Consensus Implementers’ Call #99 - 2022-12-01

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

Agenda: https://github.com/ethereum/pm/issues/667
Livestream: https://youtu.be/KFc1sWYlVZ4

Capella

Last week's All Core Devs call recap (of consensus stuff). Danny understands that we will keep withdrawals separately specified from EIP-4844, and slated for Capella. If they do end up getting implemented together, we would not combine the specs, but would just stagger the upgrade. This was confirmed to be the common understanding on this call.

testnets / progress updates

Running two testnets (no EIP-4844): one post-merge, one pre-merge (to facilitate Prysm). Many participants… (see recording for the details). Full withdrawals working fine.

Nimbus not yet participating, the consensus layer withdrawals implementation is in progress.

BLS credentials change operation is being tested with ethdo, and a subset of the clients. Gossip of change messages not yet tested, but is about to be.

Plan to move to a longer-lived single testnet when Nethermind and Besu are ready, and Prysm can start from a post-merge genesis. (I.e. soon.)

bound withdrawals sweep

Link.

[Alex] The bound is set very small, following feedback. No objections to moving forward with this. More withdrawals test cases coming to cover edge cases - by end of next week.

EIP 4844

[I probably garbled some of the below…]

[SeanA] Issue to discuss around syncing blobs. When finality is delayed (to before the prune depth), it might be complex to validate blob availability. [Danny] The link between finalisation and data availability is not clear - availability should be checked regardless. [Proto] To clarify, if the chain has not finalised for 18 days (the pruning period), there is a portion of the blob data that we no longer have and is not finalised. [Sean] Peers might lie about what the finalised epoch is, so we might need to reprocess part of the chain. [Jacek] We never trust; always verify. [Sean] Not feasible during an optimistic sync. Concern that this issue could introduce complexity. [Proto] There are a couple of edge cases, but we could simplify the spec to say just verify the last 18 days, and verify finality as we currently do. [Sean] Needs more thought. [Terence] Spec says we cannot import a block without a blob - are we thinking of changing this. [Proto] For 18 days' worth, no change, but change it if non-finality.

Action: discuss further on the PR

Could we store blobs until last finality if it is earlier? [Micah] Doesn't this lead to unbounded state growth? [Jacek] No, because the inactivity leak would lead the chain to finalise within a month or so.

KZG Ceremony

Trusted setup, Second audit with SigmaPrime being completed. Moving into public contribution period in the New Year. Starting to ramp up communication/education. If you have a community you want to reach, contact trent or Carl. Aim for it to be the largest "summoning ceremony" in crypto: 8-10k people.

For those with an interest in cryptography there will be a grants round for independent implementations. Maybe a day or two's work.

Public contribution period will be about 2 months.

[Jess - Coinbase] Aligned on the social/content side. Working on call to action in their product.

[Carl] Don't hesitate to ask if you have any questions or concerns about the process.

Research, spec, etc

(networking) attnets revamp

Link

[Age] Attnets revamp. Add a single deterministic subnet to each node, and remove random subnets. Planning to implement in Lighthouse. What is the general view?

Will run on testnets first, but these have very different node distributions, and may require more subscriptions per node.

[Jacek] Most interesting thing is that it changes sync committee subnets: currently getting onto a sync committee mesh is a problem as they are sparse. Hope that this does not make other committee subnets too sparse.

[Danny] Goal is to make sure that all nodes contribute, while reducing bandwidth overall. Hard to validate a priori.

[Age] Will post some statistics. Nodes currently subscribed to 64 subnets would be reduced to 1. But often new peers cannot connect to these nodes anyway as they are full, so they might have limited usefulness. [Danny] With the new scheme, the connectivity scales with the network size rather than the distribution of validators. [Age] It should be a strict improvement.

[Jacek] Would like to see this added to the spec (as a matter of principle), even if it is backward compatible. Also note that it is all still based on honesty: nodes can lie about their subnet subscriptions. Would be good to validate if possible: but difficult to do.

Action: Age to post stats on this.

(networking) NAT hole punching

[Age] Many peers are behind NAT. Discv5 hole punching needs some extra fields in the ENR. Lighthouse is working on this. Happy to work with other teams on this to test interop. [Jacek] Libp2p is doing AutoNAT 2.0 - how does this relate? [Age] That is enirely TCP based, and needs a relayer. We can omit the relayer as we have discovery. There are PRs for this on the Discv5 spec.

(networking) ipv6 dual-stack support

Link

[Age] IPv6 support: need IPv6 compatible boot nodes. Considering enabling v6 on the Lighthouse bootnodes. This will allow for a future upgrade in LH clients. Any issues? Are we concerned about a network split between v4 and v6 nodes? LH plans to run only dual-stack for this reason, but it opens up the possibility of a partition more generally.

[Micah] Can we make it hard for people to fall into IPv6 only?

[Micah] Related to the previous point, it would be neat if nodes could use IPv6 to establish a NAT hole for IPv4. [Age] Should be possible.

Action: Age will put up an issue for comments.

(Beacon API) checkpoint sync api status

Link

[Mikhail] Previously proposed two endpoints for checkpoints: one for state providers, one for trust providers.

