Version: latest

Disable Webhook to Customize Image Pull Behavior

Palette Edge allows you to deploy clusters to an external registry and use a local Harbor registry without manually configuring rewrites. This is possible because the Palette agent uses a webhook to redirect image pulls to the appropriate locations depending on your configuration.

While the webhook makes using an external registry or the local Harbor registry more streamlined, it can limit your flexibility to configure your own image pull behavior. This guide guides you through how to disable the Palette agent webhook and provides an example custom image pull configuration.

What Happens When You Disable the Webhook

When the agent webhook is disabled, the Palette agent will not redirect any image pull operation by default. This means that even if you specify an external registry in the user-data, the Palette agent will not pull images from that registry unless it is otherwise configured to do so. This also means that the Palette agent will not pull images from the local Harbor registry, even if the images are downloaded and stored in the registry, unless it is otherwise configured to do so. Disabling the webhook removes restrictions, but does place the burden of ensuring that images are pulled from the correct locations on yourself.

You may consider disabling the webhook if you want to configure your cluster to pull images from multiple authenticated registries, or if you do not want the default behavior that forces image pulls to be redirected to the local Harbor registry. Once the webhook is disabled, you can then take advantage of the rewrite features of some Kubernetes distributions such as K3s and RKE2, or other redirect mechanism that you implement on your own to customize the image pull behavior.

Prerequisites

The process to disable webhook is based on the EdgeForge process. We recommend that you familiarize yourself with EdgeForge and the process to build Edge artifacts.
A physical or virtual Linux machine with AMD64 (also known as x86_64) processor architecture to build the Edge artifacts. You can issue the following command in the terminal to check your processor architecture.
```
uname -m
```
Minimum hardware configuration of the Linux machine:
- 4 CPU
- 8 GB memory
- 150 GB storage
Git. You can ensure git installation by issuing the git --version command.
Docker Engine version 18.09.x or later. You can use the docker --version command to view the existing Docker version. You should have root-level or sudo privileges on your Linux machine to create privileged containers.
A Spectro Cloud account.

Procedure

Disable Webhook

Clone the CanvOS repository.

git clone https://github.com/spectrocloud/CanvOS.git

Change into the CanvOS directory.
```
cd CanvOS
```
View the available tags and use the latest available tag. This guide uses v4.5.0 as an example.
```
git tag
git checkout v4.5.0
```

In your user-data file, set stylus.imageRedirectWebhook.enable to false. This parameter defaults to true if you do not explicitly set it to false.

#cloud-config
stylus:
  site:
    edgeHostToken: XXXXXXXXXXXXXX
    paletteEndpoint: XXXXXX
    projectUid: XXXXXX
  imageRedirectWebhook:
    enable: false

Follow the rest of the Build Edge Artifact guide and build the Installer ISO with the user data configurations. The Edge clusters provisioned with the ISO will no longer automatically redirect image pull requests to the external registry or the local Harbor registry.

Redirect Image Pull

The process to redirect image pulls varies by Kubernetes distribution as well as your registry setup. This section provides an example that shows how you might customize the image pull behavior of your Edge cluster using PXK-E.

Log in to Palette.
From the left Main Menu, click Profiles. Click on the profile you use to deploy your Edge cluster.

(PXK-E Only) In the Kubernetes layer of the profile, include the following lines in the initramfs stage to adjust the containerd configuration to supports reading additional files, which you will use to configure the redirect behavior and provide registry credentials.

stages:
  initramfs:
    - name: "Manage containerd config"
      files:
        - path: /etc/containerd/config.toml
          permissions: 0644
          owner: 0
          group: 0
          content: |-
            version = 2
            imports = ["/etc/containerd/conf.d/*.toml"]
            [plugins]
              [plugins."io.containerd.grpc.v1.cri"]
                sandbox_image = "registry.k8s.io/pause:3.9"
                enable_unprivileged_ports = true
                enable_unprivileged_icmp = true
              [plugins."io.containerd.grpc.v1.cri".containerd.runtimes.runc]
                runtime_type = "io.containerd.runc.v2"
              [plugins."io.containerd.grpc.v1.cri".containerd.runtimes.runc.options]
                SystemdCgroup = true
              [plugins."io.containerd.grpc.v1.cri".registry]
                config_path = "/etc/containerd/certs.d"

info

This configuration change is only needed to PXK-E. Since it changes the initramfs stage, this will require a reboot of the node. If you are using K3s or RKE2, the ability fo read additional files is enabled by default and you don't need to add this configuration.

