This is a vanilla-compatible slim package of the upstream Zarf init components, excluding the k3s component. The package is named init-slim, which is the name it shows up under in zarf package list.
All components are made optional to allow deploying different components slowly over the course of an install.
The package uses the zarf.dev/v1beta1 API and composes the upstream component configs published at oci://ghcr.io/zarf-dev/components/init/*. Configuration is done with values, not deploy variables: either a values file (zarf package deploy -v my-values.yaml) or --set-values key.path=value. Each component exposes its chart under a top-level key - registry (the base for both registries), seedRegistry (overrides on top of registry, for the ephemeral one), agent, and gitServer.
Per zarf#5312, the registry image uses the standard s3 storage driver from environment variables, requiring no custom init package builds.
To wire the registry into an S3 bucket (such as the registry-bucket created by zarf-csi-rook-ceph):
- Set
registry.persistence.enabled: false(required so the chart does not define a filesystem storage driver alongside S3). - Pass S3 configuration through
registry.extraEnvVars.
registry:
persistence:
enabled: false
extraEnvVars:
- name: REGISTRY_STORAGE_S3_REGION
value: "us-east-1"
- name: REGISTRY_STORAGE_S3_BUCKET
value: "registry-bucket"
- name: REGISTRY_STORAGE_S3_REGIONENDPOINT
value: "http://rook-ceph-rgw-ceph-objectstore.rook-ceph.svc.cluster.local:8080"
- name: REGISTRY_STORAGE_S3_FORCEPATHSTYLE
value: "true"
- name: REGISTRY_STORAGE_S3_SECURE
value: "false"
- name: REGISTRY_STORAGE_REDIRECT_DISABLE
value: "true"
- name: REGISTRY_STORAGE_S3_ACCESSKEY
valueFrom:
secretKeyRef:
name: registry-bucket
key: AWS_ACCESS_KEY_ID
- name: REGISTRY_STORAGE_S3_SECRETKEY
valueFrom:
secretKeyRef:
name: registry-bucket
key: AWS_SECRET_ACCESS_KEYDeploy it with zarf package deploy ... -v s3-values.yaml, or point package.deploy.values at it in your zarf-config.yaml.
Any other storage.s3 setting from the distribution docs can be passed the same way, for example REGISTRY_STORAGE_S3_V4AUTH, REGISTRY_STORAGE_S3_CHUNKSIZE, REGISTRY_STORAGE_S3_MULTIPARTCOPYCHUNKSIZE, REGISTRY_STORAGE_S3_MULTIPARTCOPYTHRESHOLDSIZE, and REGISTRY_STORAGE_S3_MULTIPARTCOPYMAXCONCURRENCY. Settings left unset use the distribution defaults.
When using S3 storage, the registry does not require a ReadWriteMany volume for multi-replica operation. Scaling is configured with values:
registry:
autoscaling:
enabled: true
minReplicas: 3
maxReplicas: 7Replicas are already spread across nodes: values/init-slim.yaml sets a preferred podAntiAffinity on kubernetes.io/hostname.
values/init-slim.yaml ships with the package and is applied after the upstream component defaults, so it decides these keys unless you override them at deploy time:
| Value path | What it does |
|---|---|
registry.tolerations, registry.proxy.tolerations |
operator: Exists - registry and its proxy tolerate any taint |
registry.affinity.enabled, registry.affinity.custom |
Registry pods only schedule onto nodes labelled adrp.xyz/is-infra=true, and prefer to spread across hosts |
registry.serviceAccount |
Creates a docker-registry service account |
agent.tolerations |
operator: Exists for the Zarf agent |
seedRegistry.persistence.enabled, seedRegistry.autoscaling.enabled |
false - the seed registry is torn down once the permanent one is up, so it provisions no volume and does not scale |
The infrastructure node label is baked in at adrp.xyz/is-infra. Label the nodes first, or override registry.affinity.custom at deploy time, or the registry will not schedule.