Add blob download endpoint (getBlobs)

[jimmygchen]

Description

Addresses #282. Add /eth/v1/beacon/blobs/{block_id} endpoint to support retrieval of blobs by block id.

[jimmygchen]

From what I can see the request / response structures look similar, just some differences in paths. I think a combination of the two makes sense to me (e.g. /eth/v1/beacon/blobs/sidecar/{block_id}) for the following reasons:

1. It feels like the blob sidecar belongs under beacon as it's closely related to the beacon block
2. Having blobs/sidecars instead of blobs_sidecar may provide more flexibility, although I don't see any other potential endpoints that may fall under blobs.. so no strong opinion on this.

I'm still pretty new to this so I might be missing something - keen to hear suggestions / opinions.

[arnetheduck]

One potential issue of note: in the underlying libp2p protocol, we don't support fetching finalized items by root, instead only supporting by-slot lookups - it might be prudent to note this in this spec as well, that sidecar retrieval for finalized blocks should by done by slot else the underlying index might not be available.

Doing it this way allows clients to implement this API using linear storage which is cheaper and more efficient and more closely matches the requirements that clients have to support in order to implement the mandatory parts of the protocol.

[mcdee]

I'd be inclined to go for just /beacon/blobs as the path, I don't see the _sidecar piece adding any useful information (it's a information about how we're treating the blobs, not what they are). There's also an argument to have just /blobs, especially if the plan is to make them available from places other then the beacon node, although I'm not overly bothered about that one.

[terencechain]

Added Prysm implementation here: prysmaticlabs/prysm@50b672a

[jimmygchen]

One potential issue of note: in the underlying libp2p protocol, we don't support fetching finalized items by root, instead only supporting by-slot lookups - it might be prudent to note this in this spec as well, that sidecar retrieval for finalized blocks should by done by slot else the underlying index might not be available.

Doing it this way allows clients to implement this API using linear storage which is cheaper and more efficient and more closely matches the requirements that clients have to support in order to implement the mandatory parts of the protocol.

@arnetheduck thanks for this! I don't fully understand the implication of this on the Beacon API, would appreciate if you could elaborate a bit more - if the blob is already available in CL storage and p2p lookup isn't required, would this still be an issue?

[arnetheduck]

blob is already available in CL storage

in the core protocol, there is no requirement to index finalized data by hash, only by slot - this applies equally to blobs as it does to blocks.

Because finalized data is strictly linear, the index-by-hash is not needed and looking things up by hash is comparatively inefficient, as is building and maintaining the hash index itself.

"CL storage" for finalized data can be implemented as a flat file to serve core protocol needs - by suggesting consumers to use slots, the same efficiency can be had in the REST API as well, which saves on disk space, processing and complexity, in the CL.

edit: you're indeed right that it is also more simple to forward a request to the network when topping up local storage if that request is done by slot, since the network doesn't carry said hash index.

[jimmygchen]

I'd be inclined to go for just /beacon/blobs as the path, I don't see the _sidecar piece adding any useful information (it's a information about how we're treating the blobs, not what they are). There's also an argument to have just /blobs, especially if the plan is to make them available from places other then the beacon node, although I'm not overly bothered about that one.

Good point on _sidecar - I think I'm just getting too used to the term as a spec type now. One reason I think it could be helpful is that it tells user that the endpoint returns the BlobsSidecar spec object rather than just the blobs. If that's not a strong enough reason, I think we're ok with dropping _sidecar.

[rolfyone]

I'd actually be inclined to make this the same pattern as attestations, so the path would be more like
/eth/v1/beacon/blocks/{block_id}/blobs

[mcdee]

Wouldn't that imply that the blobs are part of the block structure, though?

[jimmygchen]

Updated to /eth/v1/beacon/blobs/{block_id}

[rolfyone]

We should add to top level CHANGES.md too

[jimmygchen]

@rolfyone @mcdee
I've taken your suggestions to remove _sidecar from the path and added the endpoint to CHANGES.md. I originally thought about having blocks in the path as well, but I think @mcdee made a good point and blobs aren't part of the block structure.

@arnetheduck thanks a lot for the explanation, If I understand you correctly, you're suggesting to not support lookup by root for blobs (and perhaps blocks too) as it would allow clients implement this more efficiently. I made the assumption we would want the same lookup (block id) for blobs as well, as looking up by root is already supported for blocks in the Beacon API. I don't really understand the libp2p protocol and storage implementations enough to make a comment about this (but interested to find out more if there's any resources you could point me to) - I'll wait for others to chime in on this.

[arnetheduck]

you're suggesting to not support lookup by root for blobs (and perhaps blocks too) as it would allow clients implement this more efficiently.

Perhaps more weakly than this, ie "recommend" that consumers look up by slot - it's possible that some clients build a by-root index for the entire history (consider archive nodes for example) but it is certainly more "compatible" to not rely on it and instead use slot (which also can offer better performance) - it would indeed be the case that the same recommendation should be added to block lookups - might get to this at some point unless someone beats me to it.

[jimmygchen]

Just going through the changes in the "Free the blobs" PR ethereum/consensus-specs#3244, and will soon be updating this branch accordingly.

[realbigsean]

@jimmygchen

I don't think this will have to change too much, I think it should probably return the same struct we return in the BlobsByRange request as this is likely how we'll store blobs in our database. This is what it looks like right now https://github.com/ethereum/consensus-specs/pull/3244/files#diff-21cd1dde2a4815e714ef42d0cefa1bbd2997a4358ecf6e9e51025332c1d642fdR222-R224

[rolfyone]

do both need to be plurals? is blob_sidecars accurate?

if the paths / files / object names can be consistent that'll make it easier to navigate

[jimmygchen]

hmm I think this is on an old commit as well, I just did a global search for blobs_ and not finding anything. I'll make sure the paths / file name are matching if we end up changing the paths 👍

[rolfyone]

So i think this is probably close, just whether namings are correct blob_sidecars v. blobs_sidecars and being consistent...

also my changes markdown comment about whatever the path ends up as :)

Then I think we should be right to merge...

[rolfyone]

Hope for one last merge to fix conflicts...

[jimmygchen]

Hope for one last merge to fix conflicts...

Thanks @rolfyone I've merged master into this branch and updated the paths.

[rolfyone]

LGTM

[mcdee]

We appear to have the type BlobSidecars that is an array of blob sidecars with the same parameters as this. Should that be used in place of redefining the array here?

[jimmygchen]

Good catch! Updated in 096aadc, thanks!

[rolfyone]

LGTM, thanks team!