bundles/sshmon: better cpu check

bundles/sshmon/files/check_cpu_stats

@@ -4,27 +4,30 @@ from re import findall
 from subprocess import check_output
 from sys import exit
 
+ITERATIONS = 10
+
 try:
     top_output = None
 
-    for line in check_output(['top', '-b', '-n1', '-d1']).decode('UTF-8').splitlines():
-        if line.lower().strip().startswith('%cpu'):
-            top_output = line.lower().split(':', 2)[1]
-            break
-
-    if not top_output:
-        print('%cpu not found in top output')
-        exit(3)
+    top_output = check_output(rf"top -b -n{ITERATIONS} -d1 | grep -i '^%cpu'", shell=True).decode('UTF-8')
 
     cpu_usage = {}
     for value, identifier in findall('([0-9\.\,]{3,5}) ([a-z]{2})', top_output):
-        cpu_usage[identifier] = float(value.replace(',', '.'))
+        if identifier not in cpu_usage:
+            cpu_usage[identifier] = 0.0
+        cpu_usage[identifier] += float(value.replace(',', '.'))
+
+    output = []
+    for identifier, value_added in cpu_usage.items():
+        value = value_added / ITERATIONS
+        output.append(f"{value:.2f} {identifier}")
+        cpu_usage[identifier] = value
+
+    print(f"Average over {ITERATIONS} seconds: " + ", ".join(output))
 
     warn = set()
     crit = set()
 
-    print(top_output)
-
     # steal
     if cpu_usage['st'] > 10:
         crit.add('CPU steal is {}% (>10%)'.format(cpu_usage['st']))

bundles/sshmon/metadata.py

@@ -19,6 +19,8 @@ defaults = {
             'services': {
                 'CPU': {
                     'command_on_monitored_host': '/usr/local/share/icinga/plugins/check_cpu_stats',
+                    # takes samples over 10 seconds
+                    'vars.sshmon_timeout': 20
                 },
                 'LOAD': {
                     'command_on_monitored_host': '/usr/lib/nagios/plugins/check_load -r -w 4,2,1 -c 8,4,2',