The first recipe on Amazon EC2 showed you how to provision multiple copies of master nodes that can survive a master node failure as well as single AZ outages. Although it is similar to what you get with the second recipe on Amazon EKS with Multi-AZ support, clusters on EC2 give you higher flexibility. When you run Amazon EKS instead, it runs a single-tenant Kubernetes control plane for each cluster, and the control plane consists of at least two API server nodes and three etcd nodes that run across three AZs within a region.

Let's take a look at the cluster options we used in step 7 with the kops create cluster command:

• --node-count=3 sets the number of nodes to create. In our example, this is 6. This configuration will deploy two nodes per zone defined with--zones=us-east-1a,us-east-1b,us-east-1c, with a total of three master nodes and six worker nodes. 
• --node-size and --master-size set the instance size for the worker and master nodes. In our example, t2.medium is used for worker nodes and t2.large is used for master nodes. For larger clusters, t2.large is recommended for a worker. 
• --zones and --master-zones set the zones that the cluster will run in. In our example, we have used three zones called us-east-1a, us-east-1b, and us-east-1c.

For additional zone information, check the AWS Global Infrastructure link in the See also section.

When deploying multi-master clusters, an odd number of master instances should be created. Also, remember that Kubernetes relies on etcd, a distributed key/value store. etcd quorum requires more than 51% of the nodes to be available at any time. Therefore, with three master nodes, our control plane can only survive a single master node or AZ outages. If you need to handle more than that, you need to consider increasing the number of master instances.