Remote Bindings - Queues - Does not work

[astanciu]

What versions & operating system are you using?

wrangler 4.24.3, node 24

Please provide a link to a minimal reproduction

https://github.com/astanciu/cf_worker_repro

Describe the Bug

Created a producer and consumer queue, added the remote binding configurations and started both with the new --x-remote-bindings, but the consumer does not receive any messages.

In the repro repo, start each worker separately:

• cd apps/producer && pnpm run dev:rb
• cd apps/consumer && pnpm run dev:rb

Please provide any relevant error logs

No response

[dario-piotrowicz]

Thank you very much for the issue @astanciu 🙏

(Also thanks so much for the minimal reproduction, I cannot stress enough how much that is appreciated and helpful! 🫶)

The problem you're running into is caused by the fact that the consumer runs locally but the producer is using the remote queue binding. This means that the producer sends messages to the remote queue but the consumer instead tries to consume messages in the local queue.

The solution here I think would need to be to run the consumer remotely via wrangler dev --remote, the issue with that is that queues aren't actually supported there:

So that's also not going to work 😕

Anyways

wrangler dev -c apps/producer/wrangler.jsonc -c apps/consumer/wrangler.jsonc --x-remote-bindings

seems to work well here, right? is this approach not good for you?

[astanciu]

Hi @dario-piotrowicz.

Unfortunately, the -c -c solution does not work. I actually tried to open a related issue here, #9795, but it was sadly closed.

Ignoring the fact that the -c -c solution is marked as experimental in the docs, it simply does not work when you have workers using Vite, or that need to be individually addressed.

Here's a simple example:
Worker A, front end app with Vite (for ex, main app dashboard)
Worker B, front end app with Vite (for ex, auth service that hosts its own UI for login pages)
Worker C, queue consumer.

Worker A and B publish to the same queue, worker C consumes.

After the other issue was closed, I was really hoping that remote bindings would allow me to run queues in a dev env, but seems like that's not the case either.

The "-c -c" solution behaves very differently than production, and just does not work in cases such as the above.

It seems like Cloudflare overall is favoring running workers with multiple dev commands. RPC (service bindings) work fine between workers running separately, and there seems to have been some investment into making that work, with the local worker dev registry.
It seems so natural to expect queues to work locally in this manner.

Why won't Cloudflare add support for this? I don't think this use case is an outlier.

TLDR: Multiple dev commands is a requirement, not optional, in some cases. And queues won't work in that scenario, locally or with remote bindings.

[astanciu]

@penalosa should we close this issue in favor of #9795 ?

[aicushman]

To tag onto this -- as far as I can tell, the -c -c pattern doesn't support the --experimental-vectorize-bind-to-prod flag.

Combined with queues only working using the -c -c pattern, this makes it impossible to locally test workers that vectorize from a queue.

Unless I'm mistaken about one or the other of those limitations, it does rise to the level of 'bug' for my purposes.