Skip to content

kona-node: retry a failed pipeline reset before deriving - #23858

Closed
joshklop wants to merge 1 commit into
developfrom
joshklop/kona-derivation-reset-retry
Closed

joshklop wants to merge 1 commit into
developfrom
joshklop/kona-derivation-reset-retry

Conversation

@joshklop

@joshklop joshklop commented Nov 3, 2026 •

Copy link
Copy Markdown
Contributor

When a derivation pipeline reset failed, kona-node's derivation actor resumed anyway. The next L1 head or EL sync completion stepped a pipeline that had never reset, and nothing sent the reset again. A failed reset now stays pending and is retried before every derivation step, waiting for the next L1 head if it fails again. The derivation actor clears its origin and finalizer and truncates the safe head database once, when the reset arrives, not on every retry. A critical pipeline error from a signal now stops the actor, as in op-node, instead of being retried forever.

Operators see no new flags. A node whose pipeline reset fails, for example during an L1 or L2 RPC outage, now recovers once the reset succeeds instead of deriving from stale pipeline state.

This leaves one known gap. Derivation can still send attributes and finalizations from its old pipeline after an engine reset, for the requests queued ahead of the reset signal. A late safe-head update that arrives after a failed reset still stops the actor, as one after a successful reset does on develop. #23859, stacked on this PR, makes the two actors count resets and drop stale work, which closes both.

Split out of #23548. Reviewed with the rust-code-reviewer and safedb-reviewer agents.

🤖 Generated with Claude Code

A derivation pipeline reset that failed left derivation in
AwaitingUpdateAfterSignal, so the next L1 head or EL sync completion
resumed derivation on a pipeline that had never reset. Nothing sent the
reset again.

Keep a failed reset pending and adopt its head, since the engine is already
there. attempt_derivation retries the pending reset before every step, and
waits for the next L1 head if the retry fails. The one-time reset work
(clearing the origin and finalizer, truncating the safe head database) runs
once when the reset arrives, not on every retry, and EL sync completion
retargets a pending reset to the head it truncated to. A critical pipeline
error from a signal now stops the actor instead of being retried forever.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@joshklop
joshklop force-pushed the joshklop/kona-derivation-reset-retry branch from 555d7ca to 4ec0486 Compare November 3, 2026 05:00
@joshklop joshklop changed the title kona-engine,kona-node!: retry failed pipeline resets and drop stale derived work kona-node: retry a failed pipeline reset before deriving Nov 3, 2026
@joshklop
joshklop added this pull request to stack #23860 November 3, 2026 05:20
@joshklop

joshklop commented Nov 3, 2026

Copy link
Copy Markdown
Contributor Author

Closing in favor of #23815, which fixes the same failed-pipeline-reset bug (ethereum-optimism/protocol-team#586) with backoff retries and a fresh engine reset when the reset head can't be walked back from. The reset-epoch change in #23859 now builds on #23815 instead of this PR.

@joshklop joshklop closed this Nov 3, 2026
@joshklop
joshklop removed this pull request from stack #23860 November 3, 2026 18:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant