Infrastructure/ansible
1
0
Fork 0
Code Issues Pull requests 6 Projects Releases Packages Wiki Activity Actions

Also, improve the other two new fedmsg checks.

Browse source
This commit is contained in:
Ralph Bean 2014-05-19 14:26:38 +00:00
parent fb03577eea
commit 7b84ea9663
2 changed files with 33 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

15
roles/nagios_client/files/scripts/check_fedmsg_consumer_backlog.py
View file

@ -12,16 +12,29 @@ try:
backlog_warning = int(sys.argv[3]) backlog_warning = int(sys.argv[3])
backlog_critical = int(sys.argv[4]) backlog_critical = int(sys.argv[4])
fname = '/var/run/fedmsg/monitoring-%s.socket' % service fname = '/var/run/fedmsg/monitoring-%s.socket' % service
if not os.path.exists(fname): if not os.path.exists(fname):
print "UNKNOWN - %s does not exist" % fname print "UNKNOWN - %s does not exist" % fname
sys.exit(3) sys.exit(3)
connect_to = "ipc:///%s" % fname connect_to = "ipc:///%s" % fname
ctx = zmq.Context() ctx = zmq.Context()
s = ctx.socket(zmq.SUB) s = ctx.socket(zmq.SUB)
s.connect(connect_to) s.connect(connect_to)
s.setsockopt(zmq.SUBSCRIBE, '') s.setsockopt(zmq.SUBSCRIBE, '')
poller = zmq.Poller()
poller.register(s, zmq.POLLIN)
timeout = 10000
events = dict(poller.poll(timeout))
if s in events and events[s] == zmq.POLLIN:
msg = s.recv() msg = s.recv()
msg = json.loads(msg) msg = json.loads(msg)
else:
print 'UNKNOWN - ZMQ timeout. No message received in %i ms' % timeout
sys.exit(3)
for consumer in msg['consumers']: for consumer in msg['consumers']:
if consumer['name'] == check_consumer: if consumer['name'] == check_consumer:
@ -38,6 +51,8 @@ try:
print 'OK: fedmsg consumer %s backlog value is %i' % (consumer['name'],consumer['backlog']) print 'OK: fedmsg consumer %s backlog value is %i' % (consumer['name'],consumer['backlog'])
sys.exit(0) sys.exit(0)
print "UNKNOWN: fedmsg consumer %s not found" % check_consumer
sys.exit(3)
except Exception as err: except Exception as err:
print "UNKNOWN:", str(err) print "UNKNOWN:", str(err)
sys.exit(3) sys.exit(3)

14
roles/nagios_client/files/scripts/check_fedmsg_consumer_exceptions.py
View file

@ -12,16 +12,28 @@ try:
exceptions_warning = int(sys.argv[3]) exceptions_warning = int(sys.argv[3])
exceptions_critical = int(sys.argv[4]) exceptions_critical = int(sys.argv[4])
fname = '/var/run/fedmsg/monitoring-%s.socket' % service fname = '/var/run/fedmsg/monitoring-%s.socket' % service
if not os.path.exists(fname): if not os.path.exists(fname):
print "UNKNOWN - %s does not exist" % fname print "UNKNOWN - %s does not exist" % fname
sys.exit(3) sys.exit(3)
connect_to = "ipc:///%s" % fname connect_to = "ipc:///%s" % fname
ctx = zmq.Context() ctx = zmq.Context()
s = ctx.socket(zmq.SUB) s = ctx.socket(zmq.SUB)
s.connect(connect_to) s.connect(connect_to)
s.setsockopt(zmq.SUBSCRIBE, '') s.setsockopt(zmq.SUBSCRIBE, '')
poller = zmq.Poller()
poller.register(s, zmq.POLLIN)
timeout = 10000
events = dict(poller.poll(timeout))
if s in events and events[s] == zmq.POLLIN:
msg = s.recv() msg = s.recv()
msg = json.loads(msg) msg = json.loads(msg)
else:
print 'UNKNOWN - ZMQ timeout. No message received in %i ms' % timeout
sys.exit(3)
for consumer in msg['consumers']: for consumer in msg['consumers']:
if consumer['name'] == check_consumer: if consumer['name'] == check_consumer:
@ -35,6 +47,8 @@ try:
print 'OK: fedmsg consumer %s exceptions value is %i' % (consumer['name'],consumer['exceptions']) print 'OK: fedmsg consumer %s exceptions value is %i' % (consumer['name'],consumer['exceptions'])
sys.exit(0) sys.exit(0)
print "UNKNOWN: fedmsg consumers %s not found" % check_consumer
sys.exit(3)
except Exception as err: except Exception as err:
print "UNKNOWN:", str(err) print "UNKNOWN:", str(err)
sys.exit(3) sys.exit(3)