Resource Scheduler Guide¶

Documentation In Progress

This service guide is a placeholder. Full documentation is being developed.

Overview¶

The Resource Scheduler is responsible for intelligent placement of LabletInstances on available CML Workers. It implements scheduling algorithms that consider resource requirements, license affinity, and time-based reservations.

Architecture¶

See Resource Scheduler Architecture for detailed design.

Core Responsibilities¶

Responsibility	Description
Scheduling Decisions	Determine optimal worker placement for LabletInstances
Timeslot Management	Handle time-windowed reservations
Capacity Tracking	Monitor available resources per worker (via etcd)
HA Coordination	Leader election for single active scheduler

Key Flows¶

Instance Scheduling Flow¶

sequenceDiagram
    participant CP as Control Plane API
    participant RS as Resource Scheduler
    participant etcd as etcd (State Store)

    CP->>etcd: Write: LabletInstance (status=PENDING)
    RS->>etcd: Watch: LabletInstances
    etcd-->>RS: Notify: New PENDING instance
    RS->>etcd: Read: Worker capacity
    RS->>RS: Calculate optimal placement
    RS->>etcd: Write: Assignment (instance→worker)
    RS->>CP: POST: Update instance status

API Endpoints¶

Internal Service

The Resource Scheduler primarily operates via etcd watches. Limited REST API for health and status.

Method	Endpoint	Description
`GET`	`/health`	Health check
`GET`	`/ready`	Readiness check
`GET`	`/metrics`	Prometheus metrics

Configuration¶

Key environment variables:

Variable	Description	Default
`SCHEDULER_ENABLED`	Enable scheduling	`true`
`SCHEDULER_LEAD_TIME_MINUTES`	Advance scheduling window	`35`
`ETCD_ENDPOINTS`	etcd cluster endpoints	`http://etcd:2379`