Scenario

Work-ledger scheduler with duplicate workers

A pool of workers pulls tasks from a ledger and writes results back. A worker that stalls on GC or a slow disk looks dead to the scheduler, so the task is handed to someone else. The stalled worker wakes up later with no idea any of this happened. A task's result must never be silently replaced by a stale one.

claimwrite resultSchedulerlockedWorkers2 decisionsWork ledger2 decisionsTask worklocked