Skip to content

Kubernetes API throttling with the K8S and NFD providers #437

Description

@dmitsh

Large Kubernetes clusters can experience significant client-side throttling during topology reconciliation.

The Kubernetes engine currently performs a per-node GET followed by a full Node update. With client-go defaults of 5 QPS and burst 10, reconciling thousands of nodes can take several minutes.
Kubernetes client rate limits are also configured independently across the DRA provider and Kubernetes-backed engines, requiring duplicated Helm values. The NFD engine does not currently accept these settings and constructs typed and dynamic clients with separate rate limiters.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions