Skip to main content

Orchestrator and Checkpointing

The orchestrator coordinates stage execution and persists checkpoints so runs can resume.

Checkpointing enables:

  • deterministic recovery after failure
  • partial reruns after upstream edits
  • stable project-level artifacts across retries

What is checkpointed

Per project run, Rilo persists:

  • stage completion state in run-state.json
  • generated outputs and paths in artifacts.json
  • local media files in assets/

Resume behavior

If a run is restarted for the same project:

  • completed stages are reused when artifacts are valid
  • missing or invalid artifacts trigger regeneration from the earliest required stage
  • downstream stages remain invalidated until prerequisites are rebuilt

Run locking

Rilo applies a project-level run lock to prevent concurrent mutations on the same project. If a project is already running, new run/regeneration requests should wait.

See: