Configuration overview and activation

English | 简体中文 | Wiki home

This reference is checked against the installer, schemas, examples, consumers, and generators at 43a88bdd62fbf200889b36932fa921439a293190. The original field inventory covers project installation flags, backend/judge YAML, per-user AI settings, environment variables, frontend bootstrap settings, and infrastructure settings generated by the project. This revision also documents database registration settings in Site management. Individual problem/contest/problem-set content fields and every possible third-party directive are outside the server configuration inventory. Open-ended dictionaries are identified with links to their sources.

Where to configure

Configuration entry	Contents
Installation	Installation entry points, every flag, and retries
Backend-Configuration	Backend YAML: accounts, security, database, mail, storage, contests, and presentation
Registration-Approval	Website registration policy, YAML fallback and delegated review
Two-Factor-Sign-In	Per-user authenticator codes and passkeys, recovery, and the two-factor.key file; no YAML setting
Judge-Configuration	Judge YAML, CPU/RAM resizing, service units, verification, and rollback
AI	Per-user AI models, search, permissions, and jobs
AI-Environment	AI directories, sandbox communication, key persistence, and environment
Environment	Runtime, installation, and build environment variables; maintenance tool arguments
Frontend-Configuration	Local frontend builds and optional CDN/bootstrap settings
Operations-Configuration	Nginx, MariaDB, Redis, MinIO, systemd, and post-install changes

Distinguish four kinds of values

• Runtime defaults are actual fallbacks implemented in code. Backend and judge YAML generally do not fill missing fields from an example.
• Example values come from repository examples or upstream bootstrap demos, which may point to another site.
• Installer-generated values are calculated at first installation from paths, ports, CPUs, and RAM. Existing private configuration is normally retained on retries.
• Effective values come from actual instance files plus systemd units and drop-ins. The database registration policy can override its YAML default; AI settings are additionally stored per user.

Examples use /opt/LLMOJ. For a custom --prefix, substitute the actual instance path everywhere. Changing one environment variable does not relocate an installation.

Applying changes

Changed setting	Activation
config/backend.yaml	Validate the complete YAML and restart libreoj-backend; refresh the browser to reload public preferences; a saved website registration override still takes precedence
Registration policy in Site management	An administrator saves or restores the server configuration in the website; effective immediately without restarting or rewriting YAML; see Registration approval
config/judge.yaml	Drain work and follow the judge maintenance procedure; slot, CPU, and directory changes also require generating and installing matching systemd units
systemd environment/resource limits/drop-ins	Run systemctl daemon-reload, then restart the affected service; reload alone does not restart its process
config/nginx.conf	Test with nginx -t, then reload; keep domains and storage signing URLs consistent
MariaDB/Redis/MinIO	Restart the corresponding service in a maintenance window and verify backend access and persistence
Frontend build settings or static assets	Rebuild/publish to the directory actually served by Nginx and refresh the browser
AI model/search settings and permissions	Save through the UI; jobs also recheck permissions while running, as described in the AI guide

Installer flags describe initial installation. Re-running a completed installation with --judge-slots, --port, or --public-url is not a reconfiguration interface: installation metadata may reject different inputs and generators retain existing YAML. Do not delete installation state, secrets.json, the AI master key, or data to bypass this behavior.

Common changes

For more or fewer CPUs/RAM, use Judge resizing. For names, ports, domains, mail, upload limits, and service memory, see Operations. For models and Tavily/MCP, see AI.

Back up the database, object storage, configuration, and AI key state together. Source code alone cannot restore a site. Keep real secrets out of Wiki pages and GitHub. When a later version adds fields, recheck its schemas and environment-variable consumers.