fix(health): report degraded when the queue is backed up with nothing active

app/api/health/route.ts

@@ -39,6 +39,21 @@ async function checkRedis(): Promise<HealthCheck> {
   }
 }
 
+// A fresh heartbeat does not mean the worker is still doing anything. The
+// heartbeat runs on its own interval, so BullMQ's consumer can stop taking jobs
+// — a dropped queue connection, a crashed consumer loop — while the process,
+// and its heartbeat, stay perfectly alive. Health then keeps answering 200
+// while the backlog grows and nobody is served. Seen in production: 385 jobs
+// waiting for hours behind an uptime monitor that never once alerted.
+//
+// A backlog with nothing in flight is the signal, and it is unambiguous: a
+// healthy worker with a concurrency of 5 never leaves jobs waiting with zero
+// active. The threshold only exists to ride out the moment between a job being
+// enqueued and the worker picking it up.
+const STUCK_QUEUE_MIN_WAITING = Number(
+  process.env.HEALTH_STUCK_QUEUE_WAITING ?? 25
+);
+
 async function checkQueue(): Promise<HealthCheck & { counts?: unknown }> {
   try {
     const counts = await getDMQueue().getJobCounts(
@@ -47,6 +62,15 @@ async function checkQueue(): Promise<HealthCheck & { counts?: unknown }> {
       "delayed",
       "failed"
     );
+    const waiting = counts.waiting ?? 0;
+    const active = counts.active ?? 0;
+    if (waiting >= STUCK_QUEUE_MIN_WAITING && active === 0) {
+      return {
+        status: "error",
+        detail: `${waiting} jobs waiting with none active — the worker is not consuming`,
+        counts,
+      };
+    }
     return { status: "ok", counts };
   } catch (error) {
     return {