add check_graylog_index_size, fix other stuff

check_graylog.py

@@ -42,8 +42,7 @@ def main():
     parser.add_argument('--crit-notif', action='store_true', help='Return critical when there are notifications')
     parser.add_argument('--ignore-update-notif', action='store_true', help='Ignore any update notifications')
     parser.add_argument('--html', action='store_true', help='Print HTML')
-    parser.add_argument('--cluster-metrics', action='store_true',
+    parser.add_argument('--cluster-metrics', action='store_true', help='Also gather cluster metrics and check for notifications')
-                        help='Also gather cluster metrics and check for notifications')
     args = parser.parse_args()
 
     base_url = args.url.strip('/')

check_graylog_index_size.py

@@ -0,0 +1,63 @@
+#!/usr/bin/env python3
+import argparse
+import sys
+import traceback
+
+import requests
+
+from checker import nagios
+from checker.result import quit_check
+from checker.string import strip_non_numbers
+
+
+def main(args):
+    data = {}  # make pycharm happy
+    try:
+        r = requests.get('http://localhost:9200/_cat/nodes?v&h=n,ip,disk*&format=json')
+        r.raise_for_status()
+        data = r.json()[0]
+    except Exception as e:
+        quit_check(f'Failed to reach Elasticsearch: {e}', nagios.STATE_CRIT)
+
+    perfdata = {
+        'disk_used': {
+            'value': data['disk.used'],
+            'min': 0,
+        },
+        'disk_total': {
+            'value': data['disk.total'],
+            'min': 0,
+        },
+        'disk_avail': {
+            'value': data['disk.avail'],
+        },
+        'disk_used_percent': {
+            'value': data['disk.used_percent'],
+            'unit': '%'
+        }
+    }
+
+    used_percent_float = float(data['disk.used_percent'])
+    warn_percent = args.size_warn_percent * 100
+    crit_percent = args.size_crit_percent * 100
+    exit_str = f"{data['disk.used_percent']}% disk used ({data['disk.used'].upper()} / {data['disk.total'].upper()}). {data['disk.avail'].upper()} available."
+    exit_code = nagios.STATE_OK
+
+    if used_percent_float >= crit_percent:
+        exit_code = nagios.STATE_CRIT
+    elif used_percent_float >= warn_percent:
+        exit_code = nagios.STATE_CRIT
+    quit_check(exit_str, exit_code, perfdata)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Check the Graylog index size.')
+    parser.add_argument('--size-warn-percent', default=50, type=float)
+    parser.add_argument('--size-crit-percent', default=65, type=float)
+    args = parser.parse_args()
+    try:
+        main(args)
+    except Exception as e:
+        print(f'UNKNOWN: exception "{e}"')
+        print(traceback.format_exc())
+        sys.exit(nagios.STATE_UNKNOWN)

check_mysql_slave.py

@@ -60,15 +60,18 @@ def main(args):
         exit_code = nagios.STATE_OK
         exit_msg = []
 
-        if slave_sql_running != 'Yes':
-            exit_code = nagios.STATE_CRIT
-            exit_msg.append('Slave SQL is not running!')
         if slave_io_running != 'Yes':
             exit_code = nagios.STATE_CRIT
             exit_msg.append('Slave IO is not running!')
         if last_io_error:
             exit_code = nagios.STATE_CRIT
             exit_msg.append(f'Last IO Error: {last_io_error.strip(".")}.')
+        if slave_sql_running != 'Yes':
+            exit_code = nagios.STATE_CRIT
+            exit_msg.append('Slave SQL is not running!')
+        if last_sql_error:
+            exit_code = nagios.STATE_CRIT
+            exit_msg.append(f'Last SQL Error: {last_sql_error.strip(".")}.')
 
         if exit_code == nagios.STATE_OK:
             # Only replication delay if everything else is healthy.

check_openwrt_bssid.py

@@ -100,7 +100,10 @@ def main(args):
 
         # ============================================
 
-        ap_data = cells[args.target_mac]
+        ap_data = cells.get(args.target_mac)
+        if not ap_data:
+            quit_check(f'AP address not found: {args.target_mac}', nagios.STATE_CRIT)
+
         computed_quality = int(try_float(ap_data['quality'][0]) / try_float(ap_data['quality'][1]) * 100)
 
         chan_width = ap_data['chan_width'].strip(' MHz')

checker/string.py

@@ -0,0 +1,2 @@
+def strip_non_numbers(input_string):
+    return ''.join(char for char in input_string if char.isdigit())