---
- name: Weekly Debian server report (with trends) → Telegram via SOCKS5
hosts: all
gather_facts: yes
become: true
vars:
telegram_token: "8820292312:AAECFO1AEWOJTh30zNuhyjC5C81o_qJHQcM"
telegram_chat_id: "407968638"
telegram_proxy: "socks5h://administrator:60NthRegT@192.168.0.218:1080"
snapshot_dir: "/var/lib/ansible-metrics"
# Известный шум — не считаем за инциденты
ignored_failed_units:
- adguard-vpn.service
- ifup@wlp2s0.service
- iptables-restore.service
- netfilter-persistent.service
- networking.service
metrics_packages:
- curl
- bc
- procps
- sysstat
- lm-sensors
- smartmontools
- gawk
- coreutils
tasks:
# ==================================================================
# 0. PREFLIGHT
# ==================================================================
- name: Ensure required packages are installed (Debian)
ansible.builtin.apt:
name: "{{ metrics_packages }}"
state: present
update_cache: yes
cache_valid_time: 3600
when: ansible_os_family == "Debian"
- name: Enable and start sysstat (needed for sar)
ansible.builtin.service:
name: sysstat
state: started
enabled: yes
when: ansible_os_family == "Debian"
failed_when: false
changed_when: false
- name: Warn if host is not Debian family
ansible.builtin.debug:
msg: "WARNING: {{ inventory_hostname }} is {{ ansible_os_family }}, не Debian — часть метрик может быть недоступна"
when: ansible_os_family != "Debian"
# ==================================================================
# 1. СНАПШОТ ПРОШЛОЙ НЕДЕЛИ
# ==================================================================
- name: Ensure snapshot dir exists on target host
ansible.builtin.file:
path: "{{ snapshot_dir }}"
state: directory
mode: "0755"
become: true
- name: Read previous week snapshot
ansible.builtin.slurp:
src: "{{ snapshot_dir }}/{{ inventory_hostname }}.json"
register: prev_snapshot_raw
failed_when: false
become: true
- name: Parse previous snapshot (or empty dict on first run)
ansible.builtin.set_fact:
prev: "{{ (prev_snapshot_raw.content | b64decode | from_json) if prev_snapshot_raw.content is defined else {} }}"
# ==================================================================
# 2. БАЗОВЫЕ МЕТРИКИ
# ==================================================================
- name: Gather uptime
ansible.builtin.command: uptime -p
register: uptime_output
changed_when: false
- name: Read load average from /proc/loadavg
ansible.builtin.slurp:
src: /proc/loadavg
register: loadavg_raw
changed_when: false
- name: Extract load averages
ansible.builtin.set_fact:
load1: "{{ (loadavg_raw.content | b64decode).split()[0] | default('N/A') }}"
load5: "{{ (loadavg_raw.content | b64decode).split()[1] | default('N/A') }}"
load15: "{{ (loadavg_raw.content | b64decode).split()[2] | default('N/A') }}"
- name: Get CPU idle via vmstat
ansible.builtin.shell: |
if command -v vmstat >/dev/null 2>&1; then
vmstat 1 2 | tail -1 | awk '{print $15}'
else
echo ""
fi
register: cpu_idle_vmstat
changed_when: false
failed_when: false
- name: Get CPU idle via mpstat
ansible.builtin.shell: |
if command -v mpstat >/dev/null 2>&1; then
mpstat 1 1 | awk '/Average/ {print $NF}'
else
echo ""
fi
register: cpu_idle_mpstat
changed_when: false
failed_when: false
when: cpu_idle_vmstat.stdout == ""
- name: Get CPU idle via top
ansible.builtin.shell: |
top -bn2 | grep "Cpu(s)" | tail -1 | sed 's/.*, *\([0-9.]*\)%* id.*/\1/'
register: cpu_idle_top
changed_when: false
failed_when: false
when: cpu_idle_vmstat.stdout == "" and cpu_idle_mpstat.stdout == ""
- name: Get CPU idle via /proc/stat
ansible.builtin.shell: |
prev_idle=$(awk '/^cpu / {print $5}' /proc/stat)
prev_total=$(awk '/^cpu / {print $2+$3+$4+$5+$6+$7+$8}' /proc/stat)
sleep 1
curr_idle=$(awk '/^cpu / {print $5}' /proc/stat)
curr_total=$(awk '/^cpu / {print $2+$3+$4+$5+$6+$7+$8}' /proc/stat)
diff_idle=$((curr_idle - prev_idle))
diff_total=$((curr_total - prev_total))
awk -v i="$diff_idle" -v t="$diff_total" 'BEGIN {printf "%.1f", (i*100)/t}'
register: cpu_idle_proc
changed_when: false
failed_when: false
when: cpu_idle_vmstat.stdout == "" and cpu_idle_mpstat.stdout == "" and cpu_idle_top.stdout == ""
- name: Set final idle with validation
ansible.builtin.set_fact:
cpu_idle_final: >-
{%- set idle = (cpu_idle_vmstat.stdout | float) if cpu_idle_vmstat.stdout != "" else
(cpu_idle_mpstat.stdout | float) if cpu_idle_mpstat.stdout != "" else
(cpu_idle_top.stdout | float) if cpu_idle_top.stdout != "" else
(cpu_idle_proc.stdout | float) if cpu_idle_proc.stdout != "" else 90 -%}
{%- if idle > 100 or idle < 0 -%}{{ 90 }}{%- else -%}{{ idle }}{%- endif -%}
- name: Get memory usage human
ansible.builtin.shell: |
free -h | awk '/Mem:/ {printf "Занято: %s, Всего: %s, Доступно: %s", $3, $2, $6}'
register: memory_usage_human
changed_when: false
- name: Get memory percentage
ansible.builtin.shell: |
free -m | awk 'NR==2{printf "%.0f", $3/$2*100}'
register: mem_percent
changed_when: false
- name: Get disk usage for root
ansible.builtin.shell: |
df -h / | awk 'NR==2{printf "Занято: %s, Всего: %s, Свободно: %s, Использовано: %s", $3, $2, $4, $5}'
register: disk_usage
changed_when: false
- name: Get CPU frequency
ansible.builtin.shell: |
if [ -f /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq ]; then
awk '{printf "%.2f GHz", $1/1000000}' /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq
elif command -v lscpu >/dev/null 2>&1; then
lscpu | grep "MHz" | awk '{print $3}' | head -1 | awk '{printf "%.2f GHz", $1/1000}'
else
echo ""
fi
register: cpu_freq
changed_when: false
failed_when: false
- name: Set CPU frequency text
ansible.builtin.set_fact:
cpu_freq_text: "{{ cpu_freq.stdout | default('') }}"
- name: Get CPU temperature from thermal zone
ansible.builtin.shell: |
cat /sys/class/thermal/thermal_zone0/temp 2>/dev/null | awk '{printf "%.1f°C", $1/1000}'
register: cpu_temp
changed_when: false
failed_when: false
- name: Fallback to sensors
ansible.builtin.shell: |
sensors 2>/dev/null | grep -i "Package id 0" | awk '{print $4}' | sed 's/+//g' | sed 's/°C//g' | awk '{printf "%.1f°C", $1}'
register: cpu_temp_sensors
changed_when: false
failed_when: false
when: cpu_temp.stdout == ""
- name: Set final CPU temperature
ansible.builtin.set_fact:
cpu_temp_final: "{{ cpu_temp.stdout if cpu_temp.stdout != '' else (cpu_temp_sensors.stdout | default('')) }}"
- name: Check if smartctl exists
ansible.builtin.command: which smartctl
register: smartctl_check
changed_when: false
failed_when: false
- name: Get first block device
ansible.builtin.shell: |
lsblk -dn -o NAME | grep -E '^(sd|hd|nvme|vd|xvd)' | head -1
register: disk_device
changed_when: false
failed_when: false
- name: Get disk temperature
ansible.builtin.shell: |
smartctl -A /dev/{{ disk_device.stdout }} 2>/dev/null \
| grep -i "Temperature" | awk '{print $10}' \
| sed 's/°C//g' | awk '{printf "%.1f°C", $1}'
register: disk_temp
changed_when: false
failed_when: false
when: smartctl_check.rc == 0 and disk_device.stdout != ""
- name: Get disk SMART health
ansible.builtin.shell: |
smartctl -H /dev/{{ disk_device.stdout }} 2>/dev/null \
| grep -i "SMART overall-health" | awk -F: '{print $2}' | xargs
register: smart_health
changed_when: false
failed_when: false
when: smartctl_check.rc == 0 and disk_device.stdout != ""
- name: Get reallocated sectors
ansible.builtin.shell: |
smartctl -A /dev/{{ disk_device.stdout }} 2>/dev/null \
| awk '/Reallocated_Sector_Ct/ {print $NF}' | head -1
register: reallocated
changed_when: false
failed_when: false
when: smartctl_check.rc == 0 and disk_device.stdout != ""
- name: Set disk-related facts
ansible.builtin.set_fact:
disk_temp_text: "{{ (disk_temp | default({})).stdout | default('') }}"
disk_device_name: "{{ (disk_device | default({})).stdout | default('') }}"
smart_health_text: "{{ (smart_health | default({})).stdout | default('N/A') }}"
reallocated_text: "{{ (reallocated | default({})).stdout | default('0') }}"
# ==================================================================
# 3. НЕДЕЛЬНЫЕ АГРЕГАТЫ (sar)
# ==================================================================
- name: Weekly CPU avg/peak from sar
ansible.builtin.shell: |
{
for i in $(seq 0 6); do
d=$(date -d "$i days ago" +%d)
f="/var/log/sysstat/sa$d"
[ -f "$f" ] && sar -u -f "$f" 2>/dev/null
done
} | awk '/^[0-9]/ {
cpu=100-$NF
sum+=cpu; n++
if (cpu>max) max=cpu
} END {
if (n>0) printf "%.1f %.1f %d", sum/n, max, n;
else print "N/A N/A 0"
}'
register: cpu_week
changed_when: false
failed_when: false
# sar -r: $5 = %memused
- name: Weekly memory peak from sar
ansible.builtin.shell: |
{
for i in $(seq 0 6); do
d=$(date -d "$i days ago" +%d)
f="/var/log/sysstat/sa$d"
[ -f "$f" ] && sar -r -f "$f" 2>/dev/null
done
} | awk '/^[0-9]/ {
if ($5+0 > max) max=$5+0
} END {
if (max>0) printf "%.1f", max;
else print "N/A"
}'
register: mem_week_peak
changed_when: false
failed_when: false
- name: Parse weekly CPU values
ansible.builtin.set_fact:
cpu_week_avg: "{{ (cpu_week.stdout.split()[0]) if cpu_week.stdout.split() | length >= 1 else 'N/A' }}"
cpu_week_peak: "{{ (cpu_week.stdout.split()[1]) if cpu_week.stdout.split() | length >= 2 else 'N/A' }}"
cpu_week_samples: "{{ (cpu_week.stdout.split()[2]) if cpu_week.stdout.split() | length >= 3 else '0' }}"
mem_week_peak_val: "{{ mem_week_peak.stdout | default('N/A') }}"
# ==================================================================
# 4. СОСТОЯНИЕ СИСТЕМЫ
# ==================================================================
- name: Count upgradable packages
ansible.builtin.shell: |
apt list --upgradable 2>/dev/null | tail -n +2 | grep -c upgradable || true
register: upgradable_count
changed_when: false
failed_when: false
- name: Count security updates
ansible.builtin.shell: |
apt list --upgradable 2>/dev/null | grep -ci security || true
register: security_count
changed_when: false
failed_when: false
- name: Check if reboot required
ansible.builtin.stat:
path: /var/run/reboot-required
register: reboot_required
- name: List failed systemd units (raw)
ansible.builtin.shell: |
systemctl --failed --no-legend 2>/dev/null | awk '{print $1}'
register: failed_units_raw
changed_when: false
failed_when: false
- name: Filter known noise from failed units
ansible.builtin.set_fact:
failed_units_list: >-
{{
failed_units_raw.stdout_lines
| reject('in', ignored_failed_units)
| list
}}
failed_units: >-
{{
failed_units_raw.stdout_lines
| reject('in', ignored_failed_units)
| list
| length
}}
failed_units_ignored_count: >-
{{
failed_units_raw.stdout_lines
| select('in', ignored_failed_units)
| list
| length
}}
- name: Count reboots in last 7 days
ansible.builtin.shell: |
last -x reboot --since "7 days ago" 2>/dev/null | grep -c "^reboot" || true
register: reboot_count
changed_when: false
failed_when: false
- name: Count failed SSH logins in last 7 days
ansible.builtin.shell: |
journalctl --since "7 days ago" -u ssh --no-pager 2>/dev/null \
| grep -c "Failed password" || true
register: failed_ssh
changed_when: false
failed_when: false
- name: Count OOM kills in last 7 days
ansible.builtin.shell: |
journalctl --since "7 days ago" -k --no-pager 2>/dev/null \
| grep -ci "out of memory" || true
register: oom_kills
changed_when: false
failed_when: false
- name: Count I/O errors in last 7 days
ansible.builtin.shell: |
journalctl --since "7 days ago" -k --no-pager 2>/dev/null \
| grep -ci "I/O error" || true
register: io_errors
changed_when: false
failed_when: false
- name: Top 5 by memory now
ansible.builtin.shell: |
ps aux --sort=-%mem | head -6 | tail -5 | awk '{print $11 " (" $4 "%)"}'
register: top_procs
changed_when: false
- name: Top 5 long-running processes
ansible.builtin.shell: |
ps -eo etimes,comm --sort=-etimes | awk 'NR>1 && $1>604800 {print $2 " (uptime " int($1/86400) "d)"}' | head -5
register: top_long
changed_when: false
failed_when: false
- name: Join top lists
ansible.builtin.set_fact:
top_procs_joined: "{{ top_procs.stdout_lines | join('\n') }}"
top_long_joined: "{{ top_long.stdout_lines | join('\n') if top_long.stdout_lines | length > 0 else '(нет процессов с аптаймом > 7 дней)' }}"
# ==================================================================
# 5. РАСЧЁТЫ, ДЕЛЬТЫ, HEALTH-SCORE
# ==================================================================
- name: Calculate CPU usage %
ansible.builtin.set_fact:
cpu_usage: "{{ 100 - (cpu_idle_final | float) }}"
- name: Extract disk used %
ansible.builtin.set_fact:
disk_used_percent: "{{ disk_usage.stdout.split('Использовано:')[1].strip().replace('%', '') | float }}"
- name: Compute deltas vs previous week
ansible.builtin.set_fact:
disk_delta: "{{ ((disk_used_percent | float) - (prev.disk_used_pct | float)) | round(1) }}"
mem_delta: "{{ ((mem_percent.stdout | float) - (prev.mem_pct | float)) | round(1) }}"
when:
- prev.disk_used_pct is defined
- prev.mem_pct is defined
- name: Format delta strings with arrows
ansible.builtin.set_fact:
disk_delta_str: >-
{%- if disk_delta is defined -%}
{%- set d = disk_delta | float -%}
{%- if d > 1 %} (+{{ d }}% ⬆️)
{%- elif d < -1 %} ({{ d }}% ⬇️)
{%- else %} (≈ без изменений){%- endif -%}
{%- else -%} (нет данных за прошлую неделю){%- endif -%}
mem_delta_str: >-
{%- if mem_delta is defined -%}
{%- set d = mem_delta | float -%}
{%- if d > 3 %} (+{{ d }}% ⬆️)
{%- elif d < -3 %} ({{ d }}% ⬇️)
{%- else %} (≈ без изменений){%- endif -%}
{%- else -%} (нет данных){%- endif -%}
- name: Set component health
ansible.builtin.set_fact:
cpu_health: "{% if (cpu_usage | float) < 50 %}🟢{% elif (cpu_usage | float) < 80 %}🟡{% else %}🔴{% endif %}"
mem_health: "{% if (mem_percent.stdout | float) < 60 %}🟢{% elif (mem_percent.stdout | float) < 85 %}🟡{% else %}🔴{% endif %}"
disk_health: "{% if (disk_used_percent | float) < 70 %}🟢{% elif (disk_used_percent | float) < 85 %}🟡{% else %}🔴{% endif %}"
- name: Compute warnings list
ansible.builtin.set_fact:
warnings_list: >-
{{
(
(['Диск > 80%'] if (disk_used_percent | float) > 80 else []) +
(['Диск вырос > 5% за неделю'] if (disk_delta is defined and (disk_delta | float) > 5) else []) +
(['Требуется перезагрузка'] if reboot_required.stat.exists else []) +
(['Security-обновления не установлены'] if (security_count.stdout | int) > 0 else []) +
(['Failed systemd units: ' ~ (failed_units_list | join(', '))] if (failed_units | int) > 0 else []) +
(['OOM-kills за неделю'] if (oom_kills.stdout | int) > 0 else []) +
(['I/O errors в ядре'] if (io_errors.stdout | int) > 0 else []) +
(['SMART не PASSED'] if (smart_health_text != 'PASSED' and smart_health_text != 'N/A' and smart_health_text != '') else [])
) | list
}}
- name: Compute overall status emoji
ansible.builtin.set_fact:
overall_status: >-
{%- set n = warnings_list | length -%}
{{ '🔴 Требует внимания' if n >= 3 else ('🟡 Есть предупреждения' if n > 0 else '🟢 Всё в порядке') }}
# ==================================================================
# 5b. ПРЕДВЫЧИСЛЕНИЕ ФРАГМЕНТОВ (лечит trim_blocks)
# ==================================================================
- name: Precompute display fragments (no block tags in template)
ansible.builtin.set_fact:
status_suffix: "{{ (' (' ~ (warnings_list | length) ~ ' ⚠️)') if (warnings_list | length) > 0 else '' }}"
cpu_freq_suffix: "{{ (' (' ~ cpu_freq_text ~ ')') if cpu_freq_text != '' else '' }}"
disk_short: "{{ disk_usage.stdout.split('Использовано:')[0].strip().rstrip(',') }}"
disk_temp_line: "{{ ('Диск ' ~ disk_device_name ~ ': ' ~ disk_temp_text ~ ' · SMART: ' ~ smart_health_text ~ ' · realloc: ' ~ reallocated_text) if disk_temp_text != '' else '' }}"
warnings_block: "{{ ('⚠️ Предупреждения:\n• ' ~ (warnings_list | join('\n• '))) if (warnings_list | length) > 0 else '' }}"
failed_units_line: "{{ (failed_units | string) ~ ((' (' ~ (failed_units_list | join(', ')) ~ ')') if (failed_units | int) > 0 else '') ~ ((' [+' ~ (failed_units_ignored_count | string) ~ ' в whitelist]') if (failed_units_ignored_count | int) > 0 else '') }}"
# ==================================================================
# 6. ФОРМИРОВАНИЕ СООБЩЕНИЯ
# ==================================================================
- name: Build weekly report message (HTML)
ansible.builtin.set_fact:
message_html: |
📅 Еженедельный отчёт: {{ ansible_hostname }}
Статус: {{ overall_status }}{{ status_suffix }}
{{ ansible_date_time.date }} {{ ansible_date_time.time }} · {{ ansible_default_ipv4.address | default('N/A') }}
{{ warnings_block }}
───────────────
📊 Средние и пиковые за неделю
💻 CPU: avg {{ cpu_week_avg }}% · peak {{ cpu_week_peak }}% ({{ cpu_week_samples }} замеров)
🧠 RAM: peak {{ mem_week_peak_val }}%
───────────────
📈 Текущее состояние
⏱️ Аптайм: {{ uptime_output.stdout }}
📉 Load avg (1,5,15): {{ load1 }} , {{ load5 }} , {{ load15 }}
💻 CPU сейчас: {{ (cpu_usage | float) | round(1) }}% {{ cpu_health }}{{ cpu_freq_suffix }}
🧠 RAM сейчас: {{ mem_percent.stdout }}% {{ mem_health }} ({{ memory_usage_human.stdout }}){{ mem_delta_str }}
💾 Диск /: {{ (disk_used_percent | float) | round(1) }}% {{ disk_health }} ({{ disk_short }}){{ disk_delta_str }}
───────────────
🌡️ Температуры
CPU: {{ cpu_temp_final | default('недоступно') }}
{{ disk_temp_line }}
───────────────
🛠️ Обслуживание
📦 Обновлений: {{ upgradable_count.stdout }} (🔴 security: {{ security_count.stdout }})
♻️ Требуется reboot: {{ 'ДА ⚠️' if reboot_required.stat.exists else 'нет' }}
🔁 Перезагрузок за неделю: {{ reboot_count.stdout }}
🚨 Failed units: {{ failed_units_line }}
───────────────
🔐 Безопасность и ошибки за неделю
🔓 Неудачных SSH-логинов: {{ failed_ssh.stdout }}
💥 OOM-kills: {{ oom_kills.stdout }}
🗄️ I/O errors: {{ io_errors.stdout }}
───────────────
🔥 Топ-5 процессов по памяти (сейчас):
{{ top_procs_joined }}
🕰️ Топ-5 долгоживущих процессов (> 7 дней):
{{ top_long_joined }}
# ==================================================================
# 7. СОХРАНЯЕМ СНАПШОТ
# ==================================================================
- name: Save snapshot for next week
ansible.builtin.copy:
content: |
{{
{
'ts': ansible_date_time.iso8601,
'disk_used_pct': disk_used_percent | float,
'mem_pct': mem_percent.stdout | float,
'cpu_now': cpu_usage | float,
'cpu_week_avg': cpu_week_avg,
'cpu_week_peak': cpu_week_peak
} | to_nice_json
}}
dest: "{{ snapshot_dir }}/{{ inventory_hostname }}.json"
mode: "0644"
become: true
# ==================================================================
# 8. ОТПРАВКА В TELEGRAM
# ==================================================================
- name: Check Telegram API reachability via SOCKS5 proxy
ansible.builtin.shell: |
curl -s -o /dev/null -w "%{http_code}" --max-time 15 \
--proxy "{{ telegram_proxy }}" \
"https://api.telegram.org/bot{{ telegram_token }}/getMe"
register: tg_check
changed_when: false
failed_when: false
delegate_to: localhost
become: false
- name: Fail if Telegram API unreachable
ansible.builtin.fail:
msg: "Telegram API недоступен через прокси {{ telegram_proxy }}. HTTP-код: {{ tg_check.stdout }}."
when: tg_check.stdout != "200"
delegate_to: localhost
become: false
- name: Send report to Telegram (via tmpfile)
ansible.builtin.shell: |
set -e
tmp=$(mktemp)
trap 'rm -f "$tmp"' EXIT
cat > "$tmp" <<'JSON_EOF'
{{ {'chat_id': telegram_chat_id, 'text': message_html, 'parse_mode': 'HTML', 'disable_web_page_preview': true} | to_json }}
JSON_EOF
response=$(curl -s -w "\n%{http_code}" -X POST \
--proxy "{{ telegram_proxy }}" \
-H "Content-Type: application/json" \
--data-binary @"$tmp" \
"https://api.telegram.org/bot{{ telegram_token }}/sendMessage")
http_code=$(echo "$response" | tail -n1)
body=$(echo "$response" | sed '$d')
echo "HTTP code: $http_code"
echo "Body: $body"
if [ "$http_code" -eq 200 ] && echo "$body" | grep -q '"ok":true'; then
echo "OK"; exit 0
fi
echo "ERROR: send failed" >&2; exit 1
changed_when: false
register: tg_response
failed_when: tg_response.rc != 0
delegate_to: localhost
become: false