Standbys that answer the client API themselves, behind router.active_active

A standby follows the leader's Ready replicas over the agent port, every
two seconds, and reports back the replicas its router used and the
on-demand models it was asked for. No engine keys travel: every control
plane derives them from the cluster secret. Budgets are raised to the
store's every 30 s. Unheard from the leader for a minute, a standby
passes /v1 on as before. Off by default.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

internal/control/adminapi/api.go

@@ -182,6 +182,7 @@ func (a *API) Register(mux Mux) {
 	a.registerPresets(mux)
 	mux.HandleFunc("GET /api/v1/models", a.listModels)
 	mux.HandleFunc("GET /api/v1/replicas", a.listReplicas)
+	a.registerRouting(mux)
 	// Listing is a read and deleting is a separate verb, so a mistyped
 	// command cannot remove anything.
 	mux.HandleFunc("GET /api/v1/weights/orphans", a.listOrphans)

internal/control/adminapi/routes.go

@@ -6,6 +6,7 @@ import (
 	llmgw "gitea.mynodes.xyz/ilker/llmgw"
 	"gitea.mynodes.xyz/ilker/llmgw/internal/bench"
 	"gitea.mynodes.xyz/ilker/llmgw/internal/bench/eval"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control"
 	"gitea.mynodes.xyz/ilker/llmgw/internal/control/catalog"
 	"gitea.mynodes.xyz/ilker/llmgw/internal/control/fabrics"
 	"gitea.mynodes.xyz/ilker/llmgw/internal/control/federation"
@@ -194,6 +195,15 @@ var Routes = []Route{
 	{Method: "GET", Path: "/api/v1/replicas", Summary: "Every replica, with its shards",
 		Query:    []Param{{Name: "model", Description: "Only this model's replicas."}},
 		Response: []ReplicaView{}},
+	{Method: "GET", Path: "/api/v1/routing", Summary: "The Ready replicas a standby routes to",
+		Description: "Read by a standby control plane when router.active_active is on: each replica's address, " +
+			"model, profile, pool and weight, without its key, and which catalog is in force.",
+		Response: control.Routing{}},
+	{Method: "POST", Path: "/api/v1/routing/activity", Summary: "A standby's router reports what it used",
+		Description: "When it last sent each replica a request, and the on-demand models it was asked for, " +
+			"so the leader does not evict a replica busy behind a standby. Answered with the routing, " +
+			"so one request a tick does both.",
+		Request: control.Activity{}, Response: control.Routing{}},
 	{Method: "GET", Path: "/api/v1/explain-placement", Summary: "Why a model could not be placed",
 		Query:    []Param{{Name: "model", Required: true, Description: "The model's name or an alias."}},
 		Response: ExplainView{}},

internal/control/adminapi/routing.go

@@ -0,0 +1,27 @@
+package adminapi
+
+import (
+	"encoding/json"
+	"io"
+	"net/http"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control"
+)
+
+// registerRouting is what a standby answering requests itself reads from the
+// leader, and reports back to it. Reached by the standby through the agent
+// port with its control-plane certificate, like the rest of the forwarded API.
+func (a *API) registerRouting(mux Mux) {
+	mux.HandleFunc("GET /api/v1/routing", func(w http.ResponseWriter, _ *http.Request) {
+		writeJSON(w, http.StatusOK, a.ctl.Routing())
+	})
+	mux.HandleFunc("POST /api/v1/routing/activity", func(w http.ResponseWriter, r *http.Request) {
+		var act control.Activity
+		if err := json.NewDecoder(io.LimitReader(r.Body, 8<<20)).Decode(&act); err != nil {
+			writeError(w, http.StatusBadRequest, err.Error())
+			return
+		}
+		a.ctl.RecordActivity(act)
+		writeJSON(w, http.StatusOK, a.ctl.Routing())
+	})
+}