What would you like to be added (User Story)?

As a production user of CAPI, I'm not comfortable working with "experimental" parts of the product, but I would like to take advantage of the native VM-grouping construct at my cloud provider.

[Stefan] We should look into the following issues before graduating MachinePools

Flaky tests with MachinePools:

• Timed out after 180.001s. waiting for cluster deletion timed out #11162

Fundamental issues / questions about the MachinePool API / contract:

• We need maintainers for the MachinePool feature (see some details below in this comment)
• We should audit the MachinePool CRD (towards v1beta2), e.g. "MachinePool has two failureDomain fields: .spec.failureDomains and .spec.template.spec.failureDomain"
  • MachinePools are currently reusing the MachineSpec struct which is also used in MD / MS. This probably has to change.
• MachinePools cannot be upgraded atomically when used with managed providers #7248
• ClusterClass MachinePool implementation is probably not able to rollout BootstrapConfig changes. #10496
• BootstrapConfig should be optional for MachinePool based ClusterClass #10943
• Note: While it sounds like some of these issues are purely because of ClusterClass, the problem is rather that there is no clear standard on how to use MachinePools, so then there is no clear way on how to manage MachinePools via ClusterClass.
  • The proper way to clarify what are the supported ways to use MachinePools is to improve the contract documentation

MachinePool Machines: (not clear to me, should be double checked by folks more familiar with MP Machines)

• Metadata propagation like for MD => MS => Machine
• from the proposal: "Allow both Cluster API-driven and externally driven interactions with individual Machines in a MachinePool"
  • MachineHealthChecks for MP Machines
  • Can MachinePool Machines today be deleted via delete machine? e.g. "Cluster Autoscaler needs to delete individual Machines" (based on the proposal only some implementations support that, how would a user know which?)
• It's not entirely clear to me how much of the proposal is implemented
• Foreground deletion for Machines (see 🌱 Foreground deletion for MachineDeployments and MachineSets #11174) (not sure how this works with MachinePools today)
• In general either feature parity between MD and MP Machines or a clear documentation about the differences would be good

Work related to the status improvements for v1beta2:

• Clarify MachinePool contract (see Documentation updates in Tracking issue for Improving status in CAPI resources - Phase 1 #11105)
• Implement v1beta2 conditions for MachinePools (sub-task "MachinePool conditions" in Tracking issue for Improving status in CAPI resources - Phase 1 #11105, more details in https://github.com/kubernetes-sigs/cluster-api/blob/main/docs/proposals/20240916-improve-status-in-CAPI-resources.md#changes-to-machinepool-resource)

(Potential) Bugs:

• MachinePool observedGeneration is updated without changing conditions on upgrades #10059
• The MachinePool remains in the "ScalingUp" phase instead of transitioning to "Running" #10541
  • This one might just be in CAPA

Misc / TBD:

• uninitialized taint should always be dropped #8258
• MachinePool.Spec.ProviderIDList may not scale well #6387
• Tracking issue for v1Beta2 - MachinePools #12178
• Kubernetes version skew check for machine pools #12732
• Check all open MachinePool issues: https://github.com/kubernetes-sigs/cluster-api/issues?q=sort%3Aupdated-desc%20state%3Aopen%20label%3A%22area%2Fmachinepool%22%20is%3Aissue

Minor improvements:

• Consider using remote.NodeProviderIDIndex in MachinePool controller #8856

As far as I can tell regarding graduation - apart from the issues listed here - the following is missing:

• Move code out of exp
• Remove the feature gate

We can/should discuss which parts of the "graduation" we want to block until which issues are resolved. I"m personally fine with moving the code out of exp, but we should then probably treat removing the feature gate as the "graduation".

In general, it would be great to see some folks signing up to be maintainers of the MachinePool feature/code. To be honest, at the moment, it's mostly up to folks which are neither very familiar with MachinePools nor are they using the feature. This is not a great state if we're planning to maintain this feature going forward, which should be the minimum bar for graduation.

/kind feature

/cc @dtzar @Jont828