Click to expand detailed summary
The team discussed the success of the Petra retrospective blog post, which accurately estimated the network’s blob count. They agreed to proceed with increasing the gas limit to 60 million, with client teams tasked to monitor metrics and signal the change asynchronously. The group decided to move all PRDOS testing to the Fusaka testing stream, with all relevant chat conversations to be moved to the interop chat channel. Barnabas provided updates on Fusaka Devnet 0, detailing the changes in maximum drop count across different epochs.
The team discussed the implementation status of validator custody and data availability features across different client implementations. Barnabas reported successful BPO testing with no issues found, while Parithosh introduced a new probe tool for network testing. Client teams provided updates on validator custody implementation, with Lighthouse expecting completion before the interop event and Members having validated custody but discussing potential issues with dynamic DA changes. Justin presented a merged PR that simplifies the custody specification by allowing implementers to wait until the custody period is passed before making changes, addressing concerns about network stability and backfilling requirements.
Manu discussed the challenges node operators face when switching between beacon nodes to minimize IP loss, highlighting that the current setup limits a node to serving only four data clubs, which undermines the purpose of validator custody. Justin Traglia suggested changing the “should not backfill” specification to “may not,” acknowledging that this could be an implementation detail. Pawan and Agnish explored the practicality of a static custody model and discussed endpoint usage in Prism for determining validator attachments, with Manu clarifying the relevant API endpoint. The group debated the pros and cons of shifting to a static custody model, with concerns raised about the impact on supernode availability and network performance.
The team discussed concerns about the accuracy and enforceability of validator custody counts in multi-BN setups, with Agnish raising questions about potential risks and attacks. Francesco explained that while custody is important for ensuring a baseline of super nodes, it’s not strictly enforceable and can be influenced by node operators’ setups. The team agreed to take the issue offline for further analysis, with Raúl and Francesco planning to collaborate on a deeper examination of edge cases. They decided to implement a MAY functionality as a temporary solution, with the goal of having a final solution for validator custody in place by next week.
The team discussed the upcoming Fusaka devnet 1 launch on Monday, June 9th, which will include EIP-7917 for deterministic proposer look-ahead and validator custody changes. They agreed to postpone the BPO networking changes and metrics implementation to devnet 2, as these changes would require more time for review and testing. Client teams reported progress on implementing the necessary changes, with most clients on track for the devnet release. The team also discussed the need for sales teams to review and approve the EIP-7917 specification.
The team discussed metrics implementation, with Barnabas taking lead on merging metrics PRs and addressing Justin’s question about BPO fork schedules. Parithosh agreed to add EL metrics changes to devnet and create a repository for EL metrics to share PRs and discuss implementation. The Sunnyside Lab team provided an update on their testing progress, having conducted 7 dead notes over the past 10 days.
The team analyzed the impact of get blob enablement on different CEOs and found interesting results. Without get blobs, block time was more stable but block counts were lower. With get blobs enabled, block counts increased by 15% for Lighthouse and 3% for Prism. They discovered that Grandine requested multiple get blobs when failing, while other CEOs requested only once. The team also found that disabling get blobs reduced data column failures on Lighthouse, leading them to suspect get blobs might be handling wrong columns. They plan to test this with Russ to confirm. Additionally, they noted that column processing time was inversely proportional to the number of requests, with Grandine having the most requests and shortest time, and Teku having the least requests and longest time.
The team discussed technical issues related to local size and CPU RAM, with J suggesting an increase in local size to 42.6 MB due to blob size concerns. Parithosh proposed creating threads on the interop chat for further discussion with respective teams. The group also confirmed migrating to the Fusaka devnet 0 spec for the next version of the net. Parithosh announced the cancellation of next week’s call due to an in-person interop event and ACD meeting. Finally, they discussed adding column site calls to the Beacon API event stream, with Dmitrii noting its absence in the backend API but presence in a Telco customer implementation.