c12s-kubespray/roles/kubernetes/master/tasks/kubeadm-upgrade.yml
Rick Haan 5cba8b1614
Wait for kube-apiserver availability before starting upgrade (#6243)
* Wait for kube-apiserver availability before starting upgrade

I am experiencing a timing issue when upgrading from kubespray
2.11.0(k8s 1.15.3) to kubespray 2.12.6(k8s 1.16.9). The certificates get
replaced in `kubeadm-secondary-legacy.yml` and kube-apiserver notices a mismatch
(for a fraction of a second) between `apiserver.crt` and `apiserver.key`
which causes it to restart. And sometimes ( ~ 1 out of 5 upgrades) the
kube-apiserver isn't back on time for the start of the upgrade task. It
fails when kubeadm checks with the kube-apiserver to start the upgrade. The
kube-apiserver returns a `connect: connection refused`. I have created this
small task to check the availability of the kube-apiserver before starting the
upgrade, so that the upgrade will run without an issue.

Signed-off-by: Rick Haan <rickhaan94@gmail.com>

* Fix markdownlint

* Remove old CI

Co-authored-by: Maxime Guyot <maxime@root314.com>
2020-06-08 07:25:53 -07:00

67 lines
2.2 KiB
YAML

---
- name: "kubeadm | Wait for master kube-apiserver"
uri:
url: "https://{{ kube_apiserver_access_address }}:{{ kube_apiserver_port }}/version"
status_code: 200
validate_certs: false
register: kube_api_server_available
until: kube_api_server_available.status == 200
retries: 180
delay: 1
when: inventory_hostname == groups['kube-master']
- name: kubeadm | Upgrade first master
command: >-
timeout -k 600s 600s
{{ bin_dir }}/kubeadm
upgrade apply -y {{ kube_version }}
--config={{ kube_config_dir }}/kubeadm-config.yaml
--ignore-preflight-errors=all
--allow-experimental-upgrades
--allow-release-candidate-upgrades
--etcd-upgrade=false
--force
register: kubeadm_upgrade
# Retry is because upload config sometimes fails
retries: 3
when: inventory_hostname == groups['kube-master']|first
failed_when: kubeadm_upgrade.rc != 0 and "field is immutable" not in kubeadm_upgrade.stderr
notify: Master | restart kubelet
# FIXME: https://github.com/kubernetes/kubeadm/issues/1498 remove stdout_lines
# check after issue is fixed
- name: kubeadm | Upgrade other masters
command: >-
timeout -k 600s 600s
{{ bin_dir }}/kubeadm
upgrade apply -y {{ kube_version }}
--config={{ kube_config_dir }}/kubeadm-config.yaml
--ignore-preflight-errors=all
--allow-experimental-upgrades
--allow-release-candidate-upgrades
--etcd-upgrade=false
--force
register: kubeadm_upgrade
when: inventory_hostname != groups['kube-master']|first
failed_when:
- kubeadm_upgrade.rc != 0
- '"field is immutable" not in kubeadm_upgrade.stderr'
- kubeadm_upgrade.stdout_lines | length > 1
notify: Master | restart kubelet
# FIXME: https://github.com/kubernetes/kubeadm/issues/1318
- name: kubeadm | scale down coredns replicas to 0 if not using coredns dns_mode
command: >-
{{ bin_dir }}/kubectl
--kubeconfig /etc/kubernetes/admin.conf
-n kube-system
scale deployment/coredns --replicas 0
register: scale_down_coredns
retries: 6
delay: 5
until: scale_down_coredns is succeeded
run_once: yes
when:
- kubeadm_scale_down_coredns_enabled
- dns_mode not in ['coredns', 'coredns_dual']
changed_when: false