Run and workload lifecycle

Independent transitions from repository resolution to useful progress

A TauGrid run crosses independent control planes:

  1. Resolution: TauGrid finds the project, workspace, and target.
  2. Rendering: TauGrid validates intent and builds a Job or RayJob.
  3. Admission: Kueue reserves quota and admits the workload.
  4. Scheduling: Kubernetes places pods on matching nodes and devices.
  5. Runtime: Ray or the Job controller starts the application.
  6. Progress: The model or data process advances and writes evidence.
  7. Completion: Results, checkpoints, metrics, and terminal state remain inspectable.

These states are not equivalent:

  • Admitted does not mean scheduled.
  • Scheduled does not mean the GPU is healthy.
  • Running does not mean the model is progressing.
  • Completed does not prove required artifacts were preserved.

Use the layered troubleshooting guide to find the first failed transition.