def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install(['python-dbus', 'lockfile-progs'])
log('Refreshing nagios checks')
if os.path.isdir(NAGIOS_PLUGINS):
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'check_ceph_status.py'),
os.path.join(NAGIOS_PLUGINS, 'check_ceph_status.py'))
script = os.path.join(SCRIPTS_DIR, 'collect_ceph_status.sh')
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'collect_ceph_status.sh'), script)
cronjob = "{} root {}\n".format('*/5 * * * *', script)
write_file(STATUS_CRONFILE, cronjob)
# Find out if nrpe set nagios_hostname
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
check_cmd = 'check_ceph_status.py -f {} --degraded_thresh {}' \
' --misplaced_thresh {}' \
' --recovery_rate {}'.format(STATUS_FILE,
config('nagios_degraded_thresh'),
config('nagios_misplaced_thresh'),
config('nagios_recovery_rate'))
if config('nagios_raise_nodeepscrub'):
check_cmd = check_cmd + ' --raise_nodeepscrub'
nrpe_setup.add_check(
shortname="ceph",
description='Check Ceph health {{{}}}'.format(current_unit),
check_cmd=check_cmd)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
log('Refreshing nagios checks')
if os.path.isdir(NAGIOS_PLUGINS):
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'check_ceph_status.py'),
os.path.join(NAGIOS_PLUGINS, 'check_ceph_status.py'))
script = os.path.join(SCRIPTS_DIR, 'collect_ceph_status.sh')
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'collect_ceph_status.sh'), script)
cronjob = "{} root {}\n".format('*/5 * * * *', script)
write_file(STATUS_CRONFILE, cronjob)
# Find out if nrpe set nagios_hostname
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe_setup.add_check(
shortname="ceph",
description='Check Ceph health {%s}' % current_unit,
check_cmd='check_ceph_status.py -f {}'.format(STATUS_FILE))
nrpe_setup.write()
def update_nrpe_config(checks_to_remove=None):
"""
Update the checks for the nagios plugin.
:param checks_to_remove: list of short names of nrpe checks to
remove. For example, pass ['radosgw'] to remove the check for
the default systemd radosgw service, to make way for per host
services.
:type checks_to_remove: list
"""
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.copy_nrpe_checks()
if checks_to_remove is not None:
log("Removing the following nrpe checks: {}".format(checks_to_remove),
level=DEBUG)
for svc in checks_to_remove:
nrpe_setup.remove_check(shortname=svc)
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
nrpe_setup.write()
def update_nrpe_config():
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
services = [service_name]
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def update_nrpe_config(unused=None):
services = ('snap.kubelet.daemon', 'snap.kube-proxy.daemon')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
log('Refreshing nagios checks')
if os.path.isdir(NAGIOS_PLUGINS):
rsync(os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'check_ceph_status.py'),
os.path.join(NAGIOS_PLUGINS, 'check_ceph_status.py'))
script = os.path.join(SCRIPTS_DIR, 'collect_ceph_status.sh')
rsync(os.path.join(os.getenv('CHARM_DIR'), 'files',
'nagios', 'collect_ceph_status.sh'),
script)
cronjob = "{} root {}\n".format('*/5 * * * *', script)
write_file(STATUS_CRONFILE, cronjob)
# Find out if nrpe set nagios_hostname
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe_setup.add_check(
shortname="ceph",
description='Check Ceph health {%s}' % current_unit,
check_cmd='check_ceph_status.py -f {}'.format(STATUS_FILE)
)
nrpe_setup.write()
def render_nrpe_checks(self):
"""Configure Nagios NRPE checks."""
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
charm_nrpe = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(charm_nrpe, self.services, current_unit)
charm_nrpe.write()
def update_nrpe_config(checks_to_remove=None):
"""
Update the checks for the nagios plugin.
:param checks_to_remove: list of short names of nrpe checks to
remove. For example, pass ['radosgw'] to remove the check for
the default systemd radosgw service, to make way for per host
services.
