So Edge Request arrives in CPH, which maps to ARN as the nearest DO colo. ARN has a DO ID assigned, you use that to create a stub, and you are good to go

I believe it was

sandbox allow-popups;

that fixed it

A question burning in my soul about DOs, what's up with the splits? Does Cloudflare have a config that randomly assigns DOs to one of X given PoPs by percentage? Or is it by request IP to try and get the request closer?

Probably better to ask brett about that. I'm a community member just like you, so I am not(yet) allowed a peek behind the curtain.

😄

Though I would assume it is based on the edge colo, rather than the location of the user

Ahh, workers don't run in every colo?

They should? Other than China, unless you have an ICP license.

On where.durableobjects.live, it has a colo and a split of where it sends requests to DOs. Or is Accra and Adelaide something other than a colo?

If it were based on colo, then each colo would send 100% of their requests to a corresponding DO colo.

No, I meant that I assume the sorting is based on the edge colo. It may still split between nearby DO-ready colos.

You know what's interesting, is workers in a colo that supports DOs aren't always sending requests that hit their colo to internal. E.g. Amsterdam sends 40% of its requests to other DO colos.

Also, I'm investigating an issue with the aggregator, as it seems to be very far behind. Currently, the data shown may be out of date by up to a few days. I should have a fix ready by tomorrow.

Is there plans to show any sort of graphs for them? It would be really interesting to see them change over time. I assume for some of those like Amsterdam, it could have been a maintenance event that sent all new DOs to diff colo

Would you mind making an issue for that? I’m already thinking about adding access to historical data(maybe via the SQL API), but to be honest, I’m not always the best with visualizations, and what works best there.

hmm yea, not sure what they would look like either lol. Maybe just the mini-graphs like they have in the dashboard next to each DO colo name. They would probably suck up your API limit though unless they were cached themselves. The amount of data you have is both a lot, and not much. If the last hour metric is right (2286), then you only create 6 DOs per source colo per hour, or 156 per day (assuming there is 350 CF Colos). Maintenance events or just minor interruptions could mess with the data a fair bit, even if it is aggregated the last over 93 days. Kind of interesting to think about

The 2k last hour metric is borked. The Cron trigger is erroring, and I think the health checks are failing too. I need to dig into it to figure out what’s going wrong. I’ll do it tomorrow when I have time.(Last hour metric should be ~20k)

They're shuffled within a small latency bound to avoid overload. Generalized/dynamic relocation will help with this in the future.

e.g. (for now) We don't want 100% of objects created in IAD to end up in one colo forever

Just to remember, is SFO-DOG publicly routable?

I don't think so, what's the context?

Just the fact that it shows up on WDL, but I’ve never been able to hit it. I’ve always hit SJC and LAX in California

Oh, showing up from the client side? That must be CFers hitting it.

SFO-DOG and SFO-TEST both 😛 those are the only 2 I think that arent normal colos you can hit that do trigger workers through health checks

I thought you meant as serving DOs, which is possible, but shouldn't happen for your namespace.

Probably the Health Checks?

?

I use Health Checks to ping from every colo once a minute. It seems to be broken rn, but I guess it is able to run from SFO-DOG?

I don't think I follow, what's broken? I'm just saying external people shouldn't be able to route to our internal colos, but CF employees do, and so that's why WDL probably has records of where those colos would make objects.

The thing that is broken is something I need to fix, not a CF issue(as far as I’m aware). And SFO-DOG seemed to appear as soon as I turned on Health Checks, so I assumed the “All Datacenters” option covered that too.