Support Machine (and MachineTemplate) but not MachinePool

[richardcase]

Support Machine (and MachineTemplate) but not MachinePool

• Status: proposed
• Date: 2021-11-1
• Authors: @richardcase
• Deciders: @jmickey @richardcase @Callisto13 @yitsushi

Context

Cluster API¹² supports a number of resources kinds that infrastructure providers can implement to create machines for use as Kubernetes nodes:

• Machine - is a declaration for an individual machine. There is always an infrastructure provider counterpart (e.g. AWSMachine, AzureMachine) that knows how to provision the infrastructure for a machine based on a spec. Machines are immutable and so the actions required will only be create/delete.
• MachineTemplate - is a declaration for a template of a machine. It doesn't relate to a specific machines but the spec it contains will be used to create instances of Machines when using the Kubeadm control plane.
• MachinePools - are an experimental feature in CAPI and provide a way to declare a dynamic pool of machines where the number of instances can scale up and down. An infrastructure provider provides this via a specific solution to the infrastructure. So CAPA uses Auto Scale Groups (ASG) and CAPZ uses Virtual Machine Scalesets.

Decision

Cluster API Provider Microvm (CAPMVM) will not support Machine Pools initially as flintlock has no way concept of auto-scaling or machine pools.

Consequences

We will need to implement the following infrastructure kinds:

• MicrovmCluster (infrastructure counterpart to Cluster)
• MicrovmMachine (infrastructure counterpart to Machine)
• MicrovmMachineTemplate (infrastructure machine template used by Kubeadm Bootstrap Provider)

We will need to revisit this decision when flintlock supports auto-scaling / pools.

Footnotes

1. https://cluster-api.sigs.k8s.io/introduction.html ↩

2. https://github.com/kubernetes-sigs/cluster-api ↩