The Checkpointz tool now provides everything we need for state provision in a more sophisticated way. And it can do trust provision as well. Some concern that Checkpointz becomes the only provider of this data. There are a couple of other proposals in flight that mitigate that risk.

Therefore, requesting to deprecate this proposal. No objections on the call. Will close it by end of the week if nobody raises an objections.

(Beacon API) Attestation and sync committee aggregation selection

Link

[Oisin - Obol] Aggregation duty for distributed validator technology. Currently aggregation duty is decided by hashing a signature over the slot. This is not MPC-friendly. Proposal is to add a couple of API endpoints to make this work with distributed validators.

Action: teams to review, and decide on next call whether to implement.

(engine api) engine api spec improvement proposal

Link

[Mikhail] Two main questions:

1. Do we need engine_getCapabilities method? Jacek's optimistic strategy of falling-back on error seems reasonable. No reason to support EngineAPI versions two or more generations old. Will not introduce this method if it's not needed. Action: consensus client devs - give feedback on your EngineAPI implementations as to whether this is needed.

2. Structure of EngineAPI spec documents. Two approaches:
   a. New spec for each fork, or
   b. Split by functionality.
   See Mikhail's examples of each. Pro and con exist for both.

AOB

[Micah] Client diversity.org has wildly divergent measures of client numbers. [Danny] One is crawler derived, the other from block data. The latter should be fairly accurate measure of validator proportions. The former of node count. The crawler data looks strange.

[Jacek] Heads up for a Nimbus release coming with a separate validator client component. Will be keeping the consolidated VC version around as well (with long term support). [Saulius] Grandine is working on similar - does Nimbus VC use the Beacon API, or an internal API? [Jacek] The internal VC uses a private non-JSON version, with a couple of shortcuts. Can mix-and-match VC/BN combos via the JSON API.

Next call 2 weeks from today.

---

Chat highlights

14:01:55 From danny To Everyone:

https://github.com/ethereum/pm/issues/667

14:11:31 From Trent To Everyone:

late addition to the agenda but myself or @Carl would like to shill the KZG Ceremony briefly when there's a moment

14:11:45 From seananderson To Everyone:

https://github.com/ethereum/consensus-specs/pull/3141

14:28:16 From danny To Everyone:

nice

14:29:28 From danny To Everyone:

https://github.com/ethereum/consensus-specs/issues/2749

14:41:22 From zahary To Everyone:

Age, have you discussed these protocol upgrades with the Portal network community?

14:44:09 From danny To Everyone:

Clients must support listening on at least one of IPv4 or IPv6. Clients that do not have support for listening on IPv4 SHOULD be cognizant of the potential disadvantages in terms of Internet-wide routability/support. Clients MAY choose to listen only on IPv6, but MUST be capable of dialing both IPv4 and IPv6 addresses.

14:45:06 From Micah Zoltu To Everyone:

What is the reason for the hard requirement of being able to communicate with IPv4?

14:50:41 From stokes To Everyone:

Backwards compatibility/reachability, I’d say

14:52:18 From Age Manning To Everyone:

I think everyone defaulted to ipv4. There were biases in discovery (at least in our discovery implementation) where given the choice the ipv4 socket would be chosen. The network is now pretty much all ipv4 now, so we have to transition to allow both to support ipv6.

14:55:56 From oisinkyne To Everyone:

https://github.com/ethereum/beacon-APIs/pull/224

14:56:43 From Nishant Das To Everyone:

+1 , also ipv4 was very straightforward for all clients to support. But ipv6 support had gotchas (at least for us). We had a few users running only on ipv6 who couldn't peer succesfully because none of the bootnodes supported it.

15:07:27 From Potuz To Everyone:

@danny just want to update on Capella testing: we just checked that valid BLS_TO_EXECUTION_CHANGES are being broadcasted in the devnet and picked up by other validators and packed into blocks (as oppposed to the validator that got it in the pool packing it itself)

15:09:04 From terence To Everyone:

Teku is working. We still have to check prysm picks them up though 🙂

15:09:31 From Enrico Del Fante (tbenr) To Everyone:

🥳

15:10:20 From Potuz To Everyone:

prysm I already checked
with another prysm

15:11:38 From terence To Everyone:

👍

15:12:32 From Potuz To Everyone:

ok checked all clients in the devnet now

From Tim Beiko to Everyone 03:13 PM

Is "Miga Labs" the crawler?

From Mikhail Kalinin to Everyone 03:14 PM

Sigma prime hides real LH usage
😱

From Tim Beiko to Everyone 03:14 PM

lmao

From danny to Everyone 03:14 PM

https://www.nodewatch.io/

From terence to Everyone 03:15 PM

🎉

From pari to Everyone 03:15 PM

Will both get LTS? Or is the split validator the long term plan?

From protolambda to Everyone 03:19 PM

Before we end this call: it's been exactly 2 years since beaconchain launch, happy anniversary everyone :)

From Tim Beiko to Everyone 03:19 PM

holy shit!!

From pari to Everyone 03:19 PM

❤️

From Tim Beiko to Everyone 03:19 PM

hbd, beacon chain

From terence to Everyone 03:19 PM

Woohoo!