monitoring stuff

bundles/icinga2/files/check_by_sshmon

@@ -2,11 +2,7 @@
 
 UNKNOWN=3
 
-if [ -z "$SSHMON_TEST" ]
+if [ -z "$SSHMON_COMMAND" ]
-then
-    echo 'check_by_sshmon: Env SSHMON_TEST missing' >&2
-    exit $UNKNOWN
-elif [ -z "$SSHMON_COMMAND" ]
 then
     echo 'check_by_sshmon: Env SSHMON_COMMAND missing' >&2
     exit $UNKNOWN
@@ -16,7 +12,14 @@ then
     exit $UNKNOWN
 fi
 
-ssh sshmon@"$SSHMON_HOST" "sudo $SSHMON_COMMAND"
+if [ -z "$SSHMON_SUDO" ]
+then
+    PREFIX=""
+else
+    PREFIX="sudo "
+fi
+
+ssh sshmon@"$SSHMON_HOST" "$PREFIX$SSHMON_COMMAND"
 
 exitcode=$?
 

bundles/icinga2/files/conf.d/commands.conf

@@ -12,9 +12,9 @@ object CheckCommand "sshmon" {
 
     command = [ "/usr/lib/nagios/plugins/check_by_sshmon" ]
 
-    env.SSHMON_TEST = "1234"
     env.SSHMON_COMMAND = "$command$"
     env.SSHMON_HOST = "$address$"
+    env.SSHMON_SUDO = "$sudo$"
 }
 
 

bundles/monitored/metadata.py

@@ -1,9 +1,10 @@
 defaults = {
     'monitoring': {
         'services': {
-            'test': {
+            # 'test': {
-                'vars.command': '/bin/ls /',
+            #     'vars.command': '/bin/ls /',
-            },
+            #     'vars.sudo': True,
+            # },
         },
     },
 }
@@ -50,6 +51,7 @@ def user(metadata):
                 conf['vars.command']
                     for conf in metadata.get('monitoring/services').values()
                     if conf['check_command'] == 'sshmon'
+                    and conf.get('vars.sudo', None)
             },
         },
     }

bundles/nginx/metadata.py

@@ -122,8 +122,8 @@ def monitoring(metadata):
     return {
         'monitoring': {
             'services': {
-                f'HTTP {hostname}': {
+                hostname: {
-                    'vars.command': f"""/usr/bin/curl -X GET {quote(hostname + vhost.get('check_path', ''))} -IL --fail"""
+                    'vars.command': f"/usr/bin/curl -X GET -IL --fail {quote(hostname + vhost.get('check_path', ''))}",
                 }
                     for hostname, vhost in metadata.get('nginx/vhosts').items()
             },

bundles/systemd-timers/metadata.py

@@ -58,7 +58,7 @@ def monitoring(metadata):
         'monitoring': {
             'services': {
                 f'{name}.timer': {
-                    'vars.command': f'/usr/lib/nagios/plugins/check_systemd_timer {name}'
+                    'vars.command': f'/usr/lib/nagios/plugins/check_systemd_timer {name}',
                 }
                     for name in metadata.get('systemd-timers')
             },

bundles/systemd/metadata.py

@@ -11,7 +11,7 @@ defaults = {
 )
 def units(metadata):
     units = {}
-    
+
     for name, config in metadata.get('systemd/units').items():
         if '/' in name:
             continue
@@ -47,10 +47,10 @@ def units(metadata):
 )
 def services(metadata):
     services = {}
-    
+
     for name, config in metadata.get('systemd/services').items():
         extension = name.split('.')[-1]
-        
+
         if extension not in ['timer', 'service', 'mount', 'swap']:
             raise Exception(f'unknown extension: {extension}')
 
@@ -59,3 +59,20 @@ def services(metadata):
             'services': services,
         }
     }
+
+
+@metadata_reactor.provides(
+    'monitoring/services',
+)
+def monitoring(metadata):
+    return {
+        'monitoring': {
+            'services': {
+                name: {
+                    'vars.command': f"/bin/sh -c '/usr/bin/systemctl is-failed {name} && /usr/bin/systemctl status {name} && exit 2 || exit 0'"
+                }
+                    for name in metadata.get('systemd/units')
+                    if name.endswith('.service')
+            },
+        },
+    }