You cannot upload a compressed Worker bundle today, what compression were you looking to use and why

You cannot upload a compressed Worker bundle today, what compression were you looking to use and why?

WWalshy You cannot upload a compressed Worker bundle today, what compression were you lo...

lusons•3/11/25, 3:28 PM

Thanks for your reply, I'm not sure if my idea is correct. In my scenario, I want to deploy the html + otherAssets + userWorker.js produced by the saas service to a worker, and then distribute it by dispatcher (similar to the publishing function of framer.com), but currently I can only use base64 for network transmission, which will be very slow. Or do I have other ways?

WalshyOP•3/11/25, 3:29 PM

using Workers Assets yeah it'd be base64 but that shouldn't be slow unless you're uploading a lot of data, for almost all sites it should take almost no time at all. We recommend concurrent uploads of these asset buckets

nclevenger•3/11/25, 3:34 PM

can you deploy the Agent SDK on Workers for Platforms?

AAhmad Awais How do y’all complie, pkg, and bundle the code meant to be deployed via workers ...

nclevenger•3/12/25, 8:28 AM

You can use wrangler to deploy to WfP, but it's also fairly straightforward to run ESBuild even on workers, including even bundling dependencies from http url imports ... there's a simple example of this w/ API at https://esb.denoflare.dev for example

esbuild as an edge service

We run esbuild so you don't have to.

rob•3/12/25, 3:04 PM

are you letting you users build via. CLI tool or ?

rob•3/12/25, 3:04 PM

Or how are they pushng the code up to your platform

AAhmad Awais Yes we can use wrangler but how? It’s users code coming to us. So code isn’t ava...

Isaac McFadyen•3/12/25, 3:06 PM

Workers for Platforms isn't meant as a fully-packaged, batteries-included solution. The idea with WfP is you upload compiled Workers to the platform which are then run, it doesn't include CI/CD because that's not the goal of the platform, really.

Isaac McFadyen•3/12/25, 3:07 PM

The expectation is that customers bring their own build system.

AAhmad Awais What am I missing here? How can anyone use WfP today in the world of TypeScript ...

DaniFoldi•3/12/25, 4:11 PM

Very easy to use in my opinion. I also think that the "open" approach of WfP is great, I can "bring my own" way of building workers - and ours is pretty unconventional, we generate the API within another worker, in one case. In another, we use fly.io, it took no more than a day to set up esbuild bundling

rob•3/13/25, 9:03 PM

how are you receving their code

Ashar Irfan•3/13/25, 9:46 PM

We're receiving their code in a CF worker

cosbgn•3/14/25, 8:57 AM

import * as esbuild from "esbuild-wasm";

await esbuild.initialize({
    worker: true,
    wasmURL: "https://unpkg.com/esbuild-wasm/esbuild.wasm",
});

const result = await esbuild.transform(code, options);

import * as esbuild from "esbuild-wasm";

await esbuild.initialize({
    worker: true,
    wasmURL: "https://unpkg.com/esbuild-wasm/esbuild.wasm",
});

const result = await esbuild.transform(code, options);

cosbgn•3/14/25, 8:57 AM

This is what you need, using eabuild wasm you can compile even client side before sending it to the worker

cosbgn•3/14/25, 8:58 AM

Or install it on a worker but it's fairly big

WalshyOP•3/14/25, 12:10 PM

It will not work for WfP, we recommend setting up a Tail Worker and having your user Workers send to that tail Worker
https://developers.cloudflare.com/workers/observability/logs/tail-workers/

spider•3/14/25, 10:13 PM

is there a local dev updated example (https://developers.cloudflare.com/cloudflare-for-platforms/workers-for-platforms/get-started/developing-with-wrangler/ doesn't work (ts or js)?
Is there typescript support or should everything be js?

Cloudflare Docs

Local development · Cloudflare for Platforms docs

To test your Dispatch Worker, user Worker and Outbound Worker before deploying to production, you can use Wrangler for development and testing.

darkpool•3/14/25, 11:10 PM

I am trying to give user workers accesses to RPC methods on "regular" worker via service bindings that make external API calls. It works locally when I call dispatch using --remote but not on a deployed dispatch worker. :/ I get this error:
Error: internal error; reference = v559jmcp0ee9g1mmqdamenrl
Error: internal error; reference = v559jmcp0ee9g1mmqdamenrl
?

It also works if the RPC method is something simple like reducing an array of numbers. The flow is: dispatch -> user worker (bound to regular worker) -> regular worker

EDIT: got it working by using this.env instead of accessing env from rpc method. Not sure why it worked with --remote flag though

Andrew0091•3/18/25, 4:07 PM

Whats the best route to take to build PR previews? I already have it set to rebuild on a merge to main, but when a dev puts in a PR I want a preview build to be generated

AAndrew0091 Whats the best route to take to build PR previews? I already have it set to rebu...

Hard@Work•3/18/25, 4:12 PM

For Workers for Platforms? I don't think it does CI at all

ajmallesh•3/19/25, 12:52 AM

Hey Cloudflare community! Need help with a very strange Workers routing issue.

We have two environments with two entirely separate domains - Playground (acmepg.com) and Sandbox (acmesbx.com)

Issue: Requests to

*.acmepg.com

*.acmepg.com

(playground env) are sometimes handled by our sandbox Worker instead of our playground Worker, even though are fully isolated in separate wrangler environments with separate domains with their own dns config.

It's intermittent but happens frequently - which makes it very confusing. For example sometimes on my laptop it routes correctly, but on my colleague's laptop it routes incorrectly and vice versa.

Setup:

Two Workers:

acme-play-router

acme-play-router

for playground,

acme-sbx-router

acme-sbx-router

for sandbox

Routes config (wrangler.toml) (note: we have workers.dev disabled and using Routes instead of custom domain):
```toml
[env.sandbox]
name = "acme-sbx-router"
routes = [{ pattern = ".acmesbx.com/", zone_name = "acmesbx.com" }]

[env.playground]
name = "acme-play-router"
routes = [{ pattern = ".acmepg.com/", zone_name = "acmepg.com" }]

- DNS config for each of our domains:

- DNS config for each of our domains:

acmesbx.com Type=CNAME, Name=, Conent=acmesbx.com
acmepg.com Type=CNAME, Name=, Conent=acmepg.com

The sandbox Worker correctly returns 403 when it sees playground domains (using Hono framework):

The sandbox Worker correctly returns 403 when it sees playground domains (using Hono framework):

// Referer validation in our middleware
if (referer) {
const refererUrl = new URL(referer);
const refererHost = refererUrl.hostname.toLowerCase();
const isValidReferer = refererHost.endsWith(baseDomain);

if (!isValidReferer) {
console.log(

❌ Referer validation failed: ${refererHost} does not match ${baseDomain}

❌ Referer validation failed: ${refererHost} does not match ${baseDomain}

);
return c.text(

Forbidden: Invalid referer

Forbidden: Invalid referer

, 403);
}
}
```

But why is the sandbox Worker handling these requests at all? The domains are completely different.

Any ideas what could cause this cross-environment routing issue? Missing something in our route definitions/wrangler config or DNS config?

Ashar Irfan•3/19/25, 11:16 PM

Does anyone know if Cloudflare has deployed this feature to Workers for Platform? Was looking forward to make this available via the compatibility flag to our customers today, but it is not working in my initial testing.

https://developers.cloudflare.com/changelog/2025-03-11-process-env-support/

Cloudflare Docs

Access your Worker's environment variables from process.env

WalshyOP•3/20/25, 2:03 AM

Yes this should work for wfp too

Palash Bansal•3/24/25, 7:33 PM

Hello, I am using the upload worker module API to deploy workers and to update bindings. It works for normal bindings, but for secrets, its not deleting the values that are not being passed in a request metadata bindings. The existing secrets are being kept even though not being passed in the bindings array.
This makes sense as you might not want to remove the secrets, but quickly checking the network calls in cloudflare dashboard, looks like on secret delete(single/multiple), it does the similar thing(create a new worker version with script code and metadata with bindings), and secrets are being deleted (unless I am missing some request). Only difference is that I am using "Workers Script Upload API" and its using "Version Upload API"

I am looking for some documentation regarding this on what the expected behaviour is with regards to secrets and other bindings. There is also

keep_bindings

keep_bindings

option in the types, but that doesn't seem to make a difference.
What I have found yet is this - https://developers.cloudflare.com/workers/configuration/multipart-upload-metadata/#bindings
Thanks

WalshyOP•3/25/25, 9:54 AM

You can delete a secret with: https://developers.cloudflare.com/api/resources/workers_for_platforms/subresources/dispatch/subresources/namespaces/subresources/scripts/subresources/secrets/methods/delete/

WalshyOP•3/25/25, 9:54 AM

Secrets inherit by default

WWalshy You can delete a secret with: <https://developers.cloudflare.com/api/resources/w...

