fix a few issues with check_media_cdn

check_media_cdn.py

@@ -18,30 +18,6 @@ from urllib3.exceptions import InsecureRequestWarning
 from checker import nagios
 from checker.synapse_client import send_image, write_login_details_to_disk
 
-parser = argparse.ArgumentParser(description='')
-parser.add_argument('--user', required=True, help='User ID for the bot.')
-parser.add_argument('--pw', required=True, help='Password for the bot.')
-parser.add_argument('--hs', required=True, help='Homeserver of the bot.')
-parser.add_argument('--admin-endpoint', required=True, help='Admin endpoint that will be called to purge media for this user.')
-parser.add_argument('--room', required=True, help='The room the bot should send its test messages in.')
-parser.add_argument('--check-domain', required=True, help='The domain that should be present.')
-parser.add_argument('--media-cdn-redirect', default='true', help='If set, the server must respond with a redirect to the media CDN domain.')
-parser.add_argument('--required-headers', nargs='*', help="If these headers aren't set to the correct value, critical. Use the format 'key=value")
-parser.add_argument('--auth-file', help="File to cache the bot's login details to.")
-parser.add_argument('--retries', type=int, default=11, help="It may take a few seconds for Synapse to send the uploaded file to S3. Retry this many times with 1 second interval between.")
-parser.add_argument('--timeout', type=float, default=90, help='Request timeout limit.')
-parser.add_argument('--warn', type=float, default=2.0, help='Manually set warn level.')
-parser.add_argument('--crit', type=float, default=2.5, help='Manually set critical level.')
-args = parser.parse_args()
-
-if args.media_cdn_redirect == 'true':
-    args.media_cdn_redirect = True
-elif args.media_cdn_redirect == 'false':
-    args.media_cdn_redirect = False
-else:
-    print('UNKNOWN: could not parse the value for --media-cdn-redirect')
-    sys.exit(nagios.UNKNOWN)
-
 
 def verify_media_header(header: str, header_dict: dict, good_value: str = None, warn_value: str = None, critical_value: str = None):
     """
@@ -70,7 +46,7 @@ def verify_media_header(header: str, header_dict: dict, good_value: str = None,
     return f'OK: {header} is present', nagios.OK  # with value "{header_value}"'
 
 
-async def main() -> None:
+async def main(args) -> None:
     exit_code = nagios.OK
 
     async def cleanup(client, test_image_path, image_event_id=None):
@@ -165,14 +141,19 @@ async def main() -> None:
         retried += 1
     end = time.time()
 
+    if args.ignore_first_attempt and retried > 0:
+        retried -= 1
+
+    exit_code = nagios.STATE_OK
     prints = []
 
     if r.status_code != 200 and not args.media_cdn_redirect:
         await cleanup(client, test_image_path, image_event_id=image_event_id)
-        prints.append(f'CRITICAL: status code is "{r.status_code}"')
+        print(f'CRITICAL: status code for CDN image is "{r.status_code}"')
+        print(f'Hosted URL is {r.url}')
         sys.exit(nagios.CRITICAL)
     else:
-        prints.append(f'OK: status code is "{r.status_code}"')
+        prints.append(f'OK: status code for CDN image is "{r.status_code}"')
 
     # Check domain
     if args.media_cdn_redirect:
@@ -228,6 +209,9 @@ async def main() -> None:
         print('WARNING: media CDN is bad.')
     elif exit_code == nagios.CRITICAL:
         print('CRITICAL: media CDN is bad.')
+    else:
+        raise Exception('No exit code matched')
+
     for msg in prints:
         print(msg)
 
@@ -235,15 +219,40 @@ async def main() -> None:
         print(clean_msg)
 
     if exit_code == nagios.STATE_OK:
-        print(f'Took {int(end - start)} seconds to with {retried} retries. | retries={retried}')
+        print(f'Took {int(end - start)} seconds to fetch the uploaded image with {retried} retries. | retries={retried}')
 
     sys.exit(exit_code)
 
 
 if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='')
+    parser.add_argument('--user', required=True, help='User ID for the bot.')
+    parser.add_argument('--pw', required=True, help='Password for the bot.')
+    parser.add_argument('--hs', required=True, help='Homeserver of the bot.')
+    parser.add_argument('--admin-endpoint', required=True, help='Admin endpoint that will be called to purge media for this user.')
+    parser.add_argument('--room', required=True, help='The room the bot should send its test messages in.')
+    parser.add_argument('--check-domain', required=True, help='The domain that should be present.')
+    parser.add_argument('--media-cdn-redirect', default='true', help='If set, the server must respond with a redirect to the media CDN domain.')
+    parser.add_argument('--required-headers', nargs='*', help="If these headers aren't set to the correct value, critical. Use the format 'key=value")
+    parser.add_argument('--auth-file', help="File to cache the bot's login details to.")
+    parser.add_argument('--retries', type=int, default=11, help="It may take a few seconds for Synapse to send the uploaded file to S3. Retry this many times with 1 second interval between.")
+    parser.add_argument('--timeout', type=float, default=90, help='Request timeout limit.')
+    parser.add_argument('--warn', type=float, default=2.0, help='Manually set warn level.')
+    parser.add_argument('--crit', type=float, default=2.5, help='Manually set critical level.')
+    parser.add_argument('--ignore-first-attempt', action='store_true', help='Ignore the first attempt at fetching the image from the media CDN because the server may not have cached the file if it has never been requested before.')
+    args = parser.parse_args()
+
+    if args.media_cdn_redirect == 'true':
+        args.media_cdn_redirect = True
+    elif args.media_cdn_redirect == 'false':
+        args.media_cdn_redirect = False
+    else:
+        print('UNKNOWN: could not parse the value for --media-cdn-redirect')
+        sys.exit(nagios.UNKNOWN)
+
     try:
-        asyncio.run(main())
+        asyncio.run(main(args))
     except Exception as e:
         print(f'UNKNOWN: exception\n{e}')
-        print(traceback.format_exc())
+        traceback.print_exc()
         sys.exit(nagios.UNKNOWN)

check_mysql_slave.py

@@ -9,6 +9,15 @@ from checker import nagios, print_icinga2_check_status
 from checker.types import try_int
 
 """
+## MySQL User Setup ##
+
+Attention: The DB-user you type in must have CLIENT REPLICATION rights on the DB-server. Example:
+    GRANT REPLICATION CLIENT on *.* TO 'nagios'@'monitoringhost' IDENTIFIED BY 'secret';
+
+If you use MariaDB 10.5 or newer, the DB user must have REPLICA MONITOR rights:
+    GRANT REPLICA MONITOR ON *.* TO 'nagios'@'monitoringhost' IDENTIFIED BY 'secret';
+
+
 ## Usage example ##
 
 Target delay = 300 seconds