How do I ensure that a request is routed to the lowest latency zone, even if it means spinning up a

How do I ensure that a request is routed to the lowest latency zone, even if it means spinning up a new container? My application is extremely latency sensitive.

I set the max instances to 10, and then hit the workers.dev endpoint for my container deployment.

I had someone from the UK hit that same endpoint, and it didn't spin up a nearer instance.

FFrozenFire How do I ensure that a request is routed to the lowest latency zone, even if it ...

Chaika•7/24/25, 10:53 PM

Right now the routing just isn't very latency/geographic aware (Edit: well, your Durable Objects it spins up with random IDs for the getRandom/load balancing don't move with sort of pre-determine your location pools and mess with overall routing). Something they're working on: https://developers.cloudflare.com/containers/scaling-and-routing/#autoscaling-and-routing-unreleased
As for right now, "load balancing" is done via getRandom/just grabbing a random container id, which will create an object near the requester and container hopefully close as well. They never move though and sometimes the placement of the container instance themselves isn't great

FrozenFireOP•7/24/25, 10:56 PM

Huh. That's rough. Are there any strategies for dealing with this? I'm primarily interested in CloudFlare because I need low-latency placement.

FrozenFireOP•7/24/25, 11:01 PM

https://developers.cloudflare.com/containers/scaling-and-routing/#scaling-container-instances-with-get

I wonder if using env.container.get(<region>) would be effective-ish.

Cloudflare Docs

Scaling and Routing

Currently, Containers are only scaled manually by calling BINDING.get() with a unique ID, then
starting the container. Unless manualStart is set to true on the Container class, each
instance will start when get() is called.

FrozenFireOP•7/24/25, 11:40 PM

This sort of seems to work

export default {
  async fetch(request, env) {
    let container_id = request.cf?.colo || request.cf?.region || "default";
    return getContainer(env.MY_CONTAINER, container_id).fetch(request);
  },
};

export default {
  async fetch(request, env) {
    let container_id = request.cf?.colo || request.cf?.region || "default";
    return getContainer(env.MY_CONTAINER, container_id).fetch(request);
  },
};

spike•7/25/25, 6:34 AM

so onStart()onStart() and the other lifecycle hooks only fire when you use startAndWaitForPorts()startAndWaitForPorts() and onStart fires when the container is started or the container and the port is ready?

Sspike so `onStart()` and the other lifecycle hooks only fire when you use `startAndWai...

John Spurlock•7/25/25, 3:40 PM

the Container helper class is open source here: https://github.com/cloudflare/containers/blob/main/src/lib/container.ts

GitHub

containers/src/lib/container.ts at main · cloudflare/containers

Contribute to cloudflare/containers development by creating an account on GitHub.

FrozenFireOP•7/25/25, 11:06 PM

https://developers.cloudflare.com/containers/examples/env-vars-and-secrets/#using-envvars-on-the-container-class

I get env is not definedenv is not defined in the worker logs when I use this

Cloudflare Docs

Env Vars and Secrets

Pass in environment variables and secrets to your container

FrozenFireOP•7/25/25, 11:09 PM

I am needing to map a secret defined in the default(?) secret store to a specific environment variable name.

itisyev•7/26/25, 1:27 AM

Is there a way to do TCP communication with a container or would it have to be wrapped with a websocket and following the pattern described here https://developers.cloudflare.com/containers/examples/websocket/

itisyev•7/26/25, 1:29 AM

Last resort would be wrapping the functionalities of interest with an HTTP API but tryna do little as possible lol

itisyev•7/26/25, 1:34 AM

According to this, it looks like websockets are generally not friendly with workers/containers https://github.com/cloudflare/workerd/issues/2319

GitHub

WebSocket Serialization Error in Durable Objects RPC · Issue #2319...

When attempting to use DurableObjects RPC with a WebSocket, an error is encountered during serialization. The error message indicates that objects of type WebSocket cannot be serialized. This issue...

Iitisyev According to this, it looks like websockets are generally not friendly with work...

Chaika•7/26/25, 1:36 AM

use fetch not containerFetch, as per: https://developers.cloudflare.com/containers/examples/websocket/

Cloudflare Docs

Websocket to Container

Forwarding a Websocket request to a Container

Chaika•7/26/25, 1:37 AM

containerFetch uses RPC which would hit that. Fetch passes it without RPC which should be fine. Ignore the bottom bit of that doc...

