def checkVolumePerfData(clusterName):
# Write command to socket
cmd = ("GET services\nColumns: description host_name "
"perf_data custom_variables\n"
"Filter: host_name = %s\n"
"Filter: description ~~ %s\n" %
(clusterName, 'Volume Utilization -'))
perf_data_output = livestatus.readLiveStatusAsJSON(cmd)
perf_data = json.loads(perf_data_output)
numvolumes = 0
totalUsed = 0.0
totalAvail = 0.0
for row in perf_data:
numvolumes += 1
if len(row) <= 3:
return 0.0, 0.0
perf_data = row[2]
if len(perf_data) > 2:
perf_arr = perf_data.split(' ')
used = perf_arr[2].split('=')[1]
avail = perf_arr[1].split('=')[1]
totalUsed += float(re.match(r'\d*\.?\d+', used).group())
totalAvail += float(re.match(r'\d*\.?\d+', avail).group())
return numvolumes, totalUsed, totalAvail
def checkVolumePerfData(clusterName):
# Write command to socket
cmd = (
"GET services\nColumns: description host_name "
"perf_data custom_variables\n"
"Filter: host_name = %s\n"
"Filter: description ~~ %s\n" % (clusterName, "Volume Utilization -")
)
perf_data_output = livestatus.readLiveStatusAsJSON(cmd)
perf_data = json.loads(perf_data_output)
numvolumes = 0
totalUsed = 0.0
totalAvail = 0.0
for row in perf_data:
numvolumes += 1
if len(row) <= 3:
return 0.0, 0.0
perf_data = row[2]
if len(perf_data) > 2:
perf_arr = perf_data.split(" ")
used = perf_arr[2].split("=")[1]
avail = perf_arr[1].split("=")[1]
totalUsed += float(re.match(r"\d*\.?\d+", used).group())
totalAvail += float(re.match(r"\d*\.?\d+", avail).group())
return numvolumes, totalUsed, totalAvail
def _getGlusterdStatus(hostname):
status = json.loads(
livestatus.readLiveStatusAsJSON(
"GET services\nColumns: state\n"
"Filter: description = Gluster Management\n"
"Filter: host_name = " + hostname + "\n"))[0][0]
return status
def _getListHosts(hostgroup):
list_hosts = []
table = json.loads(livestatus.readLiveStatusAsJSON(
"GET hostgroups\nColumns: members_with_state\n"
"Filter: name = " + hostgroup + "\n"))[0][0]
# Get the only those nodes which are UP and
# glusterd service is running
for row in table:
if row[1] == utils.HostStatusCode.UP and \
_getGlusterdStatus(row[0]) \
== utils.PluginStatusCode.OK:
list_hosts.append(row[0])
return list_hosts
def _getListHosts(hostgroup):
list_hosts = []
table = json.loads(
livestatus.readLiveStatusAsJSON(
"GET hostgroups\nColumns: members_with_state\n"
"Filter: name = " + hostgroup + "\n"))[0][0]
# Get the only those nodes which are UP and
# glusterd service is running
for row in table:
if row[1] == utils.HostStatusCode.UP and \
_getGlusterdStatus(row[0]) \
== utils.PluginStatusCode.OK:
list_hosts.append(row[0])
return list_hosts
def _getQuorumStatusOutput(hostgroup):
# get current volume quorum status
table = json.loads(livestatus.readLiveStatusAsJSON("GET services\n"
"Columns: state plugin_output\n"
"Filter: description = "
"Cluster - Quorum Status\n"
"Filter: host_name = %s\n" % hostgroup))
servicestatus = utils.PluginStatusCode.UNKNOWN
pluginoutput = ''
for row in table:
servicestatus = row[0]
pluginoutput = row[1]
if (int(servicestatus) != utils.PluginStatusCode.CRITICAL):
return _executeRandomHost(hostgroup, _getQuorumStatusNRPECommand())
else:
return servicestatus, pluginoutput
def _getQuorumStatusOutput(hostgroup):
# get current volume quorum status
table = json.loads(
livestatus.readLiveStatusAsJSON("GET services\n"
"Columns: state plugin_output\n"
"Filter: description = "
"Cluster - Quorum Status\n"
"Filter: host_name = %s\n" %
hostgroup))
servicestatus = utils.PluginStatusCode.UNKNOWN
pluginoutput = ''
for row in table:
servicestatus = row[0]
pluginoutput = row[1]
if (int(servicestatus) != utils.PluginStatusCode.CRITICAL):
