Load balancing

Spreading incoming requests across several identical servers — round-robin, least-connections, or by hash — so that no one of them is the limit and any one of them can be taken out for a deploy without readers noticing; the balancer is the one address the world knows, the instances behind it are interchangeable

Railway does it for this site at the platform level: a deploy starts the new container, health-checks it, moves traffic, then retires the old — the swap a 502 sometimes shows for a second. The precondition is that the instances hold no state a reader would miss: sessions in the database, uploads in a bucket, the cache shared or accepted as per-instance. AEM’s dispatcher is a load balancer among its other jobs; a serverless function is load balancing taken to its limit, one instance per request.