I've tested that and it was fine, the first one (1-2ms apart) got it, and if my memory serves me rig

I've tested that and it was fine, the first one (1-2ms apart) got it, and if my memory serves me right, idFromName first checks with a global coordinator before "assigning" it to the closest colo, hence the increased perceived latency of the first request

DaniFoldiOP•3/13/23, 6:19 PM

I'm assuming all colos have a map of ids stored so subsequent accesses are faster, so it's just a first-time thing

DDaniFoldi I've tested that and it was fine, the first one (1-2ms apart) got it, and if my ...

Jacob Wright•3/13/23, 10:51 PM

So you think the global coordinator is able to ensure only 1 DO by name exists, even if 2 requests come in at the same time for a new name? Thank you for your feedback. I'll also be sure to test it, but that's good to know.
It's interesting idFromNameidFromName doesn't return a promise since it can be async.

JJacob Wright So you think the global coordinator is able to ensure only 1 DO by name exists, ...

Hello, I’m Allie!•3/13/23, 10:52 PM

That’s because the check doesn’t actually occur until the stub is fetched the first time

HHello, I’m Allie!That’s because the check doesn’t actually occur until the stub is fetched the fi...

Jacob Wright•3/13/23, 10:59 PM

Ah. So if a worker in NA calls idFromName('new-name')idFromName('new-name') and a worker in EU calls idFromName('new-name')idFromName('new-name') at the same time, they'll both have different ids. When they call fetch(req)fetch(req) will they both get a different DO instance, or will the system ensure they are routed to the same DO? I'm looking at using the DO mechanism so that multiple players of an AI text game who are hitting the same un-generated branch of that game will be able to request it and have a unique DO for that branch ensure the request to generate the AI response only happens once and is returned to both players and caches the response in R2 for future players who go down that particular branch. I'm not sure if the guarantee for 1 DO in the world for a given name extended to this case when 2 clients could be asking for the same new name at the same moment.

JJacob Wright Ah. So if a worker in NA calls `idFromName('new-name')` and a worker in EU calls...

Larry•3/14/23, 12:18 PM

The intention is that the initial NA and EU idFromNameidFromName calls will meet up somewhere in the ether and one will win so that there will only ever be one DO that matches that "name". This cross coordination occurs before a fetch can be called, so they will be "routed to the same DO". That said, I don't know if anyone on this channel has seen the actual code and sequence diagrams that confirm it achieves this intention, but if it doesn't work this way, it's a bug, not expected.

Cckoeninger you don't need global coordination to get a consistent id from a given name. th...

Hello, I’m Allie!•3/14/23, 2:19 PM

Might have asked this before, but any plans to have a way to get the current “region” from within a DO?

Cckoeninger aside from the previously widely discussed unsupported hack that will likely be ...

Hello, I’m Allie!•3/14/23, 2:20 PM

Do tell?

Cckoeninger aside from the previously widely discussed unsupported hack that will likely be ...

DaniFoldiOP•3/14/23, 2:21 PM

Hello, I’m Allie!•3/14/23, 2:21 PM

I’m just wondering for having a regional coordinator, it would be nice to know “this DO is in ENAM”

Hello, I’m Allie!•3/14/23, 2:21 PM

I don’t need the exact colo, just the locationHint region

Cckoeninger oh locationHint region as opposed to the actual region an arbitary DO is running...

Hello, I’m Allie!•3/14/23, 2:22 PM

Maybe on the ID, have a location property with the region attached. Doesn’t have to be the region provided when creating the DO, just the region the DO was actually spawned in

Cckoeninger or the equivalent naming convention as locationHint, regardless of whether you g...

Hello, I’m Allie!•3/14/23, 2:24 PM

This one, I think. I don’t need it externally available, it would just be nice to have a DO be able to recognize that it spawned in North America(or whatever the region code is)

Cckoeninger what's the use case?

Hello, I’m Allie!•3/14/23, 2:26 PM

Having a regional coordinator for a group of DOs. It would allow the DOs to communicate from one region to another efficiently, without manually having to have connections to/from every DO

Hello, I’m Allie!•3/14/23, 2:27 PM

Also to apply filters to incoming messages at the regional level. Basically, a message bus, where instead of having every DO process every message, then instead have a few regional filters, that then pass to the individual DOs as necessary

Cckoeninger what's the use case?

Jacob Wright•3/14/23, 3:19 PM

I'm using the cdn-cgi hack to create connection pools for each region where DO's reside. Would there be a recommended way to manage pools of X connections for a service? locationHintlocationHint could work. Just need to route connections to a nearby pool.

Edit: I have one Coordinator DO for the DO location nearest a Worker request, referenced by colo name. It is in charge of creating new Pool DOs and assigning connections to them and keeping them from being too full or too empty. The Coordinator DOs just need to be close. So I need the information on the Worker request.cf object. The Coordinator name is arbitrary. I could have one Coordinator per Worker colo, but that would be too granular for our size. One per locationHint would work. So if a worker had request.cf.locationHintrequest.cf.locationHint I could manage pools easily that way.

Aakira.eth If I want to use websockets in cloudflare just to post updates to clients with n...

Jacob Wright•3/14/23, 3:46 PM

If you have 2+ clients that need to get messages from each other, you need durable objects, simply because you need to have each connection in memory so that when one message comes in, you can reference each other client and send the message along.

Jacob Wright•3/14/23, 3:51 PM

"the server" would need to have that list of connected clients. Do you have your own backend server, or are you referring to what you are building on the Workers platform? If the former, workers can proxy websocket connections to an origin server. If the latter, the durable object would need to be the server so that some process somewhere can know the clients to send them data.

Aakira.eth ok yeah that is basically what I thought, I have the do as a server was just won...

Jacob Wright•3/14/23, 3:54 PM

The price of a single DO for a month is about $4.15 to run it constantly, plus the cost of any messages received (sending out does not cost)

Aakira.eth ok yeah that is basically what I thought, I have the do as a server was just won...

Jacob Wright•3/14/23, 3:56 PM

There are services like https://ably.com/ but they're more expensive.

Aakira.eth ideally i would use pub sub but has a 1000 connection limit to 1 topic

Jacob Wright•3/14/23, 3:57 PM

Remember that DOs are single-threaded, so while you could have more than 1k connections to one, if there is a lot of messages being sent, it could struggle.

Jacob Wright•3/14/23, 3:58 PM

Sounds like it would work.

Jacob Wright•3/14/23, 3:58 PM

You'd want to test it, of course, to ensure it can handle it.

Cckoeninger If you're talking about eyeball worker to DO connections, you don't need the cdn...

DaniFoldiOP•3/14/23, 5:30 PM

even with that we'd need https://where.durableobjects.live to figure out where the DO could be, or is there another hack for that?

Where Durable Objects Live

Tracking where Durable Objects are created, wherever you are in the world.

DaniFoldiOP•3/14/23, 5:38 PM

sort of, different mapping pair thought - to minimize (websocket) latency I'm looking for the colo where the DO would be created, and have one global coordinator DO group connections per target colo, up to N connections, after which I let it create itself a new group DO, and assign another N-1 connections to it

Cckoeninger grouping websocket connections here is strictly for pricing reasons, there's no ...

DaniFoldiOP•3/14/23, 5:46 PM

correct, $4-5/connection/month seems wasteful for a connection that sends at most a few messages per minute

I've tested that and it was fine, the first one (1-2ms apart) got it, and if my memory serves me rig

Similar Threads