Production deployment

Processes

Process	Count	Command
scheduler	2 or 3, on different hosts	goatbeat --app myproject.beat:app scheduler or manage.py goatbeat scheduler
workers	as many as the load needs	your task backend's worker
migration	once per release	goatbeat migrate or manage.py migrate

Run schedulers under a supervisor (systemd, Kubernetes, ECS). They stop on SIGTERM, releasing the lease, and exit 0.

Configuration

Variable	Default	Meaning
GOATBEAT_APP		module:attribute of the GoatbeatApp (or --app)
GOATBEAT_DATABASE_URL		postgresql://... (or sqlite:///file for development)
GOATBEAT_SCHEDULER_ID	host:pid:random	name in logs, metrics and the lease
GOATBEAT_TIMEZONE	UTC	default cron timezone of admin-created schedules (Django); in code, pass Registry(timezone=...)
GOATBEAT_LEASE_TTL	30s	how long a lease lives without renewal
GOATBEAT_LEASE_RENEW_INTERVAL	10s	must be less than half the TTL
GOATBEAT_MAX_SLEEP	60s	longest sleep between ticks
GOATBEAT_BATCH_SIZE	500	due schedules read per query
GOATBEAT_MAX_SUBMIT_RETRIES	5	retries before an execution is failed

Invalid values stop the process at startup with every problem listed.

Health checks

• Liveness: the process is running.
• Readiness and alerting: goatbeat doctor --json (exit 1 on any failed check) and, when a metrics sink is configured, the goatbeat_leader metric, which must sum to 1 across schedulers.

Command line exit codes

0 success, 1 a check failed, 2 usage error, 3 configuration error, 4 store unavailable, 5 unexpected error.

Upgrades

1. Run the new release's migration.
2. Restart schedulers one at a time.
3. Run schedule prune with the new code if schedules were removed.