question- to use OpenAI streaming on a worker, do you need to use the openai NPM package? You ca

question- to use OpenAI streaming on a worker, do you need to use the openai NPM package?

You can't just use curl request and parse the. chunks for a SSE response to the frontend?

Can anyone elaborate for me.

Rrob question- to use OpenAI streaming on a worker, do you *need* to use the openai ...

dave•3/15/24, 4:36 PM

you don't have curl in a Worker. Use the openai npm package, it's much easier even if you aren't doing streaming.

robOP•3/15/24, 4:43 PM

not curl, i meant fetch

robOP•3/15/24, 4:43 PM

which is how im doing it now with a fetch to the cloudlfare gateway

robOP•3/15/24, 4:43 PM

its woerking fne for me now

robOP•3/15/24, 4:43 PM

just wanted to try streaming

Isaac McFadyen•3/15/24, 10:46 PM

You can manually parse the SSE, yes.

Isaac McFadyen•3/15/24, 10:46 PM

You'll just need to write the parsing logic yourself.

robOP•3/15/24, 10:58 PM

Alright cool. Thank you

shaderkyle_ | SolSentryBot•3/15/24, 11:22 PM

hey, i noticed there is no support for ciphers in node:crypto when node_compat is enabled. Is there a polyfill available, and/or an equivalent implementation on subtle crypto, specifically with aes-256 and iv (initialization vectors)?

mohamed.adham•3/16/24, 12:32 AM

Hey, I'm new to cloudflare workers,
I have a question

mohamed.adham•3/16/24, 12:33 AM

Should I use hono JS when I use the cloudflare worker ?

Unsmart•3/16/24, 12:37 AM

If you want to sure

Isaac McFadyen•3/16/24, 1:56 AM

Depends on the use-case. For simple things (i.e. single-url rewrites, single-path sites) then probably fine not to. If you have anything more complicated like multiple paths or different HTTP methods then Hono is far easier than manually writing code to route requests.

robOP•3/16/24, 4:09 AM

if i do a rollback on a worker how do i sync that rolled back code with my local?

Rrob if i do a rollback on a worker how do i sync that rolled back code with my local...

John•3/16/24, 4:29 AM

correct me if I'm wrong but can't you just perform a git rollback and pull the repo again?

John•3/16/24, 4:29 AM

and then the worker will update too

John•3/16/24, 4:30 AM

assuming it's linked to the git repo

Unsmart•3/16/24, 4:43 AM

yeah I'd go so far as to even say dont use the rollbacks and just revert commit instead and have ci push it out

Unsmart•3/16/24, 4:43 AM

unless you have a really slow build for some reason

UUnsmart If you want to sure

mohamed.adham•3/16/24, 12:10 PM

@Unsmart | Tech debt
Ok, should I use typescript with hono or JavaScript??

robOP•3/16/24, 12:47 PM

I'm building a framework like express but for cloudflare workers ... built in LLm router too and streaming support

robOP•3/16/24, 12:47 PM

Similar to hono I guess but with some baked in features

韩

韩晨•3/16/24, 5:46 PM

I saw a project on GitHub that uses worker to build a vless node. I would like to ask, is this usage within the scope of normal use?

iamtheworstdev•3/17/24, 5:28 PM

is there an updated version of this? https://github.com/cloudflare/workers-graphql-server
it doesn't work with the latest wrangler

GitHub

GitHub - cloudflare/workers-graphql-server: 🔥Lightning-fast, global...

Lightning-fast, globally distributed Apollo GraphQL server, deployed at the edge using Cloudflare Workers - cloudflare/workers-graphql-server

GitHub - cloudflare/workers-graphql-server: 🔥Lightning-fast, global...

rook•3/17/24, 6:59 PM

what is the best way to deploy a worker to a specific region? if you bind to a d1 database, does it stay there always, even if you don't use the d1 db? and can you do that with cloudflare pages, like nextjs ssr rendering?

Rrook what is the best way to deploy a worker to a specific region? if you bind to a d...

Chaika•3/17/24, 7:03 PM

Workers with a D1 binding will always be placed in a data center near the location of the D1 database they are bound to. Subrequests to other back-end services are ignored by Smart Placement in this case.

https://developers.cloudflare.com/workers/configuration/smart-placement/#d1

Should work with Pages too but I've never tried it. I know Pages has some other issues with Smart Placement like if you try to tail it, it breaks your function entirely.

Not sure why you would want that with SSR though, the nice bit about Cf's Network and SSR is it'd be close to the visitor/fast

CChaika > Workers with a D1 binding will always be placed in a data center near the loca...

rook•3/17/24, 7:10 PM

SSR without any api or db calls would be very limited no? and my databases and apis are fixed in us-east.

picture is a request to a smart placement worker, it does one api call, then SSR renders the page. this takes about 150 ms hot on vercel. but here 200-800 ms.

Chaika•3/17/24, 7:15 PM

I mean it depends what you're doing/how many requests you're making. If it's possible for you to cache that data at all, you'd benefit a lot more from CF's global scale.

That already is smart placement and it's still higher then Vercel? You can check the cf-placement header to see where it ran

Chaika•3/17/24, 7:18 PM

It will regardless of the docs saying it won't, even if you only make one request.
Edit: looks like the smart placement docs were edited a while ago remove that wording and no longer say you need more then one fetch request for smart placement, didn't notice that

rook•3/17/24, 7:22 PM

so it does work for a single request? that picture above is a single req and it looks like multiple round trips, i'm guessing tcp handshake & ssl? sometimes it is faster, not sure if it's reusing the connection or smart placement is working.

Chaika•3/17/24, 7:24 PM

Check the cf-placement response header to see where it ran. It's either local, meaning it wasn't, or remote, meaning it was. It needs some data first though. An http request is more then one roundtrip yea, for establishing the tcp connection and such.

rook•3/17/24, 7:36 PM

where is the cf-placement header? dont have it, but i have the cf-ray header and it shows europe on all requests, even the fast ones. so no smart placement, this is pages though.

Rrook where is the cf-placement header? dont have it, but i have the cf-ray header and...

Chaika•3/17/24, 7:36 PM

it's just a response header. cf-ray is only going to show the one you first connected to, even with smart placement on. Do you have Smart Placement on, and did a deployment after?

rook•3/17/24, 7:37 PM

yes

rook•3/17/24, 7:39 PM

been testing this for a week. even if smart placement did work, i don't want to use it, as the first 20-30 requests are used to place it, and 1% of requests after that don't use it, making it inconsistent.

Rrook been testing this for a week. even if smart placement did work, i don't want to ...

Chaika•3/17/24, 7:42 PM

If you get enough requests, it should just continiously work besides the 1% you mention.
If you don't see the cf-placement header at all though, means it's not enabled. I believe if you tried tailing the function that would also remove it at the moment. Would double check it's enabled for that pages project, for both preview and production, and re-deploy

If you don't want to use it at all, you have no other options for making a worker run in a specific place.

CChaika If you get enough requests, it should just continiously work besides the 1% you ...

rook•3/17/24, 7:45 PM

i triple chcked that its on in the dash at this point xd, maybe gets overwritten somewhere. what about the d1 binding, didn't it place it there right away? or do you mean you can't bind to d1 on pages. like to the nextjs api routes.

Rrook i triple chcked that its on in the dash at this point xd, maybe gets overwritten...

Chaika•3/17/24, 7:49 PM

You can bind to D1 on Pages, it would be place it near yea. You'd still have the cf-placement header if it was enabled though, and the 1% non-smart for comparsion

rook•3/17/24, 7:52 PM

it does feel like an overly engineered solution just to let us place the worker ourselves.

Chaika•3/17/24, 7:54 PM

That was feedback given to them, it would be nice to just pick a region, and plus the ability to only turn it on for specific routes/paths

Chaika•3/17/24, 7:56 PM

the whole smart placement thing is a bit against how CF works I feel, with specific dcs going off for maintenance and magical anycast routing, but a frequently requested feature nonetheless. Even with smart placement on, you still hit the closest location, you just get relayed through to the "smart" picked one

Chaika•3/17/24, 7:57 PM

weird that it doesn't work for you at all though. I would try redeploying now and make sure you don't tail or anything, if you haven't already

CChaika weird that it doesn't work for you at all though. I would try redeploying now an...

rook•3/17/24, 8:04 PM

think i fixed it, had to be in a wrangler.toml dashboard settings not used.

Rrook think i fixed it, had to be in a wrangler.toml dashboard settings not used.

Chaika•3/17/24, 8:04 PM

Are you using Pages or Workers?

rook•3/17/24, 8:04 PM

pages

Chaika•3/17/24, 8:05 PM

Pages doesn't care about wrangler.toml at all except for local dev

rook•3/17/24, 8:06 PM

probably because its a local build + deploy

rook•3/17/24, 8:24 PM

either way, got the cf-placement header now, did 300+ requests to an endpoint with 3 sub-requests, but only local- placement still.

rook•3/17/24, 8:30 PM

i am giving up

Original message was deleted

Erisa•3/17/24, 10:37 PM

Its scoped per zone, if both workers are running on the same zone then yeah they'll share cache

question- to use OpenAI streaming on a worker, do you need to use the openai NPM package? You ca

Similar Threads

Similar Threads

Similar Threads

question- to use OpenAI streaming on a worker, do you *need* to use the openai NPM package? You ca

Similar Threads

Similar Threads

Similar Threads

question- to use OpenAI streaming on a worker, do you need to use the openai NPM package? You ca