Merge pull request 'Disable HA hook dedicated listener; route HA via ctrl-agent' (#8) from benvin/ha-listener-bind into main

Reviewed-on: #8

internal/kea/config.go

@@ -241,7 +241,17 @@ func hooks(in RenderInput) []hookLib {
 		"max-response-delay":  firstNonZero(ha.MaxResponseDelay, 60000),
 		"max-ack-delay":       firstNonZero(ha.MaxAckDelay, 5000),
 		"max-unacked-clients": firstNonZero(ha.MaxUnackedClients, 5),
-		"peers":               peers,
+		// With core multi-threading enabled (Kea 2.6 default) the HA hook would
+		// open a dedicated HTTP listener bound to this server's peer url address —
+		// here a per-pod ClusterIP (virtual, kube-proxy DNAT) that is not
+		// assignable on the pod, so the bind fails. Disable it so inbound HA
+		// traffic flows via kea-ctrl-agent, which binds 0.0.0.0:CtrlAgentPort;
+		// peers stay reachable at their ClusterIP:CtrlAgentPort via the per-pod Service.
+		"multi-threading": map[string]any{
+			"enable-multi-threading":  true,
+			"http-dedicated-listener": false,
+		},
+		"peers": peers,
 	}
 	libs = append(libs, hookLib{
 		Library:    HALibrary,

internal/kea/config_test.go

@@ -114,6 +114,53 @@ func TestRenderDHCP4ReferenceSemantics(t *testing.T) {
 	}
 }
 
+// TestHADedicatedListenerDisabled guards the HA listener/bind fix: with Kea 2.6
+// core multi-threading on by default, the HA hook would otherwise open a
+// dedicated HTTP listener bound to this server's peer url — a per-pod ClusterIP
+// that is virtual (kube-proxy DNAT) and unassignable on the pod, failing with
+// "Cannot assign requested address". The rendered config must disable the
+// dedicated listener so inbound HA traffic is served by kea-ctrl-agent
+// (0.0.0.0:8000) while peers stay reachable at their ClusterIP:8000.
+func TestHADedicatedListenerDisabled(t *testing.T) {
+	out, err := RenderDHCP4(referenceInput())
+	if err != nil {
+		t.Fatalf("render: %v", err)
+	}
+	if !strings.Contains(out, `"http-dedicated-listener": false`) {
+		t.Errorf("HA hook must disable the dedicated HTTP listener (it binds this server's ClusterIP peer url, unassignable on the pod); got:\n%s", out)
+	}
+	if !strings.Contains(out, `"enable-multi-threading": true`) {
+		t.Errorf("HA multi-threading must stay enabled (CA-mediated HA traffic); got:\n%s", out)
+	}
+
+	// The CA-mediated route only works because the dedicated listener stays off
+	// on the same port the ctrl-agent binds: parse out the HA hook and assert it.
+	var root dhcp4Root
+	if err := json.Unmarshal([]byte(out), &root); err != nil {
+		t.Fatalf("unmarshal: %v", err)
+	}
+	var haParams map[string]any
+	for _, h := range root.Dhcp4.HooksLibraries {
+		if strings.Contains(h.Library, "libdhcp_ha.so") {
+			haParams = h.Parameters
+		}
+	}
+	if haParams == nil {
+		t.Fatal("HA hook library not present in rendered config")
+	}
+	rels, ok := haParams["high-availability"].([]any)
+	if !ok || len(rels) != 1 {
+		t.Fatalf("high-availability block malformed: %#v", haParams["high-availability"])
+	}
+	mt, ok := rels[0].(map[string]any)["multi-threading"].(map[string]any)
+	if !ok {
+		t.Fatalf("HA relationship missing multi-threading block: %#v", rels[0])
+	}
+	if mt["http-dedicated-listener"] != false {
+		t.Errorf("http-dedicated-listener must be false, got %#v", mt["http-dedicated-listener"])
+	}
+}
+
 // TestSubnetWithoutPoolIsDeclared verifies the pool-less subnet still appears
 // (Kea must know the subnet to service relayed requests) but carries no pools.
 func TestSubnetWithoutPoolIsDeclared(t *testing.T) {
@@ -231,6 +278,12 @@ func TestRenderCtrlAgent(t *testing.T) {
 			t.Errorf("ctrl-agent config missing %q", m)
 		}
 	}
+	// The HA hook runs with the dedicated listener disabled, so inbound HA traffic
+	// is served by the ctrl-agent; it must bind a pod-local address (0.0.0.0), not
+	// a ClusterIP, so peers reaching the per-pod Service's ClusterIP:8000 land here.
+	if !strings.Contains(out, `"http-host": "0.0.0.0"`) {
+		t.Errorf("ctrl-agent must bind 0.0.0.0 (pod-local) for CA-mediated HA, got: %s", out)
+	}
 	// Kea 2.6+ only accepts unix socket paths under /var/run/kea (exact string).
 	if !strings.Contains(out, `"socket-name": "/var/run/kea/`) {
 		t.Errorf("ctrl-agent socket-name must be under /var/run/kea, got: %s", out)