NVIDIA/k8s-test-infra v0.2.1

v0.2.1

Repository: NVIDIA/k8s-test-infra

Tag: v0.2.1

Published: 2026-06-12T12:16:02Z

Prerelease: no

Release notes: Patch release: security refresh, release-machinery hardening, and review-driven fixes. Every artifact in this release was published fully automatically — the tag-triggered publish path this release fixes was validated by this release itself.

Highlights

• Security refresh: all Go build pins now track 1.26.4 (deployment and test images, Makefiles, e2e dispatch default — the documented make docker-build works out of the box again); bundled kubectl bumped v1.32.0 → v1.36.1, removing the 17 CRITICAL/HIGH CVE findings trivy reported against that binary. (#394)
• Release machinery: chart pushes to ghcr retry through the registry's tag read-after-write lag (#390); image publishing now triggers on v* tag pushes — a paths: filter had silently suppressed every tag-triggered run — and supports workflow_dispatch for re-running a publish at any ref (#391).
• mocknvml: nvmlDeviceGetHandleByUUID returns NOT_FOUND for unknown UUIDs; the legacy default UUID scheme no longer assigns device 0 the canonical nil UUID; the tests/mocknvml integration harness builds again (broken since #269). (#395)
• mockib hardening from code review (#396): peer-registration I/O carries deadlines, so one wedged peer can no longer silently halt re-registration to every peer; recv waits are bounded and honor daemon shutdown; peer discovery and registration sweeps respect context cancellation; missing sysfs node_guid no longer renders an all-zero NodeGUID; steady-state re-register logging only fires on change.

Container Image

docker pull ghcr.io/nvidia/nvml-mock:0.2.1

Helm Chart

helm install nvml-mock oci://ghcr.io/nvidia/k8s-test-infra/chart/nvml-mock --version 0.2.1

Image and chart are cosign-signed (keyless) with SPDX SBOM attestation. Full details in CHANGELOG.md.

Full diff: https://github.com/NVIDIA/k8s-test-infra/compare/v0.2.0...v0.2.1