504 lines
19 KiB
Python
Executable File
504 lines
19 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
"""SNMP based Synology DMS / DMS UC checks"""
|
|
"""dependencys:
|
|
- pip3 install nagiosplugin
|
|
- pip3 install argparse
|
|
- pip3 install easysnmp
|
|
- https://easysnmp.readthedocs.io/en/latest/index.html
|
|
- https://easysnmp.readthedocs.io/en/latest/session_api.html
|
|
"""
|
|
|
|
__version__ = '0.9.0'
|
|
__author__ = 'anima'
|
|
|
|
# imports
|
|
import logging
|
|
import argparse
|
|
import nagiosplugin
|
|
from easysnmp import Session
|
|
|
|
# log settings
|
|
logging.basicConfig(format='[%(asctime)s] %(levelname)s %(message)s', level=logging.INFO)
|
|
|
|
def init_snmp_session(hostname: str = 'localhost', snmp_version: str | int = '2', snmp_community: str = 'public',
|
|
snmp_security_level: str = 'auth_with_privacy', snmp_security_username: str = 'monitoring',
|
|
snmp_auth_protocol: str = 'SHA256', snmp_auth_password: str | None = None,
|
|
snmp_privacy_protocol: str = 'AES', snmp_privacy_password: str | None = None
|
|
) -> Session:
|
|
"""init a easysnmp session
|
|
|
|
Args:
|
|
hostname (str, optional): hostname or ip of host. Defaults to 'localhost'.
|
|
snmp_version (str, optional): version of snmp (1, 2, 3). Defaults to 2.
|
|
snmp_community (str, only need for v1 & v2): snmp community if snmp v1 or v2 used (any). Defaults to 'public'.
|
|
snmp_security_level (str, only need for v3): security level (no_auth_or_privacy, auth_without_privacy or auth_with_privacy). Defaults to 'auth_with_privacy'.
|
|
snmp_security_username (str, only need for v3): security name (any). Defaults to 'monitoring'.
|
|
snmp_auth_protocol (str, only need for v3): authentication protocol (MD5, SHA256, SHA512). Defaults to 'SHA256'.
|
|
snmp_auth_password (str, only need for v3): authentication passphrase (any). Defaults to None.
|
|
snmp_privacy_protocol (str, only need for v3): privacy protocol (AES, DES). Defaults to 'AES'.
|
|
snmp_privacy_password (str, only need for v3): privacy passphrase (any). Defaults to None.
|
|
|
|
Return:
|
|
easysnmp.Session: session for get or walk actions
|
|
"""
|
|
# check if snmp version an valid str or int and saves as int
|
|
valid_versions: list[str, int] = ['1','2','3', 1, 2, 3]
|
|
if snmp_version not in valid_versions:
|
|
raise ValueError(f'{snmp_version=} non of {valid_versions=}')
|
|
else:
|
|
snmp_version = int(snmp_version)
|
|
|
|
if snmp_version == '3':
|
|
# check if snmp security level valid
|
|
valid_security_levels: list[str] = ['no_auth_or_privacy', 'auth_without_privacy', 'auth_with_privacy']
|
|
if snmp_security_level not in valid_security_levels:
|
|
raise ValueError(f'{snmp_security_level=} is non of {valid_security_levels=}')
|
|
|
|
# check if auth protocol valid
|
|
if snmp_security_level.startswith('auth'):
|
|
valid_auth_protocols: list[str] = ['MD5', 'SHA', 'SHA256', 'SHA512']
|
|
if snmp_auth_protocol not in valid_auth_protocols:
|
|
raise ValueError(f'{snmp_auth_protocol=} is non of {valid_auth_protocols=}')
|
|
|
|
if snmp_auth_password is None:
|
|
raise ValueError(f'{snmp_auth_password=} is not set')
|
|
|
|
# check if privacy protocol valid
|
|
if not snmp_security_level.startswith('no'):
|
|
valid_privacy_protocols: list[str] = ['AES', 'DES']
|
|
if snmp_privacy_protocol not in valid_privacy_protocols:
|
|
raise ValueError(f'{snmp_privacy_protocol=} is non of {valid_privacy_protocols=}')
|
|
|
|
if snmp_privacy_password is None:
|
|
raise ValueError(f'{snmp_privacy_password=} is not set')
|
|
|
|
|
|
if snmp_version == 3: # 2025-01-02: only 'auth_with_privacy' is tested
|
|
session = Session(hostname=hostname, version=snmp_version,
|
|
security_level=snmp_security_level, security_username=snmp_security_username,
|
|
privacy_protocol=snmp_privacy_protocol, privacy_password=snmp_privacy_password,
|
|
auth_protocol=snmp_auth_protocol, auth_password=snmp_auth_password)
|
|
else:
|
|
session = Session(hostname=hostname, version=snmp_version, community=snmp_community)
|
|
return session
|
|
|
|
|
|
#
|
|
# Synology [DSM & DSM UC] Checks disk staus (baseclass)
|
|
#
|
|
class SNMPSynologyDisk():
|
|
def __init__(self, session):
|
|
self.session = session
|
|
|
|
def get_disks(self):
|
|
disks = list()
|
|
baseoid = '.1.3.6.1.4.1.6574.2.1.1'
|
|
oids = dict()
|
|
oids['index'] = '1'
|
|
oids['id'] = '2'
|
|
oids['model'] = '3'
|
|
oids['type'] = '4'
|
|
oids['status'] = '5'
|
|
oids['temp'] = '6'
|
|
oids['role'] = '7'
|
|
oids['retry'] = '8'
|
|
oids['badSectors'] = '9'
|
|
oids['identFail'] = '10'
|
|
oids['remainLife'] = '11'
|
|
oids['name'] = '12'
|
|
oids['health'] = '13'
|
|
|
|
disk_ids = list()
|
|
results = self.session.walk(baseoid)
|
|
for result in results:
|
|
if '6574.2.1.1.1.' in result.oid:
|
|
disk_ids.append(result.value)
|
|
|
|
for disk in disk_ids:
|
|
tmp_disk = dict()
|
|
for result in results:
|
|
for name, oid in oids.items():
|
|
if '6574.2.1.1.' + oid + '.' + disk in result.oid:
|
|
tmp_disk[name] = result.value
|
|
break
|
|
disks.append(tmp_disk)
|
|
|
|
return disks
|
|
|
|
|
|
#
|
|
# Synology [DSM & DSM UC] System partition status.
|
|
#
|
|
class SNMPSynologySystemResource(nagiosplugin.Resource):
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
def probe(self) -> list:
|
|
"""check system status (normal or failed)
|
|
|
|
Returns:
|
|
nagiosplugin.Metric: single metric element (return)
|
|
"""
|
|
baseoid = '.1.3.6.1.4.1.6574.1'
|
|
oids = dict()
|
|
oids['status'] = '.1.0'
|
|
oids['model'] = '.5.1.0'
|
|
oids['sn'] = '.5.2.0'
|
|
# oids['hostname'] = '.1.0'
|
|
|
|
result = dict()
|
|
for key, oid in oids.items():
|
|
result[key] = self.session.get(baseoid + oid).value
|
|
return nagiosplugin.Metric(name='status', value=result, context='system_context')
|
|
|
|
|
|
class SNMPSynologySystemContext(nagiosplugin.Context):
|
|
def __init__(self, name):
|
|
super().__init__(name, fmt_metric='{name} is', result_cls=SNMPSynologySystemResult)
|
|
|
|
def evaluate(self, metric, resource):
|
|
if metric.value['status'] == '2':
|
|
return self.result_cls(nagiosplugin.Critical, "critical", metric)
|
|
elif metric.value['status'] == '1':
|
|
return self.result_cls(nagiosplugin.Ok, "ok", metric)
|
|
return self.result_cls(nagiosplugin.Unknown, "unknown", metric)
|
|
|
|
|
|
class SNMPSynologySystemResult(nagiosplugin.Result):
|
|
def __str__(self):
|
|
if self.metric.value['status'] == '1': status = 'normal'
|
|
else: status = 'failed'
|
|
return f'{self.metric.value["model"]} (SN: {self.metric.value["sn"]}) is in status {status}!'
|
|
|
|
|
|
#
|
|
# Synology [DSM & DSM UC] Temperature of NAS
|
|
#
|
|
class SNMPSynologyTemperatureResource(nagiosplugin.Resource, SNMPSynologyDisk):
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
def probe(self) -> list:
|
|
"""check system temperature in °C
|
|
|
|
Returns:
|
|
nagiosplugin.Metric: single metric element (return)
|
|
"""
|
|
baseoid = '.1.3.6.1.4.1.6574.1'
|
|
oids = dict()
|
|
result = self.session.get(baseoid + '.2.0').value
|
|
yield nagiosplugin.Metric(name='temp_system', value=int(result), uom='°C', context='temperature_scalar_context')
|
|
disks = self.get_disks()
|
|
for disk in disks:
|
|
yield nagiosplugin.Metric(name='temp_' + disk['id'].lower().replace(' ', '_'), value=int(disk['temp']), uom='°C', context='temperature_scalar_context')
|
|
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
|
|
#
|
|
# Synology [DSM & DSM UC] Returns error if power supplies fail
|
|
#
|
|
class SNMPSynologyPowerSupplyResource(nagiosplugin.Resource):
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
def probe(self) -> list:
|
|
"""check power supply status (normal or failed)
|
|
|
|
Returns:
|
|
nagiosplugin.Metric: single metric element (return)
|
|
"""
|
|
baseoid = '.1.3.6.1.4.1.6574.1'
|
|
result = self.session.get(baseoid + '.3.0').value
|
|
return nagiosplugin.Metric(name='status', value=result, context='powersupply_context')
|
|
|
|
|
|
class SNMPSynologyPowerSupplyContext(nagiosplugin.Context):
|
|
def __init__(self, name):
|
|
super().__init__(name, fmt_metric='{name} is', result_cls=SNMPSynologyPowerSupplyResult)
|
|
|
|
def evaluate(self, metric, resource):
|
|
if metric.value == '2':
|
|
return self.result_cls(nagiosplugin.Critical, "critical", metric)
|
|
elif metric.value == '1':
|
|
return self.result_cls(nagiosplugin.Ok, "ok", metric)
|
|
return self.result_cls(nagiosplugin.Unknown, "unknown", metric)
|
|
|
|
|
|
class SNMPSynologyPowerSupplyResult(nagiosplugin.Result):
|
|
def __str__(self):
|
|
if self.metric.value == '1': status = 'normal'
|
|
else: status = 'failed'
|
|
return f'Power Supply is in status {status}!'
|
|
|
|
|
|
#
|
|
# Synology [DSM & DSM UC] returns error if system fan fails
|
|
#
|
|
class SNMPSynologyFansResource(nagiosplugin.Resource):
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
def probe(self) -> list:
|
|
"""check fans status (normal or failed)
|
|
|
|
Returns:
|
|
Generator[nagisplugin.Metric]: multiple metric elements (yield)
|
|
"""
|
|
baseoid = '.1.3.6.1.4.1.6574.1'
|
|
fans = dict()
|
|
fans['system'] = '.4.1.0'
|
|
fans['cpu'] = '.4.2.0'
|
|
|
|
for fan, oid in fans.items():
|
|
result = self.session.get(baseoid + oid).value
|
|
yield nagiosplugin.Metric(name=fan, value=result, context='fan_context')
|
|
|
|
|
|
class SNMPSynologyFansContext(nagiosplugin.Context):
|
|
def __init__(self, name):
|
|
super().__init__(name, fmt_metric='{name} is', result_cls=SNMPSynologyFansResult)
|
|
|
|
def evaluate(self, metric, resource):
|
|
if metric.value == '2':
|
|
return self.result_cls(nagiosplugin.Critical, "critical", metric)
|
|
elif metric.value == '1':
|
|
return self.result_cls(nagiosplugin.Ok, "ok", metric)
|
|
return self.result_cls(nagiosplugin.Unknown, "unknown", metric)
|
|
|
|
|
|
class SNMPSynologyFansResult(nagiosplugin.Result):
|
|
def __str__(self):
|
|
if self.metric.value == '1': status = 'normal'
|
|
else: status = 'failed'
|
|
return f'{self.metric.name} fan is in status {status}!'
