Feature: Query CL capability

[yorickdowne]

Is your feature request related to a problem? Please describe.

Consensus Layer clients might not implement new REST API endpoints at the same time. For example, Nimbus didn’t implement /eth/v2/beacon/blinded_block until July 2024.

As a result, block production with SSV and a pure Nimbus cluster failed.

Describe the solution you'd like

Query the available endpoints / CL capabilities on startup. Vouch code can be used as inspiration, as it does that.

Then either use the capabilities as discovered or, if a missing capability is mandatory, fail with a clear error message.

Describe alternatives you've considered

Robust QA to make sure SSV works with all CLs. The trouble is that this is hard to handle when new CLs come on the market, like Grandine and Lambda. Querying capabilities doesn’t replace QA, but it does provide a baseline that makes the QA job easier.

[y0sher]

@yorickdowne , thanks for opening this issue.
It's indeed a good idea, it will be released in next version!

[moshe-blox]

@yorickdowne

i think this is a good idea, however i'm unsure we can confirm support for a route such as SubmitBlindedBeaconBlock which requires POSTing an actual block.

of course we can send an empty POST request and fail if we get back a 404, but i'm afraid some Beacon nodes would (incorrectly) return a 404 just because there was no block in the request, rather than because they don't support this route altogether

[yorickdowne]

Yeah that’s a good point. You could supplement with better error reporting. It took too long to know that the missed blocks were due to a missing endpoint. Logging more of the response on failure … code and message really … may have pinpointed that sooner

[y0sher]

@yorickdowne , From looking into the capabilities endpoint of beacon nodes, it only returns a list of supported forks (e.g capable of Capella, Deneb etc..).
There is no way to just query for the supported endpoints.

The only options which sounds cumbersome and complex is to try to call all endpoints at the beginning and crush if any is 404.

Indeed we need to change some logs to be clear errors, the case of a not functioning endpoint is none of them.