You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
kona-node: recover from rejected sequenced payloads with deposits-only forced attributes #23510
A kona-node sequencer can terminate when engine_newPayload rejects its locally built payload, because it mistakes the build's forced transactions for the complete payload transaction list. This is a pre-existing recovery bug, retained by #23509; it need not block that refactor or #23454.
Failure path
Ordinary sequencing attributes contain the L1-info deposit and any other forced deposits, with noTxPool=false. The EL adds pool transactions to the built payload. See attribute construction.
SealTask::insert_payload receives INVALID from engine_newPayload and calls handle_rejected_attributes.
The helper first calls attributes.is_deposits_only(), which checks only the forced transaction list and ignores noTxPool. It returns DepositOnlyPayloadFailed before checking whether the attributes are derived. See the helper and the predicate.
The sequencer classifies that error as fatal and exits; the node supervisor shuts down peer actors. See error handling.
Impact and scope
This causes loss of sequencing availability if the EL rejects a locally built payload. Normally the EL should accept its own payload; a builder/validation disagreement is an example of the prerequisite failure. No production incident or ordinary user-transaction trigger was established in this review. This path rejects the block before canonicalization, so the identified consequence is shutdown rather than acceptance of an invalid block. Restart or failover may restore availability, but a persistent underlying build defect can recur.
#23509 correctly prevents deposits-only replacement for sequenced attributes with non-deposit forced transactions. It does not fix this more typical deposits-only-attributes case. Its decision-table test currently expects DepositsOnlyRejected for sequenced_deposits_only.
Proposed fix and validation
At the newPayload rejection boundary in SealTask::insert_payload, handle locally sequenced payloads before applying the shared derived-attributes recovery rules: return the original PayloadInsertionFailed(UnexpectedPayloadStatus(...)) to the sequencer. The existing sequencer handler then discards the pending build and starts another attempt. Keep derived deposits-only failures critical and derived Holocene replacement/channel flushing intact. This also avoids changing forkchoiceUpdated attribute-rejection semantics merely to fix newPayload handling.
Add a regression test with is_attributes_derived=false, deposits-only forced attributes, noTxPool=false, a returned payload containing a pool transaction, and an INVALID newPayload response. Assert that there is no replacement, canonicalization, gossip, or head advancement; that the error is nonfatal to the sequencer; and that a subsequent valid build can proceed. Keep coverage for genuine derived deposits-only failure.
The separate pre-Holocene derivation stall is already tracked in #23500.
Activity