Skip to content

kona-node: resend the forkchoice update until the EL answers VALID after an initial SYNCING #24030

Open

Description

@einar-oplabs

Part of ethereum-optimism/protocol-team#545.

Summary

If kona-node's forkchoice update is answered SYNCING, the execution layer (EL) then finishes syncing, and no further unsafe payload arrives, kona-node never completes engine sync. It never resends the forkchoice update, so it never sees VALID and el_sync_finished stays false. The engine actor never reports sync completion. Under delegated derivation, the DelegateDerivationActor therefore never polls its source, and local-safe, safe and finalized stay at genesis. develop has the same behavior.

Evidence

CircleCI job 5859222 (memory-all-kona-op-reth-fusaka), test TestFollowL2_LabelsMatchELByHashWithSequencerPaused from #24007:

  • The follow-source kona-node's op-reth logged Received new payload ... number=7, then Received forkchoice updated message when syncing.
  • 16 ms later it logged Canonical chain committed number=7, so its p2p sync had finished.
  • The sequencer stopped 2 s later. After that, op-reth logged no further newPayload or forkchoiceUpdated from kona-node, and its status lines show latest_block=7 until the test ended four minutes later.
  • kona-node's local-safe stayed at 0 for 120 attempts, while the op-node sequencer and verifier reached local-safe 7.

The test passed while it waited for L2 finality before pausing: continued gossip produced a forkchoice update answered VALID.

Mechanism

  • The insert task applies the unsafe head on a SYNCING reply.
  • el_sync_finished is set only by a VALID forkchoice reply, in synchronize.
  • While el_sync_finished is false, nothing resends an unchanged forkchoice update: there is no timer, and no reset is triggered.
  • The engine actor marks sync complete only once el_sync_finished is set, and the delegate actor polls only after that.

Expected behavior / acceptance criteria

  • While el_sync_finished is false, kona-node resends the unchanged forkchoice update on a bounded interval until the EL answers VALID. Compare with op-node's handling during initial EL sync.
  • Regression test with kona-engine and mocked Engine API responses: the first forkchoice update is answered SYNCING, no further payloads arrive, and a later resend answered VALID completes engine sync.
  • TestFollowL2_LabelsMatchELByHashWithSequencerPaused passes under kona-node with the sequencer paused right after kona-node's first unsafe block. kona-node-service,op-acceptance-tests: exercise delegated derivation recovery end to end #24007 currently avoids that case by waiting for local-safe ≥ 1 before the pause.

Related

Activity

  1. self-assigned this
    on Nov 10, 2026
  2. joshklop commented on Nov 11, 2026

    @joshklop
    Contributor

    is this the same as Is this the same as https://github.com/ethereum-optimism/protocol-team/issues/681?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions