How trap-jp/trao-judge-judge works
Project type and purpose
• It comprises several microservices/command‐line apps (backend_app, judge_control_app, exec_app), shared Rust libraries (domain, infra, usecase, gRPC crates, judge core, jobapi, problem_registry), database migrations, and Python utilities for building problems, running commands, and visualizing results.
File structure
• Top‐level “app” directory holds three main binaries:
• backend_app: HTTP API server (Cargo.toml, Dockerfile, src/handler, model, di.rs)
• judge_control_app: orchestrator for judge jobs (gRPC/proto, Dockerfile, mock code)
• exec_app: standalone executor that runs judging logic (main.rs)
• “lib” directory holds shared crates implementing:
• backend_domain, backend_usecase, backend_infra (clean‐architecture layers)
• back_judge_grpc & judge_exec_grpc (gRPC definitions + client/server stubs)
• judge_core (core judging algorithms)
• jobapi (job scheduling / AWS integration)
• judge_infra_mock (local/testing mocks)
• problem_registry (problem‐metadata service)
• Python libraries under “pylib”: traopy_builder, traopy_util, traopy_visualize, plus an example problem. These are CLI tools and utilities for problem setup and result visualization.
• CI/CD manifests in .github/workflows, Docker, nix flakes, Taskfiles, docker-compose.
Key architectural insights from README
• Although the README is sparse, naming conventions and Cargo features reveal a “hexagonal/clean‐architecture” approach (domain → usecase → infra), microservices communicating via gRPC, and containerized deployment.