Part 3 - In-scope workload / The landing zone of a PCI-DSS compliant application inside Microsoft Azure

The last article was about shared responsibility between the three parties that are involved Microsoft Azure, Kubernetes (AKS) and the customer. In today's post, we tackle Azure Kubernetes Services and where the workload would need to run.

Azure Kubernetes Services (AKS) runs all the payloads where card data are decrypted and processed. It is required to run in a private cluster with no direct access to the internet. All the traffic must be limited to the private network, without direct access to and from the internet. It applies for workloads that process card data (in-scope)

A dedicated node pool with a subnet is used for in-scope components (services handling card data). The payload runs in isolated node pools, isolated from the rest of the system. Additional node pools can be created for workloads that are not out-of-scope of PCI-DSS. Using such a topology, we can move the services from out-of-scope to in-scope seamless without significant changes in the architecture of the infrastructure. 

The pod security should not be ignored we need to ensure that we:

(1) Removing the Linux capabilities

(2) Configure the securityContext of containers

(3) Use the image ID to ensure you have a reliable map of the containers

(4) Use Azure Policy to ensure that image ID is used 

For a more extensive system, where the complexity is higher, there can be an approach where the in-scope workload runs in a dedicated ASK cluster, and the out-of-scope has its own cluster (dedicated or shared). It is a more expensive solution but with a higher isolation level between the in-scope workloads and the rest of the system. For ingress, we can use controllers like NGINX or Caddy. As long as we can control the flow of data and balance it between the node pool. 

All pods need to have restricted network policies defined, and only specific traffic to be allowed.  Direct communication between in-scope and out-of-scope workloads can exist, but strict policy definitions must be implemented. Additionally, when you have direct communication between the namespaces (in the case of a mesh) or outside the mesh, a security layer like mTLS with a compatible certificate issuer needs to be used.

The AKS cluster is not publicly available on the internet. Meaning that a support engineer would not be able to connect to it. A jump box is required, combined with strict IP ranges that are allowed to access the cluster and execute commands.