Understanding VPC cluster networking

Worker-to-worker communication using VPC subnets

Before you create a VPC cluster for the first time, you must create a VPC subnet in each zone where you want to deploy worker nodes. A VPC subnet is a specified private IP address range (CIDR block) and configures a group of worker nodes and pods as if they are attached to the same physical wire.

When you create a cluster, you specify an existing VPC subnet for each zone. Each worker node that you add in a cluster is deployed with a private IP address from the VPC subnet in that zone. After the worker node is provisioned, the worker node IP address persists after a reboot operation, but the worker node IP address changes after replace and update operations.

Subnets provide a channel for connectivity among the worker nodes within the cluster. Additionally, any system that is connected to any of the private subnets in the same VPC can communicate with workers. For example, all subnets in one VPC can communicate through private layer 3 routing with a built-in VPC router. If you have multiple clusters that must communicate with each other, you can create the clusters in the same VPC. However, if your clusters don't need to communicate, you can achieve better network segmentation by creating the clusters in separate VPCs. You can also create access control lists (ACLs) for your VPC subnets to mediate traffic on the private network. ACLs consist of inbound and outbound rules that define which ingress and egress is permitted for each VPC subnet.

When you create a VPC cluster and enable both the public and private cloud service endpoints during cluster creation, the public cloud service endpoint is used by default for access to components such as the Red Hat OpenShift web console for your cluster. In order for console pods to establish a secure, public connection over the internet through the public service endpoint, you must enable a public gateway on each VPC subnet that your worker nodes are deployed to.

When you create a VPC cluster and enable only the private cloud service endpoint during cluster creation, the private cloud service endpoint is used by default to access Red Hat OpenShift components such as the Red Hat OpenShift web console or OperatorHub. You must be connected to the private VPC network, such as through a VPN connection, to access these components or run kubectl commands on your cluster.

The default IP address range for VPC subnets is 10.0.0.0 – 10.255.255.255. For a list of IP address ranges per VPC zone, see the VPC default address prefixes.

If you enable classic access when you create your VPC, classic access default address prefixes automatically determine the IP ranges of any subnets that you create. However, the default IP ranges for classic access VPC subnets conflict with the subnets for the Red Hat OpenShift on IBM Cloud control plane. Instead, you must create the VPC without the automatic default address prefixes, and then create your own address prefixes and subnets within those ranges for your cluster.

Need to create your cluster by using custom-range subnets? Check out this guidance on custom address prefixes. If you use custom-range subnets for your worker nodes, you must ensure that your worker node subnets don't overlap with your cluster's pod subnet.

Do not delete the subnets that you attach to your cluster during cluster creation or when you add worker nodes in a zone. If you delete a VPC subnet that your cluster used, any load balancers that use IP addresses from the subnet might experience issues, and you might be unable to create new load balancers.

When you create VPC subnets for your clusters, keep in mind the following features and limitations. For more information about VPC subnets, see Characteristics of subnets in the VPC.

• The default CIDR size of each VPC subnet is /24, which can support up to 253 worker nodes. If you plan to deploy more than 250 worker nodes per zone in one cluster, consider creating a subnet of a larger size.
• After you create a VPC subnet, you can't resize it or change its IP range.
• Multiple clusters in the same VPC can share subnets.
• VPC subnets are bound to a single zone region and can't span multiple zones or regions.
• After you create a subnet, you can't move it to a different zone, region, or VPC.
• If you have worker nodes that are attached to an existing subnet in a zone, you can't change the subnet for that zone in the cluster.
• The 172.16.0.0/16, 172.18.0.0/16, 172.19.0.0/16, and 172.20.0.0/16 ranges are prohibited.

Worker-to-master and user-to-master communication using Virtual private endpoints or cloud service endpoints

Red Hat OpenShift on IBM Cloud uses different types of service endpoints to establish a connection from authorized cluster users and worker nodes to the Kubernetes master. Authorized cluster users communicate with the Kubernetes master through cloud service endpoints. Depending on your cluster version, worker nodes communicate with the Kubernetes master through cloud service endpoints or VPC virtual private endpoints.

Before you create a cluster, you must enable your account to use service endpoints. To enable service endpoints, run ibmcloud account update --service-endpoint-enable true.

In VPC clusters in Red Hat OpenShift on IBM Cloud, you can't disable the private cloud service endpoint or set up a cluster with the public cloud service endpoint only.

Your VPC cluster is created with both a public and a private cloud service endpoint by default. To create clusters with worker nodes that are connected to the private network only, you must enable only the private service endpoint during cluster creation. Do not enable the public service endpoint. For example, to create a VPC cluster with only a private cloud service endpoint in the CLI, include the --disable-public-service-endpoint option. If you include this option, your cluster is created with routers and Ingress controllers that expose your apps on the private network only by default. If you later want to expose apps to a public network, you must manually create public routers and Ingress controllers.

Worker-to-master communication in VPC clusters

Worker node communication to the Kubernetes master is established differently based on your cluster version.

• Worker node communication to the Kubernetes master is established over the VPC virtual private endpoint (VPE). If the public cloud service endpoint is also enabled, worker-to-master traffic is established half over the public endpoint and half over the VPE for protection from potential outages of the public or private network.

To secure communication over public and private cloud service endpoints or VPEs, Red Hat OpenShift on IBM Cloud automatically sets up a Konnectivity connection between the Kubernetes master and worker nodes when the cluster is created. Worker nodes securely talk to the master through TLS certificates, and the master talks to workers through the Konnectivity connection.

User-to-master communication in VPC clusters

You can allow authorized cluster users to communicate with the Kubernetes master by enabling the public and private cloud service endpoints, or the private cloud service endpoint only.

• Public and private cloud service endpoints: By default, all calls to the master that are initiated by authorized cluster users are routed through the public cloud service endpoint. If authorized cluster users are in your VPC network or are connected through a VPC VPN connection, the master is privately accessible through the private cloud service endpoint.
• Private cloud service endpoint only: To access the master through the private cloud service endpoint, authorized cluster users must either be in your VPC network or connected through a VPC VPN connection.

You can secure network access to your cluster's service endpoints using context based restrictions. Only authorized requests to your cluster master that originate from subnets in the allowlist are permitted through the cluster's service endpoints. Use of context based restrictions helps prevent unauthorized scanning activities. For more information, see Using context based restrictions.

Worker communication to other services or networks

Allow your worker nodes to securely communicate with other IBM Cloud services, on-premises networks, other VPCs, and IBM Cloud classic infrastructure resources.

Communication with other IBM Cloud services over the private or public network

Your worker nodes can automatically and securely communicate with other IBM Cloud services that support private cloud service endpoints, such as IBM Cloud® Container Registry, over the private network. If an IBM Cloud service does not support private cloud service endpoints, worker nodes can securely communicate with the services over the public network through the subnet's public gateway.

Note that if you use access control lists (ACLs) for your VPC subnets, you must create inbound or outbound rules to allow your worker nodes to communicate with these services.

Communication with resources in on-premises data centers

To connect your cluster with your on-premises data center, you can use the IBM Cloud® Virtual Private Cloud VPN or IBM Cloud® Direct Link.

• To get started with the Virtual Private Cloud VPN, see Configure an on-prem VPN gateway and Create a VPN gateway in your VPC, and create the connection between the VPC VPN gateway and your local VPN gateway. If you have a multizone cluster, you must create a VPC gateway on a subnet in each zone where you have worker nodes.
• To get started with Direct Link, see Ordering IBM Cloud Direct Link Dedicated. In step 8, you can create a network connection to your VPC to be attached to the Direct Link gateway.

