2026-07-30 07:12:15 +00:00
1 changed files with 8 additions and 0 deletions
@@ -56,9 +56,17 @@ data:
        forward . /etc/resolv.conf {
           max_concurrent 1000
        }
+        # serve_stale mirrors the renner.uno:53 block (issue #393): when the
+        # upstream resolver browns out, keep answering from expired cache
+        # instead of SERVFAILing every public name cluster-wide. Seen on
+        # 2026-07-30 01:22-01:26Z (~1150 SERVFAILs) — it killed a Velero Kopia
+        # maintenance job, which runs with backoffLimit 0, on the Hetzner S3
+        # endpoint (issue #439). cluster.local is exempt from the cache above,
+        # so in-cluster Service records are never served stale.
        cache 30 {
           disable success cluster.local
           disable denial cluster.local
+           serve_stale 1h immediate
        }
        loop
        reload