github ethereum-optimism/optimism kona-node/v1.11.0
kona-node v1.11.0

36 minutes ago

Overview

Note

kona-node is not production-ready. Production deployments should use op-node.

This release changes which address kona-node advertises when it listens on all interfaces, and makes it refuse to start on alt-DA chains. Review the breaking changes before you upgrade. It also enforces peer blocks on inbound and already-open connections, holds derivation until a failed pipeline reset succeeds, and lets superroot_atTimestamp answer for blocks far below the recorded safe head.

Breaking changes

  • kona-node refuses to start on alt-DA chains (#23786). kona-node has no alt-DA support. It now exits at startup when the chain's rollup config, from the embedded registry or from --l2-config-file, has an alt-DA config, for example Redstone or Celo Sepolia. Run op-node on these chains.
  • No advertised IP until peers agree on it, when listening on all interfaces (#23874). A node listening on 0.0.0.0 or :: (the default) without --p2p.advertise.ip no longer advertises the first interface's address, usually 127.0.0.1, in its ENR. As in op-node, it advertises an IP only once enough of the peers it contacts agree on its address. Bootnodes do not count. The new --p2p.discovery.ip-votes (KONA_NODE_P2P_DISCOVERY_IP_VOTES, default 10) sets how many. On a small or private network, such as a devnet, other nodes may never find such a node: set --p2p.advertise.ip to the address peers reach it at, or listen on that address with --p2p.listen.ip. An explicit --p2p.advertise.tcp is now kept, where it used to be overwritten with the bound port.

Chain Configuration

  • sepolia-devnet-4, -5 and -6 added (#24054). These are OP Labs devnets on Sepolia used to rehearse upcoming network upgrades, listed in the superchain registry. kona-node can now start them by name with --chain, without an explicit rollup config. No other chain's configuration changes in this release.

Other changes

Derivation

  • A failed derivation pipeline reset no longer lets derivation resume from stale data. A temporary failure, such as an L2 RPC error, is retried with backoff from 100 ms up to 10 s, and derivation holds until the reset succeeds. If the L2 chain reorged under the reset head, the node asks the engine for a fresh one. Only a critical failure stops the node. The new kona_node_derivation_pending_reset_attempts gauge shows a node stuck retrying (#23815)

P2P

  • Peer, address and subnet blocks (opp2p_blockPeer, opp2p_blockAddr, opp2p_blockSubnet) now refuse inbound connections and close existing ones in both directions, including to protected and static peers. Previously they only stopped the node's own dials. Unlike op-node, a block closes connections that are already open. Score bans from --p2p.ban.peers now also refuse the banned peer in both directions until the ban expires. Refused inbound connections are counted in the new kona_node_gossip_inbound_connections_refused metric, by reason (#23801)
  • opp2p_connectPeer now waits up to 30 seconds for the connection, as op-node does, and returns the reason when the peer is blocked or the dial fails. Previously it always failed at once with "Peer not connected". opp2p_disconnectPeer now waits for the connections to close. Failures of both methods now return code -32000 with the reason, as in op-node, instead of -32602; update tooling that matches on the old code (#23801)
  • opp2p_peers now reports latency in nanoseconds, as op-node does. It used to be in whole seconds, so it always read 0 (#23801)
  • kona-node no longer serves the payload_by_number sync protocol, which op-node has retired. A peer could use it to make the node run out of memory. op-node versions that still run the sync client get "protocol not supported" from kona-node (#23801)
  • Peers connecting and disconnecting are now logged at debug level, not info and warn (#23801)

RPC

  • superroot_atTimestamp now reads the target block by number. Previously it walked back one block at a time from the recorded safe head, which could exceed the RPC timeout over a remote execution client when span batches put the target far below that head. A target removed by a reorg or rewind during the read now returns -32000, as other history changes do, instead of an internal error (#23960)

Full Changelog: kona-node/v1.10.0...kona-node/v1.11.0

🚢 Docker Image:

Don't miss a new optimism release

NewReleases is sending notifications on new releases.