def main():
args = parse_args()
percentiles = args.percentiles.split(",")
url_filters = {}
exclude_patterns = []
for expr in args.filter:
label, urlprefix = expr.split(":", 1)
url_filters[label] = urlprefix
for pattern in args.exclude:
exclude_patterns.append(pattern)
check = nagiosplugin.Check(
HAProxyLog(
args.logfile,
args.state_file,
percentiles,
url_filters,
exclude_patterns,
),
nagiosplugin.ScalarContext("http_errors", args.ew, args.ec),
HAProxyLogSummary(percentiles),
)
for pct, i in zip(percentiles, itertools.count()):
check.add(
nagiosplugin.ScalarContext(
"t_tot%s" % pct,
args.tw[i],
args.tc[i],
"total req time (%s.pct) is {valueunit}" % pct,
))
check.main(args.verbose, args.timeout)
def main():
argp = argparse.ArgumentParser(description='Nagios plugin to check Nvidia GPU status using nvidia-smi')
argp.add_argument('-w', '--gpu_warning', metavar='RANGE', default=0,
help='warning if threshold is outside RANGE')
argp.add_argument('-c', '--gpu_critical', metavar='RANGE', default=0,
help='critical if threshold is outside RANGE')
argp.add_argument('-W', '--mem_warning', metavar='RANGE', default=0,
help='warning if threshold is outside RANGE')
argp.add_argument('-C', '--mem_critical', metavar='RANGE', default=0,
help='critical if threshold is outside RANGE')
argp.add_argument('-t', '--gputemp_warning', metavar='RANGE', default=0,
help='warning if threshold is outside RANGE')
argp.add_argument('-T', '--gputemp_critical', metavar='RANGE', default=0,
help='critical if threshold is outside RANGE')
argp.add_argument('-d', '--device', default="0",
help='Device ID (starting from 0)')
argp.add_argument('-v', '--verbose', action='count', default=0,
help='increase verbosity (use up to 3 times)')
argp.add_argument('-m', '--memory', action='store_true',
help="Show memory used in % instead of utilization")
args = argp.parse_args()
check = nagiosplugin.Check(
Utilization(args),
nagiosplugin.ScalarContext('gpuutil', args.gpu_warning, args.gpu_critical),
nagiosplugin.ScalarContext('memutil', args.mem_warning, args.mem_critical),
nagiosplugin.ScalarContext('gpuTemp', args.gputemp_warning, args.gputemp_critical)
)
check.main(verbose=args.verbose)
def main():
contexts = ['temperature', 'moisture', 'light', 'conductivity', 'battery']
argp = argparse.ArgumentParser(description=__doc__)
argp.add_argument('-m', '--mac', required=True, type=str,
help='Miflora MAC address')
for context in contexts:
argp.add_argument('--%s-warning' % context, metavar='RANGE', default='',
help='return warning if %s is outside RANGE' % context)
argp.add_argument('--%s-critical' % context, metavar='RANGE', default='',
help='return critical if %s is outside RANGE' % context)
argp.add_argument('-v', '--verbose', action='count', default=0,
help='increase output verbosity (use up to 3 times)')
args = argp.parse_args()
check = nagiosplugin.Check(
Environment(args.mac),
nagiosplugin.ScalarContext('temperature', args.temperature_warning, args.temperature_critical),
nagiosplugin.ScalarContext('moisture', args.moisture_warning, args.moisture_critical),
nagiosplugin.ScalarContext('light', args.light_warning, args.light_critical),
nagiosplugin.ScalarContext('conductivity', args.conductivity_warning, args.conductivity_critical),
nagiosplugin.ScalarContext('battery', args.battery_warning, args.battery_critical),
EnvironmentSummary())
check.main(verbose=args.verbose)
def main():
argp = argparse.ArgumentParser()
argp.add_argument(
'-w',
'--warning',
metavar='RANGE',
help='warning if backup age is outside RANGE in seconds'),
argp.add_argument(
'-c',
'--critical',
metavar='RANGE',
help='critical if backup age is outside RANGE in seconds')
argp.add_argument('-v',
'--verbose',
action='count',
default=0,
help='increase output verbosity (use up to 3 times)')
argp.add_argument('-t',
'--timeout',
default=10,
help='abort execution after TIMEOUT seconds')
argp.add_argument('--base-path',
default="/srv/backup/",
help="Path to the bank of the vault (/srv/backup)")
argp.add_argument('--max-duration',
default=3600,
metavar='RANGE',
help="max time to take a backup in seconds (3600)")
argp.add_argument('--branch',
default="default",
help="Branch to check (default)")
argp.add_argument('vault', help='Name of the vault to check')
args = argp.parse_args()
check = nagiosplugin.Check(
Backup(args.vault, args.base_path, args.branch),
BoolContext(name='stale_lockfile',
critical=True,
fmt_metric=Bool_Fmt_Metric('LockFile is OK!',
'LockFile is stale!')),
BoolContext('valid_backup_found',
critical=False,
fmt_metric=Bool_Fmt_Metric('Valid backup found!',
'No valid Backup found!')),
nagiosplugin.ScalarContext(
'last_success', args.warning, args.critical,
Duration_Fmt_Metric('Last successful backup is {valueunit} old')),
nagiosplugin.ScalarContext(
'last_try', args.warning, args.critical,
Duration_Fmt_Metric('Last backup tried {valueunit} ago')),
nagiosplugin.ScalarContext(
name='duration',
warning=args.max_duration,
fmt_metric=Duration_Fmt_Metric('Last backuprun took {valueunit}')),
nagiosplugin.ScalarContext(name='running_backup_for',
warning=args.max_duration,
critical=args.max_duration * 3,
fmt_metric=Duration_Fmt_Metric(
'Running backup since {valueunit}')),
)
check.main(args.verbose, args.timeout)
def get_check(hours_warn, hours_crit, host=None, path=None, tags=None):
return Check(
ResticRepo(host=host, path=path, tags=tags),
nagiosplugin.ScalarContext("snapshot",
warning=hours_warn,
critical=hours_crit),
nagiosplugin.ScalarContext("size"),
nagiosplugin.ScalarContext("files"))
def create_contexts(interfaces, defspeed, defduplex):
yield nagiosplugin.ScalarContext("auto_spd", "", defspeed)
yield DuplexContext("auto_dup", defduplex)
for iface in interfaces:
name, speed, duplex = (iface + ",,").split(",")[0:3]
yield nagiosplugin.ScalarContext("{}_spd".format(name), "", speed
or defspeed)
yield DuplexContext("{}_dup".format(name), duplex or defduplex)
def main():
html_auth = None
args = parser()
check = nagiosplugin.Check(
Oozie(args),
nagiosplugin.ScalarContext('errors', args.warning, args.critical),
nagiosplugin.ScalarContext('coordinators', '1:', '1:'))
check.main()
def main():
parser = argparse.ArgumentParser()
parser.add_argument("hostname", help="the hostname", type=str)
parser.add_argument("username", help="the snmp user name", type=str)
parser.add_argument("authkey", help="the auth key", type=str)
parser.add_argument("privkey", help="the priv key", type=str)
parser.add_argument("mode", help="the mode", type=str,
choices=["load", "memory", "disk", "storage",
"update", "status"])
parser.add_argument("-w", help="warning value for selected mode", type=int)
parser.add_argument("-c", help="critical value for selected mode",
type=int)
parser.add_argument('-v', '--verbose', action='count', default=0)
