Meeting Summary:
The team reviewed the status of the Fusaka Devnet 3 and discussed various technical issues including peering problems, orphaned blocks, and registration workflow challenges. They addressed concerns about JSON-RPC response hashing and agreed on standardizing precompiled contract addresses, while also discussing peer connectivity issues on Mainnet and blueprint project updates. The team reviewed benchmarking results for client execution times and discussed state growth testing, with several team members reporting progress on devnets and ongoing investigations into various technical implementations.
Click to expand detailed summary
The team discussed the status of the Fusaka Devnet 3, which went live last week with 18 BPOs and 30% gigabit nodes. They identified issues with Nimbus and Nimbus EL clients experiencing peering problems, and discussed an unusual pattern of orphaned blocks due to delayed column propagation. Bharath provided updates on the MAP workflow, noting that 672 validators were registered out of 818, and highlighted two ongoing issues: timestamp problems with Lodestar registrations and state root mismatches for Nimbus block publishing. The team also discussed the implementation of private block mempools and the need to test the new configuration checks in EIP-2384.
The team discussed concerns about hashing JSON-RPC responses, with Marius expressing that the current method using RLP is brittle and suggesting alternatives like RRP. Danno agreed to remove the hash from the specification, and Parithosh confirmed that an updated PR would be submitted incorporating the EIP and execution API changes. The team also aligned on normalizing keys and values in addresses for precompiled contracts, with Danno proposing to standardize the format to “precompile:name:address” and Mario agreeing to this change.
The team discussed ongoing issues with peer connectivity on Mainnet, where Alexey and Marius reported seeing decreased visibility of peers, which Csaba attributed to recent changes in node connection logic. The team agreed to investigate further, with Csaba requesting more details about the timing and specific issues. Carlos presented updates on the blueprint project, requesting reviews of two documents containing test cases and questions for finalization, which the team agreed to review. The team also discussed the timeline for testing, with Stokes indicating they should move forward as soon as possible, pending resolution of Nimbus and map workflow issues.
Marius explained that a memory issue on Bloatnet was caused by Geth ignoring tracing flags, resulting in excessive memory usage and node crashes. The problem was resolved by updating to the current master branch, which fixed an outdated image issue. They are also reworking the tracing to use less memory. CPerezz asked about the database’s maximum payload capacity, but Marius clarified that this was a separate issue related to the log indexer. Parithosh mentioned he had started re-syncing the node with the fried database.
The team discussed state growth and state root computation testing, with Mario expressing interest in working on computationally intensive benchmarking tests. They agreed to test with a blocknet state size of 2x mainnet, with plans to expand to larger state sizes in the future. Parithosh mentioned the Netherland team’s use of a specific file system for quick snapshot copying and reverting, which could be useful for gas benchmarking and Yields integration. Mario reported a new release of benchmark tests in East, which includes a consolidated genesis file for Nevermind tool integration, though some updates to benchmark tests are still needed.
Louis presented benchmark results for calculating leading zeros in code, comparing different gas costs and transaction limits. He explained the test setup, which includes filling test cases and running benchmarks with various client configurations. The results showed that with a gas cost of 5, both 72 and 100 million gas transactions were successfully completed. The team noted that the current gas cost of 5 might be slightly conservative, but the results were still acceptable.
The team reviewed benchmarking results showing client execution times below 4 seconds without warming up, with an open issue raised for adding warming up functionality later. They discussed maintaining the current status quo of 5 clients while considering repricing in the next fork, and Mario suggested implementing EELS methodology for future hard forks. Minhyuk reported progress on two devnets, including tests with regular transactions and a 16-node network with data column propagation, while noting some issues with Teku’s backfill implementation that required further investigation with the Teku team.
Next Steps:
- Client teams to focus on investigating the orphaned block issue and any client-specific bugs found in Fusata Devnet 3.
- Parithosh to make a post on the interop channel regarding the plan for the next phase of testing.
- Client teams to reach out if they notice any peering issues with their nodes on mainnet.
- Csaba to investigate potential peering issues related to recent Geth changes.
- Nethermind, Besu, and Aragon teams to review the document shared by CPerezz regarding test cases for Bloatnet.
- CPerezz to follow up asynchronously with Kamil regarding the integration of Yul’s with the gas benchmarking tool.
- Client teams to engage with the two open metrics PRs.
- Client teams to review the hack.md document shared by Carlos.
- Danno to update the PR for the EIP and execution API regarding the eth_config changes.
- Client teams to check if they are ready for backfill implementation and reach out to Sunnyside Labs for testing if needed.
Recording Access:
