Add service_restarter.py

service_restarter.py

@@ -0,0 +1,175 @@
+#!/usr/bin/env python3
+"""
+service_restarter.py — Universal service watchdog and auto-restarter for systemd-managed services.
+
+ Features:
+- Checks if a service is listening on a given port and IP using TCP/UDP
+- Supports multiple check methods:
+    • listen   — direct socket connection (default)
+    • curl     — HTTP GET request using curl
+    • ping     — ICMP echo (TCP only, host-level reachability)
+    • nc       — use netcat to test open port
+    • cmd      — run a custom command given via --check-cmd
+- Restarts systemd service if check fails
+- Configurable retry attempts and delay between retries
+- Permission check (must be run as root)
+- Quiet mode for cron use
+
+ Examples:
+  # Basic TCP check:
+    ./service_restarter.py --proto tcp --port 631 --service cups
+
+  # With curl:
+    ./service_restarter.py --proto tcp --port 80 --host 192.168.1.10 --service apache2 --check-method curl
+
+  # With ping and retry config:
+    ./service_restarter.py --proto tcp --port 22 --service sshd --check-method ping --retries 5 --delay 2
+
+  # Using custom command:
+    ./service_restarter.py --proto tcp --port 8080 --service myapp --check-method cmd --check-cmd 'curl -fs http://localhost:8080/health'
+
+  # Cron-friendly:
+    */5 * * * * root /usr/local/bin/service_restarter.py --proto tcp --port 631 --service cups --quiet
+"""
+
+import socket
+import subprocess
+import argparse
+import sys
+import time
+import os
+import shutil
+
+def is_tcp_listening(host, port):
+    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        s.settimeout(1)
+        try:
+            s.connect((host, port))
+            return True
+        except Exception:
+            return False
+
+def is_udp_listening(host, port):
+    with socket.socket(socket.AF_INET, socket.SOCK_DGRAM) as s:
+        try:
+            s.settimeout(1)
+            s.sendto(b'', (host, port))
+            s.recvfrom(1024)
+            return True
+        except Exception:
+            return False
+
+def check_with_curl(host, port):
+    try:
+        result = subprocess.run(
+            ["curl", "-fs", f"http://{host}:{port}/"],
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.DEVNULL,
+            timeout=5
+        )
+        return result.returncode == 0
+    except Exception:
+        return False
+
+def check_with_ping(host):
+    try:
+        subprocess.run(["ping", "-c", "1", "-W", "2", host],
+                       stdout=subprocess.DEVNULL,
+                       stderr=subprocess.DEVNULL,
+                       timeout=3)
+        return True
+    except Exception:
+        return False
+
+def check_with_nc(host, port):
+    nc = shutil.which("nc")
+    if not nc:
+        return False
+    try:
+        subprocess.run([nc, "-z", "-w", "2", host, str(port)],
+                       stdout=subprocess.DEVNULL,
+                       stderr=subprocess.DEVNULL,
+                       timeout=3)
+        return True
+    except Exception:
+        return False
+
+def check_with_cmd(command):
+    try:
+        result = subprocess.run(command, shell=True,
+                                stdout=subprocess.DEVNULL,
+                                stderr=subprocess.DEVNULL,
+                                timeout=5)
+        return result.returncode == 0
+    except Exception:
+        return False
+
+def restart_service(service):
+    subprocess.run(["systemctl", "restart", service], check=False)
+    time.sleep(2)
+
+def log(msg, quiet):
+    if not quiet:
+        print(msg)
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Check if a service is reachable on a given host/proto/port and restart it via systemd if not."
+    )
+    parser.add_argument("--proto", required=True, choices=["tcp", "udp"], help="Protocol to check: tcp or udp (required)")
+    parser.add_argument("--port", required=True, type=int, help="Port number to check (required)")
+    parser.add_argument("--host", default="127.0.0.1", help="IP address to check (default: 127.0.0.1)")
+    parser.add_argument("--service", required=True, help="Systemd service name to restart if check fails (required)")
+    parser.add_argument("--check-method", choices=["listen", "curl", "ping", "nc", "cmd"], default="listen",
+                        help="How to check service: listen (default), curl, ping, nc, cmd")
+    parser.add_argument("--check-cmd", help="Custom shell command to run when --check-method is 'cmd'")
+    parser.add_argument("--retries", type=int, default=3, help="Number of retries if service is unreachable (default: 3)")
+    parser.add_argument("--delay", type=int, default=2, help="Delay between retries in seconds (default: 2)")
+    parser.add_argument("--quiet", action="store_true", help="Suppress all output (for cron)")
+
+    args = parser.parse_args()
+
+    if os.geteuid() != 0:
+        print("ERROR: This script must be run as root.")
+        sys.exit(3)
+
+    def run_check():
+        if args.check_method == "listen":
+            if args.proto == "tcp":
+                return is_tcp_listening(args.host, args.port)
+            elif args.proto == "udp":
+                return is_udp_listening(args.host, args.port)
+        elif args.check_method == "curl":
+            return check_with_curl(args.host, args.port)
+        elif args.check_method == "ping":
+            return check_with_ping(args.host)
+        elif args.check_method == "nc":
+            return check_with_nc(args.host, args.port)
+        elif args.check_method == "cmd":
+            if not args.check_cmd:
+                log("ERROR: --check-cmd is required with --check-method cmd", args.quiet)
+                sys.exit(3)
+            return check_with_cmd(args.check_cmd)
+        return False
+
+    for attempt in range(1, args.retries + 1):
+        if run_check():
+            log(f"OK: Service is responding on {args.proto.upper()} {args.host}:{args.port}", args.quiet)
+            sys.exit(0)
+        log(f"Attempt {attempt}/{args.retries} failed.", args.quiet)
+        if attempt < args.retries:
+            time.sleep(args.delay)
+
+    log(f"FAIL: Service unreachable after {args.retries} attempts. Restarting {args.service}...", args.quiet)
+    restart_service(args.service)
+
+    # One final check after restart
+    if run_check():
+        log(f"OK: {args.service} is responding after restart.", args.quiet)
+        sys.exit(0)
+    else:
+        log(f"ERROR: {args.service} is still not responding after restart.", args.quiet)
+        sys.exit(2)
+
+if __name__ == "__main__":
+    main()