:type checks_to_remove: list
"""
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.copy_nrpe_checks()
if checks_to_remove is not None:
log("Removing the following nrpe checks: {}".format(checks_to_remove),
level=DEBUG)
for svc in checks_to_remove:
nrpe_setup.remove_check(shortname=svc)
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
nrpe_setup.write()
def update_nrpe_config(unused=None):
services = ('snap.kubelet.daemon', 'snap.kube-proxy.daemon')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
log('Refreshing nrpe checks')
if not os.path.exists(NAGIOS_PLUGINS):
mkpath(NAGIOS_PLUGINS)
rsync(os.path.join(os.getenv('CHARM_DIR'), 'files', 'nrpe-external-master',
'check_swift_storage.py'),
os.path.join(NAGIOS_PLUGINS, 'check_swift_storage.py'))
rsync(os.path.join(os.getenv('CHARM_DIR'), 'files', 'nrpe-external-master',
'check_swift_service'),
os.path.join(NAGIOS_PLUGINS, 'check_swift_service'))
rsync(os.path.join(os.getenv('CHARM_DIR'), 'files', 'sudo',
'swift-storage'),
os.path.join(SUDOERS_D, 'swift-storage'))
# Find out if nrpe set nagios_hostname
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
# check the rings and replication
nrpe_setup.add_check(
shortname='swift_storage',
description='Check swift storage ring hashes and replication'
' {%s}' % current_unit,
check_cmd='check_swift_storage.py {}'.format(
config('nagios-check-params'))
)
nrpe.add_init_service_checks(nrpe_setup, SWIFT_SVCS, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install(['python-dbus', 'lockfile-progs'])
log('Refreshing nagios checks')
if os.path.isdir(NAGIOS_PLUGINS):
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'check_ceph_status.py'),
os.path.join(NAGIOS_PLUGINS, 'check_ceph_status.py'))
script = os.path.join(SCRIPTS_DIR, 'collect_ceph_status.sh')
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'collect_ceph_status.sh'), script)
cronjob = "{} root {}\n".format('*/5 * * * *', script)
write_file(STATUS_CRONFILE, cronjob)
# Find out if nrpe set nagios_hostname
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
check_cmd = 'check_ceph_status.py -f {} --degraded_thresh {}' \
' --misplaced_thresh {}' \
' --recovery_rate {}'.format(STATUS_FILE,
config('nagios_degraded_thresh'),
config('nagios_misplaced_thresh'),
config('nagios_recovery_rate'))
if config('nagios_raise_nodeepscrub'):
check_cmd = check_cmd + ' --raise_nodeepscrub'
nrpe_setup.add_check(
shortname="ceph",
description='Check Ceph health {{{}}}'.format(current_unit),
check_cmd=check_cmd)
if config('nagios_additional_checks'):
additional_critical = config('nagios_additional_checks_critical')
x = ast.literal_eval(config('nagios_additional_checks'))
for key, value in x.items():
name = "ceph-{}".format(key.replace(" ", ""))
log("Adding check {}".format(name))
check_cmd = 'check_ceph_status.py -f {}' \
' --additional_check \\\"{}\\\"' \
' {}'.format(STATUS_FILE, value,
"--additional_check_critical"
if additional_critical is True else "")
nrpe_setup.add_check(
shortname=name,
description='Additional Ceph checks {{{}}}'.format(
current_unit),
check_cmd=check_cmd)
if config('nagios_check_num_osds'):
check_cmd = 'check_ceph_status.py -f {} --check_num_osds'.format(
STATUS_FILE)
nrpe_setup.add_check(
shortname='ceph_num_osds',
description='Check whether all OSDs are up and in',
check_cmd=check_cmd)
nrpe_setup.write()
def update_nrpe_config(svc):
# python-dbus is used by check_upstart_job
fetch.apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, SVCNAME, current_unit)
nrpe_setup.write()
def update_nrpe_config(unused=None):
services = ('nginx', )
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def update_nrpe_config(unused=None):
services = ('nginx',)
