Runmoor Commands and Routing
Version note: Runmoor 0.2.0 introduced automatic setup and managed runner updates. Guided Mac VM creation during
initandimage create --ipsw latestare in the next release. Version 0.1.3 uses the explicit pinned configuration and manual image preparation also documented below. Checkrunmoor versionbefore using release-specific commands.
Use the configured scale-set label, or a matching label array, and applicable GitHub runner-group policy. Runmoor receives demand without a public webhook endpoint. Ordinary NAT networking is sufficient.
For a new ARM64 Docker pool, runs-on: [runmoor-linux, linux, ARM64] also
selects its platform and architecture; use x64 for an amd64 pool. The
single scale-set label above continues to work.
Status and doctor JSON use schema_version: 1. Status includes image revisions, calculated budgets, managed runner versions, update attempts and pending recovery. Doctor also reports the current state and briefly acquires/releases OS sleep inhibition to check permissions. Errors have stable codes, affected identifiers and a recovery action. English text is usable without color; use --no-color or NO_COLOR to disable ANSI logs. JSON never uses color.
pause stops acquisition/new capacity and preserves running jobs. drain additionally waits for jobs/local cleanup. stop drains before exiting and waits for open image setup and pending image removal, including with --force. Finish setup by shutting down its VM or sealing the revision, and retry pending removal as needed. New image work requires restarting the manager after stop; stop --pool NAME drains that pool while the manager keeps serving other pools. Only explicit --force terminates owned work. resume revalidates the pool. Pool control commands without --pool apply to all pools. A validated reload automatically resumes a suspended pool only when a setting related to its reported failure changed; a verified managed image can also recover image, version or repeated startup failures. Otherwise, correct the cause and use resume.
Unreleased pause recovery fix: A late dependency failure and corrected reload preserve an explicit
pause --pool NAMEorstop --pool NAME, including scoped force-stop. Correct the cause, then useresume --pool NAMEto enable new work.
Reload validates the requested configuration first. Managed image changes prepare before activation; preparation failures retain the previous verified environment. Existing jobs retain their original configuration and timeout. Removed/changed pools drain their previous generation; a new generation with the same GitHub scale-set identity waits until the old one retires. A failed configuration validation does not replace the last accepted configuration.
Unreleased service reload: After you install a newer Runmoor CLI,
runmoor reloadalso checks the running launchd/systemd user service. If its manager is older, reload updates the service to that installed CLI and restarts only the manager. Active jobs retain their execution environment and timeout. Equal or newer managers reload settings without a version change. Foreground managers keep their existing reload behavior. Reload does not download releases, downgrade a manager or start a stopped service. Releases through 0.2.7 reload configuration only.
Real demand receives capacity before warm runners. Round-robin allocation shares remaining resources across pools. Minimum idle is best effort; running work is never preempted. CPU/memory reservations include DinD and image setup. Low disk blocks new work without evicting active jobs or sealed images.
Complete CLI reference
All commands accept --config PATH and --no-color. Commands and flags are case sensitive.
Image commands print structured revision data. Image changes require a running manager; image list also works offline. See Tart image preparation for starting without any pools. CPU values are whole cores and memory values are MiB. Use --help to list commands. Read operations and recovery before force-stop or image removal.
Host selection and control
The unreleased macOS host backend is selected explicitly with
runmoor init --backend host. New workflows route with
runs-on: [runmoor-macos-host, macOS, ARM64]. The existing status, doctor,
runner update, reload, pause, resume, drain, stop, scoped force-stop and launchd
service commands apply to host pools. Image commands remain Tart-only.
Host work shares global admission budgets, but CPU/memory reservations do not enforce process usage limits. Force-stop retains uncertain ownership and cleanup; it never signals unrelated account processes or deletes replaced directories. Actual host execution, unsigned Xcode builds and live GitHub jobs are unvalidated.