Check health on control-plane before acting on cluster

enix · Sep 15, 2023 · d494c9d · d494c9d
1 parent c06ee1f
commit d494c9d
Show file tree

Hide file tree

Showing 4 changed files with 31 additions and 0 deletions.
diff --git a/docs/variables.md b/docs/variables.md
@@ -12,6 +12,7 @@ For hooks where a variable-per-hook is exposed, see [hooks && plugins](hooks_and
 | apiserver_manifest                | control plane       | "/etc/kubernetes/manifests/kube-apiserver.yaml" | filename to stat for presence in the process to discover already running control-plane                                                                                                                                    |
 | cluster_config                    | control plane       | {}                                              | config to be used by kubeadm for the `kind: CluserConfiguration`                                                                                                                                                          |
 | control_plane_endpoint            | control plane       | ""  (let kubeadm default)                       | control the "controlPlaneEndpoint" entry of the cluster_config. Could also be set as part of the cluster_config. Default to nothing but ansible-kubeadm will fail if not set in case of multi-control-plane nodes cluster |
+| cp_health_check_bypass            | control_plane       | false                                           | Bypass check on control-plane health                                                                                                                                                                                      |
 | enable_kubeadm_patches            | control plane       | true                                            | Deploy patches and pass `kubeadm_patch_dir` to kubeadm so that patch are applied                                                                                                                                          |
 | kube_control_plane_cidr           | control plane       | "" (let kubeadm default)                        | CIDR (eg "192.168.99.0/24") filter addresses for `_etcd_metrics_bind_address`, `_kube_apiserver_advertise_address`, `_kube_controller_manager_bind_address`, `_kube_scheduler_bind_address`                               |
 | kube_apiserver_advertise_cidr     | control plane       | "" (let kubeadm default)                        | CIDR (eg "192.168.99.0/24") filter the advertise address to `_kube_apiserver_advertise_address` (override `kube_control_plane_cidr`)                                                                                      |

diff --git a/roles/preflight_check_cp/defaults/main.yml b/roles/preflight_check_cp/defaults/main.yml
@@ -2,6 +2,7 @@
 _config_upgrade_reasons: {}
 _failure_reasons: {}
 _upgrade_reasons: {}
+cp_health_check_bypass: false
 
 kube_version:
 default_kube_version: '1.19'
diff --git a/roles/preflight_check_cp/tasks/check_control_plane_health.yml b/roles/preflight_check_cp/tasks/check_control_plane_health.yml
@@ -0,0 +1,24 @@
+---
+- name: 'get kubeadm configmap if cluster running'
+  command: kubectl get nodes -o yaml
+  changed_when: false
+  check_mode: false
+  register: _all_nodes_yaml
+  environment:
+    KUBECONFIG: '{{ kubeconfig_admin }}'
+
+- name: 'Check control-plane health'
+  set_fact:
+    _failure_reasons: >-
+      {%- set all_nodes = (_all_nodes_yaml.stdout|from_yaml)|'items']
+                  |selectattr("metadata.labels.node-role\.kubernetes\.io/control-plane", "defined") -%}
+      {%- if all_nodes|map(attribute="status.conditions")
+                      |map("selectattr", "type", "eq", "Ready")
+                      |map("first")
+                      |rejectattr("status", "eq", "True")
+                      |list|length != 0 -%}
+          {%- set _ = _failure_reasons.dict(dict(
+              cp_health = "Some control plane are not healthy")) -%}
+          {%- endif -%}
+      {%- endif -%}
+  when: not cp_health_check_bypass|bool
diff --git a/roles/preflight_check_cp/tasks/main.yml b/roles/preflight_check_cp/tasks/main.yml
@@ -32,3 +32,8 @@
 - import_tasks: check_version.yml
 
 - import_tasks: check_control_plane_endpoint.yml
+
+- import_tasks: check_control_plane_health.yml
+  when:
+    - groups.cp_running|default([])|length > 0
+    - not cp_health_check_bypass|bool