You would have to store the Server Socket somewhere, maybe in a Durable Object?

Yeah a DO would have to be what you use there

Bounded workers may be evicted after waiting for 30 seconds. Is there a way to know if/when you're about to be evicted?

Nope

TypeError: fetch failed
    at Object.processResponse (C:\---\node_modules\undici\lib\fetch\index.js:199:23)
    at C:\---\node_modules\undici\lib\fetch\index.js:928:38
    at node:internal/process/task_queues:140:7
    at AsyncResource.runInAsyncScope (node:async_hooks:204:9)
    at AsyncResource.runMicrotask (node:internal/process/task_queues:137:8)

this is happening when running wrangler with --local when it tries to respond to a request after doing some background stuff for a couple minutes. is this expected? I can't really tell where the code is throwing

actually I'm thinking this is likely due to a fetch timeout. we are requesting something that can take a few minutes to respond

but I thought fetch didn't have a default timeout

does wrangler have some sort of fetch timeout?

I think it is timing out at 120 seconds. any thoughts?

That's the default NodeJS timeout

oh! I didn't realize there was one built-in

do you know how to increase it?

Nothing user-facing that I'm aware of, that's set within Undici (https://github.com/nodejs/undici/issues/1373#issuecomment-1110260144)

Post in #891052295410835476 as that's what

--local

uses, they can take a look at the issue & fix it if needed

do production workers use different values? I could live with it for now if it will work on a real worker

Cloudflare uses 100 seconds

You definitely don't want a connection held open for that long

hmm why not? the alternative is using server sent events which isn't necessarily more reliable I think

we are calling the OpenAI API

GPT-4 + token heavy responses = long response time

I suppose I'll just switch this code to using the SSE version. there is just extra overhead that I avoid it when we don't need to stream the response back. this is returning JSON so we gotta parse the full response anyways

alright switched it to using server sent events and it is working now... the connection is still open for the same amount of time 😛

@Erisa | Support Engineer can you paste the PoC you have written so I can modify it to trigger the bug? 🙂

Uh the one in the ticket?

ya.

also I thought Ms. Erisa was funny

you should be able to copy it out of there

fudge

I missed it

I blame using an Apple Watch again for reading emails