ReferenceGlossary
Spot instance
Spare cloud capacity sold at a large discount, up to 90% off the on-demand price, that the provider can take back whenever it needs the capacity. AWS gives a two-minute interruption notice. Spot suits stateless, fault-tolerant work such as batch jobs and servers behind a load balancer with spare capacity, not a single database.
Business
In Quorum#
Spot pricing is designed but not in the game yet. The closest is the GPU shortage Incident: the provider reclaims a GPU Node, H100s first, and it stays down for the Incident; an LLM API as a fallback makes it a quarter as likely. The Instance failure Incident also kills one Node's host, three times as often on a Node with no second copy as on others.
Read next#
Operating margin, Autoscaling, Health check, Model inference