The Scheduler is the decision engine of the Public Intelligence network.
It maintains a registry of available compute nodes, receives periodic heartbeats, and selects the most appropriate node for inference requests.
Version 1 focuses on building the core scheduling infrastructure.
A walkthrough of the first working prototype is available on YouTube: Public Intelligence v1 Demo
This video demonstrates the end-to-end integration of the Website, Scheduler (showing the current v1 implementation), Node, registration, heartbeats, and local inference.
- Node registration
- Node unregistration
- Heartbeat processing
- In-memory node registry
- Deterministic scheduling algorithm
- REST API
- End-to-end demonstration
- Comprehensive test suite
Client
│
▼
Scheduler API
│
▼
Scheduling Algorithm
│
▼
Node Registry
▲
Registration / Heartbeats
│
▼
Nodes
python -m scheduler.mainor
uvicorn scheduler.main:app --reloadCurrent Release
v1.0.0
Version 2 will introduce:
- Request orchestration
- Automatic request forwarding
- Failure recovery
- Retry logic
- Multi-node inference routing
- Public Intelligence Website
- Public Intelligence Node
Apache 2.0