Distributed lock service

DesignHard45 min

Your task

Design a distributed lock service that lets thousands of stateless app servers serialize access to per-resource critical sections (e.g. "only one worker rebuilds cache key K at a time"). Targets: ~50k lock acquisitions/sec, p99 acquire latency under 10ms, locks auto-release if a holder crashes, and correctness must survive a held lock outliving a GC pause or network partition. Walk the API, how a lock is granted and reclaimed, and how you stop two clients from believing they both hold the same lock.

It covers

  • Lease
  • Fencing token
  • Split brain