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def update_nrpe_config(svc):
# python-dbus is used by check_upstart_job
fetch.apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, SVCNAME, current_unit)
nrpe_setup.write()
def update_nrpe_config(unused=None):
services = ('kube-apiserver', 'kube-controller-manager', 'kube-scheduler')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def update_nrpe_config(unused=None):
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname, primary=True)
# Add a check for the snap's systemd service
nrpe.add_init_service_checks(nrpe_setup, SNAP_SERVICES, current_unit)
nrpe_setup.write()
def update_nrpe_config(unused=None):
services = ('kube-apiserver', 'kube-controller-manager', 'kube-scheduler')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def remove_nrpe_config(nagios=None):
hostname = nrpe.get_nagios_hostname()
nrpe_setup = nrpe.NRPE(hostname=hostname)
for service in SNAP_SERVICES:
nrpe_setup.remove_check(shortname=service)
remove_state(charm_state('nrpe-initial-config'))
def main():
cmd = subprocess.Popen(['lsmod'],
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
stdout, stederr = cmd.communicate()
if re.compile('(megaraid).*').findall(stdout.decode('utf-8')):
megaraid = True
if os.path.isdir(NAGIOS_PLUGINS):
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'check_lsi_raid'),
os.path.join(NAGIOS_PLUGINS, 'check_lsi_raid'))
if os.path.isdir(SUDOERS_DIR):
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'nagios_sudoers'),
os.path.join(SUDOERS_DIR, 'nagios_sudoers'))
if os.path.isdir(NAGIOS_PLUGINS):
rsync(
os.path.join(os.getenv('CHARM_DIR'), 'files', 'nagios',
'check_bond'),
os.path.join(NAGIOS_PLUGINS, 'check_bond'))
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
# Install megaraid tools
# And add megaraid nagios check
if megaraid:
install_packages(['storcli', 'libfile-which-perl'])
nrpe_setup.add_check(shortname='lsi-raid',
description='LSI Raid Check {%s}' % current_unit,
check_cmd=(os.path.join(NAGIOS_PLUGINS,
'check_lsi_raid')))
# Install checks for the network bonds
if os.path.isfile('/proc/net/bonding/bond0') and \
os.path.isfile('/proc/net/bonding/bond1'):
nrpe_setup.add_check(
shortname='bond0',
description='Bond0 check {%s}' % current_unit,
check_cmd=(os.path.join(NAGIOS_PLUGINS, 'check_bond') +
' -i bond0 -p eth2'))
nrpe_setup.add_check(
shortname='bond0',
description='Bond1 check {%s}' % current_unit,
check_cmd=(os.path.join(NAGIOS_PLUGINS, 'check_bond') +
' -i bond1 -p eth3'))
nrpe_setup.write()
reactive.set_state('raidcheck_installed')
hookenv.status_set('active', 'Unit is ready')
def test_get_nagios_hostname(self):
rel_info = {
'nagios_hostname': 'bob-openstack-dashboard-0',
'private-address': '10.5.3.103',
'__unit__': u'dashboard-nrpe/1',
'__relid__': u'nrpe-external-master:2',
'nagios_host_context': u'bob',
}
self.patched['relations_of_type'].return_value = [rel_info]
self.assertEqual(nrpe.get_nagios_hostname(), 'bob-openstack-dashboard-0')
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.copy_nrpe_checks()
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.copy_nrpe_checks()
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
nrpe_setup.write()
def remove_nrpe_config():
hookenv.log('Removing nrpe checks for services: '
'{}'.format(MONITORED_SERVICES))
hostname = nrpe.get_nagios_hostname()
nrpe_setup = nrpe.NRPE(hostname=hostname, primary=False)
for check in MONITORED_SERVICES:
nrpe_setup.remove_check(shortname=check)
nrpe_setup.write()
remove_state('nrpe-external-master.initial-config')
def update_nrpe_config(unused=None):
# List of systemd services that will be checked
services = ('snap.etcd.etcd', )
# The current nrpe-external-master interface doesn't handle a lot of logic,
# use the charm-helpers code for now.
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname, primary=False)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, ['mysql'], current_unit)
nrpe_setup.add_check(shortname='mysql_proc',
description='Check MySQL process {%s}' % current_unit,
check_cmd='check_procs -c 1:1 -C mysqld')
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe_files_dir = os.path.join(charm_dir(), 'charmhelpers', 'contrib',
'openstack', 'files')
nrpe.copy_nrpe_checks(nrpe_files_dir=nrpe_files_dir)
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe_setup.add_check(
shortname='ceph-osd',
description='process check {%s}' % current_unit,
check_cmd=('/bin/cat /var/lib/ceph/osd/ceph-*/whoami |'
'xargs -I@ status ceph-osd id=@ && exit 0 || exit 2'))
nrpe_setup.write()
def remove_nrpe_config():
remove_state("nrpe-external-master.initial-config")
remove_nagios_plugin("check_k8s_worker.py")
# The current nrpe-external-master interface doesn't handle a lot of logic,
# use the charm-helpers code for now.
hostname = nrpe.get_nagios_hostname()
nrpe_setup = nrpe.NRPE(hostname=hostname)
for service in worker_services:
nrpe_setup.remove_check(shortname=service)
nrpe_setup.remove_check(shortname="node")
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe_files_dir = os.path.join(
charm_dir(), 'charmhelpers', 'contrib', 'openstack', 'files')
nrpe.copy_nrpe_checks(nrpe_files_dir=nrpe_files_dir)
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
nrpe_setup.write()
def configure_nrpe(unused=None):
hookenv.log('Configuring nrpe checks for services: '
'{}'.format(MONITORED_SERVICES))
# The current nrpe-external-master interface doesn't handle a lot of logic,
# use the charm-helpers code for now.
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname, primary=False)
nrpe.add_init_service_checks(nrpe_setup, MONITORED_SERVICES, current_unit)
nrpe_setup.write()
set_state('nrpe-external-master.initial-config')
def update_nrpe_checks():
if os.path.isdir(NAGIOS_PLUGINS):
rsync(os.path.join(os.getenv('CHARM_DIR'), 'scripts',
'check_rabbitmq.py'),
os.path.join(NAGIOS_PLUGINS, 'check_rabbitmq.py'))
rsync(os.path.join(os.getenv('CHARM_DIR'), 'scripts',
'check_rabbitmq_queues.py'),
os.path.join(NAGIOS_PLUGINS, 'check_rabbitmq_queues.py'))
if config('stats_cron_schedule'):
script = os.path.join(SCRIPTS_DIR, 'collect_rabbitmq_stats.sh')
cronjob = CRONJOB_CMD.format(schedule=config('stats_cron_schedule'),
timeout=config('cron-timeout'),
command=script)
rsync(os.path.join(charm_dir(), 'scripts',
'collect_rabbitmq_stats.sh'), script)
write_file(STATS_CRONFILE, cronjob)
elif os.path.isfile(STATS_CRONFILE):
os.remove(STATS_CRONFILE)
# Find out if nrpe set nagios_hostname
hostname = nrpe.get_nagios_hostname()
