How to make queues pick up messages

How to make queues pick up messages faster?

Currently I see a few seconds lag from message being inserted in queue, vs being picked up.
This is critical for UX. I want to reduce this time. Will it be better if I shift to a workflow ?

bert•10/1/25, 8:12 PM

I've noticed this as well - I've set max_batch_sizemax_batch_size to 1 and max_batch_timeoutmax_batch_timeout to 0 in my consumer Worker and I'm still seeing 10-15 seconds of delay between writing and consuming it.

switz•10/1/25, 9:59 PM

I see <= 6s using http pull

switz•10/1/25, 9:59 PM

Oct 1 21:58:26 -> 17:58:32

switz•10/1/25, 10:00 PM

yeah next one was 56->59. basically instant (clock drift, polling, etc.) probably closer to 1s realtime

switz•10/1/25, 10:03 PM

that's e2e too: so going from a remote server halfway across the world -> worker -> queue -> my server.

Tero•10/2/25, 5:13 AM

It has been discussed earlier in the channel, and I think the conclusion was that currently queues are not well suited for low latency use cases. I think there might be some improvements coming in this area, but no clue when that might happen.

ShivekOP•10/2/25, 6:50 AM

Did anyone here try workflows instead ?

I can have a one step workflow as I need to maintain object state (cannot be durable).

The 4 second pause is killing my UX.

Bbert I've noticed this as well - I've set `max_batch_size` to 1 and `max_batch_timeou...

ShivekOP•10/2/25, 6:52 AM

Max batch timeout doesn't make sense if max_batch_size is 1.

As soon as there is 1 item in the queue, the consumer will be triggered.

bert•10/2/25, 11:25 AM

Correct. I started with just max_batch_sizemax_batch_size, it didn't really work, and I added timeout just in case.

daisy•10/3/25, 5:51 PM

What would be the lower latency alternative to queue?

daisy•10/3/25, 7:40 PM

did you guys notice that there’s a clear delay whenever the queue consumer scales up and down horizontally? It sometimes take more than a minute to pick up a messages while it is scaling. Can we scale proactively?

bert•10/3/25, 11:20 PM

I'm now seeing a >10-min delay in my queue. :sad_frog: I wonder if it's related to the workers incident.

Bbert I'm now seeing a >10-min delay in my queue. :sad_frog: I wonder if it's related ...

ShivekOP•10/5/25, 10:22 AM

I'll be fired if this happens to me :/

Ddaisy What would be the lower latency alternative to queue?

ShivekOP•10/5/25, 10:25 AM

Have you tried single step workflows ?

There is a limit to number of workflows running too. But I need to check it out.

The biggest problem is that the CF ENV and Dev env don't match in terms of speed.

daisy•10/5/25, 10:40 AM

We’re getting higher than 10QPS traffic, which workflow doesn’t allow ._.

Ddaisy What would be the lower latency alternative to queue?

James•10/6/25, 5:22 PM

fire and forget another worker, if you need resiliency, youll need to roll your own guaranteed delivery/completion logic.

daisy•10/7/25, 2:24 PM

I wasn't able to keep it running when client is disconnected for worker-worker api call. (as written in cloudflare's lifecycle document)

James•10/7/25, 2:59 PM

theres a ctx waituntil method you can use to keep it running without the requestor

Sid•10/8/25, 9:51 AM

Hey all, I need a few queue IDs please to check this out. It really should not be taking a minute (10 minutes is absurd) to have messages “show up” in the queue

Sid•10/8/25, 9:53 AM

While I do agree that to get low E2E latencies you maybe should not be using queues, I generally think of it as < 2-3 seconds

SSid While I do agree that to get low E2E latencies you maybe should not be using que...

ShivekOP•10/10/25, 10:25 AM

For my usecase, I'm noticing under 5 seconds consistently. Mostly 2-3 seconds.

daisy•10/11/25, 11:49 AM

Here are my queue Ids:

d328bc53d18a492ca776d17a7036b33d
d37decf634404b038672c44a41e9abbe
d167eefc2ba44f0984c69ade3ede2ac7

Ephemeros•10/18/25, 9:32 PM

Hello, i noticed that if i set max_batch_timeout to be 0 it resets to 5 in the ui? if i set to 50 it stays as 50, so I wonder if it somehow thinks 0 is invalid and resets to 5 or something?

Ephemeros•10/18/25, 9:34 PM

yeah lol if i make it 0.001 it works, so i guess 0.0 is seen as "undefined" maybe like

==

==

comparison and then it resets to 5 as default...

Ephemeros•10/18/25, 9:49 PM

https://cf-web-app.firtina3457.workers.dev/queue here's a test

https://github.com/firtoz/cf-multiworker-boilerplate and this is the repo

agreed on these:

setting as 0 makes it 5
setting as 0.01 works but not 0.001 through wrangler.jsonc
cannot set to 0.01 from the interface, only wrangler jsonc or cli
if you set to 0.01 it still takes between 2 and 5 seconds ish, minimum was 1 second for one try only

GitHub

GitHub - firtoz/cf-multiworker-boilerplate: A modern, production-re...

A modern, production-ready boilerplate for building full-stack applications with Cloudflare Workers and Durable Objects - firtoz/cf-multiworker-boilerplate

Ephemeros•10/18/25, 10:01 PM

@Sid | Queues, R2 apologies for the ping, I hope the above helps

Ssammyboy This is super cool. We have a very similar monorepo setup for our applications, ...

Ephemeros•10/21/25, 3:26 PM

Go for it :D it's a fun experiment, and I hope it's useful!