Class Engine

async on purpose, even though the work is synchronous: resumeCompensation is async, and an operator writing recovery code under pressure reaches for Promise.allSettled([resume(id), abandon(id)]). With a sync throw that expression throws before allSettled is ever called, so the defensive form is the one that blows up. Matching the sibling costs nothing while the API is experimental.

Archive old executions to a separate table

Remove old completed/failed executions

Shut down the engine

Get archived execution count

Get execution state by ID

List executions with optional filters

Unsubscribe from a specific event type

Unsubscribe a catch-all listener

Subscribe to a specific workflow event type

Subscribe to all workflow events

Recover orphaned executions after a crash/restart.

• 'running' executions: re-enqueued at their current step
• 'waiting' executions: timeout timers re-armed (or resumed if signal arrived)
• 'compensating' executions: compensation re-run (handlers must be idempotent)

Register a workflow definition

Retry the compensation that parked a compensation-stuck run and continue the unwind. Use once the cause of the failed reversal has been fixed.

Send a signal to a waiting execution

Start a new workflow execution

Subscribe to all events for a specific execution