An online judge has to run code it can't trust, in several languages, quickly and fairly. Every submission needs the same limits, the same inputs and a deterministic verdict, and one bad program must never affect another or the host.
- 01Submissions enter a Redis-backed queue through a REST API, so accepting work is decoupled from running it.
- 02A pool of Go workers pulls jobs concurrently. Each job runs in its own Docker container with CPU, memory and wall-clock limits.
- 03The runner captures stdout and stderr, compares output against test cases one by one, and reports a verdict such as accepted, wrong answer, time limit or runtime error.
Architecture
- 1APIREST endpoint validates and enqueues submissions
- 2QueueRedis holds pending jobs and results
- 3WorkersConcurrent Go worker pool pulls jobs
- 4SandboxPer-run Docker container with resource limits
- 5VerdictOutput diffed per test case, result stored
- Isolated, containerized execution environments for secure multi-language evaluation.
- Concurrent worker pipelines and execution queues with timeout and resource-limit enforcement.
- Sandboxed runtime orchestration with Docker process isolation, stdout/stderr capture and deterministic test execution.