Private Network Instability: Internal Service Only Reachable 1/10 Times

Question

Hey! I’m running into a really strange issue with two services on the private network in a brand-new Railway environment. My trading bot calls IBeam through:

`http://ibeam-deploy.railway.internal:5000`

IBeam is definitely listening on IPv4:5000 — the public IP works instantly every time. But internally, the private domain behaves completely inconsistently.

Sometimes the trading service can reach it with no issues at all. For example, here’s a log from a deploy where everything worked perfectly:
```
2025-12-03 00:18:25 [INFO] trading_service: IBKR URL: http://ibeam-deploy.railway.internal:5000
2025-12-03 00:18:25 [INFO] trading_service: Account: **********
2025-12-03 00:18:25 [INFO] trading_service: Queue stats: {'pending': 0, 'processing': 0, 'completed': 87, 'failed': 11}
2025-12-03 00:18:25 [INFO] trading_service: Starting consumer loop (BRPOP)...
```
But then — with the exact same config — literally two deploys earlier I got nothing but timeouts, as if the IBeam service wasn’t even reachable on the private network:
```
2025-12-02T23:22:27.860451108Z [inf]  2025-12-02 23:22:21 [INFO] trading_service: IBKR URL: http://ibeam-deploy.railway.internal:5000
2025-12-02 23:22:31.885405216Z [err]  IbkrClient: Connection error detected, retrying attempt 1/3 :: HTTPConnectionPool(host='ibeam-deploy.railway.internal', port=5000): Max retries exceeded with url: /v1/api/tickle (Caused by ConnectTimeoutError(, 'Connection to ibeam-deploy.railway.internal timed out. (connect timeout=10)'))
2025-12-02 23:22:51.885723508Z [err]  IbkrClient: Connection error detected, retrying attempt 2/3 :: HTTPConnectionPool(host='ibeam-deploy.railway.internal', port=5000): Max retries exceeded with url: /v1/api/tickle (Caused by ConnectTimeoutError(, 'Connection to ibeam-deploy.railway.internal timed out. (connect timeout=10)'))
2025-12-02 23:22:56.685309728Z [err]  IbkrClient: Connection error detected, retrying attempt 3/3 :: HTTPConnectionPool(host='ibeam-deploy.railway.internal', port=5000): Max retries exceeded with url: /v1/api/tickle (Caused by ConnectTimeoutError(, 'Connection to ibeam-deploy.railway.internal timed out. (connect timeout=10)'))
2025-12-02 23:23:16.108250718Z [err]  IBKR health check failed: IbkrClient: Connection error HTTPConnectionPool(host='ibeam-deploy.railway.internal', port=5000): Max retries exceeded with url: /v1/api/tickle (Caused by ConnectTimeoutError(, 'Connection to ibeam-deploy.railway.internal timed out. (connect timeout=10)')) for POST http://ibeam-deploy.railway.internal:5000/v1/api/tickle {}
2025-12-02 23:23:16.108256553Z [inf]  2025-12-02 23:23:11 [INFO] trading_service: Queue stats: {'pending': 0, 'processing': 1, 'completed': 86, 'failed': 11}```

All I did between these deploys was restart the service and tweak an env var (the one containing the private domain). No actual configuration changes. Sometimes after that the service becomes reachable, sometimes it stays unreachable.

So the exact same internal hostname works once, then completely stops working, then works again after a random env-var update. I can’t find any pattern at all.

Is there any known issue where a *.railway.internal IPv4 service is only intermittently reachable?
Does railway.internal have any DNS or IP caching that could cause this kind of inconsistent behavior?

Any insights would be hugely appreciated — this one has been driving me crazy. 🙏