If you plan to connect your cluster to on-premises networks, check out the following helpful information:

• You might have subnet conflicts with the IBM-provided default 172.30.0.0/16 range for pods and 172.21.0.0/16 range for services. You can avoid subnet conflicts when you create a cluster from the CLI by specifying a custom subnet CIDR for pods in the --pod-subnet option and a custom subnet CIDR for services in the --service-subnet option.
• If your VPN solution preserves the source IP addresses of requests, you can create custom static routes to ensure that your worker nodes can route responses from your cluster back to your on-premises network.
• Note that the 172.16.0.0/16, 172.18.0.0/16, 172.19.0.0/16, and 172.20.0.0/16 subnet ranges are prohibited because they are reserved for Red Hat OpenShift on IBM Cloud control plane functionality.

External communication to apps that run on worker nodes

Allow private or public traffic requests from outside the cluster to your apps that run on worker nodes.

Private traffic to cluster apps

When you deploy an app in your cluster, you might want to make the app accessible to only users and services that are on the same private network as your cluster. Private load balancing is ideal for making your app available to requests from outside the cluster without exposing the app to the general public. You can also use private load balancing to test access, request routing, and other configurations for your app before you later expose your app to the public with public network services.

To allow private network traffic requests from outside the cluster to your apps, you can use private Kubernetes networking services, such as creating LoadBalancer services. For example, when you create a Kubernetes LoadBalancer service in your cluster, a load balancer for VPC is automatically created in your VPC outside of your cluster. The VPC load balancer is multizonal and routes requests for your app through the private NodePorts that are automatically opened on your worker nodes. For VPC ALBs, a security group, named in the form of kube-<vpc-id>, is automatically attached to your ALB instance.

The kube-<vpc-id> security group that is attached to your VPC ALB is the same security group used for the private VPE gateway that communicates with the Kubernetes master. Do not modify this security group, as doing so might result in disruptions to the network connectivity between the cluster and the Kubernetes master. Instead, you can remove the security group from the VPC ALB and replace it with a security group that you create and manage.

For more information, see Planning private external load balancing.

Public traffic to cluster apps

To make your apps accessible from the public internet, you can use public networking services. Even though your worker nodes are connected to private VPC subnets only, the VPC load balancer that is created for public networking services can route public requests to your app on the private network by providing your app a public URL. When an app is publicly exposed, anyone that has the public URL can send a request to your app.

You can use public Kubernetes networking services, such as creating LoadBalancer services. For example, when you create a Kubernetes LoadBalancer service in your cluster, a load balancer for VPC is automatically created in your VPC outside of your cluster. The VPC load balancer is multizonal and routes requests for your app through the private NodePorts that are automatically opened on your worker nodes. For VPC ALBs, a security group, named in the form of kube-<vpc-id>, is automatically attached to your ALB instance.

The kube-<vpc-id> security group that is attached to your VPC ALB is the same security group used for the private VPE gateway that communicates with the Kubernetes master. Do not modify this security group, as doing so might result in disruptions to the network connectivity between the cluster and the Kubernetes master. Instead, you can remove the security group from the VPC ALB and replace it with a security group that you create and manage.

Example scenarios for VPC cluster network setups

Now that you understand the basics of cluster networking, check out some example scenarios in which various VPC cluster network setups can meet your workload needs.

Scenario: Run internet-facing app workloads in a VPC cluster

In this scenario, you run workloads in a VPC cluster that are accessible to requests from the Internet. Public access is controlled by security groups so that end users can access your apps while unwanted public requests to your apps are denied. Additionally, your workers have automatic access to any IBM Cloud services that support private cloud service endpoints.

Extend your on-premises data center to a VPC cluster

In this scenario, you run workloads in a VPC cluster. However, you want these workloads to be accessible only to services, databases, or other resources in your private networks in an on-premises data center. Your cluster workloads might need to access a few other IBM Cloud services that support communication over the private network.

Next steps

Test your knowledge with a quiz.

To continue the planning process, learn about protecting sensitive information in your cluster by making decisions about the level of encryption you must configure. If you're ready to get started setting up networking, move on to Understanding Secure by Default Cluster VPC Networking.