PoS Implementers’ Call #73 - 2021-09-23

--- tags: eth2devs description: Notes from the regular proof of stake [Eth2] implementers call image: https://benjaminion.xyz/f/favicon-96x96.png --- # PoS Implementers’ Call #73 - 2021-09-23 [Quick contemporaneous notes by Ben Edgington; fka "Eth2 Implementers' Call"] Agenda: https://github.com/ethereum/eth2.0-pm/issues/237 Livestream: https://youtu.be/Pes_OaMJeDc ## Altair ### Sync committee performance Thanks to those who have been working through data analysis and bug fixes on sync committee performance. Looks like all sync committee issues are now resolved. Nimbus issue explanation: there is an anti-DoS filter in the libp2p - forgot to whitelist the sync topics (oops). Need to ensure we have decent monitoring for this. ### Upgrade epoch selection Additional testing on Pyrmont has been done (by Pari). Sync committee issues resolved. Look to get client releases at least 16 days prior, to allow for a blog post 2 weeks ahead. [Me] Would prefer 3-4 weeks lead time to allow larger staking operations more time to upgrade. Target client Mainnet releases for end of next week; blog post Monday Oct 4. Agree on Wednesday 27th. Epoch number TBD after this call. ## Client updates **Teku** - Enrico intro - Fixed an issue that came up on Prater where, in a [rare edge case](https://github.com/ConsenSys/teku/pull/4358), Teku would fail to produce blocks. A kind of race condition: see an attestation in a block before we receive it via gossip plus some other weird conditions. Very rare on Prater, and not reported on mainnet. Fixed in 21.9.1 release. - Upgraded to Blst 0.3.5 (the latest Blst release) - Support for building on JDK 17, which is the new LTS release of Java. - Working on Merge support in a dedicated branch. **Prysm** Everything merged for Altair. Some minor bugfixes to RPC endpoint. Prepping for v2 release end of next week. Lots of refactoring, renaming, etc. Also working on Merge support on a dedicated branch. **Nimbus** Working on passing all the new tests. Making sure Prater works: sync committee issues, low number of peers issues resolved. Release next week to fix everything. **Lodestar** [Post-meeting update] - Good progress on merge impl, 50% done - Achieving good consistent attester and proposer performance - Will release a guide to stake with Lodestar this week **Lighthouse** Merged 2 big PRs: checkpoint sync and batch BLS verification of attestation signatures. Pushing v1.6.0 release candidate next week. Work on fingerprinting of attestation packing to diagnose client distribution was published. Investigating a p2p bandwidth reduction optimisation. Implementing the Merge. **Grandine** Still focusing on multiple runtimes approach. Can sync Altair chain. Validator side needs more work. Should be running on Prater in a couple of weeks. ## Merge discussion Refining initial interop spec for the Merge. Planning to release a frozen spec for the consensus layer tomorrow. Stabilising EIP-3675. Should be ready for interop testnets early October. Mikhail is working on a meta document to cover loose ends. See also https://github.com/ethereum/execution-apis/pull/74. Does anyone have an execution node that is up to spec? [Proto] Many-many integrations leads to a quadratic increase in complexity. Mocked versions of the execution and consensus clients would be very useful and would simplify interop debugging. [PaulH] Looking to make a mocked execution engine on Lighthouse side. [Jacek] These APIs (execution to consensus) shouldn't normally be exposed to users, so they can be separated away from the other APIs. Using REST for these makes sense for many reasons and will be a big simplification - does everyone agree, or any objection? [Danny] Primary issue is that the execution side already has JSON-RPC. [PaulH] RPC is used in preference to REST since no load-balancing concerns - the two sides need to stay in sync, which implies RPC. [Jacek] REST and RPC are similar in this respect. Eth1 experience was an early motivator for choosing REST for the beacon node. [Danny] But that was in the context of a user API. This does not have the same requirements. [Mikhail] REST is coupled with HTTP, whereas RPC can be layered on other protocols. [Danny] From a short-term practical standpoint, it's not likely to change, but longer term let's discuss further. Tomorrow's release will have a minimal set of consensus test vectors for the Merge. ## Research updates **Leo** Paper on the Eth2 network crawler has been accepted at IEEE Estonia conference. Note that Michael Sproul's network analysis is surprisingly similar to the crawler data though derived totally independently. Other paper on Eth2 client resource analysis to be [presented next week](https://whova.com/embedded/session/cbrai1_202109/1876719/) at [this event](https://brains.dnac.org/program-2/). [Mamy] There is a crawler (Stereum) that connects primarily to Nimbus nodes. There are different behaviours between clients that make these analyses tricky. General agreement that network crawling is difficult. ## Spec discussion and AOB Skip meeting in two weeks since a number of us will be f2f working on Merge interop. Regroup on 21 October, ahead of Altair upgrade. Attestation inclusion: Prysm has optimisations around epoch boundaries, but they are behind a feature flag. In v2 release this will become the default. * * * # Chat highlights From Micah Zoltu to Everyone: 03:10 PM : What time of day is the target epoch going to be? From protolambda to Everyone: 03:11 PM : https://slots.symphonious.net/ Click upgrade-scheduler tab From danny to Everyone: 03:19 PM : https://github.com/ethereum/execution-apis/pull/74 From Leo (BSC) to Everyone: 03:29 PM : http://intelligenttech.org/BCCA2021/ https://twitter.com/sproulM_/status/1440512518242197516 https://brains.dnac.org/program-2/