diff --git a/helm-overrides/k8s-admin-prd-ase1/gitea/custom-values.yaml b/helm-overrides/k8s-admin-prd-ase1/gitea/custom-values.yaml
index 285bf3e..7fd3994 100644
--- a/helm-overrides/k8s-admin-prd-ase1/gitea/custom-values.yaml
+++ b/helm-overrides/k8s-admin-prd-ase1/gitea/custom-values.yaml
@@ -12,6 +12,19 @@ gitea:
   # persistence.size must stay 10Gi to match the already-bound PVC —
   # local-path-provisioner doesn't support volume expansion.
 
+  # Chart default is RollingUpdate with maxUnavailable: 0 — the new pod
+  # always comes up before the old one terminates. On a real multi-node
+  # cluster with real RWO block storage that's fine (the new pod just
+  # can't mount until the old one releases). On this single-node cluster,
+  # local-path-provisioner's hostPath-style volume doesn't block a second
+  # same-node mount, so old+new pods briefly run concurrently against the
+  # same /data — and Gitea's LevelDB-backed queue holds an exclusive file
+  # lock, so the new pod crashes with "unable to lock level db ...
+  # resource temporarily unavailable". Recreate forces the old pod to
+  # fully terminate (and release the lock) before the new one starts.
+  strategy:
+    type: Recreate
+
   persistence:
     size: 10Gi