HTTP Proxy Configuration

HTTP Proxy Configuration#

The AMD GPU Operator supports usage within a Kubernetes cluster behind an HTTP Proxy. Generally, the AMD GPU Operator requires Internet access for two reasons:

  1. Pulling he container images from the registry during installation.

  2. Downloading the AMD GPU driver installer.

Note

Downloading the driver installer (Step 2) can be skipped when using a pre-compiled driver image.

When users setting up a Kubernetes cluster with traffic redirected to a proxy server, ensure the Kubernetes nodes, container runtime, and GPU Operator pods are properly configured to apply the proxy network settings.

This document won’t cover all the detailed steps about how to setup proxy network, configure OS level proxy configuration and update the container runtime level networking settings, since those steps are not specific to the AMD GPU Operator. The rest of the document will show users the methods to inject the proxy configuration to AMD GPU Operator so that all the components images and driver installer can be downloaded successfully behind a HTTP proxy.

Configuration Options#

You can configure the following proxy settings:

  • HTTP_PROXY: The HTTP proxy server URL

  • HTTPS_PROXY: The HTTPS proxy server URL

  • NO_PROXY: A comma-separated list of hostnames or IP addresses that should bypass the proxy

Methods to Apply Proxy Settings#

There are two ways to apply these proxy settings:

Method 1: Using --set with helm install#

You can specify the proxy settings directly in the helm install command using the --set flag.

helm install amd-gpu-operator rocm/gpu-operator-charts \
  --namespace kube-amd-gpu \
  --create-namespace \
  --set global.proxy.env.HTTP_PROXY=http://myproxy.com:123 \
  --set global.proxy.env.HTTPS_PROXY=http://myproxy2.com:234 \
  --set global.proxy.env.NO_PROXY="10.1.2.3\,localhost"

Note: When using --set, use \, to separate multiple values in the NO_PROXY setting.

Method 2: Using a Custom values.yaml File#

  • Create a Helm Charts values YAML file named custom-values.yaml with the following content included:

global:
  proxy:
    env:
      HTTP_PROXY: "http://myproxy.com:123"
      HTTPS_PROXY: "http://myproxy2.com:234"
      NO_PROXY: "10.1.2.3,localhost"

  • Use this file when installing the Helm chart:

helm install amd-gpu-operator rocm/gpu-operator-charts \
  --namespace kube-amd-gpu \
  --create-namespace \
  -f custom-values.yaml

Verifying Proxy Settings#

After installation, you can verify the proxy settings by inspecting the environment variables of the deployed pods:

kubectl get pods -n kube-amd-gpu
kubectl exec -it <pod-name> -n kube-amd-gpu -- env | grep -i proxy

Replace <pod-name> with the name of one of the GPU Operator pods.