Sanitized mirror from private repository - 2026-03-18 10:31:50 UTC

ansible/automation/playbooks/prune_containers.yml

@@ -0,0 +1,410 @@
+---
+# Docker Cleanup and Pruning Playbook
+# Clean up unused containers, images, volumes, and networks
+# Usage: ansible-playbook playbooks/prune_containers.yml
+# Usage: ansible-playbook playbooks/prune_containers.yml -e "aggressive_cleanup=true"
+# Usage: ansible-playbook playbooks/prune_containers.yml -e "dry_run=true"
+
+- name: Docker System Cleanup and Pruning
+  hosts: "{{ host_target | default('all') }}"
+  gather_facts: yes
+  vars:
+    dry_run: "{{ dry_run | default(false) }}"
+    aggressive_cleanup: "{{ aggressive_cleanup | default(false) }}"
+    keep_images_days: "{{ keep_images_days | default(7) }}"
+    keep_volumes: "{{ keep_volumes | default(true) }}"
+    backup_before_cleanup: "{{ backup_before_cleanup | default(true) }}"
+    cleanup_logs: "{{ cleanup_logs | default(true) }}"
+    max_log_size: "{{ max_log_size | default('100m') }}"
+
+  tasks:
+    - name: Check if Docker is running
+      systemd:
+        name: docker
+      register: docker_status
+      failed_when: docker_status.status.ActiveState != "active"
+
+    - name: Create cleanup report directory
+      file:
+        path: "/tmp/docker_cleanup/{{ ansible_date_time.date }}"
+        state: directory
+        mode: '0755'
+
+    - name: Get pre-cleanup Docker system info
+      shell: |
+        echo "=== PRE-CLEANUP DOCKER SYSTEM INFO ==="
+        echo "Date: {{ ansible_date_time.iso8601 }}"
+        echo "Host: {{ inventory_hostname }}"
+        echo ""
+
+        echo "System Usage:"
+        docker system df
+        echo ""
+
+        echo "Container Count:"
+        echo "Running: $(docker ps -q | wc -l)"
+        echo "Stopped: $(docker ps -aq --filter status=exited | wc -l)"
+        echo "Total: $(docker ps -aq | wc -l)"
+        echo ""
+
+        echo "Image Count:"
+        echo "Total: $(docker images -q | wc -l)"
+        echo "Dangling: $(docker images -f dangling=true -q | wc -l)"
+        echo ""
+
+        echo "Volume Count:"
+        echo "Total: $(docker volume ls -q | wc -l)"
+        echo "Dangling: $(docker volume ls -f dangling=true -q | wc -l)"
+        echo ""
+
+        echo "Network Count:"
+        echo "Total: $(docker network ls -q | wc -l)"
+        echo "Custom: $(docker network ls --filter type=custom -q | wc -l)"
+      register: pre_cleanup_info
+      changed_when: false
+
+    - name: Display cleanup plan
+      debug:
+        msg: |
+          🧹 DOCKER CLEANUP PLAN
+          ======================
+          🖥️  Host: {{ inventory_hostname }}
+          📅 Date: {{ ansible_date_time.date }}
+          🔍 Dry Run: {{ dry_run }}
+          💪 Aggressive: {{ aggressive_cleanup }}
+          📦 Keep Images: {{ keep_images_days }} days
+          💾 Keep Volumes: {{ keep_volumes }}
+          📝 Cleanup Logs: {{ cleanup_logs }}
+
+          {{ pre_cleanup_info.stdout }}
+
+    - name: Backup container list before cleanup
+      shell: |
+        backup_file="/tmp/docker_cleanup/{{ ansible_date_time.date }}/{{ inventory_hostname }}_containers_backup.txt"
+
+        echo "=== CONTAINER BACKUP - {{ ansible_date_time.iso8601 }} ===" > "$backup_file"
+        echo "Host: {{ inventory_hostname }}" >> "$backup_file"
+        echo "" >> "$backup_file"
+
+        echo "=== RUNNING CONTAINERS ===" >> "$backup_file"
+        docker ps --format "table {{.Names}}\t{{.Image}}\t{{.Status}}\t{{.Ports}}" >> "$backup_file"
+        echo "" >> "$backup_file"
+
+        echo "=== ALL CONTAINERS ===" >> "$backup_file"
+        docker ps -a --format "table {{.Names}}\t{{.Image}}\t{{.Status}}\t{{.CreatedAt}}" >> "$backup_file"
+        echo "" >> "$backup_file"
+
+        echo "=== IMAGES ===" >> "$backup_file"
+        docker images --format "table {{.Repository}}\t{{.Tag}}\t{{.Size}}\t{{.CreatedAt}}" >> "$backup_file"
+        echo "" >> "$backup_file"
+
+        echo "=== VOLUMES ===" >> "$backup_file"
+        docker volume ls >> "$backup_file"
+        echo "" >> "$backup_file"
+
+        echo "=== NETWORKS ===" >> "$backup_file"
+        docker network ls >> "$backup_file"
+      when: backup_before_cleanup | bool
+
+    - name: Remove stopped containers
+      shell: |
+        {% if dry_run %}
+        echo "DRY RUN: Would remove stopped containers:"
+        docker ps -aq --filter status=exited
+        {% else %}
+        echo "Removing stopped containers..."
+        stopped_containers=$(docker ps -aq --filter status=exited)
+        if [ -n "$stopped_containers" ]; then
+          docker rm $stopped_containers
+          echo "✅ Removed stopped containers"
+        else
+          echo "ℹ️  No stopped containers to remove"
+        fi
+        {% endif %}
+      register: remove_stopped_containers
+
+    - name: Remove dangling images
+      shell: |
+        {% if dry_run %}
+        echo "DRY RUN: Would remove dangling images:"
+        docker images -f dangling=true -q
+        {% else %}
+        echo "Removing dangling images..."
+        dangling_images=$(docker images -f dangling=true -q)
+        if [ -n "$dangling_images" ]; then
+          docker rmi $dangling_images
+          echo "✅ Removed dangling images"
+        else
+          echo "ℹ️  No dangling images to remove"
+        fi
+        {% endif %}
+      register: remove_dangling_images
+
+    - name: Remove unused images (aggressive cleanup)
+      shell: |
+        {% if dry_run %}
+        echo "DRY RUN: Would remove unused images older than {{ keep_images_days }} days:"
+        docker images --filter "until={{ keep_images_days * 24 }}h" -q
+        {% else %}
+        echo "Removing unused images older than {{ keep_images_days }} days..."
+        old_images=$(docker images --filter "until={{ keep_images_days * 24 }}h" -q)
+        if [ -n "$old_images" ]; then
+          # Check if images are not used by any container
+          for image in $old_images; do
+            if ! docker ps -a --format "{{.Image}}" | grep -q "$image"; then
+              docker rmi "$image" 2>/dev/null && echo "Removed image: $image" || echo "Failed to remove image: $image"
+            else
+              echo "Skipping image in use: $image"
+            fi
+          done
+          echo "✅ Removed old unused images"
+        else
+          echo "ℹ️  No old images to remove"
+        fi
+        {% endif %}
+      register: remove_old_images
+      when: aggressive_cleanup | bool
+
+    - name: Remove dangling volumes
+      shell: |
+        {% if dry_run %}
+        echo "DRY RUN: Would remove dangling volumes:"
+        docker volume ls -f dangling=true -q
+        {% else %}
+        {% if not keep_volumes %}
+        echo "Removing dangling volumes..."
+        dangling_volumes=$(docker volume ls -f dangling=true -q)
+        if [ -n "$dangling_volumes" ]; then
+          docker volume rm $dangling_volumes
+          echo "✅ Removed dangling volumes"
+        else
+          echo "ℹ️  No dangling volumes to remove"
+        fi
+        {% else %}
+        echo "ℹ️  Volume cleanup skipped (keep_volumes=true)"
+        {% endif %}
+        {% endif %}
+      register: remove_dangling_volumes
+
+    - name: Remove unused networks
+      shell: |
+        {% if dry_run %}
+        echo "DRY RUN: Would remove unused networks:"
+        docker network ls --filter type=custom -q
+        {% else %}
+        echo "Removing unused networks..."
