initial version

README.md

@@ -0,0 +1,11 @@
+Простой монитор доступности виртуальных машин.
+
+Следит по IP и перезапускает если недостуна
+
+- запускать на хосте proxmox, по крону, раз в минуту (например: crontab -e; */1 * * * * /usr/bin/python3 /root/svs-proxmox-watchdog.py 1>/dev/null 2>/dev/null)
+
+- необходим первичный json со списком виртуалок для мониторинга, вида:
+
+List of dicts: [ {}, {}, ... ]
+
+Dict: { "name" : "", "id" : proxmox_vm_id, "ip" : "x.x.x.x", "down" : 0 }

svs-proxmox-watchdog.py

@@ -0,0 +1,53 @@
+#!/usr/bin/python3
+
+import subprocess as sp
+import json
+import time, re
+
+# Read DB from file
+def read_json_from_file(filepath):
+    with open(filepath, 'r') as file:
+        data = json.load(file)
+    return data
+
+# Alive check for host
+def ping(ip):
+   status,result = sp.getstatusoutput("ping -c1 -w2 " + ip)
+   return status
+
+def vmstat(id):
+   vm_status = "dnd"
+   status,result = sp.getstatusoutput("qm status 114")
+   match = re.search(r"status: (.+)", result)
+   if match:
+     vm_status = match.group(1)
+   return vm_status
+
+# timeout
+shutdown_timeout = 5
+
+# DB file
+file_path = '/root/svs-watchdog.json'
+
+# Read DB to mem
+vms = read_json_from_file(file_path)
+
+# Monitor VMs
+for vm in vms:
+         ip = vm['ip']
+         down = vm['down']
+
+         if ping(ip) == 0: 
+            vm['down'] = 0
+         else:
+            vm['down'] = down + 1
+            if vmstat(vm['id']) == 'runnnig' and vm['down'] >= shutdown_timeout:
+                 status,result = sp.getstatusoutput("qm stop " + str(vm['id']) + " --overrule-shutdown 1")
+            else:
+               if vmstat(vm['id']) == 'stopped' and vm['down'] >= shutdown_timeout:
+                 status,result = sp.getstatusoutput("qm start " + str(vm['id']))
+                 vm['down'] = 0
+
+# Update the status
+with open(file_path, "w") as file:
+    json.dump(vms, file)