In the Kubernetes layer of the profile, include the following lines in the reconcile stage. For more information, refer to Cloud-init Stages. Replace the server address and the host address with your registry and its mirror. Since you are only editing the reconcile stage, this will not result in a reboot or service restart for your cluster.

The following example will redirect image pulls for https://gcr.io to https://gcr-io-mirror.company.local.

stages:
 reconcile:
     - name: "Redirect registries"
         - path: /etc/containerd/certs.d/gcr.io/hosts.toml
           owner: 0
           group: 0
           permissions: 0644
           content: |-
             server = "https://gcr.io"
             [host."https://gcr-io-mirror.company.local"]
                 capabilities = ["pull", "resolve"]

Provide Registry Credentials

If you are using public registries that do not require authentication, you can skip this step.

If your registries require authentication, you will need to provide credentials to enable image pulls. This example uses an open source generic Kubernetes credentials provider to provide the resources. There are other resources that you can take advantage of to provide registry credentials, including using a registry.yaml file in K3s or RKE2. However, the advantage of the approach used in this example is that after installation, you will not need to restart your cluster services to update the registry paths or the registry credentials.

info

Refer to the generic-credential-provider GitHub repository for the source code for the credential provider on GitHub.

Install the generic credential provider by including the following lines in either the OS layer or the Kubernetes layer of the cluster profile. This will create the file at /usr/local/bin/generic-credential-provider, populate the content, and set the file permissions to ensure that it is executable, during the initramfs stage. Click on the box below to expand the instructions.

Install Credential Provider

stages:
  initramfs:
      - name: "Install generic-credential-provider binary"
        files:
          - path: /usr/local/bin/generic-credential-provider
            permissions: 0755
            owner: 0
            group: 0
            content: |-
              #!/usr/bin/env python3
              import os
              import sys
              import json
              import syslog
              import logging
              import argparse

              logging.basicConfig(level=logging.DEBUG, format='%(levelname)s: %(message)s')

              class generic_credential_provider:
                  def __init__(self):
                      args = generic_credential_provider_utilities.parse_args()
                      if not args.debug:
                          logging.disable(logging.DEBUG)
                      if args.version:
                          print(f"generic-credential-provider version 1.0.0")
                          exit(0)

                      base_path = args.credroot

                      syslog.openlog("generic-credential-provider", syslog.LOG_PID, syslog.LOG_USER)

                      # Read the input JSON from stdin
                      input_json = json.load(sys.stdin)
                      repository = generic_credential_provider_utilities.get_image_repository(input_json)

                      # Generate possible JSON filenames
                      possible_filenames = generic_credential_provider_utilities.generate_possible_filenames(repository)

                      # Search for the JSON file in the specified directory
                      found_json_file = generic_credential_provider_utilities.find_json_file(possible_filenames, base_path)

                      if found_json_file:
                          # If a matching JSON file is found, read the username and password
                          credentials = generic_credential_provider_utilities.read_credentials(found_json_file)
                          username = credentials.get("username", "")
                          password = credentials.get("password", "")
                          duration = credentials.get("duration", "0h5m0s")
                      else:
                          syslog.syslog(syslog.LOG_ERR, f"Failed to fulfill credential request for {repository}")
                          logging.error(f'Error running credential provider plugin: {repository} is an unknown source')
                          exit(1)

                      # Create the output JSON response
                      output_json = {
                          "kind": "CredentialProviderResponse",
                          "apiVersion": "credentialprovider.Kubelet.k8s.io/v1",
                          "cacheKeyType": "Registry",
                          "cacheDuration": duration,
                          "auth": {
                              repository: {
                                  "username": username,
                                  "password": password
                              }
                          }
                      }
                      syslog.syslog(syslog.LOG_INFO, f"Credential request fulfilled for {repository}")

                      # Print the output JSON response to stdout
                      json.dump(output_json, sys.stdout)

              class generic_credential_provider_utilities:
                  def parse_args():
                      parser = argparse.ArgumentParser(description="A generic credential provider for Kubernetes")

                      parser.add_argument('--version', '-v', action='store_true', help="version for generic-credential-provider")
                      parser.add_argument('--debug', '-d', action='store_true', help="Enable debug output")
                      parser.add_argument('--credroot', '-c', help="Provide a new credential root, only used for testing", default="/etc/kubernetes/registries/")

                      return parser.parse_args()

                  def get_image_repository(input_json):
                      image = input_json.get("image", "")
                      repository = image.split('/')[0].split(':')[0]
                      logging.debug(f"Got repository name: {repository}")
                      return repository

