Hi, does anyone knows how CPU is shared between isolates (workers) ? I am running an LZ based compr

Hi, does anyone knows how CPU is shared between isolates (workers) ?
I am running an LZ based compression algo in Rust compiled to WASM.
For the same file compressed I see high discrepancies in CPU time: most of the time it consumes about 1500ms of CPU time but sometime 2500 or 3000ms.
The worker is just calling that wasm instance, no I/O calls.

So it seems vCPU allocated to an isolate is not fixed. This seems unfair as we are billed by CPU time and sometimes we can be billed 2x,3x for the exact same compute (no I/O involved) depending on the vCPU allocated to our isolate.

Plus the fact that it adds latency to get the response, so we pay more for slower execution..

Mitya•3/5/25, 5:28 PM

CF devs may be interested to know that Cursor, the AI assistant, is giving 404 links to the docs. This is because the URLs it outputs are relative, not absolute, yet lack the prefix //, so they are served related to the base https://developers.cloudflrare.com/workers/ai/https://developers.cloudflrare.com/workers/ai/. This results in URLs like the below being generated:

https://developers.cloudflare.com/workers/ai/queues/configuration/javascript-apis/

https://developers.cloudflare.com/workers/ai/queues/configuration/javascript-apis/

Simon Colby•3/5/25, 7:49 PM

Hey guys, I have a problem that my production worker is much slower then staging one even though the only difference is the traffic (the code and configuration is exactly the same). Do you know how can I fix it?

Ttheodm Hi, does anyone knows how CPU is shared between isolates (workers) ? I am runni...

Keebs•3/5/25, 8:29 PM

feels like that is something that would be easier done external? :D

Keebs•3/5/25, 8:30 PM

like a server that is set up for compression - lz compiled with cpu flags and then its probably 10ms, you could transfer the data from worker to worker and compress in way less time then

Ttheodm Hi, does anyone knows how CPU is shared between isolates (workers) ? I am runni...

Isaac McFadyen•3/5/25, 8:33 PM

By definition CPU time is invariant of number of vCPUs. Even if two cores are allocated/working at the same time then there will be the same number of CPU (milli)seconds... but that can't happen in Workers because they're single-threaded anyway. Not sure what the reason for the variance you're seeing there is. How are you getting the specific CPU time metrics? GraphQL API?

KKeebs like a server that is set up for compression - lz compiled with cpu flags and th...

theodmOP•3/5/25, 8:37 PM

That what I currently do, those CPU heavy algos are running from GCP cloud run with allocated CPU. Waiting for cloudflare containers to see what they have to offer

Keebs•3/5/25, 8:38 PM

yeah id def outsource that part, workers always seems more of a platform for <100ms tasks :D

IIsaac McFadyen By definition CPU time is invariant of number of vCPUs. Even if two cores are al...

theodmOP•3/5/25, 8:41 PM

That is what i thought also, as CPU time should be invariant of vCPU allocation. But for those algos I see wide CPU time variations, seen from workers logs (beta) in CF dashboard.

aydrian•3/5/25, 9:41 PM

Hello. Looking to understand a little more about the lifecycle of a durable object. In the constructor of the durable object class, I want to create a store and then add a listener to it. What I'm trying to figure out is how to properly clean up the listener. It's my understanding that if the do hibernates and wakes back up, it will run the constructor again. Do I need to worry about removing that listener? Is there a method I can override to include code to remove the listener? Thanks in advance.

Aaydrian Hello. Looking to understand a little more about the lifecycle of a durable obje...

Isaac McFadyen•3/5/25, 10:03 PM

When a durable object hibernates, all memory is cleared. So in your case the store would vanish completely (and so it's fine to reattach listeners). Anything you want to persist across hibernations should be put in the DO persistent storage, either via the classic KV API or the new SQLite API.

IIsaac McFadyen When a durable object hibernates, all memory is cleared. So in your case the sto...

aydrian•3/5/25, 10:04 PM

Great. Thank you.

AGill•3/6/25, 5:58 AM

Hey peeps I'm encountering some mysterious behaviour. My Pages Function is throwing intermittent 401 errors. I traced the code back to some packages I'm using:

@propelauth/cloudflare-worker

@propelauth/cloudflare-worker

which calls

jose

jose

.

