← Back
00:00

Design Job Scheduler System

Mercor interview questions · Hard · System Design round · reported 3× · written walkthrough

This Mercor system-design round asks you to design a distributed job scheduler / work orchestration platform. Reports mention both a Marketplace-team version and a Frontier Data variant, but the core themes are the same.

Core requirements

  • Accept both one-time jobs and recurring / cron jobs.
  • Store schedule metadata, priority, retry policy, and optional dependencies.
  • Dispatch due work to a distributed worker fleet.
  • Support a DAG of job dependencies so downstream work runs only after upstream success.
  • Track lifecycle states such as PENDING, QUEUED, RUNNING, SUCCESS, and FAILED.
  • Prevent duplicate execution even during failover or retry storms.

A strong high-level design

  1. Job API + durable job store for writes, status, retry policy, and next-run metadata.
  2. Leader-elected scheduler tier so exactly one active node scans due jobs.
  3. Indexed scheduling lookup by next-run time, such as a DB index or Redis sorted set.
  4. Queue layer partitioned by priority and, if useful, by job class.
  5. Worker fleet that leases jobs, heartbeats, reports status, and retries failures.
  6. Dependency resolver that tracks unresolved upstream counts for DAG execution.

Discussion points interviewers probe

Idle