Hi Railway Support,

We are experiencing recurring PostgreSQL connection drops on our Pro plan. This is causing downtime for our production POS (Point of Sale) system.

Setup:

- Plan: Pro (serverless disabled)

- Services: 2 NestJS containers (API + Worker) + PostgreSQL (Railway native) + Redis (Bitnami)

- Database URL: Using internal URL (postgres.railway.internal:5432)

- ORM: Prisma with connection pool (10 connections)

Issue:

- PostgreSQL connections drop at random intervals (sometimes 10 min, sometimes 20 min, sometimes 57 min)

- When connections drop, all pooled connections die simultaneously

- This causes our entire application to become unresponsive (all endpoints return 499/502)

- PostgreSQL logs show: could not receive data from client: Connection reset by peer

- Recovery requires manual redeployment

What we have already tried:

- Using internal URL instead of public proxy

- Removed pgbouncer=true from connection string (not using PgBouncer)

- Removed socket_timeout parameter

- Added 30-second keepalive pings (SELECT 1)

- Added auto-reconnect logic on connection failure

- Added statement_timeout=30000 to prevent hung queries

Questions:

1. Does Railway's internal networking have an idle TCP connection timeout? If so, what is the interval?

2. Is there a recommended keepAlive interval for PostgreSQL connections on Railway?

3. Are there any known issues with long-lived database connections on Railway's internal network?

4. Is there any infrastructure maintenance or connection cycling happening at regular intervals?

Logs showing the issue:

2026-02-16 19:15:17.368 UTC [2105] LOG: could not receive data from client: Connection reset by peer

2026-02-16 19:15:17.369 UTC [2103] LOG: could not receive data from client: Connection reset by peer

2026-02-16 19:15:17.369 UTC [2107] LOG: could not receive data from client: Connection reset by peer

(8 connections dropped simultaneously)

Any guidance on maintaining stable long-lived connections would be appreciated.

Thank you.