scheduler_policies.yaml
scheduler_policies.yaml is the install-time policy catalog. The frontend reads it through Backend and displays the
available policy IDs and names on the application installation page.
- id: fixed
name: Fixed Policy
yaml: fixed-policy.yaml
dependency:
generator: generator-base.yaml
controller: controller-base.yaml
distributor: distributor-base.yaml
monitor: monitor-base.yaml
- id: fragsplice
name: FragSplice
yaml: fragsplice.yaml
dependency:
generator: generator-with-bursty.yaml
controller: controller-for-evaluation.yaml
distributor: distributor-base.yaml
monitor: monitor-queue-state.yaml
| Field | Meaning |
|---|---|
id | Stable value submitted by the frontend or API. It must be unique. |
name | Human-readable label shown to the user. |
yaml | Scheduler template under template/scheduler/. |
dependency | Generator, Controller, Distributor, and Monitor templates composed with the policy. |
The current catalog includes general fixed, cloud-only, edge-only, dynamic, and profiling policies; Dayu research families such as Hier-EI, Steady, Hedger, CASVA, CEVAS, Chameleon, and CRAVE; and experiment policies and baselines for FragSplice, IBDASH, Distream, and DTODRL. Some entries are explicitly marked as training, benchmark, cold-sample, or ablation variants and may require mounted profiles, models, or checkpoints.
Scheduler is installed before the remaining runtime because source selection and initial deployment are Scheduler API contracts. Backend validates the returned deployment plan before creating any processor routes. Adding an entry to this catalog therefore exposes a policy for installation, but does not bypass its hook, asset, or plan-validation requirements.
See Customize Scheduling for the implementation workflow.