2 months ago
We are migrating our service from Fly.io to Railway for better reliability.
Our backend is a proxy for multiple LLMs, built with Bun, Hono, and tRPC (specifically using httpBatchStreamLink; https://trpc.io/docs/client/links/httpBatchStreamLink).
We've discovered an issue where streams to our client apps are unexpectedly closing. This problem only occurs on our Railway deployment; the exact same code works perfectly on Fly.io.
We're eager to complete this migration. Are there any known issues, platform-specific configurations, or workarounds for this kind of streaming setup on Railway? We can provide a minimal reproducible repository (without LLMs) if needed.
Note that we disabled Serveless on the experiment while we enabled it now.
Please see the error log below:
[12:43:29.722] ERROR: onError(tRPC): {"error":{"cause":{},"code":"BAD_REQUEST","name":"TRPCError"},"path":"chatStream","ctx":{"req":{},"env":{}},"type":"mutation","req":{}}
{
error: 31 | var TRPCError = class extends Error {
32 | constructor(opts) {
33 | var _ref, _opts$message, _this$cause;
34 | const cause = getCauseFromUnknown(opts.cause);
35 | const message = (_ref = (_opts$message = opts.message) !== null && _opts$message !== void 0 ? _opts$message : cause === null || cause === void 0 ? void 0 : cause.message) !== null && _ref !== void 0 ? _ref : opts.code;
36 | super(message, { cause });
^
TRPCError: The connection was closed.
cause: [DOMException ...],
code: "BAD_REQUEST"
at new TRPCError (/app/node_modules/@trpc/server/dist/tracked-Bp72jHif.mjs:36:3)
at <anonymous> (/app/node_modules/@trpc/server/dist/resolveResponse-Hga1xOO1.mjs:53:11)
,
path: "chatStream",
input: undefined,
ctx: {
req: Request (0 KB) {
method: "POST",
url: "http://XXXXXX/api/trpc/chatStream?batch=1",
headers: Headers {
"host": "XXXXXX",
"user-agent": "Bun/1.2.20",
"content-length": "54604",
"accept": "*/*",
"accept-encoding": "gzip, deflate, br, zstd",
"content-type": "application/json",
"trpc-accept": "application/jsonl",
"x-forwarded-for": "XXXXXX",
"x-forwarded-host": "XXXXXX",
"x-forwarded-proto": "https",
"x-railway-edge": "railway/asia-southeast1-eqsg3a",
"x-railway-request-id": "PkHMtCQNTIK8wzXOacI7Nw",
"x-real-ip": "XXXXXX",
"x-request-start": "1755661470147",
}
},
env: HTTPServer {
address: [Object ...],
development: false,
fetch: [Function: fetch],
hostname: "0.0.0.0",
id: "",
pendingRequests: 0,
pendingWebSockets: 0,
port: 8080,
protocol: "http",
publish: [Function: publish],
ref: [Function: ref],
reload: [Function: reload],
requestIP: [Function: requestIP],
stop: [Function: stop],
subscriberCount: [Function: subscriberCount],
timeout: [Function: timeout],
unref: [Function: unref],
upgrade: [Function: upgrade],
url: [URL ...],
[Symbol(Symbol.dispose)]: [Function: dispose],
},
},
type: "mutation",
req: Request (0 KB) {
method: "POST",
url: "http://XXXXXX/api/trpc/chatStream?batch=1",
headers: Headers {
"host": "XXXXXX",
"user-agent": "Bun/1.2.20",
"content-length": "54604",
"accept": "*/*",
"accept-encoding": "gzip, deflate, br, zstd",
"content-type": "application/json",
"trpc-accept": "application/jsonl",
"x-forwarded-for": "XXXXXX",
"x-forwarded-host": "XXXXXX",
"x-forwarded-proto": "https",
"x-railway-edge": "railway/asia-southeast1-eqsg3a",
"x-railway-request-id": "PkHMtCQNTIK8wzXOacI7Nw",
"x-real-ip": "XXXXXX",
"x-request-start": "1755661470147",
}
},
}8 Replies
2 months ago
How long would you say the stream terminates after, we have a 15 minute websocket timeout for Railway generated domains.
Status changed to Awaiting User Response Railway • 2 months ago
2 months ago
Okay, it seems to be exceeding the 15-minute limit. We'll measure the exact duration to confirm. If we set up a custom domain, will that remove the 15-minute limitation?
Status changed to Awaiting Railway Response Railway • 2 months ago
2 months ago
Hmm, but, https://trpc.io/docs/client/links/httpBatchStreamLink seems not to use Websocket.
exkazuu
Okay, it seems to be exceeding the 15-minute limit. We'll measure the exact duration to confirm. If we set up a custom domain, will that remove the 15-minute limitation?
2 months ago
Yes.
Status changed to Awaiting User Response Railway • 2 months ago
2 months ago
Great! Anyway, We'll test our server with a custom domain and report the result.
Status changed to Awaiting Railway Response Railway • 2 months ago
2 months ago
I have just been informed that the limit isn't a WS specific throttle. Can you walk us through how long the connection lasts?
Status changed to Awaiting User Response Railway • 2 months ago
angelo-railway
I have just been informed that the limit isn't a WS specific throttle. Can you walk us through how long the connection lasts?
2 months ago
Okay. We cannot analyze the durations from the current logs. We'll improve our logger to measure the durations. Btw, I noticed that the error occurs both on non-streaming and streaming.
Status changed to Awaiting Railway Response Railway • 2 months ago
2 months ago
Yea, from the network request on our side we got a 200.
From the engineering team: "The request ID they gave was for a POST request that returned a 200 in 20 ms with ~50kb, so all is good on our end."
Status changed to Awaiting User Response Railway • 2 months ago
2 months ago
This thread has been marked as solved automatically due to a lack of recent activity. Please re-open this thread or create a new one if you require further assistance. Thank you!
Status changed to Solved Railway • about 2 months ago