Debugging that shouldn't be a problem. The issue is that the 401 request that I receive in my browser is not reflected in the Real Time Logs stream.

It's mysterious because it's only ever this intermittent error that does not show up. Every other request I make I can see in the logs stream. I even deployed changes with dummy errors and they show up as expected.

It's also odd that a global try catch does not work around the auth code. There shouldn't be a code path for a 401 if I wrap my auth logic in a try/catch and throw a 418 for example.

App is a Nuxt 3.15.4 app on Cloudflare Pages. Appreciate any help.

Keebs•3/6/25, 11:39 AM

I think there was a similar error a week ago

Keebs•3/6/25, 11:40 AM

hmm or maybe not, can’t find it

Aalex (he/him)I really like the new option to start multiple workers with wrangler in the same...

Murder Chicken•3/6/25, 2:57 PM

Late to the party but where did you see this option mentioned?

jackobsn•3/6/25, 4:25 PM

Hello, i have some very basic questions about Pages , is this the place to ask ?

fredg8573•3/6/25, 5:50 PM

Hello everyone!

I'm excited to share an open source tool I've developed to solve a common challenge with Cloudflare Workers: Cloudflare Microservice Secrets Manager.

What is it?
A centralized microservice for managing environment variables and shared secrets across all Workers in a project. Particularly useful for microservice architectures!

Key features:

Centralized configuration for all your variables and secrets
Environment-specific management (dev/staging/prod)
Enhanced security with authentication and attack protection
Automated deployment via a single command

GitHub Repository: cloudflare-workers-env-sync

This project is completely open source under the MIT license. I hope it can be useful to other developers working with microservice architectures on Cloudflare Workers.

Feel free to try it out, open issues, or contribute if you have ideas for improvements! I'd love to hear your feedback.

A big thank you to the Cloudflare team for their amazing solutions that make our crazy projects possible!

#CloudflareWorkers #OpenSource #Microservices

GitHub

GitHub - Fr-e-d/cloudflare-workers-env-sync: # Cloudflare Microserv...

Cloudflare Microservice Secrets - A centralized microservice for managing environment variables and shared secrets across all Cloudflare Workers in a project. Simplify configuration with environm...

EinsiTang•3/7/25, 9:22 AM

hello , have someone here , I have same question about worker dev , fetch request always pending and I have no idea to fix it

ZachHandley•3/7/25, 5:57 PM

hey so I have a CNAME for www, and a page rule, and it still effing times out on www for some stupid reason

ZZachHandley hey so I have a CNAME for www, and a page rule, and it still effing times out on...

Hard@Work•3/7/25, 5:58 PM

To confirm, you want it to redirect from

www.zach.handley

www.zach.handley

zach.handley

zach.handley

(assume the domain is your actual domain)?

ZachHandley•3/7/25, 5:59 PM

yeah so jonathansogunquit.com is my step fathers site for his restaurant I made, and www.jonathansogunquit.com times out with a 522 for some reason (immediately, might I add) despite having a CNAME and a page rule

ZachHandley•3/7/25, 6:00 PM

ZachHandley•3/7/25, 6:01 PM

idk why the hell it's giving a timeout error, I think that's the most upsetting right now, because first of all wrong, it's not timing out, cloudflare is just wrong, and second, it makes ME look bad for cloudflare's error

ZachHandley•3/7/25, 6:01 PM

nothing in the function logs either when I live stream it and try to access

ZachHandley•3/7/25, 6:02 PM

on my browser it works fine, on mobile it times out instantly. don't know why on that front either.

ZachHandley•3/7/25, 6:06 PM

mmkay, well, in order for mobile to not timeout, you have to add the www as a custom domain in addition to the other. You can't use a CNAME by default like every other. that should not be the case IMO.

ZZachHandley mmkay, well, in order for mobile to not timeout, you have to add the www as a cu...

Hard@Work•3/7/25, 6:06 PM

Have you followed this tutorial? https://developers.cloudflare.com/pages/how-to/www-redirect/

Cloudflare Docs

Redirecting www to domain apex · Cloudflare Pages docs

Learn how to redirect a www subdomain to your apex domain (example.com).