|
|
|
|
|
|
class SNMPSynologyFansSummary(nagiosplugin.Summary):
|
|
def verbose(self, results):
|
|
result_str = ''
|
|
for result in results:
|
|
result_str += f'{str(result)}\n'
|
|
return result_str
|
|
|
|
def ok(self, results):
|
|
return
|
|
|
|
def problem(self, results):
|
|
return
|
|
|
|
|
|
#
|
|
# Synology [DSM & DSM UC] Checks whether a new version or update of DSM is available
|
|
#
|
|
class SNMPSynologyFirmwareResource(nagiosplugin.Resource):
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
def probe(self) -> list:
|
|
"""check if firmware update available
|
|
|
|
Returns:
|
|
nagiosplugin.Metric: single metric element (return)
|
|
"""
|
|
baseoid = '.1.3.6.1.4.1.6574.1'
|
|
oids = dict()
|
|
oids['version'] = '.5.3.0'
|
|
oids['status'] = '.5.4.0'
|
|
oids['model'] = '.5.1.0'
|
|
oids['sn'] = '.5.2.0'
|
|
|
|
result = dict()
|
|
for key, oid in oids.items():
|
|
result[key] = self.session.get(baseoid + oid).value
|
|
return nagiosplugin.Metric(name='status', value=result, context='firmware_context')
|
|
|
|
|
|
class SNMPSynologyFirmwareContext(nagiosplugin.Context):
|
|
def __init__(self, name):
|
|
super().__init__(name, fmt_metric='{name} is', result_cls=SNMPSynologyFirmwareResult)
|
|
|
|
def evaluate(self, metric, resource):
|
|
if metric.value['status'] == '1':
|
|
return self.result_cls(nagiosplugin.Critical, "critical", metric)
|
|
elif metric.value['status'] == '2':
|
|
return self.result_cls(nagiosplugin.Ok, "ok", metric)
|
|
elif metric.value['status'] in ['3', '4']:
|
|
return self.result_cls(nagiosplugin.Warn, "warning", metric)
|
|
return self.result_cls(nagiosplugin.Unknown, "unknown", metric)
|
|
|
|
|
|
class SNMPSynologyFirmwareResult(nagiosplugin.Result):
|
|
def __str__(self):
|
|
if self.metric.value['status'] == '1': status = 'available'
|
|
elif self.metric.value['status'] == '2': status = 'up2date'
|
|
elif self.metric.value['status'] == '3': status = 'connecting'
|
|
elif self.metric.value['status'] == '4': status = 'disconnected'
|
|
else: status = 'unknown'
|
|
return f'Firmware of {self.metric.value["model"]} (SN: {self.metric.value["sn"]}; Version: {self.metric.value["version"]}) is in status {status}!'
|
|
|
|
|
|
#
|
|
# Synology [DSM] Utilization % is the sum of user and system CPU usage
|
|
#
|
|
class SNMPSynologyCPUResource(nagiosplugin.Resource):
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
def probe(self) -> list:
|
|
"""check cpu usage %
|
|
|
|
Returns:
|
|
nagiosplugin.Metric: single metric element (return)
|
|
"""
|
|
baseoid = '.1.3.6.1.4.1.6574.1'
|
|
result = self.session.get(baseoid + '.7.1.0').value
|
|
return nagiosplugin.Metric(name='cpu', value=int(result), uom='%', context='cpu_scalar_context')
|
|
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
|
|
#
|
|
# Synology [DSM] Utilization % is the sum of user and system CPU usage
|
|
#
|
|
class SNMPSynologyMemoryResource(nagiosplugin.Resource):
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
def probe(self) -> list:
|
|
"""check memory usage in %
|
|
|
|
Returns:
|
|
nagiosplugin.Metric: single metric element (return)
|
|
"""
|
|
baseoid = '.1.3.6.1.4.1.6574.1'
|
|
result = self.session.get(baseoid + '.7.2.0').value
|
|
return nagiosplugin.Metric(name='memory', value=int(result), uom='%', context='memory_scalar_context')
|
|
|
|
def __init__(self, session) -> None:
|
|
self.session = session
|
|
|
|
|
|
#
|
|
# Arguments
|
|
#
|
|
def parse_args() -> argparse.Namespace:
|
|
"""evaluates given arguments
|
|
|
|
Returns:
|
|
argsparse.Namespace: Namespace Object with all arguments insert (use: args.long_name_of_argument)
|
|
"""
|
|
argp = argparse.ArgumentParser(description=__doc__)
|
|
# Default args
|
|
argp.add_argument('-v', '--verbose', action='count', default=0,
|
|
help='increase output verbosity (use up to 3 times)')
|
|
argp.add_argument('-H', '--hostname', default='localhost',
|
|
help='IP address or hostname of device to query')
|
|
argp.add_argument('-t', '--timeout', default=30,
|
|
help='abort execution after TIMEOUT seconds')
|
|
# TEMPLATE TODO: edit choices
|
|
argp.add_argument('-m', '--check_mode',
|
|
choices=[
|
|
'system',
|
|
'temperature',
|
|
'powersupply',
|
|
'fans',
|
|
'firmware',
|
|
'cpu',
|
|
'memory',
|
|
'disk',
|
|
],
|
|
help='check mode to run')
|
|
|
|
# Nagios args / see https://nagios-plugins.org/doc/guidelines.html#THRESHOLDFORMAT
|
|
argp.add_argument('-w', '--warning', default=':80',
|
|
help='warning threshold')
|
|
argp.add_argument('-c', '--critical', default=':90',
|
|
help='critical threshold')
|
|
|
|
# SNMP args (same -* as snmpwalk / snmpget linux command; exept -C/-c & -V /-v because already used (see above))
|
|
argp.add_argument('-V', '--snmp-version', default='2',
|
|
help='Used SNMP Version. (1, 2, 3)')
|
|
argp.add_argument('-C', '--snmp_community', default='public',
|
|
help='SNMP community if snmp v1 or v2 used (any).')
