Meeting Summary:
The meeting focused on the status of DevNet 8 and preparations for testnet, with discussions about recent issues during the activation of Gloss. Barnabas reported that the network was stable until the Gloss activation, when some clients including Lighthouse, Prism, and Taku experienced problems processing builder deposits due to caching issues. Client teams provided updates on their implementations and fixes, with Lodestar and Nimbus reporting no major issues. The group discussed plans for a non-finality DevNet to allow further testing, targeting implementation either next week or the week after, depending on client readiness with trunk branches. Parithosh proposed aligning on September 28th for the Sepolia testnet fork and end of October for the GoAmsterdam fork, though Dima requested later dates for additional testing time. The meeting also covered EIP proposals including EIP 833 for aligning checkpoint routes and an engine API sync status proposal from Telsec, with discussions about the CL block retention window and its potential reduction.
Click to expand detailed summary
The meeting began with participants joining and waiting for a start time. Parithosh outlined the agenda, which included discussing DevNet 8 status, preparing for testnet, and reviewing EIPs that were submitted after the deadline. The meeting was just starting when the transcript ended, with Parithosh about to invite Barnabas or Stefan to present on DevNet 8 status.
The team discussed issues with DevNet 8 after activating Gloss, where some clients including Lighthouse, Prism, and Taku struggled with builder deposits due to caching problems, while GradD experienced a bug with proposal preferences. Dima raised concerns about Lido’s inability to properly test their oracles and tooling due to loss of finality, leading to an agreement to create a new non-finality network in the coming weeks that would be available for at least one to two days before the fork to Blue Amsterdam. The team also noted that Optimism requested another testing location, and the new devnet would be open to everyone.
The team discussed issues with client implementations following a network fork. Eitan explained that Lighthouse had implementation problems with builder deposit optimizations that were merged just before the fork, causing deposits to not get cached properly. Stefan Bratanov reported similar issues with one block taking an unusually long time for DA checks, which caused a cascade of delays affecting the entire system. Nico reported that Lodestar experienced no major issues, with an optimization reducing processing time from 20 seconds to 500 MS, allowing for a smooth fork transition.
The team discussed cache implementations and timing for mainnet deposits ahead of a fork. They clarified that while deposits can be made months in advance due to a long deposit queue, nodes shouldn’t rely on recaching on startup since most nodes won’t be restarting right before the fork. The group then aligned on having a non-finality DevNet ready either next week or the week after, with all clients including Lighthouse, Prism, Teku, Nimbus, and Lodestar confirming they have the necessary changes implemented in their unstable branches for the next week deployment.
The team discussed plans for testnet timelines, with Parithosh proposing to target next week for merging trunk branches and scheduling DevNet testing. They proposed September 28th for the Sepolia testnet fork and end of October for the Hoodi testnet, though Dima suggested moving these dates by about a week to allow more time for DevNet testing. The team agreed to continue discussions on ETHR&D and make final decisions at upcoming ACDC meetings, with the understanding that all plans depend on successful DevNet testing and stable branches across all teams.
The team discussed removing EIP 7610 from Glamstadam, with Jochem-Brouwer explaining that the EIP was about reverting creation in case of non-emples and had performance implications. The group agreed to remove the EIP, with Jochem-Brouwer requesting that Parithosh review the draft PR before approval. Cayman then presented EIP 833 from WeMeetAgain, which proposes moving the checkpoint route calculation to the boundary block leading up to the checkpoint epoch to fix an off-by-one error and improve target vote accuracy.
Dustin presented a proposal to address issues with unstable networks and highly forked networks by allowing the CL to be the sole source of truth for syncing, which would eliminate duplicative and redundant block gossip. When asked if the EIP could be optional, Dustin explained that engine API endpoints need to be mandatory for them to be useful. Kevaundray then discussed a separate EIP from him that reduces the CL block retention window, noting that the original constant needs to be halved due to the introduction of EIP-8061, and suggested standardizing on the epoch number for both CL and EL block retention.
The team discussed a proposal to modify block retention periods, with concerns raised about making the period too short (5 months) as it could be considered “dead” for longer non-finalized chains. Dustin explained that having a very conservative retention period simplifies implementation by not requiring specific finality handling, while Kevaundray suggested using separate constants for different retention periods. The group agreed to focus on the EL side of the proposal while keeping the CL side unchanged, as the current CL implementation works well without requiring changes.The meeting also covered next steps for EIP scaling, with a target for all clients to have their first version of tier lists ready by mid-September. The process will start with obvious DFI and CFI options, followed by rough consensus on the remaining EIPs, aiming to complete the process by DEFCON. Jochem emphasized the need for transparency in ranking systems, suggesting that if rankings are published publicly, they should clearly define what criteria are being used for each tier.
Next Steps:
- Parithosh: Plan and announce a non-finality DevNet (or alternative smaller devnet) for testing, targeting next week, and communicate timelines and agenda on ETHR&D.
- Client teams (Lighthouse, Prism, Teku, Lodestar, Nimbus, others): Ensure trunk branches with all DevNet 8 fixes are merged and ready for testing by next week.
- Client teams: Review and start implementing the remote signer API changes for Glamsterdam.
- Parithosh: Float the proposed Sepolia testnet fork date (28th September) for alignment, with final decision to be made in two weeks at ACDC.
- Parithosh: Approve the draft PR to remove EIP 7610 from Glamsterdam after review.
- Client teams: Provide tier rankings (A/B/C/D/FI) for Higota EIPs by mid-September, including rationale for rankings.
- Dustin: Continue discussion on EIP 3E3 (Reduce CL Block Retention Window) on ETHR&D, focusing on the trade-offs and potential separation of CL and EL retention policies.
Recording Access: