Intallation fails when waiting for k8s apiserver due to variable overlapping

[unai-ttxu]

Environment:

• OS (printf "$(uname -srm)\n$(cat /etc/os-release)\n"):

Linux 4.18.0-240.10.1.el8_3.x86_64 x86_64
NAME="CentOS Linux"
VERSION="8"
ID="centos"
ID_LIKE="rhel fedora"
VERSION_ID="8"
PLATFORM_ID="platform:el8"
PRETTY_NAME="CentOS Linux 8"
ANSI_COLOR="0;31"
CPE_NAME="cpe:/o:centos:centos:8"
HOME_URL="https://centos.org/"
BUG_REPORT_URL="https://bugs.centos.org/"
CENTOS_MANTISBT_PROJECT="CentOS-8"
CENTOS_MANTISBT_PROJECT_VERSION="8"

• Version of Ansible (ansible --version):

ansible 2.10.15
  config file = /stratio/ansible/ansible.cfg
  configured module search path = ['/root/.ansible/plugins/modules', '/usr/share/ansible/plugins/modules']
  ansible python module location = /usr/lib/python3.8/site-packages/ansible
  executable location = /usr/bin/ansible
  python version = 3.8.10 (default, May  6 2021, 00:05:59) [GCC 10.2.1 20201203]

• Version of Python (python --version):

Python 3.8.10

Kubespray version (commit) (git rev-parse --short HEAD):

92f25bf267ffd3393f6caffa588169d3a44a799c -> v2.18.0

Network plugin used:

calico

Command used to invoke ansible:

ansible-playbook cluster.yml

Output of ansible run:

TASK [kubernetes/control-plane : Create kubeadm ControlPlane config] **********************************************************************************************************************************************
» Friday 07 January 2022  12:49:32 +0000 (0:00:00.103)       0:36:48.467 ****** 
skipping: [kube-control-plane-gx03l92v]

TASK [kubernetes/control-plane : Wait for k8s apiserver] **********************************************************************************************************************************************************
» Friday 07 January 2022  12:49:32 +0000 (0:00:00.038)       0:36:48.505 ****** 
fatal: [kube-control-plane-gx03l92v]: FAILED! => changed=false 
  elapsed: 200
  msg: Timeout when waiting for kube-control-plane-gx03l92v:6443

NO MORE HOSTS LEFT ************************************************************************************************************************************************************************************************

PLAY RECAP ********************************************************************************************************************************************************************************************************
kube-control-plane-gx03l92v : ok=524  changed=117  unreachable=0    failed=1    skipped=693  rescued=0    ignored=2   
kube-node-dz6jnyww         : ok=391  changed=91   unreachable=0    failed=0    skipped=471  rescued=0    ignored=1   
kube-node-e14m35zv         : ok=391  changed=91   unreachable=0    failed=0    skipped=471  rescued=0    ignored=1   
kube-node-ey04mxwj         : ok=391  changed=91   unreachable=0    failed=0    skipped=472  rescued=0    ignored=1   
kube-node-g04l3809         : ok=391  changed=91   unreachable=0    failed=0    skipped=471  rescued=0    ignored=1   
localhost                  : ok=34   changed=11   unreachable=0    failed=0    skipped=30   rescued=0    ignored=0 

Explanation:

As we can see in the Ansible output log, the following task locating in roles/kubernetes/control-plane/tasks/kubeadm-secondary.yml is failing:

- name: Wait for k8s apiserver
  wait_for:
    host: "{{ kubeadm_discovery_address.split(':')[0] }}"
    port: "{{ kubeadm_discovery_address.split(':')[1] }}"
    timeout: 180

This is because kubeadm_discovery_address is set to kube-control-plane-gx03l92v:6443 where kube-control-plane-gx03l92v is the inventory_hostname of the first and only kube_control_plane.

I think this failure is caused by the changes of commit 27ab364. This commit sets the fact first_kube_control_plane in order to ensure task delegation to working kube_control_plane instance. This is achieved by refactoring groups['kube_control_plane']|first to first_kube_control_plane.

This main goal works smoothly but the commit also refactor the use of first_kube_master to first_kube_control_plane. But this is incorrect since first_kube_master should be an address and not the inventory_hostname. Some times both are the same but we can't ensure that. So when roles/kubernetes/control-plane/tasks/define-first-kube-control.yml is included, first_kube_control_plane fact value changes from adress to inventory_hostname generating the fail.

I think one option is to refactor the first_kube_control_plane name from roles/kubespray-defaults/defaults/main.yaml to first_kube_control_plane_address in order to avoid variable overlapping.

I will open the PR to fix this ASAP.