Artifact Manifest

The generated tables list the stack packages, charts and images selected by this docs edition. Stack, chart and image versions are independent. Use Consume an NVCF release to obtain the packages.

Publication pending means the catalog has no verified customer download location for that exact artifact. Bundled charts are included in their stack packages.

Stack archives, inventories and images are published separately. Download every selected stack archive and verify the required images before scheduling an upgrade. An available inventory or image does not establish package availability; do not substitute another stack version to bypass a pending download.

Artifacts overview

The tables group the selected artifacts by deployment plane and type. Optional components require the corresponding deployment profile.

Prepare Helm charts for Helmfile

Use the chart and image access settings to pull the selected charts directly. For a private registry, follow chart mirroring. Bundled charts remain inside their stack directories and need no separate chart publication.

Supporting images that ship from Docker Hub

These Self-Managed helpers use Docker Hub independently of global.image. Check their egress or mirror configuration before rendering an upgrade:

The generated image tables below record the selected tags. The k8s image provides a CLI toolbox used by two independently configured hooks. It is not a Kubernetes server image. Retain the selected helper versions when mirroring.

Other enabled stacks and optional components use upstream registries; inspect every image in the render and generated inventory, not just this helper table.

Verify access from cluster nodes. If Docker Hub requires authentication, add its pull secret to global.imagePullSecrets. The receipt stage forwards this list to its bundled chart. Check that each enabled workload renders the intended pull secret.

If you have mirrored these images into your own registry, redirect them from your environment file (deploy/stacks/self-managed/environments/<environment>.yaml). Set registry and repository together, because each key falls back to its own Docker Hub default rather than to global.image. You do not need to edit deploy/stacks/self-managed/global.yaml.gotmpl.

nats:
  container:
    image:
      registry: <your-registry>
      repository: <your-repository>/nats-server
      tag: "2.14.9-alpine3.22"
  natsBox:
    container:
      image:
        registry: <your-registry>
        repository: <your-repository>/nats-box
        tag: "0.20.1-nonroot"
  reloader:
    image:
      registry: <your-registry>
      repository: <your-repository>/nats-server-config-reloader
      tag: "0.25.0"
api:
  accountBootstrap:
    image:
      registry: <your-registry>
      repository: <your-repository>/alpine-k8s
      tag: "1.37.0"
upgradeReceipt:
  image:
    registry: <your-registry>
    repository: <your-repository>/alpine-k8s
    tag: "1.37.0"

Run the upstream NATS images as non-root

The stack pulls the upstream NATS server, reloader and box images from Docker Hub by default. To run them as non-root and pin the box image by digest, merge these values into deploy/stacks/self-managed/environments/<environment>.yaml. Helmfile passes these settings to the selected NATS chart; no chart or release-definition changes are needed. This example uses NATS server 2.14.9, reloader 0.25.0, and NATS box 0.20.1. The box pin identifies the upstream multi-platform index for Linux AMD64 and ARM64.

nats:
  container:
    image:
      registry: docker.io
      repository: synadia/nats-server
      tag: "2.14.9-alpine3.22"
  reloader:
    image:
      registry: docker.io
      repository: natsio/nats-server-config-reloader
      tag: "0.25.0"
  podTemplate:
    merge:
      spec:
        securityContext:
          runAsNonRoot: true
          runAsUser: 1000
          runAsGroup: 1000
          fsGroup: 1000
  natsBox:
    container:
      image:
        registry: docker.io
        repository: natsio/nats-box
        tag: "0.20.1-nonroot@sha256:9eec37227e52bf4d207ffc246284034c18dd09d51bae0bfd2af9dbbd77ba4d44"
    podTemplate:
      merge:
        spec:
          securityContext:
            runAsNonRoot: true
            runAsUser: 1000
            runAsGroup: 1000

The box image declares user/group 1000:1000. runAsNonRoot adds Kubernetes enforcement; runAsUser and runAsGroup make the identity explicit. The server pod settings also cover its reloader; neither image declares a nonroot user. Server fsGroup assigns group 1000 to the JetStream and shared PID volumes. The box image provides writable directories for CLI state and nsc keys. Its default context links to read-only mounted Secrets; use a separate XDG_CONFIG_HOME to save additional contexts. The box needs no additional volume or fsGroup for this example.

For a private mirror, set each image’s registry and repository together and use the mirrored tag or tag@sha256:.... Preserve the multi-platform index when copying the box image and verify the destination digest. Unset or empty keys retain the stack’s existing defaults; this example does not change them. Configure global.imagePullSecrets if the destination requires authentication.

Render before installing, using your normal chart source and secrets file:

cd deploy/stacks/self-managed
export HELMFILE_ENV=<environment>
helmfile -f helmfile.d/01-dependencies.yaml.gotmpl --selector name=nats template

Check the StatefulSet’s server and reloader images and the box Deployment’s full tag@digest reference and security context. Install with the normal stack workflow, then verify in the intended Kubernetes context:

kubectl --context "$KUBE_CONTEXT" -n nats-system rollout status statefulset/nats
kubectl --context "$KUBE_CONTEXT" -n nats-system rollout status deployment/nats-box
kubectl --context "$KUBE_CONTEXT" -n nats-system exec nats-0 -c nats -- id
kubectl --context "$KUBE_CONTEXT" -n nats-system exec nats-0 -c reloader -- id
kubectl --context "$KUBE_CONTEXT" -n nats-system exec deployment/nats-box -- id
kubectl --context "$KUBE_CONTEXT" -n nats-system exec deployment/nats-box -- \
  nats --context default rtt

The identities should report UID/GID 1000. The CLI check uses the mounted NKey and requires the stack’s authentication service to be ready. Expect latency output; failed or an authorization error is a failure even if rtt exits successfully. Check the reloader logs after a configuration update to confirm it reloads successfully.

Use upstream cert-manager images

Merge these overrides into the Helmfile environment file deploy/stacks/self-managed/environments/<environment>.yaml to select the upstream cert-manager images already used by this candidate. Keep all components and CRDs on the selected version. For a private mirror, replace each full repository, which takes precedence over global.image.

certManager:
  image:
    repository: "quay.io/jetstack/cert-manager-controller"
    tag: "v1.20.2"
    digest: "sha256:fe0623d7d04a382c888f03343a3a2da716e0d96ad3d5d790c0ebcbcb2a4329a5"
  webhook:
    image:
      repository: "quay.io/jetstack/cert-manager-webhook"
      tag: "v1.20.2"
      digest: "sha256:baf651128b9f05c426cbd5e60e2036bf382c99ca270f49d0757d6f7d2452f4e5"
  cainjector:
    image:
      repository: "quay.io/jetstack/cert-manager-cainjector"
      tag: "v1.20.2"
      digest: "sha256:6f5a644135887b2aa7d5cc145072fa56421560e3586ff1f184358022d490f4e1"
  startupapicheck:
    image:
      repository: "quay.io/jetstack/cert-manager-startupapicheck"
      tag: "v1.20.2"
      digest: "sha256:4e2a69b4a0cc9627905bbeecf720f95d5153ca39cacdab923d2748e73556792b"
  acmesolver:
    image:
      repository: "quay.io/jetstack/cert-manager-acmesolver"
      tag: "v1.20.2"
      digest: "sha256:cfa55e9c3dcfc6a3cba97575748c1549b04a30fd14d88a2fcb9e2dddef415e2f"

For authenticated mirrors, set global.imagePullSecrets for chart pods and separate pull credentials in the ACME Issuer’s HTTP-01 solver pod template.

Override the Cassandra images

The Cassandra server and its dynamic seed discovery container take the same override keys, but they default to global.image. Set them when your registry uses a different path or tag than the cassandra default under global.image.repository:

cassandra:
  image:
    registry: nvcr.io
    repository: nvidia/nvcf/cassandra
  dynamicSeedDiscovery:
    image:
      registry: nvcr.io
      repository: nvidia/nvcf/cassandra

Every registry, repository, and tag key is optional. Any key you leave unset keeps its default: for Cassandra, registry and repository fall back to global.image.registry and global.image.repository; for the NATS server, reloader, NATS box, account-bootstrap and upgrade receipt images they fall back to the Docker Hub coordinates above. tag falls back to the version the stack pins, or to the chart’s own default when the stack pins none.

Use the version listed in the artifact table. Verify that your cluster can reach the registry you point each image at. If it requires authentication, add its pull secret to global.imagePullSecrets.

The current Cassandra initialization hook uses the nvcf-cassandra-migrations image, and the current NATS chart renders NKeys as Secrets without an nkey job. Their legacy cassandra.initialization.image and nats.nkeyJob.image values do not control rendered workloads. Using alpine-k8s for those operations requires chart support rather than a configuration-only override.

Use the upstream Vault injector image

To use the upstream injector at the same version as the NVCF mirror, merge this opt-in example into deploy/stacks/self-managed/environments/<environment>.yaml:

openbao:
  injector:
    image:
      registry: docker.io
      repository: hashicorp/vault-k8s
      tag: "1.7.4@sha256:e2ef79b457f50d971851df836f78cf41be454644095d84a7411f93ccf3304c15"

This changes only the injector distribution source. The webhook image is separate from openbao.injector.agentImage and openbao.server.image; keep those matched NVCF OpenBao images unchanged.

Keep the selected NVCF OpenBao server and migrations images. An upstream OpenBao image does not contain the NVCF JWT lifecycle binaries and cannot replace the bridge image during an upgrade or rollback. For NVCF 1.0.4, both JWT plugin binaries remain installed: existing mounts keep their selection, and new mounts use the target plugin. Binary availability does not mean that existing mounts have migrated. Follow the upgrade procedure before changing mount selections or removing legacy binaries.

The following tables list the artifacts that the stack inventories declare. Seven upstream images that operators create at runtime are not in those inventories. See Runtime images created by operators.

Stack releases

Control plane Helm charts

Control plane services and images

Compute plane Helm charts

Compute plane services and images

Observability Helm charts

Observability services and images

Cross-stack Helm charts

Cross-stack services and images

Tools and deployment resources

Runtime images created by operators

Two operators create pods at runtime from images that no chart or Helm value lists, so the tables above do not include them. Add these images to your registry mirror or allowlist when you enable the matching component. They are third-party images that NVIDIA does not republish under nvidia/nvcf. Pull them from ghcr.io or mirror them with the other upstream images.

The versions match the operator versions that the selected stack releases pin: OpenTelemetry operator 0.158.0 and KAI Scheduler v0.17.1. If you pin a different operator version, use the matching image tags.