refactor(onboard): one router readiness predicate for both acceptance paths

Review follow-up. The startup poll and the final health snapshot each
composed the readiness rule at their own call site, which is the shape that
let the two paths carry different definitions of readiness in the first
place. Fold the complete rule -- /health answered 2xx and names at least one
healthy endpoint -- into the existing local predicate, which now takes the
snapshot itself, and have both acceptance paths call it. Retry, timeout, and
diagnostic behavior are unchanged, and production line count is neutral.

Also correct the stale comment on the isRouterHealthy stub in the
zero-healthy-endpoints case: the startup poll no longer calls isRouterHealthy
at all, so the stub's job is to trip a regression back to the old boolean
poll.

Signed-off-by: Udaya Tejas <udayatejas2004@gmail.com>

Author: udsy19

src/lib/onboard/model-router.ts

@@ -469,7 +469,7 @@ export async function startModelRouter(
     );
     healthAttempts += 1;
     const pollSnapshot = await deps.getRouterHealthSnapshot(port, healthTimeoutMs);
-    const healthy = pollSnapshot.healthy && hasHealthyEndpoint(pollSnapshot.body);
+    const healthy = isRouterSnapshotReady(pollSnapshot);
     const processAlive = deps.isProcessAlive(pid);
     if (healthy && processAlive) return pid;
     if (!processAlive) {
@@ -485,7 +485,7 @@ export async function startModelRouter(
   const finalSnapshot: RouterHealthSnapshot = childExited
     ? { healthy: false, body: null }
     : await deps.getRouterHealthSnapshot(port, ROUTER_FINAL_HEALTH_SNAPSHOT_TIMEOUT_MS);
-  if (finalSnapshot.healthy && hasHealthyEndpoint(finalSnapshot.body) && deps.isProcessAlive(pid)) {
+  if (isRouterSnapshotReady(finalSnapshot) && deps.isProcessAlive(pid)) {
     return pid;
   }
   try {
@@ -504,11 +504,11 @@ export async function startModelRouter(
   );
 }
 
-/** True when the parsed /health body names at least one healthy endpoint. */
-function hasHealthyEndpoint(body: string | null): boolean {
-  if (!body) return false;
+/** Router readiness: /health answered 2xx and names at least one healthy endpoint. */
+function isRouterSnapshotReady(snapshot: RouterHealthSnapshot): boolean {
+  if (!snapshot.healthy || !snapshot.body) return false;
   try {
-    const parsed = JSON.parse(body) as { healthy_endpoints?: readonly unknown[] };
+    const parsed = JSON.parse(snapshot.body) as { healthy_endpoints?: readonly unknown[] };
     return Array.isArray(parsed?.healthy_endpoints) && parsed.healthy_endpoints.length > 0;
   } catch {
     return false;

test/onboard-model-router.test.ts

@@ -749,8 +749,9 @@ describe("onboard Model Router setup", () => {
           }),
           resolveProviderCredential: () => null,
           buildSubprocessEnv: () => ({}),
-          // The pre-spawn port guard calls isRouterHealthy without a timeout;
-          // only the startup poll passes one. Answer 2xx for the poll alone.
+          // The pre-spawn port guard calls isRouterHealthy without a timeout.
+          // Return true for timeout-bearing calls so a regression to the old
+          // boolean startup poll cannot accept zero healthy endpoints.
           isRouterHealthy: async (_port: number, timeoutMs) => timeoutMs !== undefined,
           getRouterHealthSnapshot: async () => ({ healthy: true, body: allUnhealthyBody }),
           sleep: async () => undefined,