ZachHandley•3/7/25, 6:19 PM

yeah so I used pages

ZachHandley•3/7/25, 6:19 PM

and added a custom domain

ZachHandley•3/7/25, 6:20 PM

and then added a

www

www

-> apex domain, but it needed to be

www

www

-> cloudflare pages domain

ZachHandley•3/7/25, 6:20 PM

imo that should be offered when adding a domain, by default, because that was awful and made me look like an idiot, when normally that's what I'd do and then handle it via the Nginx

ZachHandley•3/7/25, 6:21 PM

maybe I am one, but I'd love a little bit of idiot-proof-added if possible, not that it really matters anymore haha

ZZachHandley and then added a `www` -> apex domain, but it needed to be `www` -> cloudflare p...

Hard@Work•3/7/25, 6:32 PM

I'm confused, redirecting

www

www

-> apex doesn't directly involve Pages at all. You just need a dummy record, and a Redirect Rulw

ZachHandley•3/7/25, 7:25 PM

the www needs to be pointed to the pages domain

ZachHandley•3/7/25, 7:25 PM

and old people still use it

ZachHandley•3/7/25, 7:25 PM

so to me that's either something that should be put in as an option when adding a custom domain or

ZZachHandley the www needs to be pointed to the pages domain

Hard@Work•3/7/25, 8:58 PM

Ok, I'm even more confused now. If I visit

www

www

, should I be redirected to the apex? What if I visit the Apex? Should I be redirected to

www

www

? Should both be serving the same content, without any redirects?

Kaua•3/8/25, 5:51 AM

Very simple question: can I just straight up upload a node.js website to cloudflare workers or do I have to change something?

KKaua Very simple question: can I just straight up upload a node.js website to cloudfl...

Brett Willis•3/8/25, 6:56 AM

You'll almost certainly have to change something. At the very least it needs to be bundled into a single JS file which exports a fetch handler: https://developers.cloudflare.com/workers/runtime-apis/handlers/fetch.

Beyond that, it depends entirely on the implementation of your site. Workers supports many but not all Node.JS APIs. I'm guessing your Node.JS site uses

fs

fs

to serve files, which will not work in Workers.

BBrett Willis You'll almost certainly have to change something. At the very least it needs to ...

Kaua•3/8/25, 7:10 AM

It's a very simple app, a budget manager. If by fs you mean the website has file upload/download support, the only file related option there is a .csv export of the table data.

husen•3/8/25, 12:51 PM

I just upgraded the zola version to 0.20.0 but getting the error as in the picture

husen•3/8/25, 12:51 PM

husen•3/8/25, 12:53 PM

I am using Cloudflare pages build system 1

husen•3/8/25, 1:00 PM

Oh I forgot, Happy Weekend! Don't forget to take some time for yourself

xeon06•3/8/25, 5:01 PM

Has anyone ever attempted to serve hashed assets from previous deployments in the current deployment? As an idea to fix issues where a dynamic route's JS file doesn't exist anymore at the main domain after a deployment, but where a user might already have the website open

Walshy•3/8/25, 5:19 PM

As part of Pages or Workers Assets?

xeon06•3/8/25, 5:19 PM

Pages, which seems to be the issue

Walshy•3/8/25, 5:19 PM

Pages, you cannot
Workers Assets, you'd need to run the Worker first and use

env.ASSETS.unstable_getByETag()

env.ASSETS.unstable_getByETag()

xeon06•3/8/25, 5:19 PM

I can omit the static assets from _routes.json for my worker to hit it, but then I don't see a way to then serve it

xeon06•3/8/25, 5:19 PM

Gotcha. Sounds like a better idea might be to try tweaking the build so it uses an absolute URL to the deployed version for those assets

Hi, does anyone knows how CPU is shared between isolates (workers) ? I am running an LZ based compr

Cloudflare Microservice Secrets - A centralized microservice for managing environment variables and shared secrets across all Cloudflare Workers in a project. Simplify configuration with environm...

Similar Threads

Similar Threads

Cloudflare Microservice Secrets - A centralized microservice for managing environment variables and shared secrets across all Cloudflare Workers in a project. Simplify configuration with environm...

Similar Threads