args = parser.parse_args()
warning = args.w
critical = args.c
check = None
if args.mode == 'load':
check = Check(
Load(args.hostname, args.username, args.authkey, args.privkey),
nagiosplugin.ScalarContext('load', warning, critical),
LoadSummary())
elif args.mode == 'memory':
check = Check(
Memory(args.hostname, args.username, args.authkey,
args.privkey),
nagiosplugin.ScalarContext('memory', warning, critical),
MemorySummary())
elif args.mode == 'disk':
check = Check(
Disk(args.hostname, args.username, args.authkey, args.privkey),
nagiosplugin.ScalarContext('temp', warning, critical),
DiskContext('disk_status'),
DiskSummary())
elif args.mode == 'storage':
check = Check(
Storage(args.hostname, args.username, args.authkey,
args.privkey),
nagiosplugin.ScalarContext('storage', warning, critical),
StorageSummary())
elif args.mode == 'update':
check = Check(
Update(args.hostname, args.username, args.authkey,
args.privkey),
UpdateContext('update'),
UpdateSummary())
elif args.mode == 'status':
check = Check(
Status(args.hostname, args.username, args.authkey,
args.privkey),
nagiosplugin.ScalarContext('temp', warning, critical),
StatusContext('status'),
StatusSummary())
else:
raise CheckError('Unknown mode', args.mode)
check.main(args.verbose)
def main():
args = parse_args()
percentiles = args.percentiles.split(',')
check = nagiosplugin.Check(
HAProxyLog(args.logfile, args.state_file, percentiles),
nagiosplugin.ScalarContext('error_rate', args.ew, args.ec))
for pct, i in zip(percentiles, itertools.count()):
check.add(nagiosplugin.ScalarContext(
'ttot%s' % pct, args.tw[i], args.tc[i],
'total time (%s.pct) is {valueunit}' % pct))
check.main(args.verbose, args.timeout)
def main():
argp = SNMPSkeletton.default_args('Check memory on Cisco.')
args = argp.parse_args()
check = nagiosplugin.Check(
CiscoMem(args.host, args.port, args.community, args.version),
nagiosplugin.ScalarContext('alert_mem_percent',
args.warning, args.critical),
nagiosplugin.ScalarContext('mem_used'),
nagiosplugin.ScalarContext('mem_total'))
check.main()
def create_check(args):
"""
Creates and configures a check for the sessioninfo command.
:return: the sessioninfo check.
"""
return np.Check(
SessInfo(args.host, args.token),
np.ScalarContext('session', args.warn, args.crit),
np.ScalarContext('throughput'),
SessSummary())
def main():
argp = SNMPSkeletton.default_args('Check memory on Sun.')
args = argp.parse_args()
check = nagiosplugin.Check(
SunMem(args.host, args.port, args.community, args.version),
nagiosplugin.ScalarContext('alert_mem',
args.warning,
args.critical,
fmt_metric='alert_mem is {value}%'),
nagiosplugin.ScalarContext('graph_swap'))
check.main()
def check_backup(Collector, config):
return (
Collector(
config['config'],
config['facility'],
),
nagiosplugin.ScalarContext('age',
config['warning'],
config['warning']),
nagiosplugin.ScalarContext('size', "1:", "1:"),
)
def main():
timeout = 10 # default
args = parser()
check = nagiosplugin.Check(
ZookeeperZnode(args), stringContext.StringContext('true', True),
stringContext.StringContext('false', False),
nagiosplugin.ScalarContext('total_topics', args.warn_topics,
args.crit_topics),
nagiosplugin.ScalarContext('hosts', args.warn_hosts, args.crit_hosts))
if args.test == "kafka":
timeout = 0
check.main(timeout=timeout)
def main():
argp = SNMPSkeletton.default_args('Check storage on Windows file systems.')
