check_greylog: alert notifications, better metric names, better filesizes

check_graylog.py

@@ -10,6 +10,7 @@ from urllib3.exceptions import InsecureRequestWarning
 import checker.nagios
 from checker import print_icinga2_check_status
 from checker.linuxfabric.base import get_state
+from checker.units import human_readable_size
 
 
 def main():
@@ -20,6 +21,7 @@ def main():
     parser.add_argument('--warn-mem', type=int, default=75, help='Percentage of JVM memory used for warm')
     parser.add_argument('--crit-mem', type=int, default=100, help='Percentage of JVM memory used for critical')
     parser.add_argument('--insecure', action='store_false', help="Don't verify SSL")
+    parser.add_argument('--crit-notif', action='store_true', help='Return critical when there are notifications')
     args = parser.parse_args()
 
     if not args.insecure:
@@ -78,27 +80,19 @@ def main():
             '.').replace('-', '_').replace('.', '_')
         value = None
         if 'value' in metric['metric']:
-            # perfdata.append(f'{name}={metric["metric"]["value"]}')
             value = metric["metric"]["value"]
         elif 'count' in metric['metric']:
-            # perfdata.append(f'{name}={metric["metric"]["count"]}')
             value = metric["metric"]["count"]
         elif 'rate' in metric['metric']:
-            # perfdata.append(f'{name}_total={metric["metric"]["rate"]["total"]}')
-            # perfdata.append(f'{name}_mean={metric["metric"]["rate"]["mean"]}')
-            # perfdata.append(f'{name}_five_minute={metric["metric"]["rate"]["five_minute"]}')
-            # perfdata.append(f'{name}_fifteen_minute={metric["metric"]["rate"]["fifteen_minute"]}')
-            # perfdata.append(f'{name}_one_minute={metric["metric"]["rate"]["one_minute"]}')
             value = metric["metric"]["rate"]["one_minute"]
             name = f'{name}_one_minute'
-        # if isinstance(value, float):
-        #     value = round(value, 1)
         value = int(value)
         metrics_data[name] = value
 
     jvm_mem_usage_warn = int(metrics_data['jvm_memory_heap_max'] / int(100 / args.warn_mem))
     jvm_mem_usage_crit = int(metrics_data['jvm_memory_heap_max'] / int(100 / args.crit_mem))
 
+    # Some metric names are changed for better readability
     perfdata = {
         'throughput_input_1_sec_rate': {
             'value': int(metrics_data['throughput_input_1_sec_rate']),
@@ -108,28 +102,34 @@ def main():
             'value': int(metrics_data['throughput_output_1_sec_rate']),
             'min': 0,
         },
-        'incoming_messages_one_minute': {
+        'incoming_messages_rate_per_sec_one_minute': {
             'value': metrics_data['incomingMessages_one_minute'],
             'min': 0,
         },
-        'open_connections': {
+        'connections': {
             'value': metrics_data['open_connections'],
             'min': 0,
         },
-        'total_connections': {
+        'network_out_total_1sec': {
-            'value': metrics_data['total_connections'],
-            'min': 0,
-        },
-        'written_bytes_1sec': {
             'value': metrics_data['written_bytes_1sec'],
             'min': 0,
             'unit': 'B',
         },
-        'read_bytes_1sec': {
+        'network_out_total_total': {
+            'value': metrics_data['written_bytes_total'],
+            'min': 0,
+            'unit': 'B',
+        },
+        'network_in_1sec': {
             'value': metrics_data['read_bytes_1sec'],
             'min': 0,
             'unit': 'B',
         },
+        'network_in_total': {
+            'value': metrics_data['read_bytes_total'],
+            'min': 0,
+            'unit': 'B',
+        },
         'entries_uncommitted': {
             'value': metrics_data['journal_entries_uncommitted'],
             'min': 0,
@@ -173,9 +173,15 @@ def main():
     if jvm_mem_usage_state != checker.nagios.STATE_OK:
         text_result += f' JVM memory usage is high!'
 
-    text_result = text_result + f' JVM memory usage: {int((perfdata["jvm_memory_used"]["value"] / metrics_data["jvm_memory_heap_max"]) * 100)}%, incoming_messages_one_minute: {perfdata["incoming_messages_one_minute"]["value"]}, open_connections: {perfdata["open_connections"]["value"]}' + '\n' + notif_str
-
     exit_code = max(checker.nagios.STATE_OK, jvm_mem_usage_state)
+
+    if notifications['total'] > 0:
+        text_result += f' There are notifications!'
+        if args.crit_notif:
+            exit_code = checker.nagios.STATE_CRIT
+
+    text_result = text_result + f' JVM memory usage: {int((perfdata["jvm_memory_used"]["value"] / metrics_data["jvm_memory_heap_max"]) * 100)}%, incoming rate (events/second for last minute): {perfdata["incoming_messages_rate_per_sec_one_minute"]["value"]}, connections: {perfdata["connections"]["value"]}, total network in: {human_readable_size(perfdata["network_in_total"]["value"], decimal_places=0)}' + '\n' + notif_str
+
     print_icinga2_check_status(text_result, exit_code, perfdata)
     sys.exit(exit_code)
 

checker/units.py

@@ -1,3 +1,6 @@
+from math import log2, log10
+from typing import Union
+
 from hurry.filesize import size
 
 
@@ -18,3 +21,39 @@ def filesize(bytes: int, spaces: bool = True, formatter: bool = True):
         return x
     else:
         return x.replace(' ', '')
+
+
+def human_readable_size(size: Union[int, float], bits=False, decimal_places: int = 2, base: int = 10):
+    # Define the units
+    units = {False: {2: ['bytes', 'KiB', 'MiB', 'GiB', 'TiB', 'PiB', 'EiB', 'ZiB', 'YiB'],
+                     10: ['bytes', 'KB', 'MB', 'GB', 'TB', 'PB', 'EB', 'ZB', 'YB']},
+             True: {2: ['bits', 'Kib', 'Mib', 'Gib', 'Tib', 'Pib', 'Eib', 'Zib', 'Yib'],
+                    10: ['bits', 'Kb', 'Mb', 'Gb', 'Tb', 'Pb', 'Eb', 'Zb', 'Yb']}}
+
+    # Convert bytes to bits if needed
+    if bits:
+        size *= 8
+
+    # Determine the unit
+    if size == 0:
+        return '0 ' + units[bits][base][0]
+    else:
+        if base == 2:
+            log = int(log2(size))
+            exp = log // 10
+        elif base == 10:
+            log = int(log10(size))
+            exp = log // 3
+        else:
+            raise ValueError("Invalid base. Use either 2 or 10.")
+
+        if exp >= len(units[bits][base]):
+            exp = len(units[bits][base]) - 1
+        size /= base ** (exp * (10 if base == 2 else 3))
+
+        if decimal_places == 0:
+            size = int(size)
+        else:
+            round(size, decimal_places)
+
+        return f'{size} {units[bits][base][exp]}'