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():
timeout = 30 # default
args = parser()
check = nagiosplugin.Check(
KafkaTopics(args),
stringContext.StringContext(
'Under Replication',
None,
fmt_metric='{value} partitions are under replicated'),
stringContext.StringContext('Unavailability',
None,
fmt_metric='{value} are unavailables'))
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(
Zookeeper(args.hosts),
stringContext.StringContext('running',
True,
fmt_metric='ZK Server running {value}'),
stringContext.StringContext('writable',
True,
fmt_metric='Path writable {value}'),
stringContext.StringContext('version',
args.version,
fmt_metric='Runnng version is {value}'),
nagiosplugin.ScalarContext('latency', args.latency_warning,
args.latency_critical),
stringContext.StringContext('mode',
'1/%d' % (len(args.hosts.split(',')) - 1),
fmt_metric='leader/followers {value}'))
check.main()
def main():
args = parser()
check = nagiosplugin.Check(
Hdfs(args),
stringContext.StringContext('active nn',
1,
fmt_metric='{value} active namenodes'),
stringContext.StringContext('standby nn',
len(args.namenodes.split(',')) - 1,
fmt_metric='{value} standby namenodes'),
nagiosplugin.ScalarContext('used',
args.warning_used,
args.critical_used,
fmt_metric='{value}% storage in use'),
nagiosplugin.ScalarContext('total_blocks',
args.warning_blocks,
args.critical_blocks,
fmt_metric=' is using {value} blocks'),
nagiosplugin.ScalarContext(
'ureplicated',
args.warning_ureplicated,
args.critical_ureplicated,
fmt_metric='{value} under replicated blocks'),
nagiosplugin.ScalarContext('missing',
args.warning_missing,
args.critical_missing,
fmt_metric='{value} missing blocks'),
nagiosplugin.ScalarContext('corrupt',
args.warning_corrupt,
args.critical_corrupt,
fmt_metric='{value} corrupted blocks'),
nagiosplugin.ScalarContext(
'balanced',
args.warning_balanced,
args.critical_balanced,
fmt_metric='There are {value}% usage difference between datanodes'
),
stringContext.StringContext('datanodes',
args.critical_datanodes,
fmt_metric='{value} living datanodes'),
HdfsSummary())
check.main()
def main():
args = parser()
check = nagiosplugin.Check(
Httpfs(args),
stringContext.StringContext('type',
args.type,
fmt_metric=' is a {value}'),
stringContext.StringContext('owner',
args.owner,
fmt_metric=' owner is {value}'),
stringContext.StringContext('group',
args.group,
fmt_metric=' group is {value}'),
stringContext.StringContext('permission',
args.permission,
fmt_metric=' has {value} permission'),
stringContext.StringContext('writable',
True,
fmt_metric='HTTPFS writable: {value}'))
check.main()
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():
args = parser()
check = nagiosplugin.Check(Historyserver(args),
stringContext.StringContext('hs_status',
True))
check.main()
def main():
args = parser()
check = nagiosplugin.Check(Hbase(args),
stringContext.StringContext('status',
'OK'))
check.main()