Monitoring and logs

Container health checks

Container	Built-in check	What it tests
api	php artisan health:check every 30 s	Platform database connection and Redis (app/Providers/HealthCheckProvider.php)
h5p	GET http://127.0.0.1:8080/h5p/health	Liveness and Redis (loopback host, no tenant)
pdf	GET http://127.0.0.1:3000/health	Process up, fonts loaded
web	GET http://127.0.0.1:4321/healthz	Process up (answers ok)
admin, front	none	Apache serving static files

docker compose ps shows the state; restart policies in the example restart crashed containers but not unhealthy ones. An external monitor should probe the HTTP endpoints below.

HTTP endpoints to probe

URL	Answer
https://api.<domain>/api/health	200 or 503 from the last stored health:check result (spatie/laravel-health)
https://api.<domain>/api/healthcheck	JSON with each check’s status
https://<slug>.api.<domain>/h5p/health	{ok, tenant, db, redis, s3} for that tenant; 503 if a dependency is down, 404 for an unknown tenant
https://<slug>.app.<domain>/healthz	ok from the learner front
https://<slug>.api.<domain>/api/settings	The tenant’s public settings: a cheap end-to-end check of routing, the tenant env file and its database

Probe at least the platform API, one tenant’s /h5p/health (it covers the tenant database, Redis and the bucket) and the front.

Workers and the scheduler

• Horizon runs the platform queues (default and the long-job queue for video). Its dashboard at /horizon is open only when APP_ENV is local or stage, or for users allowed by the viewHorizon gate, which lists nobody by default. Keep APP_ENV=production.
• Tenant queues are processed by queue.sh, which loops over all tenant hosts and runs queue:work --stop-when-empty --domain=<host> for each. Watch for growing Redis lists under the tenant prefixes (ulams_<slug>_queues:default) if jobs seem stuck.
• Scheduler: scheduler.sh runs schedule:run for the platform and every tenant once a minute. Failed scheduled jobs show up in the api log.

Logs

• Container stdout/stderr: supervisord forwards php-fpm, Horizon, the queue loops and the scheduler to the api container log; the Node services log JSON (pino) to stdout; Caddy logs requests with the H5P _token parameter and auth headers redacted. The example caps Docker’s json-file logs at 5 x 20 MB per container.
• Laravel application logs go to the daily file channel by default: storage/logs/laravel-<date>.log for the platform and storage/<host with dots as underscores>/logs/ for each tenant, on the api_storage volume. Rotate or ship them; set LOG_CHANNEL=stderr to send them to the container log instead (not the project’s default).

Error tracking

Component	Variables
API	SENTRY_LARAVEL_DSN (or SENTRY_DSN), SENTRY_ENVIRONMENT, SENTRY_RELEASE, SENTRY_TRACES_SAMPLE_RATE, SENTRY_PROFILES_SAMPLE_RATE, SENTRY_SEND_DEFAULT_PII (default off)
Admin	REACT_APP_SENTRYDSN on the container; the release is taken from the image version
Older React front	VITE_APP_SENTRYDSN on the container

Note

Tenants inherit the platform’s Sentry settings through their env files. Keep SENTRY_SEND_DEFAULT_PII off unless your data-processing agreements cover it: errors can carry learner data.