Highly Available Clusters
This chapter contains high level information about the types of highly available (HA) Kubernetes clusters you can deploy using Oracle CNE.
Kubernetes can be deployed with more than one replica of the control plane node. Automated failover to those replicas provides a more scalable and resilient service. This type of cluster deployment is referred to in this document as an HA cluster.
Important:
To create an HA cluster you need at least three control plane nodes and two worker nodes.
Creating an HA cluster with three control plane nodes ensures replication of configuration
data between them through the distributed key store, etcd, so an HA cluster
is resilient to a single control plane node failing without any loss of data or up time. If
more than one control plane node fails, you can restore the control plane nodes in the cluster
from a backup file to avoid data loss.
Oracle CNE implements the Kubernetes stacked
etcd topology, where etcd runs on the control plane nodes.
For more information on this topology, see the upstream documentation.
Load Balancer
An HA cluster needs a load balancer to provide high availability of the control plane nodes. A load balancer communicates with the Kubernetes API Server to maintain high availability of the control plane nodes.
You can use an external load balancer instance, or have the CLI install a load balancer on the control plane nodes.
Highly Available Cluster with External Load Balancer
When you set up an HA cluster to use an external load balancer, the load balancer is used to manage the resource availability and efficiency of control plane nodes to ensure instances of the Kubernetes API Server on control plane nodes can fail without impacting cluster availability.
To use an external load balancer, it must be set up and ready to use before you perform an HA cluster deployment. The load balancer hostname and port is entered as an option with the CLI when you create the Kubernetes cluster.
Highly Available Cluster with Internal Load Balancer
When you set up an HA cluster to use an internal load balancer, the CLI installs NGINX and Keepalived on the control plane nodes as a systemd service to enable the deployment of the load balancer. The internal load balancer configures the native active-active high availability solution for the Kubernetes API Server.
NGINX improves the resource availability and efficiency of control plane nodes to ensure instances of the Kubernetes API Server on control plane nodes can fail without impacting cluster availability.
If you use the internal load balancer, you must set aside an IP address on the control plane network to use as a virtual IP address. The Keepalived instance makes sure the virtual IP address is always reachable by monitoring the health of other control plane nodes and appropriating the IP address if a node fails. Keepalived is used to fail over automatically to a standby control plane node if problems occur.