MachinePool observedGeneration is updated without changing conditions on upgrades

[dkoshkin]

What steps did you take and what happened?

Create an AKS cluster with machinepools with CAPZ then upgrade the spec.version in the MachinePool.

What did you expect to happen?

We rely on status.observedGeneration and different status fields and conditions to determine if the MachinePool is still being upgraded. This works great with MachineDeployments with lots of different infra providers, but for MachinePools there is no clear signal when the upgrade has started.

I've captured the MachinePool objects during an upgrade:

1. spec.version was upgraded to v1.28.3 but observedGeneration is still 2, this is expected the controllers haven't acted yet.

[2024-01-16 15:54:09] ---
apiVersion: cluster.x-k8s.io/v1beta1
kind: MachinePool
metadata:
  creationTimestamp: "2024-01-16T23:45:43Z"
  finalizers:
  - machinepool.cluster.x-k8s.io
  generation: 3
  name: dkoshkin-az-upgrade-8
  namespace: default
  resourceVersion: "4509"
  uid: 612e7d87-b0d2-4da9-8174-9ca890642246
spec:
  template:
    spec:
      version: v1.28.3
status:
  availableReplicas: 3
  bootstrapReady: true
  conditions:
  - lastTransitionTime: "2024-01-16T23:51:06Z"
    status: "True"
    type: Ready
  - lastTransitionTime: "2024-01-16T23:45:43Z"
    status: "True"
    type: BootstrapReady
  - lastTransitionTime: "2024-01-16T23:51:06Z"
    status: "True"
    type: InfrastructureReady
  - lastTransitionTime: "2024-01-16T23:45:43Z"
    status: "True"
    type: ReplicasReady
  infrastructureReady: true
  observedGeneration: 2
  phase: Running
  readyReplicas: 3
  replicas: 3

The controller picks up the spec change and reconciles it, updating observedGeneration to 3 which matches generation.
This is where I would expect to some status change that the spec is outdated and will be upgraded.

[2024-01-16 15:54:09] ---
apiVersion: cluster.x-k8s.io/v1beta1
kind: MachinePool
metadata:
  creationTimestamp: "2024-01-16T23:45:43Z"
  finalizers:
  - machinepool.cluster.x-k8s.io
  generation: 3
  name: dkoshkin-az-upgrade-8
  namespace: default
  resourceVersion: "4510"
  uid: 612e7d87-b0d2-4da9-8174-9ca890642246
spec:
  template:
    spec:
      version: v1.28.3
status:
  availableReplicas: 3
  bootstrapReady: true
  conditions:
  - lastTransitionTime: "2024-01-16T23:51:06Z"
    status: "True"
    type: Ready
  - lastTransitionTime: "2024-01-16T23:45:43Z"
    status: "True"
    type: BootstrapReady
  - lastTransitionTime: "2024-01-16T23:51:06Z"
    status: "True"
    type: InfrastructureReady
  - lastTransitionTime: "2024-01-16T23:45:43Z"
    status: "True"
    type: ReplicasReady
  infrastructureReady: true
  observedGeneration: 3
  phase: Running
  readyReplicas: 3
  replicas: 3

Then about 10 seconds later we get a status change with Ready and InfrastructureReady conditions changing to False. By this point our wait code has exited since it checks for observedGeneration==generation and the Ready condition.

[2024-01-16 15:54:21]  ---
apiVersion: cluster.x-k8s.io/v1beta1
kind: MachinePool
metadata:
  creationTimestamp: "2024-01-16T23:45:43Z"
  finalizers:
  - machinepool.cluster.x-k8s.io
  generation: 3
  name: dkoshkin-az-upgrade-8
  namespace: default
  resourceVersion: "4574"
  uid: 612e7d87-b0d2-4da9-8174-9ca890642246
spec:
  template:
    spec:
      version: v1.28.3
status:
  availableReplicas: 3
  bootstrapReady: true
  conditions:
  - lastTransitionTime: "2024-01-16T23:54:21Z"
    message: agentpools creating or updating
    reason: Creating
    severity: Info
    status: "False"
    type: Ready
  - lastTransitionTime: "2024-01-16T23:45:43Z"
    status: "True"
    type: BootstrapReady
  - lastTransitionTime: "2024-01-16T23:54:21Z"
    message: agentpools creating or updating
    reason: Creating
    severity: Info
    status: "False"
    type: InfrastructureReady
  - lastTransitionTime: "2024-01-16T23:45:43Z"
    status: "True"
    type: ReplicasReady
  infrastructureReady: true
  observedGeneration: 3
  phase: Running
  readyReplicas: 3
  replicas: 3

Cluster API version

v1.5.3

Kubernetes version

No response

Anything else you would like to add?

We would like to avoid solving this and not have "sleeps" to wait for changes to happen (or not happen) and instead would like to use the status.

I'm looking for some guidance

1. Confirming that this is a bug
2. and any pointers if this is already handled in the topology reconcilers, as I believe there would be a similar need to know when a MachinePool is outdated and is going to be upgraded

Label(s) to be applied

/kind bug
One or more /area label. See https://github.com/kubernetes-sigs/cluster-api/labels?q=area for the list of labels.

[k8s-ci-robot]

This issue is currently awaiting triage.

If CAPI contributors determines this is a relevant issue, they will accept it by applying the triage/accepted label and provide further guidance.

The triage/accepted label can be added by org members by writing /triage accepted in a comment.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[fabriziopandini]

@jackfrancis @willie-yao could you kindly triage this issue?

[willie-yao]

/triage accepted
Will take a look and try to reproduce!

[sbueringer]

Cluster topology controller only waits for Kubernetes upgrades. We check if all Machines have been upgraded via versions on the corresponding Nodes.

[chrischdi]

/retitle MachinePool observedGeneration is updated without changing conditions on upgrades

[fabriziopandini]

/priority important-longterm

[fabriziopandini]

@willie-yao @mboersma @Jont828 PTAL

/remove-triage accepted