Both services are in Southeast Asia (Singapore), same project

(triumphant-commitment): service "web" (Django) and service "Postgres".

A bare TCP connect from web to postgres.railway.internal:5432 takes ~220ms.

I would expect sub-millisecond on the private network.

REPRODUCTION

Run in Railway -> web -> Console:

python -c "import socket,time; t=time.time(); s=socket.create_connection(('postgres.railway.internal',5432),5); print(round((time.time()-t)*1000,2)); s.close()"

Returns ~221ms every time, over many runs, at different times of day.

DNS

postgres.railway.internal resolves to fd12:b2b5:498a:1:d000:a1:ec3b:e6b2 (IPv6).

ALREADY TRIED

• "Enable Outbound IPv6" turned on for the web service. No change.

• Both services confirmed in the same region.

• Django CONN_MAX_AGE=60 set, so this is not per-request connection churn

  alone, but the underlying connect cost is still the floor.

IMPACT

Every Django query costs roughly 170ms. An unauthenticated endpoint that

touches no database responds in 103ms; an endpoint doing 4 queries responds

in 1,400ms. Query counts are already minimised (dashboard reduced 13 -> 4),

so application-side optimisation cannot go further.

QUESTION

Is this expected latency for the private network, or is something misrouted

for this project? Is there a way to connect over IPv4 internally, or to pin

both services to the same host/AZ?