Capacity Planning

Design a system that runs jobs at a chosen time: "run job X at time T" once, or on a recurring schedule like "every day at 09:00." The work must run reliably at scale, even when the machines running it crash, and it must run effectively once — never silently skipped, never harmfully duplicated.

In scope: submitting one-off and recurring jobs, dispatching due jobs to workers, crash recovery, and effectively-once execution. Out of scope: the workers' own job logic, and dependencies between jobs (a workflow/DAG engine, covered briefly as a variant).