CChaika use fetch not containerFetch, as per: https://developers.cloudflare.com/containe...

itisyev•7/26/25, 1:38 AM

So if I have a TCP endpoint listening on a container (with defaultPort set to the one to be listening at) and wrap in a Worker with container.fetchcontainer.fetch that should do the trick?

Iitisyev So if I have a TCP endpoint listening on a container (with defaultPort set to th...

Chaika•7/26/25, 1:40 AM

In theory. I haven't tried websockets before w/ containers, all I know for sure is avoid rpc/containerFetch to avoid the seralization issues you linked

CChaika In theory. I haven't tried websockets before w/ containers, all I know for sure ...

itisyev•7/26/25, 1:44 AM

I've tried a few instances of switchPort and container.fetch with little success but will see if doing defaultPort helps. Thanks!

Iitisyev I've tried a few instances of switchPort and container.fetch with little success...

Chaika•7/26/25, 1:47 AM

My understanding is you should only have to use switchPort if the default port isn't set

defaultPort?: Optional default port to use when communicating with the container. If not set, you must specify port in containerFetch calls, or use switchPort.

So if not using containerFetch (and not using default port), use normal fetch like:

container.fetch(switchPort(request, port))

As per README.md of the containers lib

**Do not use 'containerFetch' when trying to send a Request object with a websocket, until this issue is addressed. You can overcome this limitation by doing: container.fetch(switchPort(request, port))

https://github.com/cloudflare/containers/tree/main

Alexander•7/26/25, 4:15 PM

Anyone else hitting

AssertionError: Build ID should be set if containers are defined and enabled

AssertionError: Build ID should be set if containers are defined and enabled

when running integration tests with containers enabled?

�

😈 Donkey 💫•7/27/25, 11:14 AM

My container cannot complete any request, but gets stuck at the middle of request and returns error: Container suddenly disconnected. Try again. Any one encountered the same?

�

😈 Donkey 💫•7/27/25, 11:15 AM

metrics show that it never reach the limit of memory/cpu

Ghost Furry•7/27/25, 5:25 PM

Does uploading to R2 count as egress?

GGhost Furry Does uploading to R2 count as egress?

Hello, I’m Allie!•7/27/25, 5:30 PM

If you are doing so from a cloud that charges egress, yes

Hello, I’m Allie!•7/27/25, 5:30 PM

We won’t bill you for it, but they might

HHello, I’m Allie!If you are doing so from a cloud that charges egress, yes

Ghost Furry•7/27/25, 5:32 PM

From where i will bet object is different things

Take a example

I got the object file in container

Then upload that to R2 like we do using workers by binding

Then it will charge for uploading to R2 ?

GGhost Furry From where i will bet object is different things Take a example I got the ob...

Hello, I’m Allie!•7/27/25, 5:37 PM

You will be charged a single Class A operation(assuming you don’t use a Mulitpart Upload). R2 doesn’t bill by the gigabyte for uploads, just for storage and Infrequent Access Retrieval

HHello, I’m Allie!You will be charged a single Class A operation(assuming you don’t use a Mulitpar...

Ghost Furry•7/27/25, 5:38 PM

Yeah that i get but talking about egress charges in Cloudflare container

So putting file in r2 doesn't count as leaving out the network

GGhost Furry Yeah that i get but talking about egress charges in Cloudflare container So pu...

Hello, I’m Allie!•7/27/25, 5:39 PM

Sorry about that, I completely missed which channel we are in. R2 does not bill for Ingress, but Containers does bill requests to R2 for egress

1•7/28/25, 6:52 AM

Would containerFetch be the reason why my container takes a long time between requests to many routes? Making many containers - one for each pathname - makes things fast but every weird, that’s one container per pathname

Marcell•7/29/25, 7:44 PM

Is it possible to route request to a container without javascript fetch?

King Kazu•7/30/25, 4:07 AM

why is my container stuck as "degraded"? I can't even debug why the container is not starting???

King Kazu•7/30/25, 4:50 AM

would love to see some visibility into the container provisioning process so I could at least see why its erroring?

King Kazu•7/30/25, 5:27 AM

so I think my issue had to due with my issue being too large, but there was never any warnings / errors from wrangler

Marcell•7/30/25, 7:24 AM

Is there any plan to run containers without workers script? It would be nice to deploy containers without requiring to write the glue code that slows down the requests anyway.

