From f2f4a2580ff26fb47c8d55fa92eef9bba878e649 Mon Sep 17 00:00:00 2001
From: Story Crater Bot <19826264+Riotpiaole@users.noreply.github.com>
Date: Tue, 21 Jul 2026 11:14:11 -0700
Subject: [PATCH] =?UTF-8?q?fix(prometheus):=20pin=20to=20az-a=20+=20longho?=
 =?UTF-8?q?rn-wffc=20SC=20=E2=80=94=20RWO=20PVC=20failed=20to=20attach=20o?=
 =?UTF-8?q?n=20cp-2=20(sole=20Longhorn=20node=20is=20cp-1)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 k8s/platform/monitoring/prometheus-values.yaml | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/k8s/platform/monitoring/prometheus-values.yaml b/k8s/platform/monitoring/prometheus-values.yaml
index 47a819d..d0828bb 100644
--- a/k8s/platform/monitoring/prometheus-values.yaml
+++ b/k8s/platform/monitoring/prometheus-values.yaml
@@ -36,12 +36,18 @@ prometheus:
     scrapeTimeout: 60s
     evaluationInterval: 30s
 
+    # Pin to az-a (talos-cp-1) — sole Longhorn node; else the RWO PVC can't
+    # attach on cp-2/cp-3 (CSINode lacks driver.longhorn.io).
+    nodeSelector:
+      topology.kubernetes.io/zone: az-a
+
     # Persistent storage — metrics survive node reboots and pod restarts.
-    # Longhorn reattaches the PVC automatically when the pod reschedules.
+    # longhorn-wffc (WaitForFirstConsumer) binds after scheduling so the volume
+    # lands on the node the pod runs on.
     storageSpec:
       volumeClaimTemplate:
         spec:
-          storageClassName: longhorn
+          storageClassName: longhorn-wffc
           accessModes: ["ReadWriteOnce"]
           resources:
             requests: