6884ca0c-a0e2-4c3a-bb76-e0f29c956f83

There is no such limit. Is there anything in your service logs that might explain it?

there definitely is a limit, but it is orders of magnitude higher than 16 RPS

so aleks has the right idea, this would be an application level issue

According to this https://docs.railway.app/reference/public-networking#domain-rate-limits 3000 RPS

that is very outdated

I created a new service that is only returning data of the request to test and then I run a load-test, my service receive 1107 requests and then it just stop receiving traffic from Railway
(the errors are not errors, they are just request log)

then the limit is under the 3k rps, maybe max 1k

after that all the request are ERRCONNECTIONTIMED_OUT

and there is no network activity on the service

when I did the test previously, my service was not able to receive any traffic until 1 hour

-- this is the result of a load test of only 1k requests to a dummy golang service:

the network throughput is really bad, is this expected???

The RPS limit is lower than 3K, yes.

network throughput would be impacted since you have either hit the RPS or concurrent connection limits.

During my test, the effective RPS was only 32, and I was blocked from accessing the service for a while. I discovered that when I use higher concurrency (around 400 concurrent calls), Railway starts giving me timeouts for almost an hour. However, if I keep the concurrency under 50 calls, the latency is affected but the service remains accessible.

the limits seem to be significantly below what is documented, to the point where something deployed on the Hobby plan is only useful for a very limited demo environment. I hope this isn't the case with the other plans 🙂

the limits are the same across all plans, and I can assure you they are higher than 32 RPS

I remember getting IP ban when doing a lot of requests with a single IP

yep in this case it looks like they are getting a soft ban

this has sense… thanks!