Hi, is it true that I will get billed for compute for all of the 10s idle time before the DO gets wi

Hi, is it true that I will get billed for compute for all of the 10s idle time before the DO gets wiped from memory and hibernated? My use case for DO is low frequency RPCs per DO (e.g. 1 req every few minutes), if I'm implementing one DO per user. Is it possible to make DOs hibernate faster? (e.g. 1s or less).

References:

From lifecycle page: "After 10 seconds of no incoming request or event, and all the above conditions satisfied, the Durable Object will transition into the hibernated state."
From pricing page: "A Durable Object incurs duration charges as long as the JavaScript object has to be in memory, either because it is actively handling a request, or because it cannot hibernate.".

Eelunesarrow Hi, is it true that I will get billed for compute for all of the 10s idle time b...

boop•10/5/25, 10:22 AM

You will not get billed for the idle time

Eelunesarrow Hi, is it true that I will get billed for compute for all of the 10s idle time b...

Larry•10/5/25, 11:42 AM

The key word is “has” in “has to be”. Notice it does not say “is in memory”. If you look on the lifecycle page there is a “idle, hibernatable” state that is NOT marked as billable.

elunesarrowOP•10/5/25, 4:04 PM

Oh right that blue box at the end of the page "...incurs charges only when it is actively running in-memory, or when it is idle in-memory and non-hibernateable (indicated as green rectangles in the diagram).". I skimmed this line and thought the "and" was not in bold, and missed the bracket completely. It's clearer now, thanks for the clarification!

zilch•10/6/25, 12:03 AM

why are DOs charged on wall time and not cpu time like workers? if i am making an LLM api call from DO vs workers, there would be an exponential difference in cost due to the two pricing strategies?

Zzilch why are DOs charged on wall time and not cpu time like workers? if i am making a...

1984 Ford Laser•10/6/25, 12:13 AM

You could call back out to a Worker to handle the LLM call then notify the DO when done?

Zzilch why are DOs charged on wall time and not cpu time like workers? if i am making a...

Larry•10/6/25, 12:19 AM

+1 to what @1984 Ford Laser said. I'd also encourage you to think about the different models. Workers are pure stateless. Every new request requires you to refetch all context from wherever it's stored. With DOs it's right there and stays there. DOs maintain WebSocket connections. DO's have input/output gates that allow you to intermix database reads, manipulate data with code, more database operations, etc. They hold on to cache differently. There are certain contexts where DOs actually cost less. However, even when they cost more, under load, I'm pretty sure the difference is not exponential. It's only exponential if you compare one request in isolation.

LLarry +1 to what @1984 Ford Laser said. I'd also encourage you to think about the diff...

João Castro•10/6/25, 12:25 AM

I'm also pretty sure DOs can be cheaper if you are doing CPU intensive workloads

JJoão Castro I'm also pretty sure DOs can be cheaper if you are doing CPU intensive workloads

1984 Ford Laser•10/6/25, 12:26 AM

Yeah Im pretty sure it costs ~$1.70 to run a Worker for an entire day (in pure CPU time only) whereas a DO is like ~$0.34 per day

1984 Ford Laser•10/6/25, 4:12 AM

Are Durable Object outgoing websocket connections part of the 6 concurrent outgoing connection limit of the runtime?

Zzilch why are DOs charged on wall time and not cpu time like workers? if i am making a...

gruntlord6•10/6/25, 3:38 PM

Durable Objects are stateful so while active they have to have everything running in reserved memory and storage, and they have a reasonable expectation to be accessible in that time (they could still disconnect or be evicted for other reasons). A worker is ethereal and there is never any requirement or expectation that you are served by the same worker. So it makes sense that workers are billed only for what you use since the server can kill your worker anytime it isn't processing requests and another one anywhere in the network can take that request in its place

gruntlord6•10/6/25, 3:40 PM

Like the others said, idle isn't counted towards this, and you can have workers handle things on behalf of the durable object if your worried about the time the request would be running

Zetax•10/7/25, 12:28 AM

Any updates on this? Would be a game changer https://github.com/cloudflare/workerd/issues/4864

GitHub

Feature Request: Add hibernation support for outgoing WebSocket con...

Summary Add hibernation support for outgoing WebSocket connections created by Durable Objects using new WebSocket(). Currently, hibernation only works for incoming WebSocket connections accepted vi...

Tero•10/7/25, 9:19 AM

Is there some known issues regarding durable object metrics / observability? When I navigate to durable objects in console, already in the listing I see error rates that make no sense. Selecting one of the durable object with seemingly high error rates I see a diagram like this.

