Monitoring/nagios-plugins
1
0
Fork 0
mirror of https://github.com/opinkerfi/nagios-plugins.git synced 2024-11-22 10:23:46 +01:00
Code Issues Releases Wiki Activity

Added check for suspended services -Z

Browse Source
This commit is contained in:
Tomas Edwardsson 2012-12-13 15:07:18 +00:00
parent 745c719ac8
commit 824665b6d9
1 changed files with 12 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
check_rhcs/check_rhcs
View File

@ -40,6 +40,8 @@ Usage: """ + sys.argv[0] + """ ([-s serviceName] | [-c])
Gathers the overall cluster status for the local node Gathers the overall cluster status for the local node
-s, --service -s, --service
Gets the stats of the named service Gets the stats of the named service
-Z, --suspended
Checks whether there are any suspended services
-h, --help -h, --help
Display this Display this
""" """
@ -99,23 +101,27 @@ def getServiceState(dom, service):
if group.attributes['name'].value in (service,"service:"+service,"vm:"+service): if group.attributes['name'].value in (service,"service:"+service,"vm:"+service):
serviceState['owner'] = group.attributes['owner'].value serviceState['owner'] = group.attributes['owner'].value
serviceState['state'] = group.attributes['state_str'].value serviceState['state'] = group.attributes['state_str'].value
serviceState['flags'] = group.attributes['flags_str'].value
return serviceState return serviceState
def main(): def main():
try: try:
opts, args = getopt.getopt(sys.argv[1:], 's:ch', ['service=', 'cluster', 'help']) opts, args = getopt.getopt(sys.argv[1:], 's:cZh', ['service=', 'cluster', 'supsended', 'help'])
except getopt.GetoptError: except getopt.GetoptError:
usage() usage()
sys.exit(2) sys.exit(2)
check_suspend = False
for o, a in opts: for o, a in opts:
if o in ('-c', '--cluster'): if o in ('-c', '--cluster'):
typeCheck = 'cluster' typeCheck = 'cluster'
if o in ('-s', '--service'): if o in ('-s', '--service'):
typeCheck = 'service' typeCheck = 'service'
serviceName = a serviceName = a
if o in ('-Z', '--suspended'):
check_suspend = True
if o in ('-h', '--help'): if o in ('-h', '--help'):
usage() usage()
sys.exit() sys.exit()
@ -124,8 +130,8 @@ def main():
clustatOutput = os.popen('/usr/sbin/clustat -fx') clustatOutput = os.popen('/usr/sbin/clustat -fx')
dom = xml.dom.minidom.parse(clustatOutput) dom = xml.dom.minidom.parse(clustatOutput)
except Exception, e: except Exception, e:
print "Error: could not parse output of : '/usr/sbin/clustat -fx': ", e print "Error: could not parse output of : '/usr/sbin/clustat -fx': ", e
sys.exit(3) sys.exit(3)
if typeCheck == 'cluster': if typeCheck == 'cluster':
# First we query for the state of the cluster itself. # First we query for the state of the cluster itself.
@ -156,6 +162,9 @@ def main():
if serviceState['state'] != 'started': if serviceState['state'] != 'started':
print "CRITICAL: Service " + serviceName + " on " + serviceState['owner'] + " is in " + serviceState['state'] + " state" print "CRITICAL: Service " + serviceName + " on " + serviceState['owner'] + " is in " + serviceState['state'] + " state"
sys.exit(2) sys.exit(2)
elif check_suspend is True and serviceState['flags'] == 'frozen':
print "WARNING: Service " + serviceName + " on " + serviceState['owner'] + " is in " + serviceState['flags'] + " state"
sys.exit(1)
else: else:
print "OK: Service " + serviceName + " on " + serviceState['owner'] + " is in " + serviceState['state'] + " state" print "OK: Service " + serviceName + " on " + serviceState['owner'] + " is in " + serviceState['state'] + " state"
sys.exit(0) sys.exit(0)