Release v0.1.0

NVCRE v0.1.0

The NVIDIA Cluster Readiness Engine (NVCRE) is a Kubernetes controller that certifies GPU clusters before production use. It runs real training and communication workloads across topology-aware node groups, measures goodput and bandwidth, detects hardware failures, and reports every failed node with a reason. See the README for the full feature list.

Install nvcrectl

Install this release:

curl -fsSL https://github.com/NVIDIA/cluster-readiness-engine/releases/download/v0.1.0/installer | bash -s -- -v v0.1.0

Or install the newest stable release (releases/latest resolves only to stable releases, never pre-releases):

curl -fsSL https://github.com/NVIDIA/cluster-readiness-engine/releases/latest/download/installer | bash

Helm Chart

helm install nvcre \
  oci://ghcr.io/nvidia/cluster-readiness-engine \
  --version v0.1.0 \
  --namespace nvcre \
  --create-namespace

The controller image for this release is ghcr.io/nvidia/cluster-readiness-engine/manager:v0.1.0.

nvcrectl CLI

Download the binary for your platform from the assets below and place it on your $PATH.

Platform	Binary
Linux amd64	nvcrectl-linux-amd64
Linux arm64	nvcrectl-linux-arm64
macOS amd64	nvcrectl-darwin-amd64
macOS arm64 (Apple Silicon)	nvcrectl-darwin-arm64

Verify your download

Every binary asset (the installer and the nvcrectl-* binaries) is listed in checksums.txt with its SHA-256 digest. After you download an asset, verify it:

sha256sum --check --ignore-missing checksums.txt

Third-Party Notices

Third-party Go module licenses are listed in THIRD_PARTY_NOTICES.md and attached as a release asset.

Support

• Ask questions in GitHub Discussions.
• Report bugs with the issue templates.
• Report security issues per SECURITY.md, never through GitHub issues.

What's Changed

Dependency Updates

• build(deps): bump docker/library/golang from 1.26.6-trixie to 1.27.0-trixie in the docker-images group by @dependabot[bot] in #225
• build(deps): bump the actions group with 6 updates by @dependabot[bot] in #226

Other Changes

• docs: resolve the newest release dynamically instead of pinning tags by @ndipebot in #224
• fix(threshold): fail loudly on unknown threshold keys instead of skipping validation by @ndipebot in #227
• fix(controller): surface validation failures on WorkloadRun as ValidationFailed by @ndipebot in #229
• fix(setup): apply chart CRDs on every init run to prevent schema drift on upgrade by @ndipebot in #236
• fix(cli): implement workloadrun run --cleanup instead of silently discarding it by @ndipebot in #228
• fix(platform): pass spec.env to MPI containers instead of silently dropping it by @ndipebot in #230
• fix(controller): exclude nodes with insufficient allocatable GPUs from partitioning by @ndipebot in #231
• fix(setup): verify the installed Kubeflow Trainer version in setup status by @ndipebot in #237
• fix(controller): apply testScale node sizing in the WorkloadRun reconciler by @ndipebot in #233
• fix(controller): wait for workload pods to drain before deleting scoped dependencies by @ndipebot in #235
• fix(workload): treat a suspended TrainJob as pending, not running by @ndipebot in #238
• fix(controller): apply orchestration state mutations inside the status retry closure by @ndipebot in #234
• fix(cli): report workloadrun state and write results on wait timeout by @ndipebot in #241
• fix(api): make Certification and WorkloadRun specs immutable after creation by @ndipebot in #240
• feat(catalog): make training entry CPU/memory overridable via CategoryOptions by @ndipebot in #232
• fix(controller): certify the majority GPU architecture, not the first node's by @ndipebot in #251
• fix(controller): verify ownership before adopting or deleting child resources by @ndipebot in #239
• fix(helm): add missing user-facing ClusterRole trios for three CRDs by @ndipebot in #249
• docs(readme): add WorkloadRun to the "How it works" diagram by @ndipebot in #247
• test(integration): sort validation causes for deterministic goldens by @ndipebot in #253
• fix(controller): use majority GPU architecture detection on every path, not just certification by @ndipebot in #254
• fix(controller): wire event recorders into the four recorder-less reconcilers by @ndipebot in #250
• docs: switch install and release instructions to public download URLs by @ndipebot in #255
• chore: public-launch scrub of docs, CLI hints, tests, and CI by @ndipebot in #256
• fix(helm): rename chart to cluster-readiness-engine and publish to the repo-linked GHCR package by @ndipebot in #257

Full Changelog: v0.1.0-rc.13...v0.1.0