Yikes, calling out amazon for the data transfer costs then matching it for DO Not a good look at all 😕

is there a way kill an existing object

Really frightned with this new pricing. Im really curious whether if you leave a websocket connected but no messages are sent for say 1 hour, youre charged for 1 hour of compute?

i’m sure we’ll get clarification from the team, but to calibrate, an idle websocket on AWS is about a penny a month ($0.25 per million connection minutes, per https://aws.amazon.com/api-gateway/pricing/)

Is that an apples to apples comparison though? A DO could be conceptualised as a hybrid API Gateway/Lambda (obviously it is not as simple as that but I'm sure you get the point that a long lived lambda connection would not be the same cost...)

the point is that an idle websocket does not incur compute costs in AWS, since the lifecycle of the lambda is detached from the lifecycle of the websocket.

also of note if i understand correctly is that durable objects are charging by wall time a la unbound, not by CPU time a la bundled.

The pricing questions are gonna need a response from our product manager, since I'm not sure of all the details, but he's unfortunately out today. Hopefully we'll have clarifications for you all soon.

Just to clarify, we confirmed in DMs that this was due to CPU exceeded errors. A websocket request currently just gets the one 50ms CPU budget for its entire lifetime, so if it's used regularly it'll run out after a while and get disconnected due to exceeding the limit. This is not at all the desired behavior and we'll work on fixing it soon, but in the meantime reconnecting does get you a new CPU budget

They're in a couple dozen PoPs spread around the world at the moment, but we plan to continue adding more over time.

No rust examples yet. If you get one working it'd be great if you could share it

What exactly do you mean? If you stop sending requests to an objcet and disconnect any websockets to it, it'll automatically be stopped by the system. Objects are only kept live in memory so long as they're actively in use

Oh okay that’s good to know, I was just checking what colo they were being deployed to. How long until they will be automatically stopped by the system when there’s no websocket? Seconds, Minutes or hours?

Generally around a minute, although there's no specific guarantee there

Thanks 👍

anyone have an example of how to upload a Durable Object class script in a module file by itself? i.e. not mixed in with the calling worker.

I'm using just the api, and I'm almost there, but in the end it forces me to specify a "main_module" in metadata, and then if I do, it says: "The uploaded script has no registered event handlers.", which of course it doesn't

are there any typescript examples for getting types in for

WebSocketPair

? any help is appreciated 🙂 - is this the thing? https://github.com/cloudflare/workers-types

@User right now we still require an event handler with all uploads, even if you're not going to use it. We hope to remove this restriction soon 🙂

Congrats on the open beta! 😄

has anyone benchmarked DO yet?

Thanks! got it working - but yea would love to be able to remove this unneeded fetch handler, ideally in this scenario the worker script (containing just the DO) should not really have a public route, right?

Usually, but you may want a public route w/ E.g. access or some sort of access control in front of it for debugging

I see how that would be useful, once I'm past this hello world example, but what are the bandwidth charge implications? i.e. would I start paying the bandwidth charge for all incoming calls to that DO fetch handler? vs a normal worker

when DO is out of beta, will any amount be included in the Workers Bundled plan?

Congrats on the beta launch! I've been running some latency tests as part of the private beta and wanted to check these latency times look right? I'm just grabbing a single counter item from the data store:

time_namelookup:  0.001117s
        time_connect:  0.019944s
     time_appconnect:  0.068345s
    time_pretransfer:  0.068393s
       time_redirect:  0.000000s
  time_starttransfer:  0.123801s
                     ----------
          time_total:  0.123859s

These times seem high?

its only possible to upload DOs using wrangler now right?

no web gui? like for workers

Nope, only wrangler right now

Does DO data egress pricing apply to transfers out to Bandwidth Alliance members?