probably can't. we charge based on requests and usage of our service, but our api doesn't have thing

probably can't. we charge based on requests and usage of our service, but our api doesn't have things like image streams etc. is your service really bandwidth heavy? like an image api or smth? maybe you should look into tail workers or something like that. if your saas product is more like an api to connect services etc. than you should probably charge by invocation or something like that

SteffTekOP•6/18/24, 5:41 PM

maybe 0.003ct per invocation

SSteffTek probably can't. we charge based on requests and usage of our service, but our ap...

Nick•6/18/24, 5:44 PM

Yea it’s bandwidth heavy, basically a wrapper around r2 with some value added features. Tail workers are the same as event trace logging and don’t get the outgoing bandwidth sadly… I might have to compress at another layer of the stack and then calc the bandwidth ahead of time

NNick Yea it’s bandwidth heavy, basically a wrapper around r2 with some value added fe...

SteffTekOP•6/18/24, 5:45 PM

if it's r2 based you should easily get the file size you're serving, thats a good baseline for bandwidth usage

Unsmart•6/18/24, 5:46 PM

Cloudflare doesnt charge you bandwidth so why do you want to charge by bandwidth instead of by request?

UUnsmart Cloudflare doesnt charge you bandwidth so why do you want to charge by bandwidth...

SteffTekOP•6/18/24, 5:46 PM

https://tenor.com/view/money-mr-krabs-gif-18326632

Tenor

UUnsmart Cloudflare doesnt charge you bandwidth so why do you want to charge by bandwidth...

Nick•6/18/24, 5:48 PM

That’s my pricing model

Nick•6/18/24, 5:48 PM

I have to work harder on larger files so it’s a decent abstraction without getting into charging for cpu time

Nick•6/18/24, 5:51 PM

The issue is I don’t want someone to count the bytes and be like “you’re overcharging me for bandwidth”

NNick The issue is I don’t want someone to count the bytes and be like “you’re overcha...

SteffTekOP•6/18/24, 5:52 PM

be transparent, say the bytes come from the files on your storage backend

Nick•6/18/24, 5:53 PM

I guess if I get enough $ to afford enterprise then I can offer a “discount” on bandwidth

SteffTekOP•6/18/24, 5:53 PM

i always tell my business partners: "there are some problems that should be solved with tech solutions, others should be solved with ToS" :D

Nick•6/18/24, 5:53 PM

Yeah good point! if there’s no easy solution then I’ll go that route, just wondering if I was missing something. Seems not. Thanks anyway

Erisa•6/18/24, 5:58 PM

Yeah for that you wouldnt want to think in terms of bandwidth and rather "bytes retrieved from storage"

SteffTekOP•6/18/24, 6:03 PM

all your bytes are belong to us

f8•6/19/24, 5:23 AM

Why am i only getting a CORS error when i try calling worker endpoint from localhost? Works fine calling it in browser and i set Access-Control-Allow-Origin to * as well as added preflight

simons2889•6/19/24, 8:00 AM

At this point, is it possible to communicate via grpc directly from the worker to the client?

Ff8 Why am i only getting a CORS error when i try calling worker endpoint from local...

neoney•6/19/24, 10:23 AM

Are you properly handling OPTIONS requests?

EErisa I raised this discrepancy to the team so it can be fixed. The $0.30/mo on the do...

Evil Morty•6/19/24, 12:54 PM

Hello, is this another typo ?

EEvil Morty Hello, is this another typo ?

Erisa•6/19/24, 1:08 PM

Seems so, I've raised that as well

Nicolas•6/19/24, 4:09 PM

Hey all, does someone know if it's possible to use a more advanced http client than the provided

fetch

fetch

? I would like more control on timeouts and ssl which could be done with access to undiciundici or node:httpnode:http but these two don't work on workers

Nicolas•6/19/24, 4:17 PM

The only alternative I see is to wait for Python packages deployment to be available and use httpxhttpx or aiohttpaiohttp, but something in js should surely be available

NNicolas Hey all, does someone know if it's possible to use a more advanced http client t...

Chaika•6/19/24, 4:21 PM

fetch goes out through the cdn request pipeline, you'd have to use the tcp socket support https://developers.cloudflare.com/workers/runtime-apis/tcp-sockets/ with some library to do completely arbitrary ssl/timeout stuff. There was work being done on making node networking/sockets work iirc, just slow
fetch() = through cdn, cloudflare.com/ips
TCP Socket = using WARP IPs (unlisted), arbitrary, can't connect to Cloudflare IPs though

NNicolas The only alternative I see is to wait for Python packages deployment to be avail...

