Design an online coding judge
Company: Meta
Role: Software Engineer
Category: System Design
Difficulty: hard
Interview Round: Onsite
##### Question
Design an online coding practice and judging platform (think LeetCode / HackerRank / Codeforces) where users submit solutions that are compiled and executed against hidden test cases, then graded asynchronously. Walk through the end-to-end architecture, APIs, data models, scaling strategy, and failure handling.
Address the following:
1. **User-facing features** – user accounts/auth, problem browsing, full-text search and filtering, and per-user submission history.
2. **Submissions** – multi-language support (e.g., Python, Java, C++), an intake API that returns a submission id, and idempotent handling of duplicate/retried submits.
3. **Secure execution** – compile and run *untrusted* code in sandboxed environments with strict CPU, memory, and wall-clock limits, no outbound network, and isolation strong enough to contain a malicious submission.
4. **Asynchronous grading** – a distributed judge worker fleet pulling from a task queue, per-test execution, result aggregation/verdict computation (AC/WA/TLE/MLE/RE/CE), and live status updates (queued / running / accepted / failed).
5. **Storage** – data models for users, problems, submissions, and per-test results; object storage for source code, compile/run logs, and test-data bundles.
6. **Test-data versioning and re-judging** – immutable, versioned test bundles; pinning the exact version a submission ran against; and re-running cohorts of submissions when tests, checkers, or limits change.
7. **Anti-cheat / plagiarism detection** – scalable code-similarity clustering plus behavioral signals.
8. **Rate limiting and fairness** – per-user/IP submission limits, contest priority, and queue backpressure so contest spikes don't starve regular users.
9. **Caching, observability (metrics/tracing/logging), and security isolation.**
10. **Scaling strategy and failure handling** – how the system scales to contest spikes (e.g., 1k submissions/min) and stays correct when workers crash or the queue redelivers a job.
Discuss the key trade-offs you make along the way (e.g., container vs. microVM isolation, managed queue vs. event stream).
Overview: A Meta software-engineer onsite system-design question: design an online coding judge (LeetCode/HackerRank-style) that compiles and executes untrusted user submissions against hidden tests and grades them asynchronously. It tests secure sandboxed execution with CPU/memory/time limits, a distributed queue-backed worker fleet, data modeling and APIs, test-data versioning and re-judging, anti-cheat, rate limiting/fairness, caching, observability, and contest-scale failure handling.