1
"""Compatibility with the nrpe-external-master charm"""
2
# Copyright 2012 Canonical Ltd.
5
# Matthew Wedgwood <matthew.wedgwood@canonical.com>
15
from charmhelpers.core.hookenv import (
24
from charmhelpers.core.host import service
26
# This module adds compatibility with the nrpe-external-master and plain nrpe
27
# subordinate charms. To use it in your charm:
29
# 1. Update metadata.yaml
33
# nrpe-external-master:
34
# interface: nrpe-external-master
42
# interface: local-monitors
46
# 2. Add the following to config.yaml
52
# Used by the nrpe subordinate charms.
53
# A string that will be prepended to instance name to set the host name
54
# in nagios. So for instance the hostname would be something like:
56
# If you're running multiple environments with the same services in them
57
# this allows you to differentiate between them.
58
# nagios_servicegroups:
62
# A comma-separated list of nagios servicegroups.
63
# If left empty, the nagios_context will be used as the servicegroup
65
# 3. Add custom checks (Nagios plugins) to files/nrpe-external-master
67
# 4. Update your hooks.py with something like this:
69
# from charmsupport.nrpe import NRPE
71
# def update_nrpe_config():
72
# nrpe_compat = NRPE()
73
# nrpe_compat.add_check(
74
# shortname = "myservice",
75
# description = "Check MyService",
76
# check_cmd = "check_http -w 2 -c 10 http://localhost"
78
# nrpe_compat.add_check(
80
# "Check for widget failures",
81
# check_cmd = "/srv/myapp/scripts/widget_check"
85
# def config_changed():
87
# update_nrpe_config()
89
# def nrpe_external_master_relation_changed():
90
# update_nrpe_config()
92
# def local_monitors_relation_changed():
93
# update_nrpe_config()
95
# 5. ln -s hooks.py nrpe-external-master-relation-changed
96
# ln -s hooks.py local-monitors-relation-changed
99
class CheckException(Exception):
104
shortname_re = '[A-Za-z0-9-_]+$'
105
service_template = ("""
106
#---------------------------------------------------
107
# This file is Juju managed
108
#---------------------------------------------------
111
host_name {nagios_hostname}
112
service_description {nagios_hostname}[{shortname}] """
114
check_command check_nrpe!{command}
115
servicegroups {nagios_servicegroup}
119
def __init__(self, shortname, description, check_cmd):
120
super(Check, self).__init__()
121
# XXX: could be better to calculate this from the service name
122
if not re.match(self.shortname_re, shortname):
123
raise CheckException("shortname must match {}".format(
125
self.shortname = shortname
126
self.command = "check_{}".format(shortname)
127
# Note: a set of invalid characters is defined by the
128
# Nagios server config
129
# The default is: illegal_object_name_chars=`~!$%^&*"|'<>?,()=
130
self.description = description
131
self.check_cmd = self._locate_cmd(check_cmd)
133
def _locate_cmd(self, check_cmd):
135
'/usr/lib/nagios/plugins',
136
'/usr/local/lib/nagios/plugins',
138
parts = shlex.split(check_cmd)
139
for path in search_path:
140
if os.path.exists(os.path.join(path, parts[0])):
141
command = os.path.join(path, parts[0])
143
command += " " + " ".join(parts[1:])
145
log('Check command not found: {}'.format(parts[0]))
148
def write(self, nagios_context, hostname, nagios_servicegroups=None):
149
nrpe_check_file = '/etc/nagios/nrpe.d/{}.cfg'.format(
151
with open(nrpe_check_file, 'w') as nrpe_check_config:
152
nrpe_check_config.write("# check {}\n".format(self.shortname))
153
nrpe_check_config.write("command[{}]={}\n".format(
154
self.command, self.check_cmd))
156
if not os.path.exists(NRPE.nagios_exportdir):
157
log('Not writing service config as {} is not accessible'.format(
158
NRPE.nagios_exportdir))
160
self.write_service_config(nagios_context, hostname,
161
nagios_servicegroups)
163
def write_service_config(self, nagios_context, hostname,
164
nagios_servicegroups=None):
165
for f in os.listdir(NRPE.nagios_exportdir):
166
if re.search('.*{}.cfg'.format(self.command), f):
