MachineDeployment status does not show the number of ReadyReplicas in a Management cluster

[DheerajSShetty]

Create a MachineDeployment in ManagementCluster with n Replica
n nodes are created.
Check status of MachineDeployment, the ReadyReplica wont be set to n.

[vincepri]

/kind bug

This issue is related to the AWS provider not setting the NodeRef field when running in a Management cluster. The nodeRef was being set before, but the logic was revisited in kubernetes-sigs/cluster-api-provider-aws#744 because Cluster API wouldn't properly delete the cluster in a management scenario.

Going to close this one here, given it's specific to the AWS provider.
/close

[k8s-ci-robot]

@vincepri: Closing this issue.

In response to this:

/kind bug

This issue is related to the AWS provider not setting the NodeRef field when running in a Management cluster. The nodeRef was being set before, but the logic was revisited in kubernetes-sigs/cluster-api-provider-aws#744 because Cluster API wouldn't properly delete the cluster in a management scenario.

Going to close this one here, given it's specific to the AWS provider.
/close

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.

[detiber]

@vincepri This would also affect other providers that do not set NodeRef when run in a management cluster (which I believe is all)

/reopen

[k8s-ci-robot]

@detiber: Reopened this issue.

In response to this:

@vincepri This would also affect other providers that do not set NodeRef when run in a management cluster (which I believe is all)

/reopen

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.

[liztio]

/assign

[liztio]

Reproduction steps:

1. Create a new management cluster
2. Use the example yaml and generate.sh scripts as per capa setup instructions
3. Wait for machines to run
4. Observe lack of ReadyRequests:

$ kubectl get machinedeployments   -o=jsonpath='{.items[0].status}'
map[observedGeneration:2 replicas:1 unavailableReplicas:1 updatedReplicas:1]%

[liztio]

I'm curious how we can make this work in the general case. The only status information reported by the machine is provider-specific:

$ kubectl get machine brew-69b776f86c-8s2tl -o json | jq '.status'
{
  "providerStatus": {
    "instanceID": "i-0d833b1e56938a3bb",
    "instanceState": "running",
    "metadata": {
      "creationTimestamp": null
    }
  }
}

If we look at the structure itself, we see that there's a Phase field, currently underpopulated. But it's a free-form text field, not an enum.

There's the Conditions field, vwhich includes a ready status. This is also not currently not populated by CAPA, I don't know if other things do populated it.

I think perhaps the issue here is that the MachineSet doesn't get updated with ReadyReplicas either. If it did, it'd be trivial to propagate that up to MachineDeployment

[ncdc]

@liztio FYI @vincepri is working on a POC to handle setting remote node references. If that is approved, this should be achievable as a follow-up.

[ncdc]

/area machine

[vincepri]

/lifecycle active