myunit = nrpe.get_nagios_unit_name()
# create unique user and vhost for each unit
current_unit = local_unit().replace('/', '-')
user = '******' % current_unit
vhost = 'nagios-%s' % current_unit
password = rabbit.get_rabbit_password(user, local=True)
rabbit.create_vhost(vhost)
rabbit.create_user(user, password)
rabbit.grant_permissions(user, vhost)
nrpe_compat = nrpe.NRPE(hostname=hostname)
nrpe_compat.add_check(
shortname=rabbit.RABBIT_USER,
description='Check RabbitMQ {%s}' % myunit,
check_cmd='{}/check_rabbitmq.py --user {} --password {} --vhost {}'
''.format(NAGIOS_PLUGINS, user, password, vhost)
)
if config('queue_thresholds'):
cmd = ""
# If value of queue_thresholds is incorrect we want the hook to fail
for item in yaml.safe_load(config('queue_thresholds')):
cmd += ' -c "{}" "{}" {} {}'.format(*item)
nrpe_compat.add_check(
shortname=rabbit.RABBIT_USER + '_queue',
description='Check RabbitMQ Queues',
check_cmd='{}/check_rabbitmq_queues.py{} {}'.format(
NAGIOS_PLUGINS, cmd, STATS_DATAFILE)
)
nrpe_compat.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe_setup.add_check(
shortname='ceph-osd',
description='process check {%s}' % current_unit,
check_cmd=('/bin/cat /var/lib/ceph/osd/ceph-*/whoami |'
'xargs -I@ status ceph-osd id=@ && exit 0 || exit 2')
)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
if not install_nrpe_deps():
return
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe_setup.add_check(
shortname='elasticsearch-cluster-status',
description='Elasticsearch cluster status check {%s}' % current_unit,
check_cmd=('/usr/local/bin/check-elasticsearch'))
nrpe_setup.write()
def remove_nrpe_config(nagios=None):
remove_state('nrpe-external-master.initial-config')
# List of systemd services for which the checks will be removed
services = ('snap.kubelet.daemon', 'snap.kube-proxy.daemon')
# The current nrpe-external-master interface doesn't handle a lot of logic,
# use the charm-helpers code for now.
hostname = nrpe.get_nagios_hostname()
nrpe_setup = nrpe.NRPE(hostname=hostname)
for service in services:
nrpe_setup.remove_check(shortname=service)
def remove_nrpe_config(nagios=None):
remove_state('nrpe-external-master.initial-config')
# List of systemd services for which the checks will be removed
services = ('snap.kubelet.daemon', 'snap.kube-proxy.daemon')
# The current nrpe-external-master interface doesn't handle a lot of logic,
# use the charm-helpers code for now.
hostname = nrpe.get_nagios_hostname()
nrpe_setup = nrpe.NRPE(hostname=hostname)
for service in services:
nrpe_setup.remove_check(shortname=service)
def remove_nrpe_config(nagios=None):
remove_state(NPRE_EXTERNAL_RELATION + ".initial-config")
# List of systemd services for which the checks will be removed
services = ("snap.etcd.etcd", )
