Skip to main content
Version: Next

Kubernetes MarbleRun deployment

This guide walks you through setting up MarbleRun in your Kubernetes cluster.

The Kubernetes deployment is managed through the use of a Helm chart, which can be found in our source repository and installed via our Helm repository. The installation consists of a deployment for the Coordinator and an admission controller. For more details see our section on Kubernetes Integration.

Prerequisites

SGX device plugin on Kubernetes

Kubernetes manages hardware resources like Intel SGX through its device plugin framework. The SGX device plugin can either be deployed manually or as a DaemonSet in the cluster. Different vendors provide open-source device plugins for SGX:

info

If you are using a confidential computing-enlightened, managed Kubernetes cluster, you will usually already have an SGX device plugin installed. For example, creating a confidential computing cluster on AKS comes with a pre-configured SGX device plugin.

Manually deploying an SGX device plugin

For different reasons, you may want to deploy the device plugin manually. Intel provides a guide to install their SGX plugin. In any case, you will need to adjust your deployments to request the SGX resources provided by the plugin:

apiVersion: apps/v1
kind: Deployment
metadata:
name: oe-deployment
spec:
selector:
matchLabels:
app: oe-app
replicas: 1
template:
metadata:
labels:
app: oe-app
spec:
tolerations:
- key: sgx.intel.com/epc
operator: Exists
effect: NoSchedule
containers:
- name: <image_name>
image: <image_reference>
command: <exec>
resources:
limits:
sgx.intel.com/epc: 10Mi
sgx.intel.com/enclave: 1
sgx.intel.com/provision: 1

Note, that every plugin uses its own way of injecting SGX resources into deployments. Please refer to the documentation for your plugin of choice. This is an example of the Intel plugin.

MarbleRun supports automatic injection of those values for a selection of popular plugins:

  • Intel using sgx.intel.com/epc, sgx.intel.com/enclave, and sgx.intel.com/provision
  • Alibaba Cloud using alibabacloud.com/sgx_epc_MiB
  • You can use the --resource-key flag, during installation with the CLI, to declare your own SGX resource key for injection
tip

If you are using a different plugin please let us know, so we can add support!

Out-of-process attestation

Intel SGX supports two modes for obtaining remote attestation quotes:

  • In-process: The software generating the quote is part of the enclave application
  • Out-of-process: The software generating the quote isn't part of the actual enclave application. This requires the Intel SGX Architectural Enclave Service Manager (AESM) to run on the system

While Marbles built with Ego perform in-process attestation, other frameworks, such as Gramine, use out-of-process attestation. If your confidential application uses out-of-process attestation, you will need to expose the AESM device to your container.

You can follow the AKS guide to make your deployments able to use AESM for quote generation. Note, that in this case, your Kubernetes nodes need the AESM service installed. See the Intel installation guide for more information.

Option 1: Install with the MarbleRun CLI

Use MarbleRun's CLI that facilitates the administrative tasks. You can install MarbleRun using the CLI as follows:

  • For a cluster with SGX support:

    marblerun install --domain=mycluster.uksouth.cloudapp.azure.com
  • For a cluster without SGX support:

    marblerun install --domain=mycluster.uksouth.cloudapp.azure.com --simulation

This command will pull the latest Helm chart from our repository and manages the installation of said chart.

By default --domain is set to localhost. The domain is used as the CommonName in the Coordinator's TLS certificate. This certificate is used for the HTTPS communication of the Coordinator's client API. The HTTPS endpoint is exposed via a Kubernetes ClusterIP Service. If you plan on exposing the endpoint on a public IP, make sure that the domain set via --domain matches the one configured for the public IPs provisioned in your cluster. On Azure, you can use a static public IP address with the Azure Kubernetes Service (AKS) load balancer. The client API can be used by users/clients of your application to obtain one concise remote attestation statement for your cluster.

The Coordinator is now in a pending state, waiting for a manifest. See the how to add a service documentation for more information on how to create and set a manifest.

Option 2: Install with Helm

Make sure that you are using the latest release of Helm and have access to the MarbleRun Helm repositories. For upgrade instructions, see the Helm install docs. For more information on configuring and using Helm, see Install applications with Helm in Azure Kubernetes Service (AKS).

Adding MarbleRun's Helm repository

helm repo add edgeless https://helm.edgeless.systems/stable
helm repo update

Installing the chart

Note that installing MarbleRun with the marble-injector webhook enabled using Helm requires cert-manager to be installed in your cluster. Review the values.yaml file of the chart for a full list of available configuration options.

Update the hostname with your cluster's FQDN.

  • For a cluster with SGX support:

    helm install marblerun edgeless/marblerun \
    --create-namespace \
    -n marblerun \
    --set coordinator.hostname=mycluster.uksouth.cloudapp.azure.com \
    --set marbleInjector.start=true \
    --set marbleInjector.useCertManager=true
  • For a cluster without SGX support:

    helm install marblerun edgeless/marblerun \
    --create-namespace \
    -n marblerun \
    --set coordinator.resources=null \
    --set coordinator.simulation=1 \
    --set tolerations=null \
    --set coordinator.hostname=mycluster.uksouth.cloudapp.azure.com \
    --set marbleInjector.start=true \
    --set marbleInjector.useCertManager=true

By default coordinator.hostname is set to localhost. The domain is used as the CommonName in the Coordinator's TLS certificate. This certificate is used for the HTTPS communication of the Coordinator's client API. The HTTPS endpoint is exposed via a Kubernetes ClusterIP Service. If you plan on exposing the endpoint on a public IP, make sure that the domain set via --domain matches the one configured for the public IPs provisioned in your cluster. On Azure, you can use a static public IP address with the Azure Kubernetes Service (AKS) load balancer. The client API can be used by users/clients of your application to obtain one concise remote attestation statement for your cluster.

The Coordinator is now in a pending state, waiting for a manifest. See the how to add a service documentation for more information on how to create and set a manifest.

(Optional) Exposing the client API

The Coordinator creates a ClusterIP service called coordinator-client-api exposing the client API on the default port 4433. Depending on your deployment type you may want to deploy an Ingress Gateway forwarding the traffic or create a LoadBalancer service to expose the endpoint on a public IP.

You can also use kubectl to forward the port to your local system:

kubectl -n marblerun port-forward svc/coordinator-client-api 4433:4433 --address localhost

Ingress/Gateway configuration

If you're using an ingress-controller or gateway for managing access to the coordinator-client-api service, make sure you're enabling SNI for your TLS connections.

DCAP configuration

By default, the Coordinator's quote provider is configured to generate its quote using the Azure PCCS. If you're running on Azure, no additional steps are required. Otherwise, set the necessary configuration during installation:

  • Using the CLI

    marblerun install --dcap-pccs-url <PCCS_URL> --dcap-secure-cert <TRUE/FALSE>
  • Using Helm

    helm install marblerun edgeless/marblerun \
    --create-namespace \
    -n marblerun \
    --set coordinator.hostname=mycluster.uksouth.cloudapp.azure.com \
    --set dcap.pccsUrl=<PCCS_URL> \
    --set dcap.useSecureCert=<TRUE/FALSE>