+        docker network prune -f
+        echo "✅ Removed unused networks"
+        {% endif %}
+      register: remove_unused_networks
+
+    - name: Clean up container logs
+      shell: |
+        {% if dry_run %}
+        echo "DRY RUN: Would clean up container logs larger than {{ max_log_size }}"
+        find /var/lib/docker/containers -name "*-json.log" -size +{{ max_log_size }} 2>/dev/null | wc -l
+        {% else %}
+        {% if cleanup_logs %}
+        echo "Cleaning up large container logs (>{{ max_log_size }})..."
+
+        log_count=0
+        total_size_before=0
+        total_size_after=0
+
+        for log_file in $(find /var/lib/docker/containers -name "*-json.log" -size +{{ max_log_size }} 2>/dev/null); do
+          if [ -f "$log_file" ]; then
+            size_before=$(stat -f%z "$log_file" 2>/dev/null || stat -c%s "$log_file" 2>/dev/null || echo 0)
+            total_size_before=$((total_size_before + size_before))
+
+            # Truncate log file to last 1000 lines
+            tail -1000 "$log_file" > "${log_file}.tmp" && mv "${log_file}.tmp" "$log_file"
+
+            size_after=$(stat -f%z "$log_file" 2>/dev/null || stat -c%s "$log_file" 2>/dev/null || echo 0)
+            total_size_after=$((total_size_after + size_after))
+
+            log_count=$((log_count + 1))
+          fi
+        done
+
+        if [ $log_count -gt 0 ]; then
+          saved_bytes=$((total_size_before - total_size_after))
+          echo "✅ Cleaned $log_count log files, saved $(echo $saved_bytes | numfmt --to=iec) bytes"
+        else
+          echo "ℹ️  No large log files to clean"
+        fi
+        {% else %}
+        echo "ℹ️  Log cleanup skipped (cleanup_logs=false)"
+        {% endif %}
+        {% endif %}
+      register: cleanup_logs_result
+      when: cleanup_logs | bool
+
+    - name: Run Docker system prune
+      shell: |
+        {% if dry_run %}
+        echo "DRY RUN: Would run docker system prune"
+        docker system df
+        {% else %}
+        echo "Running Docker system prune..."
+        {% if aggressive_cleanup %}
+        docker system prune -af --volumes
+        {% else %}
+        docker system prune -f
+        {% endif %}
+        echo "✅ Docker system prune complete"
+        {% endif %}
+      register: system_prune_result
+
+    - name: Get post-cleanup Docker system info
+      shell: |
+        echo "=== POST-CLEANUP DOCKER SYSTEM INFO ==="
+        echo "Date: {{ ansible_date_time.iso8601 }}"
+        echo "Host: {{ inventory_hostname }}"
+        echo ""
+
+        echo "System Usage:"
+        docker system df
+        echo ""
+
+        echo "Container Count:"
+        echo "Running: $(docker ps -q | wc -l)"
+        echo "Stopped: $(docker ps -aq --filter status=exited | wc -l)"
+        echo "Total: $(docker ps -aq | wc -l)"
+        echo ""
+
+        echo "Image Count:"
+        echo "Total: $(docker images -q | wc -l)"
+        echo "Dangling: $(docker images -f dangling=true -q | wc -l)"
+        echo ""
+
+        echo "Volume Count:"
+        echo "Total: $(docker volume ls -q | wc -l)"
+        echo "Dangling: $(docker volume ls -f dangling=true -q | wc -l)"
+        echo ""
+
+        echo "Network Count:"
+        echo "Total: $(docker network ls -q | wc -l)"
+        echo "Custom: $(docker network ls --filter type=custom -q | wc -l)"
+      register: post_cleanup_info
+      changed_when: false
+
+    - name: Generate cleanup report
+      copy:
+        content: |
+          🧹 DOCKER CLEANUP REPORT - {{ inventory_hostname }}
+          ===============================================
+
+          📅 Cleanup Date: {{ ansible_date_time.iso8601 }}
+          🖥️  Host: {{ inventory_hostname }}
