pluto
b9be45dc71
Complete IaC covering: - 14 roles: common, docker, traefik, vaultwarden, monitoring, media, matrix, home_assistant, frigate, adguard, gitea, semaphore, openclaw, security, backup - Inventory for all hosts (Peggy, Bart, Yoda, NAS, Firewalla) - Stack playbooks: core-infrastructure, monitoring, media, communication, automation, smarthome, security-hardening - Maintenance playbooks: update-containers, backup, health-check - Production Docker Compose templates with Traefik labels - Security hardening (sysctl, auditd, UFW, fail2ban) - Automated backups with retention policy
88 lines
2.5 KiB
YAML
88 lines
2.5 KiB
YAML
---
|
|
- name: Health check all services
|
|
hosts: docker_hosts
|
|
become: true
|
|
tasks:
|
|
- name: Check Docker daemon is running
|
|
ansible.builtin.systemd:
|
|
name: docker
|
|
state: started
|
|
check_mode: true
|
|
register: docker_status
|
|
tags: [health]
|
|
|
|
- name: Get running containers
|
|
community.docker.docker_host_info:
|
|
containers: true
|
|
register: docker_info
|
|
tags: [health]
|
|
|
|
- name: Display container count
|
|
ansible.builtin.debug:
|
|
msg: "{{ inventory_hostname }}: {{ docker_info.containers | length }} containers running"
|
|
tags: [health]
|
|
|
|
- name: Check for unhealthy containers
|
|
ansible.builtin.shell: |
|
|
docker ps --filter health=unhealthy --format '{{ '{{' }}.Names{{ '}}' }}'
|
|
register: unhealthy
|
|
changed_when: false
|
|
tags: [health]
|
|
|
|
- name: Report unhealthy containers
|
|
ansible.builtin.debug:
|
|
msg: "WARNING: Unhealthy containers on {{ inventory_hostname }}: {{ unhealthy.stdout_lines }}"
|
|
when: unhealthy.stdout | length > 0
|
|
tags: [health]
|
|
|
|
- name: Check disk usage
|
|
ansible.builtin.shell: |
|
|
df -h / /data | tail -n +2 | awk '{print $1, $5, $6}'
|
|
register: disk_usage
|
|
changed_when: false
|
|
tags: [health]
|
|
|
|
- name: Display disk usage
|
|
ansible.builtin.debug:
|
|
msg: "{{ disk_usage.stdout_lines }}"
|
|
tags: [health]
|
|
|
|
- name: Check memory usage
|
|
ansible.builtin.shell: |
|
|
free -h | grep Mem | awk '{print "Total:", $2, "Used:", $3, "Free:", $4}'
|
|
register: mem_usage
|
|
changed_when: false
|
|
tags: [health]
|
|
|
|
- name: Display memory usage
|
|
ansible.builtin.debug:
|
|
msg: "{{ mem_usage.stdout }}"
|
|
tags: [health]
|
|
|
|
- name: Check key services via HTTP
|
|
ansible.builtin.uri:
|
|
url: "https://{{ item }}.{{ domain }}"
|
|
validate_certs: false
|
|
status_code: [200, 301, 302, 303, 401, 403]
|
|
timeout: 10
|
|
loop:
|
|
- traefik
|
|
- vault
|
|
- grafana
|
|
- jellyfin
|
|
- gitea
|
|
- ha
|
|
register: http_checks
|
|
ignore_errors: true
|
|
tags: [health]
|
|
when: inventory_hostname == 'peggy'
|
|
|
|
- name: Report HTTP check results
|
|
ansible.builtin.debug:
|
|
msg: "{{ item.item }}.{{ domain }}: {{ item.status | default('FAILED') }}"
|
|
loop: "{{ http_checks.results | default([]) }}"
|
|
loop_control:
|
|
label: "{{ item.item | default('unknown') }}"
|
|
when: inventory_hostname == 'peggy'
|
|
tags: [health]
|