Hi! We're setting up a self-hosted monitoring stack (Prometheus + Grafana) inside the same Railway project as our API, scraping over the private network to avoid egress.

Setup:

• NestJS API service running 4 replicas in production, exposing GET /metrics (prom-client)
• Prometheus deployed as a separate service in the same project/environment
• Scrape target: ${{api.RAILWAY_PRIVATE_DOMAIN}}:3000 over private networking

The problem: with 4 replicas behind a single private domain, a plain static scrape target would hit a different replica on each scrape interval, mixing per-process metrics (heap, GC, event loop, counters) into one useless series.

Questions:

1. When a service has N replicas, does a DNS lookup on its RAILWAY_PRIVATE_DOMAIN return one AAAA record per replica, or a single record that's load-balanced at connection time?
2. If it returns all replicas: is Prometheus dns_sd_configs (type AAAA) a supported/recommended pattern on Railway? Anything to know about DNS TTL / propagation after redeploys or scaling events, so targets refresh correctly?
3. If it does NOT return all replicas: is there any supported way to enumerate or address individual replicas over the private network (per-replica hostnames, headless-service-style DNS, etc.)?
4. Is there an official recommended pattern for scraping per-replica metrics on Railway?

Thanks! Happy to share our final setup back with the community once it works.