|
|
argp.add_argument('-l', '--snmp-security-level', default='auth_with_privacy',
|
|
help='Security level (no_auth_or_privacy, auth_without_privacy or auth_with_privacy)')
|
|
argp.add_argument('-u', '--snmp-security-username', default='monitoring',
|
|
help='Security name (any)')
|
|
argp.add_argument('-a', '--snmp-auth-protocol', default='SHA256',
|
|
help='Authentication protocol (MD5, SHA256, SHA512)')
|
|
argp.add_argument('-A', '--snmp-auth-pass', default=None,
|
|
help='Authentication passphrase (any)')
|
|
argp.add_argument('-x', '--snmp-privacy-protocol', default='AES',
|
|
help='Privacy protocol (AES, DES)')
|
|
argp.add_argument('-X', '--snmp-privacy-pass', default=None,
|
|
help='Privacy passphrase (any)')
|
|
|
|
args = argp.parse_args()
|
|
return args
|
|
|
|
#
|
|
# init
|
|
#
|
|
def main():
|
|
args = parse_args()
|
|
if args.verbose >= 3:
|
|
logging.getLogger().setLevel(logging.DEBUG)
|
|
|
|
session = init_snmp_session(hostname=args.hostname, snmp_version=args.snmp_version, snmp_community=args.snmp_community,
|
|
snmp_security_level=args.snmp_security_level, snmp_security_username=args.snmp_security_username,
|
|
snmp_auth_protocol=args.snmp_auth_protocol, snmp_auth_password=args.snmp_auth_pass,
|
|
snmp_privacy_protocol=args.snmp_privacy_protocol, snmp_privacy_password=args.snmp_privacy_pass)
|
|
|
|
# dice which check will be run bases on check_mode
|
|
match args.check_mode:
|
|
case 'system':
|
|
check = nagiosplugin.Check(SNMPSynologySystemResource(session=session),
|
|
SNMPSynologySystemContext(name='system_context'),
|
|
nagiosplugin.Summary())
|
|
check.name = "System Status"
|
|
case 'temperature':
|
|
check = nagiosplugin.Check(SNMPSynologyTemperatureResource(session=session),
|
|
nagiosplugin.ScalarContext(name='temperature_scalar_context', warning=args.warning, critical=args.critical),
|
|
nagiosplugin.Summary())
|
|
check.name = "System Temperature"
|
|
case 'powersupply':
|
|
check = nagiosplugin.Check(SNMPSynologyPowerSupplyResource(session=session),
|
|
SNMPSynologyPowerSupplyContext(name='powersupply_context'),
|
|
nagiosplugin.Summary())
|
|
check.name = "Power Supply Status"
|
|
case 'fans':
|
|
check = nagiosplugin.Check(SNMPSynologyFansResource(session=session),
|
|
SNMPSynologyFansContext(name='fan_context'),
|
|
SNMPSynologyFansSummary())
|
|
check.name = "Fans Status"
|
|
case 'firmware':
|
|
check = nagiosplugin.Check(SNMPSynologyFirmwareResource(session=session),
|
|
SNMPSynologyFirmwareContext(name='firmware_context'),
|
|
nagiosplugin.Summary())
|
|
check.name = "Firmware"
|
|
case 'cpu':
|
|
check = nagiosplugin.Check(SNMPSynologyCPUResource(session=session),
|
|
nagiosplugin.ScalarContext(name='cpu_scalar_context', warning=args.warning, critical=args.critical),
|
|
nagiosplugin.Summary())
|
|
check.name = "CPU Usage"
|
|
case 'memory':
|
|
check = nagiosplugin.Check(SNMPSynologyMemoryResource(session=session),
|
|
nagiosplugin.ScalarContext(name='memory_scalar_context', warning=args.warning, critical=args.critical),
|
|
nagiosplugin.Summary())
|
|
check.name = "Memory Usage"
|
|
case 'disk':
|
|
# check = nagiosplugin.Check(SNMPSynologyDiskResource(session=session),
|
|
# nagiosplugin.ScalarContext(name='memory_scalar_context', warning=args.warning, critical=args.critical),
|
|
# nagiosplugin.Summary())
|
|
# check.name = "Memory Usage"
|
|
pass
|
|
case _:
|
|
raise nagiosplugin.CheckError(f'Unknown check mode: {args.check_mode}')
|
|
|
|
check.main(args.verbose, args.timeout)
|
|
|
|
if __name__ == '__main__':
|
|
main() |