fix: converge load-spread rollouts

infrastructure/core/coredns-deployment.yaml

@@ -95,6 +95,8 @@ spec:
                     operator: In
                     values:
                       - "true"
+                  - key: node-role.kubernetes.io/storage-backbone
+                    operator: DoesNotExist
           preferredDuringSchedulingIgnoredDuringExecution:
             - weight: 100
               preference:

services/monitoring/helmrelease.yaml

@@ -70,7 +70,6 @@ metadata:
 spec:
   interval: 15m
   upgrade:
-    force: true
     disableWait: true
   chart:
     spec:
@@ -100,12 +99,10 @@ spec:
 
       persistentVolume:
         enabled: true
-        # A year of Atlas cardinality is ~210Gi today, and this 100Gi never
+        # The claim has already been expanded to 400Gi. Keep the original
-        # matched the 200Gi volume actually bound in the cluster. That volume
+        # StatefulSet template size here because Kubernetes forbids changing
-        # filled on 2026-08-18: VictoriaMetrics went read-only, ingestion
+        # volumeClaimTemplates; the bound PVC retains its larger capacity.
-        # stopped, and every public dashboard panel rendered "No data" while
+        size: 100Gi
-        # all 119 scrape targets stayed healthy.
-        size: 400Gi
       resources:
         requests:
           cpu: 500m