+          🔍 Dry Run: {{ dry_run }}
+          💪 Aggressive Mode: {{ aggressive_cleanup }}
+          📦 Image Retention: {{ keep_images_days }} days
+          💾 Keep Volumes: {{ keep_volumes }}
+          📝 Log Cleanup: {{ cleanup_logs }}
+
+          📊 BEFORE CLEANUP:
+          {{ pre_cleanup_info.stdout }}
+
+          🔧 CLEANUP ACTIONS:
+
+          🗑️  Stopped Containers:
+          {{ remove_stopped_containers.stdout }}
+
+          🖼️  Dangling Images:
+          {{ remove_dangling_images.stdout }}
+
+          {% if aggressive_cleanup %}
+          📦 Old Images:
+          {{ remove_old_images.stdout }}
+          {% endif %}
+
+          💾 Dangling Volumes:
+          {{ remove_dangling_volumes.stdout }}
+
+          🌐 Unused Networks:
+          {{ remove_unused_networks.stdout }}
+
+          {% if cleanup_logs %}
+          📝 Container Logs:
+          {{ cleanup_logs_result.stdout }}
+          {% endif %}
+
+          🧹 System Prune:
+          {{ system_prune_result.stdout }}
+
+          📊 AFTER CLEANUP:
+          {{ post_cleanup_info.stdout }}
+
+          💡 RECOMMENDATIONS:
+          - Schedule regular cleanup: cron job for this playbook
+          - Monitor disk usage: ansible-playbook playbooks/disk_usage_report.yml
+          - Consider log rotation: ansible-playbook playbooks/log_rotation.yml
+          {% if not aggressive_cleanup %}
+          - For more space: run with -e "aggressive_cleanup=true"
+          {% endif %}
+
+          ✅ CLEANUP COMPLETE
+
+        dest: "/tmp/docker_cleanup/{{ ansible_date_time.date }}/{{ inventory_hostname }}_cleanup_report.txt"
+
+    - name: Display cleanup summary
+      debug:
+        msg: |
+
+          ✅ DOCKER CLEANUP COMPLETE - {{ inventory_hostname }}
+          =============================================
+
+          🔍 Mode: {{ 'DRY RUN' if dry_run else 'LIVE CLEANUP' }}
+          💪 Aggressive: {{ aggressive_cleanup }}
+
+          📊 SUMMARY:
+          {{ post_cleanup_info.stdout }}
+
+          📄 Full report: /tmp/docker_cleanup/{{ ansible_date_time.date }}/{{ inventory_hostname }}_cleanup_report.txt
+
+          🔍 Next Steps:
+          {% if dry_run %}
+          - Run without dry_run to perform actual cleanup
+          {% endif %}
+          - Monitor: ansible-playbook playbooks/disk_usage_report.yml
+          - Schedule regular cleanup via cron
+
+          =============================================
+
+    - name: Restart Docker daemon if needed
+      systemd:
+        name: docker
+        state: restarted
+      when:
+        - restart_docker | default(false) | bool
+        - not dry_run | bool
+      register: docker_restart
+
+    - name: Verify services after cleanup
+      shell: |
+        echo "Verifying critical services are still running..."
+        sleep 10  # Give services time to start
+
+        critical_services=("plex" "immich-server" "vaultwarden" "grafana" "prometheus")
+        failed_services=()
+
+        for service in "${critical_services[@]}"; do
+          if ! docker ps --filter "name=$service" --format "{{.Names}}" | grep -q "$service"; then
+            failed_services+=("$service")
+          fi
+        done
+
+        if [ ${#failed_services[@]} -eq 0 ]; then
+          echo "✅ All critical services verified running"
+        else
+          echo "⚠️  Some services may need attention: ${failed_services[*]}"
+        fi
+      register: service_verification
+      when:
+        - not dry_run | bool
+        - verify_services | default(true) | bool
+
+    - name: Display service verification
+      debug:
+        msg: "{{ service_verification.stdout }}"
+      when: service_verification is defined