argp.add_argument('--filesystem', '-f',
help='ID Filesystem'),
args = argp.parse_args()
check = nagiosplugin.Check(
WinStorage(args.host, args.port, args.community,
args.version, args.filesystem),
nagiosplugin.ScalarContext('alert_fs_percent',
args.warning, args.critical),
nagiosplugin.ScalarContext('fs_used'),
nagiosplugin.ScalarContext('fs_total'))
check.main()
def main():
args = parser()
check = nagiosplugin.Check(
Resourcemanager(args), stringContext.StringContext('state', 'STARTED'),
stringContext.StringContext('nodeState', 'RUNNING'),
nagiosplugin.ScalarContext('unhealthy', args.unhealthy_warn,
args.unhealthy_crit),
nagiosplugin.ScalarContext('lost', args.lost_warn, args.lost_crit),
nagiosplugin.ScalarContext('rebooted', args.rebooted_warn,
args.rebooted_crit),
nagiosplugin.ScalarContext('appsPending', args.apps_warn,
args.apps_crit))
check.main()
def main():
args = parser()
check = nagiosplugin.Check(
Flume(args),
nagiosplugin.ScalarContext('io',
args.batch_size,
fmt_metric='{value} msg differs'),
nagiosplugin.ScalarContext(
'LastMsg',
str(int(time.time()) * 1000 - args.timeout_war * 1000) + ':',
str(int(time.time()) * 1000 - args.timeout_crit * 1000) + ':',
fmt_metric='Last msg recived in {value}'))
check.main()
def main():
args = parse_args()
check = nagiosplugin.Check(
PureFAalert(args.endpoint, args.apitoken),
nagiosplugin.ScalarContext(
'critical', args.warning_crit, args.critical_crit,
fmt_metric='{value} critical messages'),
nagiosplugin.ScalarContext(
'warning', args.warning_warn, args.critical_warn,
fmt_metric='{value} warning messages'),
nagiosplugin.ScalarContext(
'info', args.warning_info, args.critical_info,
fmt_metric='{value} info messages'))
check.main(args.verbose, args.timeout)
def main():
args = parser()
check = nagiosplugin.Check(
QJM(args),
nagiosplugin.ScalarContext('processing', args.process_warn,
args.process_crit),
nagiosplugin.ScalarContext('sync', args.sync_threshold_warn,
args.sync_threshold_crit),
stringContext.StringContext('connection', "OK"),
nagiosplugin.ScalarContext(
'quorum',
nagiosplugin.Range("%s:" % str(len(args.qjm.split(',')) / 2)),
nagiosplugin.Range("%s:" % str(len(args.qjm.split(',')) / 2))))
check.main()
def main():
timeout = 10 # default
args = parser()
check = nagiosplugin.Check(
StormTopology(args),
stringContext.StringContext('connected',
True,
fmt_metric='Connection status: {value}'),
nagiosplugin.ScalarContext('1d load', args.load_1d_warn,
args.load_1d_crit),
nagiosplugin.ScalarContext('3h load', args.load_3h_warn,
args.load_3h_crit),
nagiosplugin.ScalarContext('10m load', args.load_10m_warn,
args.load_10m_crit))
check.main(timeout=timeout)
def main():
"""
:return:
"""
argp = argparse.ArgumentParser(description='Nagios/Icinga check for Kubernetes Nodes')
argp.add_argument('-v', '--version', action='version', version='%(prog)s ' + version)
argp.add_argument('--kube-config', help='Kubernetes Config File')
args = argp.parse_args()
check = nagiosplugin.Check(
Nodes(args.kube_config),
nagiosplugin.ScalarContext('problem_nodes', 1, 2),
nagiosplugin.ScalarContext('all_nodes')
)
check.main()
def main():
argp = argparse.ArgumentParser(description=__doc__)
argp.add_argument('-l',
'--logfile',
metavar='FILE',
default='/var/log/mail.log',
help='Postfix logfile (default: /var/log/mail.log)')
argp.add_argument('-m',
'--mode',
metavar='MODE',
default='minute',
help='mode to check: minute, hour, day, week \
(default: minute)')
argp.add_argument('-w',
'--warning',
metavar='RANGE',
default='',
help='return warning if value is outside RANGE')
argp.add_argument('-c',
'--critical',
metavar='RANGE',
default='',
help='return critical if value is outside RANGE')
args = argp.parse_args()
check = nagiosplugin.Check(
Postfix(logfile=args.logfile, mode=args.mode),
nagiosplugin.ScalarContext(args.mode, args.warning, args.critical),
LoadSummary(args.mode))
check.main(timeout=120)
def main():
arg_parser = argparse.ArgumentParser()
