[WIP]: Feature lifecycle and graduation criteria #931
Description
Issue: #877 — Policy on Feature Gates
Motivation
The NVIDIA DRA Driver for GPUs uses Kubernetes-style feature gates to manage
the lifecycle of new capabilities. Community members and users need clear guidance on:
- How do I propose a new feature or enhancement?
- What does it take for a feature gate to graduate from Alpha to Beta to Stable?
- When can I rely on a feature in production?
This document establishes a lightweight proposal and graduation process adapted
from the Kubernetes Enhancement Proposal Process and tailored for the pace and scope of this project. It draws on graduation
patterns from various upstream DRA KEPs.
Scope
This process applies to changes that meet any of the following criteria:
- Introduce a new feature gate
- Promote an existing feature gate to a new stability level
- Make a significant behavioral change to an existing feature
- Add or modify existing APIs, CLI flags, or Helm chart configuration
- Introduce or modify external dependencies (e.g., DCGM, NVSentinel, Virtualized environment etc)
- Require coordination with upstream Kubernetes features or KEPs
Changes that do not require a proposal:
- Bug fixes to existing behavior
- Documentation-only changes
- Internal refactoring that does not change observable behavior
- Dependency version bumps with no API impact