Add check_systemd_services.py

check_systemd_services.py

@@ -0,0 +1,92 @@
+#!/usr/bin/env python3
+
+import subprocess
+import json
+import sys
+import argparse
+from datetime import datetime
+from fnmatch import fnmatch
+
+DEFAULT_WHITELIST = [
+    "emergency.service",
+    "rescue.service",
+    "rc-local.service",
+    "getty@tty1.service",
+    "systemd-udev-settle.service",
+    "systemd-pcrphase*.service",
+    "apt-daily*.service",
+    "fstrim.service",
+    "man-db.service",
+    "modprobe@*.service"
+]
+
+def get_all_services():
+    """Pobiera wszystkie jednostki systemd typu service jako JSON."""
+    cmd = [
+        "systemctl", "list-units", "--type=service", "--all",
+        "--no-legend", "--no-pager", "--output=json"
+    ]
+    try:
+        output = subprocess.check_output(cmd, text=True)
+        services = json.loads(output)
+        return services
+    except Exception as e:
+        print(f"UNKNOWN - systemctl JSON parse failed: {e}")
+        sys.exit(3)
+
+def is_whitelisted(name, whitelist):
+    return any(fnmatch(name, pattern) for pattern in whitelist)
+
+def check_services(services, whitelist):
+    problems = []
+    for svc in services:
+        name = svc.get("unit")
+        load = svc.get("load")
+        active = svc.get("active")
+        sub = svc.get("sub")
+
+        if not name or name == "●":
+            continue  # Pomijamy nieczytelne/uszkodzone wpisy
+
+        if is_whitelisted(name, whitelist):
+            continue
+
+        if active == "failed" or sub == "failed":
+            problems.append((name, active, sub))
+        elif active not in ["active", "inactive"]:
+            problems.append((name, active, sub))
+        elif active == "inactive" and sub not in ["dead", "exited"]:
+            problems.append((name, active, sub))
+    return problems
+
+def format_output(problems):
+    if not problems:
+        return "OK - All systemd services are healthy", 0
+    summary = ", ".join(f"{name}: {active}/{sub}" for name, active, sub in problems)
+    return f"CRITICAL - Problematic services: {summary}", 2
+
+def log_result(message, logfile="/var/log/nagios_systemd_check.log"):
+    try:
+        with open(logfile, "a") as f:
+            f.write(f"[{datetime.now()}] {message}\n")
+    except Exception:
+        pass
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("-w", "--whitelist", nargs="*", default=DEFAULT_WHITELIST,
+                        help="Whitelist of service patterns to ignore (fnmatch supported)")
+    parser.add_argument("--log", action="store_true", help="Log result to /var/log/nagios_systemd_check.log")
+    args = parser.parse_args()
+
+    services = get_all_services()
+    problems = check_services(services, args.whitelist)
+    message, code = format_output(problems)
+
+    print(message)
+    if args.log:
+        log_result(message)
+    sys.exit(code)
+
+if __name__ == "__main__":
+    main()