Kubernetes DRA Extended Resources Let GPUs Run Without Device Plugins
Kubernetes v1.37 moved DRA extended resources to general availability, giving platform teams a stable path off device plugins. The bridge lets existing manifests keep requesting nvidia.com/gpu while the scheduler allocates specific devices through DRA. A DRA driver publishes device details, and a DeviceClass maps the extended resource name to those devices. Migration is done node pool by node pool, with the device plugin removed so a GPU is managed by only one mechanism. Teams should also confirm that quota and cost tooling still reads correctly once devices are allocated through DRA.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in