# The current nrpe-external interface doesn't handle a lot of logic,
# use the charm-helpers code for now.
hostname = nrpe.get_nagios_hostname()
nrpe_setup = nrpe.NRPE(hostname=hostname, primary=False)
for service in services:
nrpe_setup.remove_check(shortname=service)
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, 'mysql', current_unit)
nrpe_setup.add_check(
shortname='mysql_proc',
description='Check MySQL process {%s}' % current_unit,
check_cmd='check_procs -c 1:1 -C mysqld'
)
nrpe_setup.write()
def update_nrpe_checks():
if os.path.isdir(NAGIOS_PLUGINS):
rsync(os.path.join(os.getenv('CHARM_DIR'), 'scripts',
'check_rabbitmq.py'),
os.path.join(NAGIOS_PLUGINS, 'check_rabbitmq.py'))
rsync(os.path.join(os.getenv('CHARM_DIR'), 'scripts',
'check_rabbitmq_queues.py'),
os.path.join(NAGIOS_PLUGINS, 'check_rabbitmq_queues.py'))
if config('stats_cron_schedule'):
script = os.path.join(SCRIPTS_DIR, 'collect_rabbitmq_stats.sh')
cronjob = "{} root {}\n".format(config('stats_cron_schedule'), script)
rsync(os.path.join(charm_dir(), 'scripts',
'collect_rabbitmq_stats.sh'), script)
write_file(STATS_CRONFILE, cronjob)
elif os.path.isfile(STATS_CRONFILE):
os.remove(STATS_CRONFILE)
# Find out if nrpe set nagios_hostname
hostname = nrpe.get_nagios_hostname()
myunit = nrpe.get_nagios_unit_name()
# create unique user and vhost for each unit
current_unit = local_unit().replace('/', '-')
user = '******' % current_unit
vhost = 'nagios-%s' % current_unit
password = rabbit.get_rabbit_password(user, local=True)
rabbit.create_vhost(vhost)
rabbit.create_user(user, password)
rabbit.grant_permissions(user, vhost)
nrpe_compat = nrpe.NRPE(hostname=hostname)
nrpe_compat.add_check(
shortname=rabbit.RABBIT_USER,
description='Check RabbitMQ {%s}' % myunit,
check_cmd='{}/check_rabbitmq.py --user {} --password {} --vhost {}'
''.format(NAGIOS_PLUGINS, user, password, vhost)
)
if config('queue_thresholds'):
cmd = ""
# If value of queue_thresholds is incorrect we want the hook to fail
for item in yaml.safe_load(config('queue_thresholds')):
cmd += ' -c "{}" "{}" {} {}'.format(*item)
nrpe_compat.add_check(
shortname=rabbit.RABBIT_USER + '_queue',
description='Check RabbitMQ Queues',
check_cmd='{}/check_rabbitmq_queues.py{} {}'.format(
NAGIOS_PLUGINS, cmd, STATS_DATAFILE)
)
nrpe_compat.write()
def update_nrpe_config():
"""
:return: None
"""
# List of systemd services that will be checked.
services = ['docker']
# The current nrpe-external-master interface doesn't handle a lot of logic,
# use the charm-helpers code for now.
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
monitored_services = services()
try:
# qemu-kvm is a one-shot service
monitored_services.remove('qemu-kvm')
except ValueError:
pass
nrpe.add_init_service_checks(nrpe_setup, monitored_services, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.copy_nrpe_checks()
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
if service_enabled('api'):
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
else:
nrpe.remove_deprecated_check(nrpe_setup,
["haproxy_servers", "haproxy_queue"])
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
monitored_services = services()
try:
# qemu-kvm is a one-shot service
monitored_services.remove('qemu-kvm')
except ValueError:
pass
nrpe.add_init_service_checks(nrpe_setup, monitored_services, current_unit)
nrpe_setup.write()
def update_nrpe_config():
scripts_src = os.path.join(os.environ["CHARM_DIR"], "files",
"nrpe")
scripts_dst = "/usr/local/lib/nagios/plugins"
if not os.path.exists(scripts_dst):
os.makedirs(scripts_dst)
for fname in glob.glob(os.path.join(scripts_src, "*")):
if os.path.isfile(fname):
shutil.copy2(fname,
os.path.join(scripts_dst, os.path.basename(fname)))
sudoers_src = os.path.join(os.environ["CHARM_DIR"], "files",
"sudoers")
sudoers_dst = "/etc/sudoers.d"
for fname in glob.glob(os.path.join(sudoers_src, "*")):
if os.path.isfile(fname):
shutil.copy2(fname,
os.path.join(sudoers_dst, os.path.basename(fname)))
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
apt_install('python-dbus')
# corosync/crm checks
nrpe_setup.add_check(
shortname='corosync_rings',
description='Check Corosync rings {%s}' % current_unit,
check_cmd='check_corosync_rings')
nrpe_setup.add_check(
shortname='crm_status',
description='Check crm status {%s}' % current_unit,
check_cmd='check_crm')
# process checks
nrpe_setup.add_check(
shortname='corosync_proc',
description='Check Corosync process {%s}' % current_unit,
check_cmd='check_procs -c 1:1 -C corosync'
)
nrpe_setup.add_check(
shortname='pacemakerd_proc',
description='Check Pacemakerd process {%s}' % current_unit,
check_cmd='check_procs -c 1:1 -C pacemakerd'
)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
# fasteners is used by apt_install collect_ceph_osd_services.py
pkgs = ['python3-dbus']
if CompareHostReleases(lsb_release()['DISTRIB_CODENAME']) >= 'bionic':
pkgs.append('python3-fasteners')
apt_install(pkgs)
# copy the check and collect files over to the plugins directory
charm_dir = os.environ.get('CHARM_DIR', '')
nagios_plugins = '/usr/local/lib/nagios/plugins'
# Grab nagios user/group ID's from original source
_dir = os.stat(nagios_plugins)
uid = _dir.st_uid
gid = _dir.st_gid
for name in ('collect_ceph_osd_services.py', 'check_ceph_osd_services.py'):
target = os.path.join(nagios_plugins, name)
shutil.copy(os.path.join(charm_dir, 'files', 'nagios', name), target)
os.chown(target, uid, gid)
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
# BUG#1810749 - the nagios user can't access /var/lib/ceph/.. and that's a
# GOOD THING, as it keeps ceph secure from Nagios. However, to check
# whether ceph is okay, the check_systemd.py or 'status ceph-osd' still
# needs to be called with the contents of ../osd/ceph-*/whoami files. To
# get around this conundrum, instead a cron.d job that runs as root will
# perform the checks every minute, and write to a tempory file the results,
# and the nrpe check will grep this file and error out (return 2) if the
# first 3 characters of a line are not 'OK:'.
cmd = ('MAILTO=""\n'
'* * * * * root '
'/usr/local/lib/nagios/plugins/collect_ceph_osd_services.py'
' 2>&1 | logger -t check-osd\n')
with open(CRON_CEPH_CHECK_FILE, "wt") as f:
f.write(cmd)
nrpe_cmd = '/usr/local/lib/nagios/plugins/check_ceph_osd_services.py'
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe_setup.add_check(
shortname='ceph-osd',
description='process check {%s}' % current_unit,
check_cmd=nrpe_cmd
)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
log('Updating NRPE configuration')
status_set('maintenance', 'Updating NRPE configuration')
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.copy_nrpe_checks()
_services = []
for service in services():
if service.startswith('snap.'):
service = service.split('.')[1]
_services.append(service)
nrpe.add_init_service_checks(nrpe_setup, _services, current_unit)
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.copy_nrpe_checks()
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
nrpe.add_haproxy_checks(nrpe_setup, current_unit)
conf = nrpe_setup.config
check_http_params = conf.get('nagios_check_http_params')
if check_http_params:
nrpe_setup.add_check(
shortname='vhost',
description='Check Virtual Host {%s}' % current_unit,
check_cmd='check_http %s' % check_http_params