arg_parser.add_argument('-v', '--verbose', action='count', default=0)
arg_parser.add_argument('-H', '--host', required=True, nargs=1)
arg_parser.add_argument('-p', '--portscan-range', nargs=1,
metavar='PORT_RANGE', help='port range, e.g.: 1-1024')
arg_parser.add_argument('-a', '--allowed-ports', nargs=1, default='',
metavar='PORTS', help='list of ports split by "," (comma)')
arg_parser.add_argument('-n', '--nmap-extra-args', nargs=1)
args = arg_parser.parse_args()
checker = OpenPortsResource('PORT SECURITY SCANNER')
checker.target_host = args.host[0]
if args.portscan_range:
checker.portscan_range = args.portscan_range[0]
else:
checker.portscan_range = '1-65535'
if args.allowed_ports and len(args.allowed_ports) > 0:
checker.allowed_ports = args.allowed_ports[0].split(',')
if args.nmap_extra_args:
checker.nmap_extra_args = args.nmap_extra_args[0]
check = nagiosplugin.Check(
checker,
OpenPortsContext('ports'),
nagiosplugin.ScalarContext('unknown_ports_count', critical='0:0'),
)
check.main(args.verbose, timeout=0)
def main():
args = parse_args()
check = nagiosplugin.Check(
Rubrikrunway(args.rubrik_ip, args.rubrik_user, args.rubrik_pass))
check.add(
nagiosplugin.ScalarContext('rk_runway', args.warning, args.critical))
check.main(args.verbose, args.timeout)
def main():
args = parse_args()
check = nagiosplugin.Check(
CohesityAlerts(args))
check.add(
nagiosplugin.ScalarContext(
'critical',
critical='~:0'
))
check.add(
nagiosplugin.ScalarContext(
'warning',
warning='~:0'
))
check.main(args.verbose, args.timeout)
def half_installed_check(config):
return (HalfRemoved(),
nagiosplugin.ScalarContext(
'halfinstalled',
config['warning'],
config['warning'],
fmt_metric='{value} half-installed packages'))
def main():
args = get_argparser().parse_args()
objects = []
if args.unit:
objects.append(SystemctlIsActiveResource(unit=args.unit))
else:
objects += [
SystemdctlListUnitsResource(excludes=args.exclude),
PerformanceDataContext(),
SystemdAnalyseResource(),
]
objects += [
UnitContext(),
nagiosplugin.ScalarContext(
name='startup_time',
warning=args.warning,
critical=args.critical,
),
SystemdSummary()
]
check = nagiosplugin.Check(*objects)
check.main(args.verbose)
def main():
argp = SNMPSkeletton.default_args('Check temperature on Dell.')
args = argp.parse_args()
check = nagiosplugin.Check(
DellTemp(args.host, args.port, args.community, args.version),
nagiosplugin.ScalarContext('alert_temp', args.warning, args.critical))
check.main()
def main():
argp = argparse.ArgumentParser(description=__doc__)
argp.add_argument('-w',
'--warning',
metavar='RANGE',
default='',
help='return warning if activeVisitors is outside RANGE')
argp.add_argument(
'-c',
'--critical',
metavar='RANGE',
default='',
help='return critical if activeVisitors is outside RANGE')
argp.add_argument('-D', '--data', action='store', required=True)
argp.add_argument('-V', '--view', action='store', required=True)
argp.add_argument('-v',
'--verbose',
action='count',
default=0,
help='increase output verbosity (use up to 3 times)')
args = argp.parse_args()
check = nagiosplugin.Check(
RealtimeVisitors(args.data, args.view),
nagiosplugin.ScalarContext('activeVisitors', args.warning,
args.critical))
check.main(verbose=args.verbose)
def main():
"""
:return:
"""
argp = argparse.ArgumentParser(
description='Nagios/Icinga check for Kubernetes Pods')
argp.add_argument('-v',
'--version',
action='version',
version='%(prog)s ' + version)
argp.add_argument('--kube-config', help='Kubernetes Config File')
for phase in Pods.phases:
argp.add_argument('--warning-' + phase.lower())
argp.add_argument('--critical-' + phase.lower())
args = argp.parse_args()
checks = [Pods(args.kube_config)]
for phase in Pods.phases:
checks.append(
nagiosplugin.ScalarContext(
phase, getattr(args, 'warning_' + phase.lower()),
getattr(args, 'critical_' + phase.lower())))
checks.append(PodsSummary())
check = nagiosplugin.Check(*checks)
check.main()