Palash Bansal•3/25/25, 4:24 PM

Hey thanks for replying, but this only allows to delete one secret at a time, which is not ideal as it requires deploying several worker versions which is not atomic and some rollback flow would be required incase one of the request failed.
We are giving users option to edit and remove secrets similar to the cloudflare dashboard, where they can perform edits and commit to the worker. So is it possible to do it in a single call?
Also is the functional difference(relating to secret inheritance) I am seeing between the version and upload api an actual api spec difference, or I am doing something incorrectly?
Or is there a "type" value like "inherit" which can be used to delete a secret?

Palash Bansal•3/25/25, 4:30 PM

Another question I have related to bindings in workers for platforms is How do access control for bindings work?
If we try to bind a D1 database or R2 bucket to a worker in user space, would that work? Does it require additional setting/configuration? Or is that not possible and we need to use SQLite-in-DO for that?
The docs do mention -

You can use any Workers bindings with the dynamic dispatch Worker or any user Workers.

You can use any Workers bindings with the dynamic dispatch Worker or any user Workers.

, but then how does it control which bindings it can bind to? If it doesn't then is it on the developers to ensure that the users are not able to edit the bindings as then they can get access to any resource on the main CF account

Original message was deleted

Palash Bansal•3/25/25, 5:01 PM

Can we assume this won't change in the future and bindings made to the user workers now will keep the access forever?

rob•3/27/25, 2:48 AM

Whats the limit on WfP for SQLite DB on DO? is it still 50 per account ?

Vvmg I'm trying to work with local development mode and get the following error: `[...

daniel•3/30/25, 7:35 PM

did u find a fix? i'm experiencing the same thing

Ddaniel did u find a fix? i'm experiencing the same thing

daniel•3/30/25, 9:07 PM

yeah looks like local dev is too outdated and most of it is broken, --remote with latest wrangler works perfectly though

Alex99like•3/31/25, 7:30 AM

Hey folks!

Need help! Getting a strange error in a Cloudflare Worker in production

I’m seeing the following error on prod:

sync error RuntimeError: memory access out of bounds

What’s confusing is that I’m not using any WebAssembly (WASM) modules in my project. It’s all plain TypeScript and regular dependencies.
Has anyone run into something like this? Could it be that one of the packages secretly uses WASM under the hood, or is this something else entirely?
Would really appreciate any ideas or pointers

AAlex99like Hey folks! ⚠️ Need help! Getting a strange error in a Cloudflare Worker in produ...

DaniFoldi•3/31/25, 10:07 AM

likely one of the packages is using wasm under the hood - can you share your list of dependencies?

DDaniFoldi likely one of the packages is using wasm under the hood - can you share your lis...

Alex99like•3/31/25, 10:13 AM

The worker is a member of the orchestra and uses hono to scale the hono.js prisma it has no more dependencies

Alex99like•3/31/25, 10:15 AM

could there be a problem that one of the databases starts to degrade and the query time increases in the worker crashes, or is it the worker's problem?

DaniFoldi•3/31/25, 10:17 AM

hmm, do you happen to be using prisma's driver adapters?

Alex99like•3/31/25, 10:18 AM

Yes, I use planetscale

DaniFoldi•3/31/25, 10:18 AM

we've also hit this issue, and so have many others, see: https://github.com/prisma/prisma/issues/25714

unfortunately, we ended up having to switch to drizzle which has worked perfectly from the first moment - so that's the only thing I can comfortably recommend

GitHub

Possible memory leak using driver adapters with cloudflare D1 · Is...

Bug description Around the 16th november we started to get errors when making prisma calls in our production prisma/cloudflare workers/d1 setup. The errors consisted of either: "memory access ...

DDaniFoldi we've also hit this issue, and so have many others, see: https://github.com/pris...

Alex99like•3/31/25, 10:20 AM

The problem occurs under heavy load when there is a simultaneous request for 500 records. The worker is supposed to fetch data from the primary database and move it to the D1 replica. This issue does not occur when requesting fewer than 500 records.

Alex99like•3/31/25, 10:22 AM

or maybe the problem is that I access one worker 500 requests at the same time through bindings, does it have limits?

AAlex99like or maybe the problem is that I access one worker 500 requests at the same time t...

DaniFoldi•4/1/25, 12:36 PM

I don't think there is a limit for that

DaniFoldi•4/1/25, 12:36 PM

likely it's just the prisma driver code leaking memory :/

import * as esbuild from "esbuild-wasm"; await esbuild.initialize({ worker: true, wasmURL: "https://unpkg.com/esbuild-wasm/esbuild.wasm", }); const result = await esbuild.transform(code, options);

You cannot upload a compressed Worker bundle today, what compression were you looking to use and why

Similar Threads

Similar Threads

Similar Threads