Searcing the logs I find no errors or exceptions, and this 96% error rate does not correspond at all to what is happening in reality with the durable object.

TTero Is there some known issues regarding durable object metrics / observability? Whe...

boop•10/7/25, 11:35 AM

Not seeing the same

lnow•10/7/25, 3:30 PM

What are the costs of calling `storage.deleteAll()` ?Let's say I have 1M rows in a table, will I be billed for 1M rows written as all 1M rows must be deleted?
Are there any plans for read replicas for DOs?

Llnow 1. What are the costs of calling `storage.deleteAll()` ? Let's say I have 1M row...

lambrospetrou•10/7/25, 3:48 PM

Are there any plans for read replicas for DOs?

Yes, once D1 read replication is Generally Available, we will open up replicas for DOs too (or around that time), since the underlying infrastructure is the same. D1 read replication uses DO read replicas.

Burrito•10/7/25, 4:17 PM

That's exciting.

BBrayden Durable Objects will have a built in Studio in the dashboard next week if this f...

Jacob Marshall•10/7/25, 8:55 PM

Any chance we'll be seeing it this week?

Llnow 1. What are the costs of calling `storage.deleteAll()` ? Let's say I have 1M row...

1984 Ford Laser•10/7/25, 9:02 PM

To answer the first one:

Deletes are counted as rows written.

Deletes are counted as rows written.

https://developers.cloudflare.com/durable-objects/platform/pricing/#sqlite-storage-backend

JJacob Marshall Any chance we'll be seeing it this week? 🙂

Brayden•10/7/25, 9:11 PM

So.... it's ready, but...

just finishing an internal audit before we release

Jacob Marshall•10/7/25, 9:12 PM

Very exciting! I assume you can't really say how long those usually take haha?

Jacob Marshall•10/7/25, 9:13 PM

It's not like it's urgent to me, but it's still very exciting to see - coz it's long overdue

JJacob Marshall It's not like it's urgent to me, but it's still very exciting to see - coz it's ...

Brayden•10/7/25, 9:17 PM

am I allowed to say "this week" again?

Jacob Marshall•10/7/25, 9:43 PM

ZZetax Any updates on this? Would be a game changer https://github.com/cloudflare/worke...

plgingras88•10/7/25, 9:47 PM

I am in a hard release too, can't blame you Brayden ! Be proud it's a game changer feature.

BBrayden am I allowed to say "this week" again?

lambrospetrou•10/7/25, 10:18 PM

Brayden didn't learn yet not to give out ETAs

If there isn't an ETA, you cannot miss it.

11984 Ford Laser To answer the first one: `Deletes are counted as rows written.` https://develope...

lnow•10/8/25, 6:45 AM

I know how deletes are billed. What I don't know is if calling

deleteAll()

deleteAll()

which effectively removes DO is treated the same or not.

grrowl•10/8/25, 7:38 AM

does anyone know how to override the RPC Timeout? We're streaming data using ReadableStream over RPC, but the 100s time limit is too low for long-running AI responses.

ProxyReadTimeout

ProxyReadTimeout

is defined here: https://developers.cloudflare.com/api/resources/zones/subresources/settings/models/proxy_read_timeout/

Cloudflare API | Zones › Settings › ProxyReadTimeout

Interact with Cloudflare's products and services via the Cloudflare API

Marak•10/8/25, 8:09 AM

Is it possible to view an account wide “Service Map” or “Topology View” showing all Workers, Durable Objects, and their bindings/dependencies in the dashboard?

Currently, each individual Worker page in the dashboard displays its bindings and dependencies (Durable Objects, KV, R2, etc.), which is helpful, but there’s no system-wide view that connects them all.

MMarak Is it possible to view an account wide “Service Map” or “Topology View” showing ...

lambrospetrou•10/8/25, 10:01 AM

I don't think this exists right now, although it's been discussed a couple of times.

Llambrospetrou I don't think this exists right now, although it's been discussed a couple of ti...

Marak•10/8/25, 10:21 AM

With many Worker and Durable Object classes, it’s challenging to get a high-level picture of how everything fits together.

MMarak With many Worker and Durable Object classes, it’s challenging to get a high-leve...

lambrospetrou•10/8/25, 11:10 AM

The UI gets a lot of love lately, so I am sure (or I hope) this will get some attention at some point as well.

Nick•10/8/25, 1:48 PM

Is there any way to access the request ID of the worker that called the durable object from inside the handler somehow? Without having to manually plumb it in.

