feat(k3s-upgrade): add podSelector to ignore lognhorn-manager and instance-manager pods

- because of storageclass longhorn-static with numberOfReplicas=1, it is impossible to drain nodes without violating pod disruption budget
- dataloss may occur on volume-level, but postgres-operator should replicate data from primary, if primary is evicted, new will be selected

manifests/system-upgrade/k3s-upgrade-v1.28.yaml

@@ -16,6 +16,7 @@ spec:
       - {key: k3os.io/mode, operator: DoesNotExist}
       - {key: node-role.kubernetes.io/control-plane, operator: Exists}
   serviceAccountName: system-upgrade
+  jobActiveDeadlineSecs: 3600
   cordon: true
   drain:
     timeout: 900s # 15 minutes
@@ -23,5 +24,18 @@ spec:
     ignoreDaemonSets: true
     deleteLocalData: true
     skipWaitForDeleteTimeout: 180
+    # ! We ignore longhorn-manager and instance-manager pods
+    # ! because with storageclass longhorn-static which has only single replica, it is impossible to drain the node
+    # ! without violating the pod disruption budget.
+    podSelector:
+      matchExpressions:
+        - key: app
+          operator: NotIn
+          values:
+            - longhorn-manager
+        - key: longhorn.io/component
+          operator: NotIn
+          values:
+            - instance-manager
   upgrade:
     image: rancher/k3s-upgrade