Alexander•7/30/25, 7:59 AM

Hey

If you accept feature requests here - Windows support for containers local dev would be great! I tried running through WSL but the install / startup time was so slow I couldn't launch the dev server

MMike Nomitch Hello again 👋 - Here with a general update on some common snags people are see...

mattderron•7/30/25, 8:26 PM

Hi Mike - wanted to follow up on the Cold Start PR - did that get rolled out already? I've notice some pretty long cold starts (10s+ in some cases), and I'm sure there are things I can do to improve it code-wise. But was wondering if that PR was already out so I know what baseline I'm starting from to try to improve.

PPonjimon How do you pass build args to the container image when building it?

henri•7/31/25, 12:27 AM

Has anyone figured out how to pass build args to the container image during the build? I need to access an environment variable at build time, but I haven’t found a way to pass

--build-arg

--build-arg

s or something similar when building a Dockerfile

Hhenri Has anyone figured out how to pass build args to the container image during the ...

kian•7/31/25, 12:32 AM

It sounds like you want

image_vars

image_vars

in the Wrangler config under

containers

containers

kian•7/31/25, 12:32 AM

{
  "containers": [
    {
      "class_name": "MyContainer",
      "image": "./Dockerfile",
      "max_instances": 10,
      "instance_type": "basic",
      "image_vars": {
        "FOO": "BAR"
      }
    }
  ]
}

{
  "containers": [
    {
      "class_name": "MyContainer",
      "image": "./Dockerfile",
      "max_instances": 10,
      "instance_type": "basic",
      "image_vars": {
        "FOO": "BAR"
      }
    }
  ]
}

kian•7/31/25, 12:32 AM

I don't think there's anything for reading it from an environment variable though.

Kkian I don't *think* there's anything for reading it from an environment variable tho...

henri•7/31/25, 5:28 AM

I see... I was hoping I would be able to read it using

ARG

ARG

or something like that

Codename_404•7/31/25, 8:39 AM

You are billed for your usage of both Workers and Durable Objects.

Does this mean, i will be charged for the entire execution time for all 3? DO, container and worker?

Kkian ```json { "containers": [ { "class_name": "MyContainer", "imag...

Yash•7/31/25, 9:01 AM

Does it support a minimum instance setting, so that at least one instance stays active to avoid delays caused by container cold starts?

Hhenri I see... I was hoping I would be able to read it using `ARG` or something like t...

kian•7/31/25, 12:25 PM

Can’t you use ENV?

Julian•8/1/25, 12:38 PM

After reducing the number of max_instances i get the Error "The maximum amount of containers allowed to run in the Durable Object namespace is the same to the amount of running Durable Object Containers, we won't be able to assign an instance to this Durable Object Container" when trying to start a container. how can i fix that

CCodename_404 You are billed for your usage of both Workers and Durable Objects. Does this me...

John Spurlock•8/1/25, 2:01 PM

Yes, the DO and container based on actual time, and the entry worker based on cpu time only (which will be much less if it is just calling a DO and waiting for IO)

HHello, I’m Allie!Sorry about that, I completely missed which channel we are in. R2 does not bill ...

gruntlord6•8/1/25, 2:52 PM

what if it was proxied through the durable object? It's very odd that egress is charged when its not leaving cloudflare, especially since thats not how any other cloudflare compute works

JJohn Spurlock Yes, the DO and container based on actual time, and the entry worker based on cp...

Codename_404•8/1/25, 3:45 PM

So it will also charge for DO IO, as well as CPU time?
of so it should be very expensive for large project, right?

dustiiin•8/1/25, 5:26 PM

Hello, where do i see billable usage for containers?

Ddustiiin Hello, where do i see billable usage for containers?

John Spurlock•8/1/25, 5:33 PM

hmm, good question - it's not currently on the account-level billable usage page

https://dash.cloudflare.com/<account-id>/billing/billable-usage

https://dash.cloudflare.com/<account-id>/billing/billable-usage

so the best thing it seems would be to use the GraphQL analytics api

https://developers.cloudflare.com/analytics/graphql-api/

How do I ensure that a request is routed to the lowest latency zone, even if it means spinning up a

Similar Threads

Similar Threads

Similar Threads