Less on the client connection and more on our end

We'd only have to do the metadata lookup once & then we issue a bunch of get requests on your behalf to the storage nodes in parallel automatically.

ah makes sense, that sounds ideal

also interesting would be automatic handling of compressed request bodies during uploads (either keep it compressed in object storage, or decompress it before storing it, that decision could be configurable on a per-request basis). i wonder... do browsers even support

Content-Encoding: deflate

on POST/PUT request bodies?

(e.g.

fetch()

in browsers)

Already discussed this previously. Compressing/decompressing things on upload is difficult. Our storage nodes currently force us to tell them the content-length up-front. That means we'd have to buffer the payload up-front. Eventually I'm hoping we can lift that requirement from our storage nodes but it's going to take a while (year+, not weeks/months of effort).

ah right right. i vaguely remember this now

i'd probably also test the auto-S3-to-R2 feature with some B2 buckets as well

that sounds rather convenient rather than doing a full migration of all objects upfront

is there somewhere in the dashboard/api to specify the the key/secret for the S3 bucket it's pulling from, or does it only work with public S3 buckets?

We don't have the migrator yet at open beta & I don't think we have timelines to share on that yet. We'll have something else which is kind of related but fits a slightly different niche.

I can't speculate as to how the migrator works but yeah, putting the keys to your S3 bucket in the UI or via APIGW sounds reasonable

Ok, yeah. i can't imagine it's a dealbreaker for many so it seems like a good thing to save for later. nice to have, though

I agree. There's so many nice to haves. Some are easier than others and these days I'm having to force myself to focus on the absolutely critical things for launch even if it's only a day or two do add a feature like conditional PutObject :\.

i know how that goes ^_^

Reading this again with the benefit of morning coffee, noticed the options for

list

is missing a

startAfter

name option

Also it should probably be

R2Object.key

instead of name anyway, to match the parameter names in the bucket methods

Also, in 2022,

R2HttpMetadata

should have CORS headers options (e.g.

R2HttpMetadata { accessControlAllowOrigin?: string; }

)

just dropping by to say that r2 is pretty cool, although i havent tried it yet. looking forward to q2. kian brought me here

We don't actually support startAfter yet in the bindings. I'm not confident of our implementation in the S3 API either. No ETA yet. Good call on R2Object.key - will try to get a fix up for that

Unfortunately, for open beta we didn't have enough time to flush out CORS. Nothing stops you from applying it yourself in a Worker of course but that's not a satisfactory response. S3 has CORS at the bucket level not the object level btw https://docs.aws.amazon.com/AmazonS3/latest/API/API_PutBucketCors.html. Some people want object level policies, some people want bucket level. I think S3's approach with using DSLs as an entire approach has been mistaken, as is picking sides between bucket and object (which they're having to choose because of architectural decisions specific to when S3 started and the product suite available to them at the time). I'm trying to get a blog post up around R2 launch outlining what I'm thinking of instead.

i.e.

accessControlAllowOrigin

would be as as much of a DSL as S3's XML API. I'm not saying we won't offer you way to specify it via a DSL. It has a benefit of simplicity which can't be ignored.

+1 to the Cloudflare Pages approach with the _headers file: https://developers.cloudflare.com/pages/platform/headers/

wait, couldnt you just serve r2 from your own domain and use transform rules to add the CORS header? (if serving r2 from a custom domain is even possible without workers)

just for curiosity, how much is the free tier?

or, is there going to be a free tier?

Yes I believe you could

hm.. is it possible to create a cname that points to an r2 bucket’s s3 endpoint hostname?

For those in the beta, there's a new release going out shortly. The main change(s) are: * DeleteObjects of > 128 keys now explicitly returns MalformedXML and v4 code 10027 instead of InternalError. We don't support deleting > 128 keys at a time currently. We'll be working on relaxing this to 1000 keys between Open Beta being announced and GA.

short question, what is GA?