Consensus Implementers’ Call #100 - 2022-12-15

[Quick contemporaneous notes by Ben Edgington; fka "Eth2 Implementers' Call"]

Agenda: https://github.com/ethereum/pm/issues/688
Livestream: https://www.youtube.com/watch?v=UazJO0fQ3Ho

The last call of 2022.

Capella

Goat Star

Goat Star is a spec release made yesterday, with new test vectors.

Shanghai testnet (Zhejiang)

Broader than withdrawals only (timestamp-based forking, warm coinbase, push0, etc.). Testnet spec.

What EngineAPI updates do we want to include? [Lightclient] is happy to make an EngineAPI spec release. [Pari] Would that include PR#314 (GetPayloadv2 and block value return)? Exec clients could (optionally) mock the block value to 0 for now to reduce their work. No objections.

PR#146, PR#218 - rough consensus is that CL teams don't need these right now.

Do we want to include the forkID logic for this testnet? Nice to have, but not a required dependency.

Validator set size? Withdrawals sweep limit implies that we should have more than 16k validators to test that properly. Similar to Kintsugi (~100k validators, 40-50 nodes) is good. [Danny] Suggest doing a leak at some point to make all validators <32 ETH so that a maximum sweep occurs.

Plan to set all validators to 0x00 creds at genesis, and selectively test upgrading to 0x01.

BLS credential change submissions: array vs single in the Beacon API? Is the array API implemented in clients? Lodestar and Teku support this.

Timeline: is pre-Christmas ok to start the testnet? No objections. All clients seem confident of being ready (or nobody said otherwise).

Spec Items

historical batch revamp

[Jacek] Complexity is minimal. Requires changes to beacon state, but the logic changes are very simple.

Saves you from having to compute the state roots and run the state transition function (due to empty slots) when you have 8000 blocks you want to verify belong to the state. Great for archive nodes. Also for the Portal network. Note that state growth will double from 10KB/yr to 20KB/yr.

Adds ~2 days overhead for test generation.

Action: will slate for Jan testnets, and Capella in a week's time unless any teams object between now and then.

EL block header vs CL ExecutionPayloadHeader fields

[Etan] Slide

Two fields are encoded differently between the EL block header and the CL executionPayloadHeader. The former uses a hexary trie root, the latter SSZ. Use case for unifying these is for light clients/wallets, otherwise they need to download both, which doubles the work.

Could extend EngineAPI to get these quantities from the EL for storage in the CL. Or could change the EL so that it stores SSZ for these fields. There seems to be a move away from hexary tries and RLP in general.

Changing the EL is probably a deeper change. Need a broader conversation that might not be feasible before Capella.

Discussion is happening on PR#3078. Need to discuss with EL people, which won't happen before January.

EIP-4844

Protocol upgrade name

Generally agreed at the last ACD to put 4844 in a follow-on upgrade from Shanghai/Capella.

Need a star name! Proto suggests Dubhe (the Great Bear). [What's the canonical pronunciation???]

[HWW] Might need better overall names for describing upgrades to the media. Unified names. [Marius] Drop the EL names, and go with the CL naming scheme.

Action: participate in the Magicians' discussion thread. Hsiao-Wei to coordinate.

DA on historic blocks

Consensus seems to be that, beyond the pruning window we assume that data_is_available is false (unless you already validated its availability yourself). Nodes will not need to store older blobs in the case of long periods of non-finality.

[Sean] What would recovery look like in this edge case? [Micah] Do clients treat their checkpoint sync state as functionally final, ie. that they would never reorg it? [Jacek] Nimbus does, and would never revert the checkpoint. Also treats the checkpoint immediately as the finalised state in the API. Similar to a genesis state.

[Danny] Action: would like to get this merged to close the edge case. We can discuss in January if the engineering effort looks terrible.

No blob available error code

The PR will be merged soon. Has been discussed on the 4844 calls.

Research, spec, etc

Nothing.

Open Discussion/Closing Remarks

No ACD next week or AC/DC (ACD Consensus) the following week. Next ACD Jan 5; next AC/DC on Jan 12.

Look out for news from Pari about the testnet next week.

Chat highlights

From pari to Everyone 02:05 PM: https://notes.ethereum.org/@parithosh/zhejiang-testnet-spec
From Etan (Nimbus) to Everyone 02:28 PM: https://github.com/ethereum/pm/files/10237672/hexssz.pdf
From Hsiao-Wei Wang to Everyone 02:30 PM: Etan’s PR: https://github.com/ethereum/consensus-specs/pull/3078
From Hsiao-Wei Wang to Everyone 02:41 PM: I feel that people generally agreed to keep using Devcon City names for EL and Star names for CL. But we might need to choose a better name to describe the hard fork when we are talking to the media later.
From Barnabas Busa to Everyone 02:42 PM: What will happen if we run out of devcon city names ?
From Etan (Nimbus) to Everyone 02:42 PM: Host more devcons 🙂
From Łukasz Rozmej to Everyone 02:42 PM: Deneb
From Hsiao-Wei Wang to Everyone 02:42 PM: So, would like to check (1) are CL ppl happy with star name? (2) any ideas other than Duhbe? (3) if not, I can seek for community feedback after the call
From stokes to Everyone 02:42 PM: Stars good
From Pooja Ranjan to Everyone 02:43 PM: Happy to support with name!
From Łukasz Rozmej to Everyone 02:43 PM: https://en.wikipedia.org/wiki/Deneb
From Hsiao-Wei Wang to Everyone 02:43 PM: 👍
From arnetheduck to Everyone 02:43 PM: or call it dname
From Pooja Ranjan to Everyone 02:43 PM: FEM thread - https://ethereum-magicians.org/t/rfc-post-merge-network-upgrade-naming-schemes/11977/9?u=timbeiko
From Micah Zoltu to Everyone 02:44 PM: I like Shangpella or Caphai.
We can keep two names, but communicate with the combined name.
From Micah Zoltu to Everyone 02:52 PM: Big fan of defaulting to safe (assume false).
From stokes to Everyone 02:52 PM: +1
From Micah Zoltu to Everyone 02:53 PM: Resync with weak subjectivity checkpoint.
From Barnabas Busa to Everyone 03:01 PM: Devnet-0 will be relaunching today.
Hopefully all clients can join in 🙂