Merge "Add new healthcheck-service-status validation"

doc/source/roles/role-healthcheck-service-status.rst

@@ -0,0 +1,6 @@
+==========================
+healthcheck-service-status
+==========================
+
+.. ansibleautoplugin::
+   :role: roles/healthcheck-service-status

playbooks/healthcheck-service-status.yaml

@@ -0,0 +1,14 @@
+---
+- hosts: undercloud, overcloud
+  vars:
+    metadata:
+      name: Healthcheck systemd services Check
+      description: >
+        Check for failed healthcheck systemd services.
+      groups:
+        - post-deployment
+    retries_number: 1
+    delay_number: 1
+    inflight_healthcheck_services: []
+  roles:
+    - healthcheck-service-status

roles/healthcheck-service-status/defaults/main.yml

@@ -0,0 +1,8 @@
+---
+# The default values for the two following variables are set up to wait up to
+# 300s for the first healthcheck to run.
+retries_number: 10
+delay_number: 30
+# Please use inflight_healtcheck_services when using this role through the
+# inflight validations.
+inflight_healthcheck_services: []

roles/healthcheck-service-status/tasks/main.yml

@@ -0,0 +1,34 @@
+---
+- name: Get the healthcheck services list enabled on node
+  shell: >
+    systemctl list-unit-files | grep "^tripleo.*healthcheck.*enabled" | awk -F'.' '{print $1}'
+  changed_when: False
+  register: healthcheck_services_list
+  when: inflight_healthcheck_services | length < 1
+
+- name: Set hc_services
+  set_fact:
+    hc_services: >
+       {%- if inflight_healthcheck_services | length > 0 -%}
+         {{ inflight_healthcheck_services }}
+       {%- else -%}
+         {{ healthcheck_services_list.stdout_lines }}
+       {%- endif -%}
+
+- name: Get systemd healthcheck service status
+  systemd:
+    name: "{{ item }}"
+  retries: "{{ retries_number|int }}"
+  delay: "{{ delay_number|int }}"
+  until:
+    - systemd_healthcheck_state.status.ExecMainPID != '0'
+    - systemd_healthcheck_state.status.ActiveState in ['inactive', 'failed']
+  ignore_errors: True
+  register: systemd_healthcheck_state
+  with_items: "{{ hc_services }}"
+
+- name: Fail if systemd healthcheck services are in failed status
+  fail:
+    msg: "Failed systemd healthcheck service detected: {{ item.item }}"
+  when: item.status.ExecMainStatus != '0'
+  loop: "{{ systemd_healthcheck_state.results }}"

roles/healthcheck-service-status/vars/main.yml

@@ -0,0 +1,7 @@
+---
+metadata:
+  name: Healthcheck systemd services Check
+  description: >
+    Check for failed healthcheck systemd services.
+  groups:
+    - post-deployment