I mentioned them because Lars intro'd me to them. The frequency bias is still one layer, not two.

Yeah, wrote this before I listened.

1. if you want geo-replication with R2 you better enjoy writing edge worker Cache API code because R2 doesn't do it by default. The marketing says yes. Threads on forum says no.
2. Yes, for buckets. Also can be actively told to cache eagerly which is rad. Not for your pages though unless you render them out to disk.
3. No. Not that I am aware.

It essentially rounds out the Fly offering of putting your apps close to your users by also putting your files close to your users.

if you ran more app instances in a geo-distributed fashion and had some light caching in-app on the heavily used pages you'd probably get good results.

But most likely you'll find one or two pathological cases where moving your app and DB too far apart makes something awfully slow.

Also, cool to be a bias with frequency. Signal all on my own :D

Just to double-check my understanding:

1. Yes, with extras (e.g. object geo-replication that just works)
2. Not for files that do not exist in the bucket. Caching is configured on the object store, not on origins.
3. 👍

Based on the above, this still sounds like an R2 alternative with the added benefit that we could serve those files directly from the object store cache which can be configured to replicate closer to end-users. Tigris doesn't behave as a CDN in the sense that we can't declare origins, request rules, etc.

Yeah, it is more of an R2 alternative than a CDN in that regard. It will CDN your media files and all that but unless you render out your pages to file on change it will not solve your landing page etc.