Jeff Kao

And because people have built libraries to essentially cache S3 on disk, what we're thinking about is we're going to build a storage layer that almost caches S3.

And then while the service is booting up and hasn't pulled in the assets to the local NVMe disk,

it can temporarily sort of pull in ranges of bytes from these like storage services.

So it can immediately start serving queries.

Obviously there's going to be a latency hit, but in terms of being able to deal with things like, oh, we have a traffic spike.

That's where like, that's sort of one of the problems we're trying to solve.

And I think the fact that there's so much development around, you know, data infrastructure projects and, you know, we use Spark at Radar and like so many companies use Radar.

Just this whole concept of like,

data infrastructure in the cloud and using object storage like we're seeing like you know new streaming frameworks written in rust or a lot of these sort of new iceberg tools in in rust and that that's going to help us a lot for building something where we want to separate where it's going to accelerate essentially like what we want to build where we separate storage and compute how long did it take you to build and ship the first version of horizon db

Right.

So as I mentioned before, like our first use case was address validation.

We had a customer who was asking for it like and we built it in about a month.

Actually, I think less than that, like to actually get something working, but to get it closer to like full production with this customer and.

Maybe it was like two months, but it was like with a lot of QA and like a lot of feedback, things like that.

But I would say it took about like two months.

Not really.

One was because we were only onboarding a single customer and like the QPS wasn't like that high to begin with.

And two, because we had this property of everything was sort of self-contained, it was very trivial for us to do blue-green deploys, which also includes the data as well.

So like, hey, you process the string in a wrong way and you're starting to see that, you just flip it back to the previous service and it's all self-contained.

Whereas that might be harder to do with like a single backing database or things like that.