Chaika•6/19/24, 4:21 PM

httpx/aiohttp are just patched to normal js fetch https://github.com/cloudflare/workerd/blob/main/src/pyodide/internal/patches/aiohttp.py, would have same limitations with ssl/timeouts

Nesk•6/19/24, 4:45 PM

Hi everyone, is there any way to use the Rate Limiting API with a custom duration period? The documentation says:

Period: the duration of the period, in seconds. Must be either 10 or 60

Is this some kind of limitation during the beta? I would like to use a 1 hour duration period, or more.

CChaika httpx/aiohttp are just patched to normal js fetch <https://github.com/cloudflare...

Nicolas•6/19/24, 4:50 PM

@Chaika thank you for the details that's interesting

Original message was deleted

Nesk•6/19/24, 4:57 PM

Yes, as in “yes, you can use a custom period duration”?
Or yes, as in “yes, this is a beta limitation”?

Nesk•6/19/24, 4:58 PM

thank you for your help

cron•6/19/24, 6:36 PM

possibly a silly question - i have a worker that i created 5 months ago and it defaults to a file named index.js for fetch

but i created a worker now, which defaults to worker.js for fetch calls, even if i try to change it to index.js - has something changed with the archetype between the last five months, or did i use a different template while setting up and have forgotten about it?

cron•6/19/24, 6:39 PM

AH. I figured it out - I guess it defaults to wrangler projects now when you're trying to create a blank worker?

cron•6/19/24, 6:39 PM

has it always done that? i feel like i am having amnesia lol

madi. ™•6/19/24, 6:58 PM

i am crazy new to server hosting. i have a discord bot that is already coded up and i just need to upload it to cloudlfare, is there a way to do that..?

Mmadi. ™i am crazy new to server hosting. i have a discord bot that is already coded up ...

Hard@Work•6/19/24, 7:00 PM

What language is the bot written in? Does it use the Gateway, or just Slash Commands?

HHard@Work What language is the bot written in? Does it use the Gateway, or just Slash Comm...

madi. ™•6/19/24, 7:01 PM

i havent enabled slash commands, not sure about the gateway thing, im basically self hosting a log bot that recently went down. called Logger

madi. ™•6/19/24, 7:02 PM

written in javascript

Mmadi. ™i havent enabled slash commands, not sure about the gateway thing, im basically ...

Hard@Work•6/19/24, 7:03 PM

Yeah, I don't think that would be able to be run on any Cloudflare products at the moment. You would need a VPS or something like Fly/AWS Lambda(or another service that can run Node)

madi. ™•6/19/24, 7:03 PM

ah i see, ok thank you :)

Brett Willis•6/19/24, 9:11 PM

Hey, can anyone give a gauge of the hardware performance underpinning the Workers platform? Given that "CPU time" is a billed metric, and we have no gauge of how many ops/sec or something the isolate/thread can do, it is quite arbitrary and hard to gauge. Any hardware acceleration for crypto operations? I'm guessing the hardware is constantly evolving and different depending on the data centre... but perhaps someone can comment or improve my understanding...?

For example, verifying JWTs using the SubtleCrypto API is common on every request for an API, and also a computationally heavy operation. And just from experience I'm getting the feeling that the CPU available to Workers feels pretty slow compared to other "single vCPU" runtimes I've used... any insights?

Original message was deleted

Brett Willis•6/19/24, 9:31 PM

Ah thanks! I see in https://blog.cloudflare.com/cloudflare-gen-12-server-bigger-better-cooler-in-a-2u1n-form-factor their Gen 11 fleet was AMD Milan 7713, but doesn't look like they actually confirmed what their Gen 12 (2024) fleet is going to be...

Brett Willis•6/19/24, 9:39 PM

Also I've been finding that our Workers get evicted crazy fast. Like it gets evicted in under 5 seconds a lot of the time, and then does a cold start for a request to the same data centre 5 seconds later. Is that normal or what? On a cold start it also has to do a bit of CPU work like importing crypto keys etc.
The docs say that the

env

env

object can be equal by reference until the env variables or bindings change, so we cache our state on the env object but it gets evicted crazy fast.

Original message was deleted

Brett Willis•6/19/24, 9:42 PM

I've got one too that verifies a JWT and sends a response, and it tends to use upwards of 80ms CPU per request which feels like a lot. But I've got a feeling that a lot of that CPU is related to spinning up a cold environment as it goes down significantly when I ramp up the RPS to keep the instance alive.

Original message was deleted

Brett Willis•6/19/24, 9:44 PM

Using

jose

jose

Original message was deleted

Brett Willis•6/19/24, 9:46 PM

Ah I see, that is interesting... It felt natural because the application state is explicitly tied to the env config, and the docs says it's the same object reference between requests. If using global vars I'd have to reason about invalidating when env config changes.
Caching on env does seem to work, but it just seems to get evicted pretty promptly...

Original message was deleted

Brett Willis•6/19/24, 9:46 PM

Are there any docs about this env being a software update that causes eviction?

Brett Willis•6/19/24, 9:52 PM

Hmm ok. Well it's worth a try, I'll try caching on