This guide focuses on configuring access to a private Docker registry with self-signed certificates.

If focuses on:

• Secure automatic mounting of a registry certificate to notebook containers, and
• Using the SDK API to configure the properties necessary to access a private registry.

Using custom certificates with a private Docker registry requires the configuration of Docker access credentials using Kubernetes Secrets and PodDefaults as described in the Accessing Docker and Cloud Storage guide.

Pre-requisites

• A private Docker registry reachable from the cluster over the network.
• Configuration of the Nodes of the cluster where Kaptain is running to accept the self-signed certificates used by the registry. More information is available in the Konvoy air-gapped installation guide

Creating Docker credentials and registry certificate secrets

Access to a private Docker registry with self-signed certificate requires:

• A Docker configuration file (known as Docker config.json) which contains the registry URL and authentication information (login and password encoded in base64 format), and
• A CA certificate file for the private Docker registry to validate its authenticity.

Example Docker config.json:

{
    "auths": {
            "https://my.private.registry:5000/": {
                    "auth": "<username and password in base64>"
            }
    }
}

Example CA certificate file:

-----BEGIN CERTIFICATE-----
MIIFATCCAumgAwIBAgIJAL4N40kdtHKvMA0GCSqGSIb3DQEBCwUAMBcxFTATBgNV
BAMMDDEwLjAuMTkyLjIzMDAeFw0yMTAyMjMxODEyMTBaFw00ODA3MTExODEyMTBa
... <omitted for better readability> ...
uvoXCJ1qhxxkeTFwTGz78xkjZAGRxPzuOSPtnFFfidbE3i+WEbj6PTbz4AbPamjn
0dFfxN97eKhnCVQpZ/ijcwZWbPvaXg+QkIMOG3B9wjmDzN+20g==
-----END CERTIFICATE-----

To create a Secret from the credentials file config.json run the following command:

kubectl create secret generic docker-secret -n <kaptain_namespace> --from-file=config.json=config.json

Be sure to replace <kaptain_namespace> with the namespace you use for creating notebooks. In this example, we used a namespace named ‘user’.

Verify the Secret is created:

kubectl get secret docker-secret -o yaml

# the output should look like this:

apiVersion: v1
data:
  config.json: ewogICJhdXRocyI6IH...
kind: Secret
metadata:
  name: docker-secret
  namespace: user
type: Opaque

To create a Secret from the certificate file certificate.crt run the following command:

kubectl create secret generic docker-registry-certificate -n <kaptain_namespace> --from-file=certificate.crt=certificate.crt

Verify the Secret is created:

kubectl get secret -n <kaptain_namespace> docker-registry-certificate -o yaml

# the output should look like this:

apiVersion: v1
data:
  certificate.crt: LS0tLS1CRUdJTiBDRVJUSUZJQ0FURS0tLS0tCk1JSUZBVENDQXVtZ0F3SUJBZ0lKQUw0T...
kind: Secret
metadata:
  name: docker-registry-certificate
  namespace: user
type: Opaque

NOTE: It is recommended to name the files with the Secret config.json for Docker credentials and certificate.crt for the registry CA certificate, as this will allow the SDK to automatically use them from the default locations. It is also possible to use custom names for the files in a secret but it will require custom configuration of the SDK Config object.

Create PodDefaults for mounting secrets to Jupyter Notebooks

To make the Docker credentials Secret available for selection in the Notebook creation dialogue, create a PodDefault referencing them. Create a file named pod-defaults.yaml with the following contents:

apiVersion: "kubeflow.org/v1alpha1"
kind: PodDefault
metadata:
  name: notebook-configuration
  namespace: user
spec:
  selector:
    matchLabels:
      notebook-configuration: "true"
  desc: "Private Docker Registry"
  volumeMounts:
    - name: docker-secret-volume
      mountPath: /home/kubeflow/.docker/
    - name: docker-registry-certificate
      mountPath: /home/kubeflow/.tls/
  volumes:
    - name: docker-secret-volume
      secret:
        secretName: docker-secret
    - name: docker-registry-certificate
      secret:
        secretName: docker-registry-certificate

WARNING: Volume name and mountPath must be unique across all PodDefault objects to avoid conflicts when mounting Secrets to Pods.

Create a PodDefault resource from file using the following command:

kubectl create -f pod-defaults.yaml

After that, the Docker credentials and registry certificate secrets will be available for selection in the Notebook Spawner UI and, if selected, will be mounted as /home/kubeflow/.docker/config.json and /home/kubeflow/.tls/certificate.crt correspondingly:

Configure the SDK to use the private Docker registry with self-signed certificates

In order for the Kaptain SDK to authenticate with a private Docker registry with self-signed certificates, a custom Config object is required:

from kaptain.config import Config
from kaptain.platform.config.certificates import DockerRegistryCertificateProvider

config = Config.default()
config.docker_registry_url = "my.private.registry:5000"
config.docker_registry_certificate_provider = DockerRegistryCertificateProvider.default()

The code above creates a default Config object which automatically uses a Docker credentials file (config.json) mounted to /home/kubeflow/.docker/config.json. This configuration works for both self-hosted Docker registries as well as public registries.

In order to add a certificate verification option to the model training and tuning steps, it is required to provide two additional properties: docker_registry_url and docker_registry_certificate_provider.

DockerRegistryCertificateProvider.default() is a utility method which reads the registry certificate from the default location /home/kubeflow/.tls/certificate.crt. If the certificate is mounted in the non-default location, use the DockerRegistryCertificateProvider.from_file(<path/to/certificate.crt>) method to specify the location.

Once the Config properties are set, it is sufficient to provide it to the Model constructor to override the defaults and to enable private registry support. Note in the example below that the image_name uses a private registry for pushing and pulling images with the model trainer code and dependencies:

model = Model(
    id="Model id",
    name="Model name",
    description="Model description",
    version="Model version",
    framework="tensorflow",
    framework_version="2.4.0",
    main_file="trainer.py",
    base_image="mesosphere/kubeflow:1.3.0-tensorflow-2.5.0",
    image_name="my.private.registry:5000/kubeflow:mnist-sdk-example",
    config=config,
)