Merge pull request 'fix(talos): route LoadBalancer replies through table 47' (#1549) from lb-replies-table-47 into main

Reviewed-on: #1549

kubernetes/NETWORK-traffic-classes.md

@@ -22,6 +22,13 @@ Two rules, one per family, send every packet with a pod source to table 47
 (`talos/patches/site-homelab.yaml`, priorities 1000/1001, below main). The
 table decides by longest prefix; no rule carves out exceptions.
 
+A reply from a LoadBalancer VIP is pod traffic as well, but it reaches the
+routing decision with the VIP as its source. Two more rules (1002/1003) send
+the LoadBalancer pool to table 47, so the reply leaves over `ingress0`, where
+the request arrived as the BGP next hop. Without them main sent it out
+`mgmt0`. The pool is a site fact: the site patch's `lb-pool-site` ConfigMap,
+which `bgp-pool` takes by Flux substitution, and the rules beside it.
+
 bird (`kubernetes/infrastructure/bird/`) owns table 47 and fills it with:
 
 - the node's attached networks — `mgmt0`, `ingress0`, `pod0`, `node0` —

kubernetes/apps/home-assistant-iot/shelly-manager-deployment.yaml

@@ -64,7 +64,7 @@ spec:
           type: RuntimeDefault
       containers:
         - name: shelly-manager
-          image: gitea.renner.uno/konairius/shelly-manager:2026-10-30-1343 # {"$imagepolicy": "flux-system:shelly-manager"}
+          image: gitea.renner.uno/konairius/shelly-manager:2026-11-02-0659 # {"$imagepolicy": "flux-system:shelly-manager"}
           imagePullPolicy: IfNotPresent
           ports:
             - name: http

kubernetes/clusters/homelab/infrastructure.yaml

@@ -89,6 +89,9 @@ spec:
     substituteFrom:
       - kind: ConfigMap
         name: cluster-settings
+      # The LoadBalancer pool, from the site's Talos patch (site-*.yaml).
+      - kind: ConfigMap
+        name: lb-pool-site
   dependsOn:
     - name: infrastructure
 ---

kubernetes/infrastructure/post-crds/cilium-bgp.yaml

@@ -8,6 +8,10 @@
 # Selector is opt-in via `lb-pool: bgp`. There is no default pool —
 # Services without this label do not get a LoadBalancer IP.
 #
+# The ranges are a site fact: lb-pool-site from the site's Talos patch
+# (talos/patches/site-*.yaml), by Flux substitution. A CIDR block uses all its
+# addresses (allowFirstLastIPs defaults to Yes).
+#
 # IPv4 + IPv6 ranges keep matching last bytes (10.2.99.x ↔ fd99::x)
 # so each LB has the same "tail" in both families.
 apiVersion: cilium.io/v2alpha1
@@ -19,10 +23,8 @@ spec:
     matchLabels:
       lb-pool: bgp
   blocks:
-    - start: "10.2.99.0"
-      stop: "10.2.99.255"
-    - start: "fd99::"
-      stop: "fd99::ff"
+    - cidr: "${LB_POOL_V4}"
+    - cidr: "${LB_POOL_V6}"
 ---
 # Cilium BGPv2 control-plane config peering each node with the UDM over the
 # ingress VLAN 47 — never over v30iot, which is the untrusted IoT segment.

kubernetes/talos/patches/site-hcloud.yaml

@@ -67,6 +67,18 @@ cluster:
                 nameservers:
                   - 185.12.64.1
                   - 185.12.64.2
+    # The LoadBalancer pool (see site-homelab.yaml). Services pin their
+    # addresses from the production pool, so a test run carries the same one.
+    - name: lb-pool-site
+      contents: |
+        apiVersion: v1
+        kind: ConfigMap
+        metadata:
+          name: lb-pool-site
+          namespace: flux-system
+        data:
+          LB_POOL_V4: 10.2.99.0/24
+          LB_POOL_V6: fd99::/120
 ---
 # Host DNS asks k8s-gateway first, so containerd resolves in-cluster names
 # (gitea.renner.uno) in the cluster; bootstrap.sh installs k8s-gateway right

kubernetes/talos/patches/site-homelab.yaml

@@ -85,6 +85,18 @@ cluster:
                 nameservers:
                   - fd02:47::1
                   - 10.2.47.1
+    # The LoadBalancer pool (bgp-pool, infrastructure/post-crds/cilium-bgp.yaml,
+    # by Flux substitution). The routing rules 1002/1003 below name it too.
+    - name: lb-pool-site
+      contents: |
+        apiVersion: v1
+        kind: ConfigMap
+        metadata:
+          name: lb-pool-site
+          namespace: flux-system
+        data:
+          LB_POOL_V4: 10.2.99.0/24
+          LB_POOL_V6: fd99::/120
 ---
 # All pod traffic goes to table 47, and the table decides — longest prefix
 # match, no exceptions to carve out. bird fills it with the node's attached
@@ -115,6 +127,21 @@ name: "1001"
 src: 10.244.0.0/16
 table: "47"
 ---
+# A LoadBalancer reply is pod traffic too, but it reaches the routing decision
+# with the VIP as its source, so the rules above miss it and main sends it out
+# mgmt0 while the request came in over ingress0. The pool is lb-pool-site above.
+apiVersion: v1alpha1
+kind: RoutingRuleConfig
+name: "1002"
+src: fd99::/120
+table: "47"
+---
+apiVersion: v1alpha1
+kind: RoutingRuleConfig
+name: "1003"
+src: 10.2.99.0/24
+table: "47"
+---
 # Host DNS asks k8s-gateway first, so containerd resolves in-cluster names
 # (gitea.renner.uno) in the cluster; bootstrap.sh installs k8s-gateway right
 # after the CNI, before anything else needs DNS. The UDM is the site's

kubernetes/talos/patches/site-qemu.yaml

@@ -47,3 +47,15 @@ cluster:
           values.yaml: |
             deployment:
               dnsPolicy: Default
+    # The LoadBalancer pool (see site-homelab.yaml). Services pin their
+    # addresses from the production pool, so a test run carries the same one.
+    - name: lb-pool-site
+      contents: |
+        apiVersion: v1
+        kind: ConfigMap
+        metadata:
+          name: lb-pool-site
+          namespace: flux-system
+        data:
+          LB_POOL_V4: 10.2.99.0/24
+          LB_POOL_V6: fd99::/120