Pools and scaling

A pool defines which Forgejo jobs it serves, which VM it creates, and how many instances it can run. Pools can serve a repository, organization, user, or the whole Forgejo instance, subject to the token's permissions.

The pool's first label routes waiting jobs. It must be unique across pools and first in the job's runs-on list; the job must request every pool label.

FARM targets enough ready or bootstrapping VMs for waiting jobs plus min_idle. max_instances limits the pool's non-finished VMs, including those running or cleaning up. max_provisioning limits simultaneous starts.

After bootstrap_attempt_limit unsuccessful starts, FARM pauses new VMs in that pool. Existing runners continue. After bootstrap_retry_interval, it starts one probe VM. A successful probe resumes provisioning; a failed probe starts another delay. The pause persists across FARM restarts and does not make /healthz fail: reconciliation is still enforcing the configured limit.

See scaling fields for defaults and limits.