A from-scratch distributed task scheduler in Go. Workers register and heartbeat; the control plane places queued tasks by Filter → Score → Bind, and reschedules them when a worker stops answering. This cluster runs itself: it submits work continuously and kills a worker every so often, so the reschedule happens on its own. This is a recording of a real session — a worker dies partway through, and you can watch its tasks land somewhere else. Source · Live cluster · Recorded session
Each bar is committed capacity, not measured load: the scheduler only ever places a task where the requested CPU and memory still fit. A worker goes red when its heartbeat lease runs out, and whatever it was running goes back in the queue.
| Task | Status | Worker | CPU | Mem | Priority | Command |
|---|