// KEP-4671 landed gang scheduling in Kubernetes 1.37 beta, behind a feature gate that is off by default. A four-node Kind cluster with six fake DRA GPUs shows the partial-placement deadlock, the single field that fixes it, and the two limits nobody documents: the Job field is pruned silently without a second alpha gate, and workload-aware preemption cannot reclaim DRA devices.
posts tagged #Gpu
// Part two of the Modelplane fleet inference series, covering the layer above the scheduler. ModelService is a routing policy rather than a load balancer, spec.stack Dynamo swaps the whole gang-scheduling machinery, and four documented gotchas will cost you an afternoon each.
// A hands-on tour of Modelplane, the Crossplane-based control plane for AI inference across a fleet of GPU clusters. Two regions on Kind with fake DRA GPUs, the platform/developer contract written in DRA's own vocabulary, fleet-level drain, and the release trap that costs you two APIs.