can you share the url you are using? pay attention that for internal request you should use http not https , I'm able to send request using the internal url + the port, something like:
http://XXX-YYY.railway.internal:8080

The error happens because the Next.js app is trying to fetch data from the Convex backend using the wrong URL. For internal requests, you should use the internal URL with http and the correct port, not https. Using https or a public URL can cause timeouts (ETIMEDOUT) even if the backend shows a 200 status. Updating the fetch URL to the internal one usually fixes this error.

const data = await (await fetch${process.env.API_URL}/api/data, { cache: "no-store" })).json();

I am not self hosting convex my self, the convex URL is provided via the convex.dev website, and it works locally, and if I spam refresh I can get it to work sometimes. I asked in the convex discord, they said its an issue with IPV6.

I am not fetching like this, I am using the usePreloadQuery provided by convex.

I am using the convex.cloud URL, not an internal URL, I am not self hosting convex my self.

can you try to connect to your container via ssh and try to ping convex? At least you can understand if it is blocked by railway or not

How do you ping inside of railways docker container? there is no ping, curl or wget nor busybox installed, and installing ping via apt, says I dont have permission to run it, even after setting the chmod

I just installed curl via apt, and ran it on my production url, and it processes the response with success and I can see its the correct response text as well.

I don’t have direct SSH access to the Railway container. However, I tested network connectivity from within the container by installing curl via apt and making a request to the Convex production URL. The request completed successfully, and the response content matches the expected output.

This indicates that outbound connectivity to Convex from Railway is working and does not appear to be blocked.

ok, now we know that railway doesn't block the request, can you please console.log the url passed when you initialise the client? I wonder if by any reason the env is empty or wrong.

How could that be when the error is a timeout error if the .env is empty or wrong? and in the convex backend logs, I can see the request passes with success. (if the URL was wrong, then the convex backend wouldn't log which functions were called and showing a success audit log)

I feel like the solution should be this, https://github.com/vercel/next.js/discussions/70423, I monitored the CURL request and it took 750ms to get back to the railway ssh, and from what I can see the default timeout is 250ms, I'll try setting it to 1 second to see if that fixes it, and additionality, if that does fix it, I'll self host convex my self via railway and use the internal networking.

sorry I misunderstood part of the message, if the request hit convex then the env it's fine. my idea was that maybe the code didn't picked up the convex url and was sending the request to the wrong url....but i you see the request in convex then it means the code is fine, yeah try to increase the timeout on the client

Increasing the timeout fixed the issue, but the website is really slow, I am guessing I need to selfhost convex my self.