                  def generate_possible_filenames(repository):
                      possible_filenames = []
                      possible_filename = ""
                      parts = repository.split(".")
                      parts.reverse()
                      for part in parts:
                          if possible_filename == "":
                              possible_filename = part
                          else:
                              possible_filename = f"{part}.{possible_filename}"
                          possible_filenames.append(possible_filename)
                      possible_filenames.reverse()
                      return possible_filenames

                  def find_json_file(possible_filenames, base_path):
                      for filename in possible_filenames:
                          json_file_path = os.path.join(base_path, f"{filename}.json")
                          logging.debug(f"Testing json_file_path: {json_file_path}")
                          if os.path.exists(json_file_path):
                              logging.debug(f"Got it")
                              return json_file_path
                      return None

                  def read_credentials(json_file_path):
                      with open(json_file_path, "r") as json_file:
                          credentials = json.load(json_file)
                          logging.debug(f"Got credentials: {credentials}")
                          return credentials

              if __name__ == "__main__":
                  generic_credential_provider()

In the Kubernetes layer of your cluster, add the following lines to the kubeadmconfig.KubeletExtraArgs field. This tells Kubernetes to use the credential provider you installed in the previous step.

KubeletExtraArgs:
  cluster:
    config: |
      initConfiguration:
        nodeRegistration:
          KubeletExtraArgs:
            image-credential-provider-bin-dir: /usr/local/bin
            image-credential-provider-config: /opt/kubernetes/generic-credential-provider-config.json
      joinConfiguration:
        discovery: {}
        nodeRegistration:
          KubeletExtraArgs:
            image-credential-provider-bin-dir: /usr/local/bin
            image-credential-provider-config: /opt/kubernetes/generic-credential-provider-config.json

In the Kubernetes or OS layer of your cluster profile, use a reconcile stage to define the JSON file with the CredentialProviderConfig for Kubelet. This configuration specifies the registries that will use the credential provider.

stages:
  reconcile:
    - name: "Registry credential management"
      files:
        - path: /opt/kubernetes/generic-credential-provider-config.json
          owner: 0
          group: 0
          permissions: 0644
          content: |-
            {
              "apiVersion": "Kubelet.config.k8s.io/v1",
              "kind": "CredentialProviderConfig",
              "providers": [
                {
                  "name": "generic-credential-provider",
                  "matchImages": [
                    "*.io",
                    "*.*.io"
                  ],
                  "defaultCacheDuration": "5m",
                  "apiVersion": "credentialprovider.Kubelet.k8s.io/v1"
                }
              ]
            }

Registry URLs that match the patterns in the mathImages field will use this provider for credentials. For example, *.io would match docker.io, quay.io, gcr.io and *.*.io would cover URLs like us.gcr.io. Refer to Kubernetes documentation about the parameters you can use to configure credential providers for Kubelet.

tip

We suggest that you define a broad pattern, as updating this file requires Kubelet to restart. There are no adverse effects when the pattern matches a registry URL for which there is no defined credentials: the Palette agent will still perform the image pull, but it will not use the credential provider.

With the credential provider installed and configured, you can now provide your registry credentials. Similar to other configurations, you will also perform this step during the reconcile stage. Add the following lines to your cluster profile in the OS or the Kubernetes layer. Replace <registry-url> with the URL of your image registry. Replace the username and password fields with your credentials.
```
stages:
  reconcile:
    - name: "Registry credential management"
      files:
        - path: /etc/kubernetes/registries/<registry-url>.json
          owner: 0
          group: 0
          permissions: 0644
          content: |-
            {
              "username": "proxy-access",
              "password": "*****",
              "duration": "0h5m0s"
            }
```
warning
Avoid entering sensitive information like passwords directly into your cluster profile in plain text. Instead, you can either use a cluster profile variable or a macro. For more information, refer to Macros and Define and Manage Profile Variables.

The script will lookup the JSON file by registry naming and allows partial matches. For example, gcr.io.json would match gcr.io as well as us.gcr.io, and those image pulls to those registries will use the credentials your provided.

Validate

Use the ISO to install Palette Edge on an Edge host. For more information, refer to Installation.
Create a cluster profile that references the image registries you configured.
Deploy a cluster with the cluster profile. Confirm that the cluster is able to download the necessary images.

What Happens When You Disable the Webhook​

Prerequisites​

Procedure​

Disable Webhook​

Redirect Image Pull​

Provide Registry Credentials​

Validate​