cyberes
/
icinga2-checks
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

check_scrutiny_disks: fix not no data

This commit is contained in:
Cyberes 2023-06-12 10:21:05 -06:00
parent 3e86cc3614
commit 847999f43b
Signed by: cyberes
GPG Key ID: 6B4A33836A9500FE
1 changed files with 41 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

55
check_scrutiny_disks.py
View File

@ -20,7 +20,8 @@ def get_disk_wwn_ids(ignore_non_smart: bool = False) -> List[str] or bool:
device = parts[0]
device_type = parts[2].replace('scsi', 'sat,auto')
try:
smart_supported = subprocess.check_output(["sudo", "smartctl", "-i", device, "-d", device_type]).decode("utf-8")
smart_supported = subprocess.check_output(
["sudo", "smartctl", "-i", device, "-d", device_type]).decode("utf-8")
if "SMART support is: Enabled" in smart_supported:
wwn_line = [line for line in smart_supported.split("\n") if "LU WWN Device Id" in line]
wwn_id = '0x' + wwn_line[0].replace('LU WWN Device Id: ', '').replace(' ', '')
@ -45,20 +46,26 @@ def get_smart_health(wwn_id: str, scrutiny_endpoint: str) -> dict:
if response.status_code == 200:
return response.json()
elif response.status_code == 404:
print(f"Disk {wwn_id} not found on Scrutiny")
return {}
return {
'fetch_error': f"Disk {wwn_id} not found on Scrutiny"
}
else:
print(f"Scrutiny Error {response.status_code} for disk {wwn_id}: {response.text}")
return {}
return {
'fetch_error': f"Error {response.status_code} for disk {wwn_id}: {response.text}"
}
def main(args):
results = {}
wwn_ids = get_disk_wwn_ids(args.ignore_non_smart)
metics_out_of_date = False
metrics_out_of_date = False
for wwn_id in wwn_ids:
smart_health = get_smart_health(wwn_id, args.scrutiny_endpoint)
if smart_health.get('fetch_error'):
print('UNKNOWN -', smart_health.get('fetch_error'))
sys.exit(nagios.UNKNOWN)
name = f'/dev/{smart_health["data"]["device"]["device_name"]} {wwn_id}' # differentiate disks in RAID arrays
results[name] = {
@ -73,7 +80,7 @@ def main(args):
last_updated = datetime.strptime(smart_health['data']['device']['UpdatedAt'][:-4] + 'Z', '%Y-%m-%dT%H:%M:%S.%fZ')
if datetime.utcnow() - timedelta(hours=args.time_delta_limit) > last_updated:
metics_out_of_date = True
metrics_out_of_date = True
if smart_health and len(smart_health['data']['smart_results']):
try:
@ -91,10 +98,10 @@ def main(args):
# continue
values['attribute_name'] = metadata[attribute_id]['display_name']
values['metadata'] = metadata[attribute_id]
if 'observed_thresholds' in values['metadata'].keys():
del values['metadata']['observed_thresholds']
results[name]['failed_attributes'].append(values)
results[name]['status'] = 'good'
else:
results[name]['status'] = 'no data'
@ -130,7 +137,7 @@ def main(args):
dd = '<dd>' if args.html else '\t- '
dds = '</dd>' if args.html else ''
out_of_date_str = f'metrics are >{args.time_delta_limit} hrs out of date ' if metics_out_of_date else ''
out_of_date_str = f'metrics are >{args.time_delta_limit} hrs out of date ' if metrics_out_of_date else ''
return_code = nagios.OK
if len(crit_disks):
@ -139,7 +146,8 @@ def main(args):
x = f' and {len(warn_disks)} {"warnings" if len(results) > 1 else "warning"}'
else:
x = ''
print(f'CRITICAL: {out_of_date_str + "and " if len(out_of_date_str) else ""}{len(crit_disks)} {"errors" if len(crit_disks) > 1 else "error"}{x}')
print(
f'CRITICAL: {out_of_date_str + "and " if len(out_of_date_str) else ""}{len(crit_disks)} {"errors" if len(crit_disks) > 1 else "error"}{x}')
print('<dl>')
print(f'{dt}Disks with Errors:{dts}')
@ -180,19 +188,38 @@ def main(args):
sys.exit(return_code)
def is_smartmontools_installed():
try:
result = subprocess.run(['dpkg', '-s', 'smartmontools'], stdout=subprocess.PIPE, stderr=subprocess.PIPE,
check=True, text=True)
if 'Status: install ok installed' in result.stdout:
return True
else:
return False
except subprocess.CalledProcessError:
return False
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='')
parser.add_argument('--scrutiny-endpoint', required=True, help='Base URL for scrutiny.')
parser.add_argument('--time-delta-limit', default=24, type=int, help='The Scrutiny data must not be older than this many hours. Default: 24.')
parser.add_argument('--warn-non-critical', action='store_true', help='Warn when a non-critical metric is marked as failed.')
parser.add_argument('--time-delta-limit', default=24, type=int,
help='The Scrutiny data must not be older than this many hours. Default: 24.')
parser.add_argument('--warn-non-critical', action='store_true',
help='Warn when a non-critical metric is marked as failed.')
parser.add_argument('--html', action='store_true', help='Print HTML.')
parser.add_argument('--pretty-url', help='The pretty URL to link to when printing HTML.')
parser.add_argument('--ignore-non-smart', action='store_true', help="Ignore any non-SMART devices and any devices that error when reading SMART.")
parser.add_argument('--ignore-non-smart', action='store_true',
help="Ignore any non-SMART devices and any devices that error when reading SMART.")
parser.add_argument('--dont-warn-no-data', action='store_true', help="Don't warn if there is no data for a disk.")
args = parser.parse_args()
if args.html and not args.pretty_url:
print('UKNOWN: when using --html you must also set --pretty-url')
print('UKNOWN - when using --html you must also set --pretty-url')
sys.exit(nagios.UNKNOWN)
if not is_smartmontools_installed():
print('UNKNOWN - smartmontools is not installed.')
sys.exit(nagios.UNKNOWN)
args.scrutiny_endpoint = args.scrutiny_endpoint.strip('/')