NNick Is there any way to access the request ID of the worker that called the durable ...

Larry•10/8/25, 2:12 PM

Are you using RPC for the call?

NNick Is there any way to access the request ID of the worker that called the durable ...

Larry•10/8/25, 2:13 PM

Where is the request ID coming from in the Worker? Headers on an HTTP Request?

Ivan•10/8/25, 2:49 PM

I am trying to create a worker with a durable object attached via Terraform. However, I keep on getting “Cannot create binding for class 'Aggregator' that is not exported by the script.” However, I do export it:

import { DurableObject } from 'cloudflare:workers';
export class Aggregator extends DurableObject { /* ... */ }
export default { async fetch() { /* ... */ } }

import { DurableObject } from 'cloudflare:workers';
export class Aggregator extends DurableObject { /* ... */ }
export default { async fetch() { /* ... */ } }

Is there something that is typically missed?

LLarry Are you using RPC for the call?

Nick•10/8/25, 2:52 PM

yes using RPC for the call. The requestID is the one that we get from the cloudflare header, which also ends up in the observability logs somehow

Ivan•10/8/25, 3:10 PM

Just maybe putting Terraform aside for a second, when I go via the UI and try to bind, it does not seem to see any durable objects exported in the script. I guess there is some misunderstanding from my side. What is going on here? How does one create durable objects?

IIvan Just maybe putting Terraform aside for a second, when I go via the UI and try to...

1984 Ford Laser•10/8/25, 10:32 PM

Have you added the binding to wrangler config?

IIvan Just maybe putting Terraform aside for a second, when I go via the UI and try to...

Marak•10/9/25, 12:15 AM

You'll want to ensure that the Durable Object class is exported in your main workerfile

// Export Durable Object
export { MyDurable } from './durables/MyDurable.js';

// Export Durable Object
export { MyDurable } from './durables/MyDurable.js';

You also need to configure the

wrangler

wrangler

config file in the project to have the binding.

  "durable_objects": {
    "bindings": [
      {
        "name": "MY_DURABLE",
        "class_name": "MyDurable"
      }
   ]

  "durable_objects": {
    "bindings": [
      {
        "name": "MY_DURABLE",
        "class_name": "MyDurable"
      }
   ]

xormask•10/9/25, 12:32 AM

I have a question about durable object pricing with RpcTarget. If my durable object has returned an RpcTarget (allowing me to amortize multiple calls within the same RPC session) am I correct to assume that while the RpcTarget stub is alive (hasn't been disposed of), the durable object must be actively running and billing me? I have a worker that makes multiple calls to a durable object, but between each call there's slow calls to an external service. If the durable object will be billing me for the whole rpc session, my guess is it would be cheaper to break into multiple shorter rpc sessions. Appreciate any clarifications!

Avi•10/9/25, 2:48 AM

what can be done to prevent a malicious actor from sending too many websocket messages to a durableobject? there seems to be some internal queue of incoming messages that's getting flooded (even before the requests reach my DO's code)

AAvi what can be done to prevent a malicious actor from sending too many websocket me...

Marak•10/9/25, 2:59 AM

You could try the Firewall API and create a new rule, https://developers.cloudflare.com/api/resources/firewall/subresources/rules/methods/create/

Cloudflare API | Firewall › Rules › Create Firewall Rules

Interact with Cloudflare's products and services via the Cloudflare API

Marak•10/9/25, 2:59 AM

What malicious actor is causing the problem @Avi?

MMarak You could try the Firewall API and create a new rule, https://developers.cloudfl...

Avi•10/9/25, 3:01 AM

this is apparently deprecated? to be clear the issue is the frequency of sending of traffic. i'm looking for a ratelimit, not to block a particular actor indefiniteyl

MMarak What malicious actor is causing the problem @Avi?

Avi•10/9/25, 3:01 AM

it's a videogame, so people who are looking to cheat

AAvi this is apparently deprecated? to be clear the issue is the frequency of sending...

Marak•10/9/25, 3:02 AM

Thanks. Yes. The new API appears to be: https://developers.cloudflare.com/ruleset-engine/rulesets-api/create/

We are using the legacy API in production at the moment. Idea is to make API call to CF to dynamicaly manage firewall rules to prevent connection from happening in the first place.

MMarak Thanks. Yes. The new API appears to be: https://developers.cloudflare.com/rulese...

Avi•10/9/25, 3:03 AM

right, i wouldn't want to prevent the connection entirely

Avi•10/9/25, 3:03 AM

just to say, limit the number of messages being sent to X per Y amount of time