Scaling the cluster with --limit and without access_ip or ip doesn't work

[sirkrypt0]

What happened?

I have a cluster with 3 nodes, 2 control and 1 worker. I want to scale my cluster with another worker. I have only specified ansible_host in my inventory and neither access_ip nor ip.
I tried to run the scale playbook with --limit to scale my cluster, but it failed with:

TASK [kubernetes_sigs.kubespray.kubernetes/node : Nginx-proxy | Write nginx-proxy configuration] ***
An exception occurred during task execution. To see the full traceback, use -vvv. The error was: ansible.errors.AnsibleUndefinedVariable: 'dict object' has no attribute 'k8s-02'. 'dict object' has no attribute 'k8s-02'
fatal: [k8s-04]: FAILED! => {"changed": false, "msg": "AnsibleUndefinedVariable: 'dict object' has no attribute 'k8s-02'. 'dict object' has no attribute 'k8s-02'"}

The reason for that to happen is that the nginx loadbalancer template iterates over the groups['kube_control_plane'] and tries to set the upstream for all control plane nodes:

kubespray/roles/kubernetes/node/templates/loadbalancer/nginx.conf.j2

Lines 16 to 18 in f9ebd45

	{% for host in groups['kube_control_plane'] -%}
	server {{ hostvars[host]['access_ip'] | default(hostvars[host]['ip'] | default(fallback_ips[host])) }}:{{ kube_apiserver_port }};
	{% endfor -%}

If no access_ip or ip are defined, as in my case, it uses the fallback_ips. As the error message indicates, the second control node k8s-02 doesn't have a fallback_ip.

I think that this issue was only recently introduced in #11370 which restricted the gathering of the fallback_ip to only the first control node when --limit is specified.

What did you expect to happen?

Scaling should work, just as creating the cluster did. Code-wise, a fallback_ip for all nodes required for the playbook to work with --limit should be gathered.

How can we reproduce it (as minimally and precisely as possible)?

Create a cluster using an inventory like the following. Things to note:

• Only ansible_host, no access_ip nor ip
• Two control nodes

all:
  hosts:
    k8s-01:
      ansible_host: 10.0.0.11
    k8s-02:
      ansible_host: 10.0.0.12
    k8s-03:
      ansible_host: 10.0.0.13
    # k8s-04:
    #   ansible_host: 10.0.0.14
  children:
    kube_control_plane:
      hosts:
        k8s-01:
        k8s-02:
    kube_node:
      hosts:
        k8s-01:
        k8s-02:
        k8s-03:
        # k8s-04:
    etcd:
      hosts:
        k8s-01:
        k8s-02:
        k8s-03:
    k8s_cluster:
      children:
        kube_control_plane:
        kube_node:

Then, uncomment the 4th node and run the scale playbook with --limit k8s-04.

OS

Linux 6.6.47-1-MANJARO x86_64
NAME="Manjaro Linux"
PRETTY_NAME="Manjaro Linux"
ID=manjaro
ID_LIKE=arch
BUILD_ID=rolling

Version of Ansible

ansible [core 2.16.11]

Version of Python

3.12.5

Version of Kubespray (commit)

v2.26.0

Network plugin used

calico

Full inventory with variables

See sample inventory above. I use all the defaults from the sample inventory.

Command used to invoke ansible

ansible-playbook --become --ask-become-pass -i inventory/hosts.yaml kubernetes_sigs.kubespray.scale --limit=k8s-04

Output of ansible run

TASK [kubernetes_sigs.kubespray.kubernetes/node : Nginx-proxy | Write nginx-proxy configuration] ***
An exception occurred during task execution. To see the full traceback, use -vvv. The error was: ansible.errors.AnsibleUndefinedVariable: 'dict object' has no attribute 'k8s-02'. 'dict object' has no attribute 'k8s-02'
fatal: [k8s-04]: FAILED! => {"changed": false, "msg": "AnsibleUndefinedVariable: 'dict object' has no attribute 'k8s-02'. 'dict object' has no attribute 'k8s-02'"}

Anything else we need to know

No response