PoS Implementers’ Call #82 - 2022-02-24

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

Agenda: https://github.com/ethereum/pm/issues/484
Livestream: https://youtu.be/fqPk576t5iw

Kiln office hours

merge-devnet-4 updates

All client teams participating (Besu coming soon) in devnet-4. 99.7% participation. But we see lower sync committee participation (~80%), why? Maybe a client bug/issue - please check your clients.

Kiln public testnet readiness

Poll was taken as to readiness of client teams for a testnet launch next Tuesday, per Kiln v1 spec (same config as devnet-4). Unanimously ready.

Kiln v2 spec

v2 spec is out.

Two new features (athentication, exchange) as previously documented, and renaming RANDOM to PREVRANDAO. Need to get renaming done on both Eth1 and Eth2 sides for things to work.

But should we do v2 for the testnet rather than v1? Geth expects to be v2 ready soon - needs testing. Teku is fully v2-ready and falls back gracefully on naming. Prysm needs to test authentication. Lodestar has authentication. Nethermind is close. Lighthouse does not have authentication or renaming yet.

Action: All clients to get renaming done by Monday. This is sufficient to get v1 and v2 clients to talk to each other, so we can launch with this and upgrade clients to v2 spec in slower time.

Potential Kiln launch date

Pari to confirm readiness with client teams on Monday and launch testnet Tuesday or Wednesday accordingly. Contact Pari if you want to run any genesis validators. Otherwise the EF will take the load.

[Marius] It would be good to have a longer PoW period ahead of the Merge event. [Pari] Planning to 100x the TTD, so will be a few days.

Q. What's the situation on authenticated vs. non-authenticated ports? 8550 is unauthenticated. 8551 is authenticated. This differs from Geth's implementation (although the goal is to make it configurable in due course). Geth supports only the authenticated API currently.

Q. Can you still access the eth_ endpoints on the authenticated port of Geth? Don't recall, but it is available on the normal port. Geth may be able to offer this. [Update in chat: all namespaces are available on the authenticated port.]

Recommendation: execution clients should use the port numbers from the spec as defaults.

Over the weekend Danny will make a summary table for client teams to report readiness. (There is already the milestones doc.)

Kurtosis

Kurtosis testing breakout room. A tool for setting up testnets. Please send somebody from each team to the session tomorrow.

Shadow forking Goerli

Plan to do this again with Kiln v2. If anyone has ideas for automated tests to run while shadow forking Goerli please contact Pari.

Beacon APIs

[Mikhail] Optimistic flag status?

[PaulH] Teku and Lighthouse plan to go ahead with it, and there seem to be no objections from others, so will probably become the standard.

Nimbus has an issue switching between v1 and v2 API versions. To be discussed on the PR.

There is also an open PR to better define the eth_ endpoints to be served over the engine API.

Other client updates

Nimbus

[Jacek] Implemented both client and server for light client sync. It takes the trust relationship out of the REST API. [Danny] How do you trustlessy bootstrap the light client committee? [Jacek] Unsure on details just now.

Spec discussion and AOB

Benchmarks

[Leonardo] Plan to re-assess client benchmarks on Merge testnets, and will be contacting teams in respect of this.

Withdrawals

EF research team is actively discussing different withdrawal designs. E.g. push vs. pull. It was determined that pull is the only viable method on the execution layer for 0x00 withdrawal credentials, but 0x01 credentials might support either. Danny is making two sample PRs, one for each case. [Some technical discussion ensues…]

Safe head block

Aditya and Dankrad have made a version of the "safe-head" confirmation rule. [Dankrad did a walk through the doc.]

Question: can this algorithm be efficiently implemented using only information in the protoarray? Pari will summarise the issues and discuss with client teams.

There may need to be some extra accounting in the clients to handle the FFG votes side of things: a running count of the support for any checkpoint after that last justified.

Other

[Proto] Team made a demo of "blob transactions" (data availability) at EthDenver. Preparing an EIP, and it will also involve consensus layer work.

Chat highlights

From danny to Everyone 02:02 PM: https://github.com/ethereum/pm/issues/484
From danny to Everyone 02:23 PM: https://notes.ethereum.org/@timbeiko/kiln-milestones
From pari to Everyone 02:25 PM: Current versions used in merge-devnet-4 can be found here: https://notes.ethereum.org/D5xK4XrmTb6MKGudf_hcrg
Would be great if client teams can update it if/when there are new versions
We use that doc as the source of truth for other setup docs
From Marius Van Der Wijden (M) to Everyone 02:34 PM: All namespaces (eth etc) are available on the authenticated port
cc Paul
From Dankrad Feist to Everyone 02:40 PM: Next Adias and I would like to discuss the safe head rule: https://notes.ethereum.org/@adiasg/safe-head
From Micah Zoltu to Everyone 02:42 PM: We still expose the head to execution layer, just via unsafe_head.
From danny to Everyone 02:43 PM: yes
safe takes over latestis the idea
and a new keyword is added — unsafe
From pari to Everyone 03:05 PM: EF would exit and shut down our Pyrmont nodes soon, please yell if we shouldn’t