Proposal for Node Lifecycle WG

[Jonathan Innis]

Hey everyone 👋

SIG Autoscaling and Karpenter have been involved in Working Group discussions around Karpenter’s proposal to join the SIG (see https://github.com/kubernetes/org/issues/4258). Part of that discussion has involved thinking around how parts of Cluster Autoscaler’s (CAS) behaviors and APIs have parallels in Karpenter, prompting a discussion around API alignment between the two projects.

This has surfaced a separate question, which we are hoping to answer with SIG Arch’s help: What should this new namespace be? Karpenter does perform auto-scaling behaviors on a user cluster, but it also has a lot of overlap with what we would classify as “node lifecycle behaviors” i.e. behaviors that have been delegated to the individual cloudprovider APIs with CAS but are managed through the K8s layer with Karpenter.

One of our thoughts around this was that CAS and Karpenter could align their APIs into a broader namespace that deals with node lifecycle behaviors in general e.g. node-lifecycle.kubernetes.io , though we recognize that this has broader scope than what SIG Autoscaling can handle alone. In particular, we think we can make some progress in resolving some of the gaps outlined here: https://github.com/kubernetes/kubernetes/issues/115139.

We’re planning to attend the SIG Arch Working Group session on Oct. 5 to start some discussion around this, with the intent to propose a working group session across SIGs that would want to be involved here e.g. SIG Arch, SIG Autoscaling, SIG Node, SIG Cluster Lifecycle, etc.

Best,
Jonathan