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
Persist and recover committor intent execution stages #1702
A restart shouldn't mean losing pending intents or repeating transactions that already succeeded. Record execution progress in outbox accounts so the committor can pick up where it left off.
Why
The tricky part is knowing what happened before the restart. An intent might not have started, a transaction might still be pending, or the commit might have succeeded while finalization is unfinished.
A confirmation timeout doesn't tell us which happened. Before retrying, check the previous transaction's status so we don't submit the same work twice.
Scope
Replace the committor persister with outbox-backed recovery, following MIMD-0025 and MIMD-0027.
Keep enough information to resume each stage safely:
Record accepted intents and execution-stage changes, including pending signatures and blockhashes.
Recover both single-stage and two-stage execution, checking pending transactions before replacing an attempt.
Validate stage transitions. If only finalization needs retrying, retain the confirmed commit signature.
Clean up completed intents so recovery doesn't pick them up again.
Before rollout, we still need to decide how to recover work stored by the old persister. Shutdown coordination is tracked separately in #1262.
Activity