feat: heartbeat 4h via HEARTBEAT_INTERVAL_MINUTES (2.0.2-SAC)

Co-authored-by: Cursor <cursoragent@cursor.com>

ssh-monitor

@@ -7,7 +7,7 @@ IFS=$'\n\t'
 # ============================================
 
 CONFIG_FILE="/etc/ssh-monitor.conf"
-SSH_MONITOR_VERSION="2.0.1-SAC"
+SSH_MONITOR_VERSION="2.0.2-SAC"
 
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 if [ -f "$SCRIPT_DIR/sac-client.sh" ]; then
@@ -51,6 +51,8 @@ MAIL_SMTP_SSL="0"
 # Конфигурация логов
 LOG_FILE="/var/log/ssh_monitor.log"
 LAST_HEARTBEAT_FILE="/var/log/last_heartbeat.txt"
+# Интервал SAC agent.heartbeat (минуты). 240 = 4 ч.
+HEARTBEAT_INTERVAL_MINUTES=240
 LAST_REPORT_FILE="/var/log/last_daily_report.txt"
 DAILY_REPORT_ENABLED=1
 DAILY_REPORT_HOUR=9
@@ -183,6 +185,7 @@ load_config() {
  local cfg_backup_webhook_url="${BACKUP_WEBHOOK_URL}"
  local cfg_log_file="${LOG_FILE}"
  local cfg_last_heartbeat_file="${LAST_HEARTBEAT_FILE}"
+ local cfg_heartbeat_interval_minutes="${HEARTBEAT_INTERVAL_MINUTES:-240}"
  local cfg_last_report_file="${LAST_REPORT_FILE}"
  local cfg_daily_report_enabled="${DAILY_REPORT_ENABLED:-1}"
  local cfg_daily_report_hour="${DAILY_REPORT_HOUR}"
@@ -243,6 +246,10 @@ load_config() {
  BACKUP_WEBHOOK_URL="${BACKUP_WEBHOOK_URL:-$cfg_backup_webhook_url}"
  LOG_FILE="${LOG_FILE:-$cfg_log_file}"
  LAST_HEARTBEAT_FILE="${LAST_HEARTBEAT_FILE:-$cfg_last_heartbeat_file}"
+ HEARTBEAT_INTERVAL_MINUTES="${HEARTBEAT_INTERVAL_MINUTES:-$cfg_heartbeat_interval_minutes}"
+ if ! [[ "${HEARTBEAT_INTERVAL_MINUTES:-240}" =~ ^[0-9]+$ ]] || [ "${HEARTBEAT_INTERVAL_MINUTES:-0}" -lt 1 ]; then
+ HEARTBEAT_INTERVAL_MINUTES=240
+ fi
  LAST_REPORT_FILE="${LAST_REPORT_FILE:-$cfg_last_report_file}"
  LAST_SSH_CHECK_FILE="${LAST_SSH_CHECK_FILE:-$cfg_last_ssh_check_file}"
  LAST_SUDO_CHECK_FILE="${LAST_SUDO_CHECK_FILE:-$cfg_last_sudo_check_file}"
@@ -1911,8 +1918,9 @@ check_heartbeat() {
  local current_epoch
  current_epoch=$(date '+%s')
  local diff=$((current_epoch - last_heartbeat_epoch))
+ local heartbeat_interval_sec=$((HEARTBEAT_INTERVAL_MINUTES * 60))
 
- if [ "$last_heartbeat_epoch" -eq 0 ] || [ "$diff" -ge 43200 ]; then
+ if [ "$last_heartbeat_epoch" -eq 0 ] || [ "$diff" -ge "$heartbeat_interval_sec" ]; then
  send_heartbeat
  fi
 }

ssh-monitor-watchdog

@@ -5,7 +5,7 @@ IFS=$'\n\t'
 CONFIG_FILE="/etc/ssh-monitor.conf"
 WATCHDOG_LOG_FILE="/var/log/ssh_monitor_watchdog.log"
 LAST_HEARTBEAT_FILE="/var/log/last_heartbeat.txt"
-WATCHDOG_MAX_HEARTBEAT_AGE=5400
+WATCHDOG_MAX_HEARTBEAT_AGE=18000
 WATCHDOG_SERVICE_NAME="ssh-monitor.service"
 WATCHDOG_NOTIFY_ON_RECOVERY="0"
 

ssh-monitor.conf.example

@@ -49,6 +49,8 @@ MAIL_SMTP_SSL="0"
 
 LOG_FILE="/var/log/ssh_monitor.log"
 LAST_HEARTBEAT_FILE="/var/log/last_heartbeat.txt"
+# Интервал SAC agent.heartbeat (минуты). 240 = 4 ч.
+HEARTBEAT_INTERVAL_MINUTES="240"
 LAST_REPORT_FILE="/var/log/last_daily_report.txt"
 LAST_SSH_CHECK_FILE="/var/log/last_ssh_check"
 LAST_SUDO_CHECK_FILE="/var/log/last_sudo_check"
@@ -102,7 +104,8 @@ PROMETHEUS_TEXTFILE_DIR=""
 # JSON-снимок состояния на каждой итерации (timestamp, hostname, dry_run). Пусто — отключено.
 HEALTHCHECK_STATUS_FILE=""
 
-WATCHDOG_MAX_HEARTBEAT_AGE="5400"
+# Порог устаревания heartbeat для watchdog (сек): интервал + 1 ч запас (240 мин → 18000).
+WATCHDOG_MAX_HEARTBEAT_AGE="18000"
 WATCHDOG_LOG_FILE="/var/log/ssh_monitor_watchdog.log"
 WATCHDOG_SERVICE_NAME="ssh-monitor.service"
 WATCHDOG_NOTIFY_ON_RECOVERY="0"

update_ssh_monitor.sh

@@ -24,6 +24,8 @@ SUMMARY_UPDATER="—"
 
 CONFIG_FILE="/etc/ssh-monitor.conf"
 DEPLOY_INSTALL_PATH="/opt/scripts/update_ssh_monitor.sh"
+EXPECTED_HEARTBEAT_INTERVAL_MINUTES=240
+EXPECTED_WATCHDOG_MAX_HEARTBEAT_AGE=$((EXPECTED_HEARTBEAT_INTERVAL_MINUTES * 60 + 3600))
 SAC_SPOOL_DIR_DEFAULT="/var/lib/ssh-monitor/sac-spool"
 SYSTEMD_UNIT_PATH="/etc/systemd/system/ssh-monitor.service"
 
@@ -155,6 +157,40 @@ ensure_ssh_monitor_config_from_example() {
     return 0
 }
 
+set_config_kv() {
+    local key="$1" val="$2" file="$3"
+    if grep -qE "^[[:space:]]*${key}=" "$file" 2>/dev/null; then
+        sed -i -E "s/^[[:space:]]*${key}=.*/${key}=\"${val}\"/" "$file"
+    else
+        printf '%s="%s"\n' "$key" "$val" >>"$file"
+    fi
+}
+
+sync_heartbeat_interval_config() {
+    local file="$CONFIG_FILE"
+    local current_interval current_watchdog
+
+    [ -f "$file" ] || return 1
+
+    current_interval=$(config_conf_value "HEARTBEAT_INTERVAL_MINUTES" "$file" 2>/dev/null || true)
+    current_watchdog=$(config_conf_value "WATCHDOG_MAX_HEARTBEAT_AGE" "$file" 2>/dev/null || true)
+
+    if [ "$current_interval" = "$EXPECTED_HEARTBEAT_INTERVAL_MINUTES" ] \
+        && [ "$current_watchdog" = "$EXPECTED_WATCHDOG_MAX_HEARTBEAT_AGE" ]; then
+        return 1
+    fi
+
+    cp "$file" "${file}.bak.$(date +%Y%m%d_%H%M%S)" 2>/dev/null || true
+    set_config_kv "HEARTBEAT_INTERVAL_MINUTES" "$EXPECTED_HEARTBEAT_INTERVAL_MINUTES" "$file"
+    set_config_kv "WATCHDOG_MAX_HEARTBEAT_AGE" "$EXPECTED_WATCHDOG_MAX_HEARTBEAT_AGE" "$file"
+    log_message "INFO" "Config $file: HEARTBEAT_INTERVAL_MINUTES=$EXPECTED_HEARTBEAT_INTERVAL_MINUTES, WATCHDOG_MAX_HEARTBEAT_AGE=$EXPECTED_WATCHDOG_MAX_HEARTBEAT_AGE (deploy policy on script upgrade)."
+    if [ "$DEPLOY_MODE" -eq 1 ]; then
+        say_info "Heartbeat: ${EXPECTED_HEARTBEAT_INTERVAL_MINUTES} мин, watchdog ${EXPECTED_WATCHDOG_MAX_HEARTBEAT_AGE} с."
+    fi
+    CONFIG_CHANGED=1
+    return 0
+}
+
 sync_use_sac_fallback_mode() {
     local file="$CONFIG_FILE"
     [ -f "$file" ] || return 1
@@ -968,6 +1004,7 @@ update_script() {
 
     if [ "$script_changed" -eq 1 ]; then
         sync_use_sac_fallback_mode || true
+        sync_heartbeat_interval_config || true
     fi
 
     if [ "$script_changed" -eq 1 ] || [ "${SAC_CLIENT_CHANGED:-0}" -eq 1 ] || [ "${CONFIG_CHANGED:-0}" -eq 1 ]; then

version.txt

@@ -1 +1 @@
-2.0.1-SAC
+2.0.2-SAC