How do Kubernetes QoS classes work?

Kubernetes assigns each Pod one of three Quality-of-Service classes based on its resource requests and limits, and the kubelet uses that class to decide which Pods to evict first under memory pressure:

Guaranteed: every container has CPU and memory requests equal to its limits. Last to be evicted.
Burstable: at least one container has a requests or limits set, but the Pod doesn’t meet the Guaranteed criteria (for example, requests are lower than limits, or only one of the two is specified). Evicted before Guaranteed.
BestEffort: no requests or limits set on any container. Evicted first.

For long-running training and inference workloads on CoreWeave, prefer Guaranteed by setting requests equal to limits, so a job is not evicted before its peers. For full details, see the Kubernetes Pod QoS reference.

Workload Scheduling