-prefer-noschedule-cronjob.yaml

@@ -24,8 +24,17 @@ spec:
                 - bash
                 - -ceu
                 - |
+                  KUBE_TOKEN_PATH="/var/run/secrets/kubernetes.io/serviceaccount/token"
+                  KUBE_CA_PATH="/var/run/secrets/kubernetes.io/serviceaccount/ca.crt"
+                  KUBE_SERVER="https://${KUBERNETES_SERVICE_HOST}:${KUBERNETES_SERVICE_PORT_HTTPS:-443}"
+
                   k() {
-                    kubectl --request-timeout=10s "$@"
+                    kubectl \
+                      --server="${KUBE_SERVER}" \
+                      --certificate-authority="${KUBE_CA_PATH}" \
+                      --token="$(cat "${KUBE_TOKEN_PATH}")" \
+                      --request-timeout=10s \
+                      "$@"
                   }
 
                   clear_worker() {
@@ -33,9 +42,9 @@ spec:
                     local hardware="${2}"
                     if k get node "${node}" >/dev/null 2>&1; then
                       k label node "${node}" node-role.kubernetes.io/worker=true "hardware=${hardware}" --overwrite=true || true
+                      k label node "${node}" node-role.kubernetes.io/storage-backbone- || true
                       k label node "${node}" atlas.bstein.dev/spillover- || true
-                      k taint node "${node}" node.kubernetes.io/unschedulable:NoSchedule- || true
+                      # Recovery cordons are owned by Ananke, not this role reconciler.
-                      k uncordon "${node}" || true
                     else
                       echo "skipping missing node ${node}"
                     fi
@@ -71,7 +80,12 @@ spec:
 
                   for node in titan-13 titan-15 titan-17 titan-19; do
                     if k get node "${node}" >/dev/null 2>&1; then
-                      k label node "${node}" atlas.bstein.dev/spillover=true longhorn-host=true --overwrite=true || true
+                      k label node "${node}" \
+                        atlas.bstein.dev/spillover=true \
+                        longhorn-host=true \
+                        node-role.kubernetes.io/worker=true \
+                        node-role.kubernetes.io/storage-backbone=true \
+                        --overwrite=true || true
                       k taint node "${node}" longhorn=true:PreferNoSchedule --overwrite=true || true
                       k taint node "${node}" atlas.bstein.dev/spillover=true:PreferNoSchedule --overwrite=true || true
                     else

infrastructure/core/node-prefer-noschedule-rbac.yaml

@@ -6,7 +6,7 @@ metadata:
 rules:
   - apiGroups: [""]
     resources: ["nodes"]
-    verbs: ["get", "list", "patch"]
+    verbs: ["get", "list", "patch", "update"]
 ---
 apiVersion: rbac.authorization.k8s.io/v1
 kind: ClusterRoleBinding