Add preflight role — disk, load, reboot, change freeze checks
This commit is contained in:
@@ -1,2 +1,5 @@
|
|||||||
---
|
---
|
||||||
# preflight default variables
|
min_free_disk_percent: 20
|
||||||
|
max_load_multiplier: 2
|
||||||
|
change_freeze: false
|
||||||
|
host_reboot_required: false
|
||||||
|
|||||||
@@ -1,6 +1,81 @@
|
|||||||
---
|
---
|
||||||
# preflight tasks
|
- name: Check change freeze
|
||||||
# Implementation to follow
|
ansible.builtin.assert:
|
||||||
- name: Placeholder
|
that:
|
||||||
|
- change_freeze == false
|
||||||
|
fail_msg: "FROZEN: {{ inventory_hostname }} is under a change freeze. Aborting."
|
||||||
|
success_msg: "Change freeze check passed"
|
||||||
|
|
||||||
|
- name: Check disk space on critical mountpoints
|
||||||
|
ansible.builtin.shell: |
|
||||||
|
df {{ item }} --output=pcent | tail -1 | tr -d ' %'
|
||||||
|
register: disk_usage
|
||||||
|
changed_when: false
|
||||||
|
loop:
|
||||||
|
- /
|
||||||
|
- /var
|
||||||
|
- /boot
|
||||||
|
ignore_errors: true
|
||||||
|
|
||||||
|
- name: Assert disk space is sufficient
|
||||||
|
ansible.builtin.assert:
|
||||||
|
that:
|
||||||
|
- item.stdout | int < (100 - min_free_disk_percent)
|
||||||
|
fail_msg: "DISK: {{ item.item }} is {{ item.stdout }}% full — minimum {{ min_free_disk_percent }}% free required"
|
||||||
|
success_msg: "Disk OK: {{ item.item }} is {{ item.stdout }}% used"
|
||||||
|
loop: "{{ disk_usage.results }}"
|
||||||
|
when: item.rc == 0
|
||||||
|
|
||||||
|
- name: Get load average
|
||||||
|
ansible.builtin.shell: |
|
||||||
|
cat /proc/loadavg | awk '{print $3}'
|
||||||
|
register: load_avg
|
||||||
|
changed_when: false
|
||||||
|
|
||||||
|
- name: Get CPU count
|
||||||
|
ansible.builtin.shell: |
|
||||||
|
nproc
|
||||||
|
register: cpu_count
|
||||||
|
changed_when: false
|
||||||
|
|
||||||
|
- name: Assert load average is acceptable
|
||||||
|
ansible.builtin.assert:
|
||||||
|
that:
|
||||||
|
- load_avg.stdout | float < (cpu_count.stdout | int * max_load_multiplier)
|
||||||
|
fail_msg: "LOAD: 15min load average {{ load_avg.stdout }} exceeds threshold ({{ cpu_count.stdout }} CPUs x {{ max_load_multiplier }})"
|
||||||
|
success_msg: "Load average OK: {{ load_avg.stdout }} ({{ cpu_count.stdout }} CPUs)"
|
||||||
|
|
||||||
|
- name: Check for pending reboot (Debian/Ubuntu)
|
||||||
|
ansible.builtin.stat:
|
||||||
|
path: /var/run/reboot-required
|
||||||
|
register: reboot_required
|
||||||
|
when: ansible_os_family == "Debian"
|
||||||
|
|
||||||
|
- name: Warn if reboot is pending
|
||||||
ansible.builtin.debug:
|
ansible.builtin.debug:
|
||||||
msg: "preflight role - tasks to be implemented"
|
msg: "WARNING: {{ inventory_hostname }} has a pending reboot. Patching will proceed but reboot required afterwards."
|
||||||
|
when:
|
||||||
|
- ansible_os_family == "Debian"
|
||||||
|
- reboot_required.stat.exists
|
||||||
|
|
||||||
|
- name: Set reboot_required fact (Debian)
|
||||||
|
ansible.builtin.set_fact:
|
||||||
|
host_reboot_required: "{{ reboot_required.stat.exists | default(false) }}"
|
||||||
|
when: ansible_os_family == "Debian"
|
||||||
|
|
||||||
|
- name: Check for pending reboot (RHEL/CentOS)
|
||||||
|
ansible.builtin.shell: |
|
||||||
|
needs-restarting -r
|
||||||
|
register: rhel_reboot_check
|
||||||
|
changed_when: false
|
||||||
|
failed_when: false
|
||||||
|
when: ansible_os_family == "RedHat"
|
||||||
|
|
||||||
|
- name: Set reboot_required fact (RHEL)
|
||||||
|
ansible.builtin.set_fact:
|
||||||
|
host_reboot_required: "{{ rhel_reboot_check.rc == 1 }}"
|
||||||
|
when: ansible_os_family == "RedHat"
|
||||||
|
|
||||||
|
- name: Preflight complete
|
||||||
|
ansible.builtin.debug:
|
||||||
|
msg: "Preflight passed for {{ inventory_hostname }} — proceeding with maintenance"
|
||||||
|
|||||||
Reference in New Issue
Block a user