)
nrpe_setup.write()
def update_nrpe_config():
# python-dbus is used by check_upstart_job
apt_install('python-dbus')
hostname = nrpe.get_nagios_hostname()
current_unit = nrpe.get_nagios_unit_name()
nrpe_setup = nrpe.NRPE(hostname=hostname)
nrpe.add_init_service_checks(nrpe_setup, services(), current_unit)
cronpath = '/etc/cron.d/nagios-netns-check'
cron_template = ('*/5 * * * * root '
'/usr/local/lib/nagios/plugins/check_netns.sh '
'> /var/lib/nagios/netns-check.txt\n'
)
f = open(cronpath, 'w')
f.write(cron_template)
f.close()
nrpe_setup.add_check(
shortname="netns",
description='Network Namespace check {%s}' % current_unit,
check_cmd='check_status_file.py -f /var/lib/nagios/netns-check.txt'
)
nrpe_setup.write()
def update_nrpe_checks():
if os.path.isdir(NAGIOS_PLUGINS):
rsync(os.path.join(charm_dir(), 'scripts',
'check_rabbitmq.py'),
os.path.join(NAGIOS_PLUGINS, 'check_rabbitmq.py'))
rsync(os.path.join(charm_dir(), 'scripts',
'check_rabbitmq_queues.py'),
os.path.join(NAGIOS_PLUGINS, 'check_rabbitmq_queues.py'))
if config('management_plugin'):
rsync(os.path.join(charm_dir(), 'scripts',
'check_rabbitmq_cluster.py'),
os.path.join(NAGIOS_PLUGINS, 'check_rabbitmq_cluster.py'))
if config('stats_cron_schedule'):
script = os.path.join(SCRIPTS_DIR, 'collect_rabbitmq_stats.sh')
cronjob = CRONJOB_CMD.format(schedule=config('stats_cron_schedule'),
timeout=config('cron-timeout'),
command=script)
rsync(os.path.join(charm_dir(), 'scripts',
'collect_rabbitmq_stats.sh'), script)
write_file(STATS_CRONFILE, cronjob)
elif os.path.isfile(STATS_CRONFILE):
os.remove(STATS_CRONFILE)
# Find out if nrpe set nagios_hostname
hostname = nrpe.get_nagios_hostname()
myunit = nrpe.get_nagios_unit_name()
# create unique user and vhost for each unit
current_unit = local_unit().replace('/', '-')
user = '******' % current_unit
vhost = 'nagios-%s' % current_unit
password = rabbit.get_rabbit_password(user, local=True)
rabbit.create_vhost(vhost)
rabbit.create_user(user, password, ['monitoring'])
rabbit.grant_permissions(user, vhost)
nrpe_compat = nrpe.NRPE(hostname=hostname)
if config('ssl') in ['off', 'on']:
cmd = ('{plugins_dir}/check_rabbitmq.py --user {user} '
'--password {password} --vhost {vhost}')
cmd = cmd.format(plugins_dir=NAGIOS_PLUGINS, user=user,
password=password, vhost=vhost)
nrpe_compat.add_check(
shortname=rabbit.RABBIT_USER,
description='Check RabbitMQ {%s}' % myunit,
check_cmd=cmd
)
if config('ssl') in ['only', 'on']:
log('Adding rabbitmq SSL check', level=DEBUG)
cmd = ('{plugins_dir}/check_rabbitmq.py --user {user} '
'--password {password} --vhost {vhost} '
'--ssl --ssl-ca {ssl_ca} --port {port}')
cmd = cmd.format(plugins_dir=NAGIOS_PLUGINS,
user=user,
password=password,
port=int(config('ssl_port')),
vhost=vhost,
ssl_ca=SSL_CA_FILE)
nrpe_compat.add_check(
shortname=rabbit.RABBIT_USER + "_ssl",
description='Check RabbitMQ (SSL) {%s}' % myunit,
check_cmd=cmd
)
if config('queue_thresholds'):
cmd = ""
# If value of queue_thresholds is incorrect we want the hook to fail
for item in yaml.safe_load(config('queue_thresholds')):
cmd += ' -c "{}" "{}" {} {}'.format(*item)
nrpe_compat.add_check(
shortname=rabbit.RABBIT_USER + '_queue',
description='Check RabbitMQ Queues',
check_cmd='{}/check_rabbitmq_queues.py{} {}'.format(
NAGIOS_PLUGINS, cmd, STATS_DATAFILE)
)
if config('management_plugin'):
# add NRPE check
nrpe_compat.add_check(
shortname=rabbit.RABBIT_USER + '_cluster',
description='Check RabbitMQ Cluster',
check_cmd='{}/check_rabbitmq_cluster.py --port {} --user {} --password {}'.format(
NAGIOS_PLUGINS,
rabbit.get_managment_port(),
user,
password
)
)
nrpe_compat.write()
