Overview
Important
This is a recommended release for sequencer operators, and an optional one for other node operators. It contains a change to how sequencers publish new blocks, which removes the block-building stalls caused by slow remote signers, and the removal of the server-side RPC metrics.
Other changes
Sequencer
- Sequencers now insert each new block into their own execution client before publishing it, and publishing runs in the background instead of holding up the next block. A slow remote signer (
--signer.endpoint) no longer delays block production, and peers only receive blocks the sequencer has already adopted as its head, so a block the execution client rejects can no longer cause a reorg on every verifier. Failed publishes are retried about once a second until the block leaves the gossip window set by--p2p.gossip.timestamp.threshold, after which it is dropped with aGiving up on publishing blockerror log.publishing_errors_totalnow counts every failed attempt rather than one per block. New metrics:sequencer_insert_seconds,sequencer_publish_handoff_seconds,publish_queue_len,publish_delay_seconds,dropped_publishes_total, andunsafe_payload_arrival_delay_seconds, which every node records when it receives a block over p2p (#22967)
Metrics
- The server-side RPC metrics
op_node_default_rpc_server_*(requests, durations, responses, parameter and result sizes, notifications sent) andop_node_default_rpc_client_notifications_received_totalare no longer exported. Dashboards and alerts that query them need updating; theop_node_default_rpc_client_*request metrics are unchanged (#23114, #23055)
Full Changelog: op-node/v1.19.8...op-node/v1.19.9
🚢 Docker Image: