Gloas devnet planning

[michaelsproul]

Glamsterdam Devnet 0 (EL devnet 3 with BAL)

• Port remaining BAL changes that aren't in Gloas spec v1.7.0-alpha.5 and beacon_chain tests #8998:
  • Gloas engine api updates #9150
• Defer payload processing spec change (@michaelsproul )
  • Defer payload processing to next block ethereum/consensus-specs#5094
  • Gloas spec v1.7.0-alpha.5 and beacon_chain tests #8998
• Revert this change:
  • Gloas serve post block state for finalized/justified state requests #9092
  • Revert PR: Revert Gloas API changes from 9092 #9151
• Review previous PRs and make sure it's consistent with the new paradigm
• Ensure by range only serves canonical envelopes for non-finalized slots
  • Ensure payload envelope streamer always serves canonical envelopes after the split slot #9085
• Range sync + checkpoint sync (@pawanjay176)
  • Review Lion's lookup PR
  • Blocked on deciding whether to send payload / check block hash to EL for finalized sync
• Use attestation.data.index to indicate payload status (@eserilev)
  • Gloas set AttestationData.index  #9100
• PTC
  • PTC validator duties (@eserilev) Gloas vc ptc duty #8338
  • PTC beacon api endpoints for vc duties (@eserilev)
  • PTC Gossip verification @hopinheimer Gloas PayloadAttestation gossip verification #9145
  • PTC op pool & Pack PTC into blocks
• (Optional) VC publish proposer preference
• (Optional) Use in-protocol builder bids when proposing blocks
• (Optional) DA checker (@eserilev)
  • Gloas data availability checker #8716
• Enable gloas http API tests on CI
• Enable gloas kurtosis tests on CI

Future work:

• Stateless block production
• Multi BN changes for VC
  • gloas VC block production #8313
• Sync (@eserilev, @pawanjay176, @dapplion)

Old stuff

Devnet 1

• Alpha spec 4 changes (@michaelsproul): Update to spec v1.7.0-alpha.4 #9046
• Gossip of builder bids/proposer prefs (@eserilev) Gloas bid and preference verification #9036
• Update beacon API endpoint to serve pending states for finalized and justified checkpoint (needed to correctly serve checkpoint sync states) Gloas serve post block state for finalized/justified state requests #9092
• Fork choice (@hopinheimer @dapplion @michaelsproul @eserilev) Gloas fork choice redux #9025
• Required for Dora:
• SSE event for envelopes: Add Gloas SSE event boilerplate #9053
• Beacon API GET payload envelope (@eserilev) Gloas get payload envelope beacon API #9038
• RPC by range/root responses for envelopes (@eserilev) Gloas serve envelope rpc #8896

Local devnet

• Gossip of envelopes (@eserilev, @jimmygchen): to be separated from range sync.
  • Gloas gossip boilerplate #8700
  • Gloas payload envelope processing #8806
• Consensus changes including containers (@michaelsproul and @jimmygchen):
  • Gloas consensus v1.7.0-alpha.1 #8677
  • Implement Gloas withdrawals and refactor #8692
  • Move KZG commitments from payload envelope to payload bid and spec alpha.2 #8725
• Envelope processing & beacon chain changes
  • (outdated) Gloas(EIP-7732): Envelope Processing #8279
  • Gloas payload envelope processing #8806
• Block production support (@michaelsproul, @eserilev, @shane-moore when avail):
  • Gloas local block building MVP #8754
  • Once this is done, we shoud be able to enable test-beacon-chain-gloas
• Bare minimum fork choice (import payload and building on top of payload) (@hopinheimer)
• Database changes
  • Payload envelope db operations #8717

Devnet-0

• Gossip verification changes for attestations and blocks (@jimmygchen).
  • Gloas attestation verification  #8705
  • Implement gloas block gossip verification changes #8878

[michaelsproul]

We've bumped RPC and sync support out of scope for devnet 0 so we are more likely to meet the goal of interop on Feb 18 (as discussed offline with @jimmygchen and @eserilev). If we have time we can consider RPC responses for envelopes (a stretch goal).

[eserilev]

Potuz wrote:

ah yeah the entire forkchoice is irrelevant IMO for devnet-0, it's not enforceable, As long as you can follow empty/full chains and build on full if you have the payload I think we're fine
My moto for devnet-0 is "the minimum possible to separate the block from the payload"

So it sounds like we can punt most fork choice complexities to devnet 1.

[eserilev]

It seems some clients may be relying on rpc payload by root requests in order to prevent themselves from stalling out. I think introducing the rpc boilerplate and serving payloads over rpc is a relatively small lift. I'll aim to get that done by the devnet deadline

[eserilev]

I've moved PTC BN/VC support to the future work section. It's not needed for devnet 0. All we need is to be able to gossip payload attestations (so peers dont downscore us).

We also wont be including any payload attestations in blocks that LH proposes.

[eserilev]

We wont be propogating any data columns during devnet 0, so I'm moving data column gossip verification to the futures work section

[michaelsproul]

The DB stuff (#8998) might be more important than previously thought, Pawan got some breakages related to the DB while testing sync.