167
os.remove(os.path.join(NRPE.nagios_exportdir, f))
169
if not nagios_servicegroups:
170
nagios_servicegroups = nagios_context
173
'nagios_hostname': hostname,
174
'nagios_servicegroup': nagios_servicegroups,
175
'description': self.description,
176
'shortname': self.shortname,
177
'command': self.command,
179
nrpe_service_text = Check.service_template.format(**templ_vars)
180
nrpe_service_file = '{}/service__{}_{}.cfg'.format(
181
NRPE.nagios_exportdir, hostname, self.command)
182
with open(nrpe_service_file, 'w') as nrpe_service_config:
183
nrpe_service_config.write(str(nrpe_service_text))
186
subprocess.call(self.check_cmd)
190
nagios_logdir = '/var/log/nagios'
191
nagios_exportdir = '/var/lib/nagios/export'
192
nrpe_confdir = '/etc/nagios/nrpe.d'
194
def __init__(self, hostname=None):
195
super(NRPE, self).__init__()
196
self.config = config()
197
self.nagios_context = self.config['nagios_context']
198
if 'nagios_servicegroups' in self.config:
199
self.nagios_servicegroups = self.config['nagios_servicegroups']
201
self.nagios_servicegroups = 'juju'
202
self.unit_name = local_unit().replace('/', '-')
204
self.hostname = hostname
206
self.hostname = "{}-{}".format(self.nagios_context, self.unit_name)
209
def add_check(self, *args, **kwargs):
210
self.checks.append(Check(*args, **kwargs))
214
nagios_uid = pwd.getpwnam('nagios').pw_uid
215
nagios_gid = grp.getgrnam('nagios').gr_gid
217
log("Nagios user not set up, nrpe checks not updated")
220
if not os.path.exists(NRPE.nagios_logdir):
221
os.mkdir(NRPE.nagios_logdir)
222
os.chown(NRPE.nagios_logdir, nagios_uid, nagios_gid)
225
monitors = {"monitors": {"remote": {"nrpe": nrpe_monitors}}}
226
for nrpecheck in self.checks:
227
nrpecheck.write(self.nagios_context, self.hostname,
228
self.nagios_servicegroups)
229
nrpe_monitors[nrpecheck.shortname] = {
230
"command": nrpecheck.command,
233
service('restart', 'nagios-nrpe-server')
235
for rid in relation_ids("local-monitors"):
236
relation_set(relation_id=rid, monitors=yaml.dump(monitors))
239
def get_nagios_hostcontext(relation_name='nrpe-external-master'):
241
Query relation with nrpe subordinate, return the nagios_host_context
243
:param str relation_name: Name of relation nrpe sub joined to
245
for rel in relations_of_type(relation_name):
246
if 'nagios_hostname' in rel:
247
return rel['nagios_host_context']
250
def get_nagios_hostname(relation_name='nrpe-external-master'):
252
Query relation with nrpe subordinate, return the nagios_hostname
254
:param str relation_name: Name of relation nrpe sub joined to
256
for rel in relations_of_type(relation_name):
257
if 'nagios_hostname' in rel:
258
return rel['nagios_hostname']
261
def get_nagios_unit_name(relation_name='nrpe-external-master'):
263
Return the nagios unit name prepended with host_context if needed
265
:param str relation_name: Name of relation nrpe sub joined to
267
host_context = get_nagios_hostcontext(relation_name)
269
unit = "%s:%s" % (host_context, local_unit())
275
def add_init_service_checks(nrpe, services, unit_name):
277
Add checks for each service in list
279
:param NRPE nrpe: NRPE object to add check to
280
:param list services: List of services to check
281
:param str unit_name: Unit name to use in check description
284
upstart_init = '/etc/init/%s.conf' % svc
285
sysv_init = '/etc/init.d/%s' % svc
286
if os.path.exists(upstart_init):
289
description='process check {%s}' % unit_name,
290
check_cmd='check_upstart_job %s' % svc
292
elif os.path.exists(sysv_init):
293
cronpath = '/etc/cron.d/nagios-service-check-%s' % svc
294
cron_file = ('*/5 * * * * root '
295
'/usr/local/lib/nagios/plugins/check_exit_status.pl '
296
'-s /etc/init.d/%s status > '
297
'/var/lib/nagios/service-check-%s.txt\n' % (svc,
300
f = open(cronpath, 'w')
305
description='process check {%s}' % unit_name,
306
check_cmd='check_status_file.py -f '
307
'/var/lib/nagios/service-check-%s.txt' % svc,