# Build a Concurrent Recurring Job Scheduler Design a thread-safe `JobScheduler` that can schedule a callable to run repeatedly at a requested interval and later deschedule it. The public API should support `schedule(job_id, fn, interval_seconds)` and `deschedule(job_id)`. Explain the core data structures, worker coordination, cancellation semantics, shutdown, and how you would test races. ### Constraints & Assumptions - Multiple threads may schedule and deschedule jobs concurrently. - A descheduled job must not start again, but a callable already running is allowed to finish. - An invocation is considered started when the scheduler atomically transitions it to `RUNNING` under its registry lock. If that transition wins the race, deschedule may return before the executor enters the user callable, but that already-started invocation is allowed to run. - Callables may raise or run longer than their configured interval. - Waiting should be efficient; busy polling is not acceptable. ### Clarifying Questions to Ask - Is the recurrence fixed-rate or fixed-delay, and should missed runs be skipped or caught up? - May two invocations of the same job overlap? - Does scheduling an existing `job_id` replace it or fail? - What guarantees should `deschedule` provide when racing with dispatch? ### Solving Hints Separate deciding what is due from executing user callables. Use monotonic time, and make stale queued entries harmless when a job is replaced or canceled. ### What a Strong Answer Covers - A min-heap of next-run times, a job registry, and condition-variable coordination. - A precise linearization point for schedule and deschedule operations. - Generation tokens or equivalent protection against stale heap entries and cancellation races. - Execution outside locks, bounded worker resources, exception isolation, and shutdown behavior. - Deterministic concurrency tests using fake time, barriers, and events. ### Follow-up Questions - How would you persist schedules across process restarts? - How would you prevent one long-running job from starving others? - What changes in a distributed scheduler with multiple scheduler instances?