Repository navigation
Conversation
This was referenced Nov 3, 2026
6 of 7 tasks
joshklop
force-pushed
the
joshklop/kona-reset-epoch
branch
from
November 3, 2026 18:06
dbfc333 to
e96990b
Compare
joshklop
changed the base branch from
joshklop/kona-derivation-reset-retry
to
einar/kona-failed-reset
November 3, 2026 18:06
einar-oplabs
self-requested a review
November 5, 2026 10:09
joshklop
force-pushed
the
joshklop/kona-reset-epoch
branch
from
November 19, 2026 23:32
e96990b to
d988d28
Compare
Derivation keeps handling the requests queued ahead of an engine reset with its old pipeline, so attributes and finalizations derived before the reset reach the engine after it. Consolidated, those attributes moved the safe head past blocks that were never re-derived, recorded the wrong block in the safe head database after an L1 reorg, and could leave derivation waiting forever for a confirmation; a stale finalization above the rewound safe head stopped the engine. Now: - both actors count engine resets: the engine counts the reset signals it sends, derivation the ones it handles, and derivation stamps its attributes and finalizations with its count. The engine drops work stamped before its latest reset, and drops consolidations and finalizations queued when a reset succeeds; - derived attributes whose parent is not the local-safe head mean the actors disagree, so the engine resets instead of consolidating them; - the engine reports the local-safe head before starting a reset or backing off, so a confirmation earlier in the same drain is not lost; - derivation ignores a safe head that does not confirm the attributes it is waiting for, instead of adopting it and leaving a gap in the safe head database. Derivation counts a reset when the engine's reset signal arrives, not on each retry of its pipeline reset. BREAKING CHANGE: EngineActorRequest::ProcessSafeL2Signal is replaced by ProcessDerivedAttributes and ProcessDerivedFinalization, which carry the derivation actor's reset epoch, and ProcessDelegatedSafeHead. DerivationEngineClient::send_safe_l2_signal is replaced by send_derived_attributes, send_derived_finalization and send_delegated_safe_head, and Engine::drop_derived_tasks is new. ConsolidateTaskError has a new DoesNotExtendLocalSafe variant, which requests a reset. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
joshklop
force-pushed
the
joshklop/kona-reset-epoch
branch
from
November 21, 2026 23:13
d988d28 to
61370c3
Compare
joshklop
changed the base branch from
develop
to
joshklop/kona-el-sync-completion
November 21, 2026 23:13
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #23883 (base
joshklop/kona-el-sync-completion), so the diff shows only the reset-epoch change. Stack, in merge order: #23990 → #23883 → #23859.Closes https://github.com/ethereum-optimism/protocol-team/issues/587
After an engine reset, kona-node's engine and derivation actors could disagree. Derivation keeps handling the requests queued ahead of the reset signal with its old pipeline. The attributes and finalizations it produces then reach the engine after the reset. Consolidating those attributes moved the safe head past blocks that were never re-derived, and after an L1 reorg it recorded the wrong block in the safe head database. Disagreements stalled derivation instead of recovering. A stale finalization above the rewound safe head stopped the engine.
Operators see no new flags. After a reset, a node that would have stalled, recorded the wrong safe head or crashed now recovers.
The two actors' counts stay equal because the channel between them is FIFO, a failed reset send stops the engine, and the node stops every actor when one exits. Derivation counts a reset when the engine's reset signal arrives, not when #23815 retries its pipeline reset; the parity tests check both the retry and the fresh-reset path. A debug assertion catches derivation counting ahead of the engine. A reviewer should check that this still holds for any future path that sends a reset signal.
Reviewed with the
rust-code-reviewerandsafedb-revieweragents, including after rebasing onto #23815. Not done:BREAKING CHANGE: In kona-engine,
EngineActorRequest::ProcessSafeL2Signalis replaced byProcessDerivedAttributesandProcessDerivedFinalization, which carry the derivation actor's reset epoch, and byProcessDelegatedSafeHead.ConsolidateTaskErrorhas a newDoesNotExtendLocalSafevariant, which requests a reset, andEngine::drop_derived_tasksis new. In kona-node-service,DerivationEngineClient::send_safe_l2_signalis replaced bysend_derived_attributes,send_derived_finalizationandsend_delegated_safe_head. Code that builds these requests or implements the trait must switch to the new variants and methods.🤖 Generated with Claude Code