How useful is the Kubernetes Control Plane health feature we released recently?

[sirishkr]

Looking for some feedback and discussion on this feature that we released recently.

Motivation
Spot today runs its control planes on a centralized infrastructure. The majority of users also start out with the free control plane, which we allocate limited capacity to (to control our costs). Some of these users go on to run relatively large clusters (50+ nodes) with that smallish control plane. Our goal was to provide more visibility into the API server response times from the K8s control plane.

Questions

1. Is it clear what Kubernetes control plane health is referring to?
2. Does it help you better understand how your cluster is performing?
3. Any suggestions or changes to make this more useful?
4. Any examples of other implementations from other Managed Kubernetes offerings that do this better?

[incidentally-nameless]

Slightly unrelated: it'd be useful to understand the value proposition behind the "production" control plane, given the high relative cost of the control plane vs the actual worker nodes

[sirishkr]

@incidentally-nameless there's the current implementation and upcoming features for the production control plane. I'll include both below.

Current implementation:

1. Production control plane runs multiple replicas
2. Provides higher resource allocations for each of those replicas
3. Is backed up and we have the ability to restore from backup

Upcoming roadmap:

1. Control plane that is co-located or more closely located to the infrastructure where you are running servers. For e.g. London servers being managed by a control plane running in London
2. 30-day retention for events and audit log
3. Vertical auto-scaling for resource allocation as Cloudspaces grow larger

It costs money to host and operate these control planes; we feel confident we're offering a great price point vs the value for the production control planes.

[OrangeDrangon]

I would be happy to pay for a slightly more premium experience than free but my cluster is not large enough to justify a 40/mo fixed cost. Something that is in the ~5 - 10 dollar range that ensures backups exist and maybe some slightly more performance or something would be cool.

I would also encourage communicating the limitations somewhere. The fact that free tier api servers have no backups or anything is spooky and was not aware of that.

[sirishkr]

@OrangeDrangon thank you for the feedback. We will certainly incorporate this into our planning as well as improve our documentation about the distinctions.