return _executeRandomHost(hostgroup, _getQuorumStatusNRPECommand())
else:
return servicestatus, pluginoutput
def _getVolumeStatusOutput(hostgroup, volume):
status, output = _executeRandomHost(hostgroup,
_getVolStatusNRPECommand(volume))
if status == utils.PluginStatusCode.OK:
brick_details = json.loads(
livestatus.readLiveStatusAsJSON(
"GET services\n"
"Columns: description state host_address host_name\n"
"Filter: host_groups >= %s\n"
"Filter: custom_variable_values >= %s\n"
"Filter: description ~ Brick - \n" % (hostgroup, volume)))
# output will be as below:
# [[u'Brick - /root/b3', 0, u'10.70.42.246', u'nishanth-rhs-2']]
# parse this to find the no of critical/ok bricks and list of
# critical bricks
bricks_ok = 0
bricks_critical = 0
brick_list_critical = []
for brick_detail in brick_details:
if brick_detail[1] == utils.PluginStatusCode.OK:
bricks_ok += 1
elif brick_detail[1] == utils.PluginStatusCode.CRITICAL:
bricks_critical += 1
# get the critical brick's host uuid if not present
# int the list
custom_vars = json.loads(
livestatus.readLiveStatusAsJSON(
"GET hosts\n"
"Columns: custom_variables\n"
"Filter: groups >= %s\n"
"Filter: name = %s\n" % (hostgroup, brick_detail[3])))
brick_dict = {}
brick_dict['brick'] = brick_detail[2] + ":" + \
brick_detail[0][brick_detail[0].find("/"):]
brick_dict['uuid'] = custom_vars[0][0]['HOST_UUID']
brick_list_critical.append(brick_dict)
# Get volume details
nrpeStatus, nrpeOut = _executeRandomHost(
hostgroup, _getVolDetailNRPECommand(volume))
if nrpeStatus != utils.PluginStatusCode.OK:
return utils.PluginStatusCode.UNKNOWN, nrpeOut
volInfo = json.loads(nrpeOut)
# Get the volume type
vol_type = volInfo[volume]['type']
if bricks_ok == 0 and bricks_critical > 0:
status = utils.PluginStatusCode.CRITICAL
output = "CRITICAL: Volume : %s type - All bricks " \
"are down " % (vol_type)
elif bricks_ok > 0 and bricks_critical == 0:
status = utils.PluginStatusCode.OK
output = "OK: Volume : %s type - All bricks " \
"are Up " % (vol_type)
elif bricks_critical > 0:
if (vol_type == "DISTRIBUTE"):
status = utils.PluginStatusCode.CRITICAL
output = "CRITICAL: Volume : %s type \n Brick(s) - <%s> " \
"is|are down " % \
(vol_type, ', '.join(dict['brick']for dict in
brick_list_critical))
elif (vol_type == "DISTRIBUTED_REPLICATE"
or vol_type == "REPLICATE"
or vol_type == "DISTRIBUTED_DISPERSE"
or vol_type == "DISPERSE"):
status, output = _getDisperseOrReplicaStatus(
vol_type, brick_list_critical, volInfo, volume)
else:
output = "WARNING: Volume : %s type \n Brick(s) - <%s> " \
"is|are down" % (vol_type,
', '.join(dict['brick']
for dict in
brick_list_critical))
status = utils.PluginStatusCode.WARNING
return status, output
return status, output
def _getGlusterdStatus(hostname):
status = json.loads(livestatus.readLiveStatusAsJSON(
"GET services\nColumns: state\n"
"Filter: description = Gluster Management\n"
"Filter: host_name = " + hostname + "\n"))[0][0]
return status
def _getVolumeStatusOutput(hostgroup, volume):
status, output = _executeRandomHost(hostgroup,
_getVolStatusNRPECommand(volume))
if status == utils.PluginStatusCode.OK:
brick_details = json.loads(livestatus.readLiveStatusAsJSON(
"GET services\n"
"Columns: description state host_address host_name\n"
"Filter: host_groups >= %s\n"
"Filter: custom_variable_values >= %s\n"
"Filter: description ~ Brick - \n"
% (hostgroup, volume)))
# output will be as below:
# [[u'Brick - /root/b3', 0, u'10.70.42.246', u'nishanth-rhs-2']]
# parse this to find the no of critical/ok bricks and list of
# critical bricks
bricks_ok = 0
bricks_critical = 0
brick_list_critical = []
for brick_detail in brick_details:
if brick_detail[1] == utils.PluginStatusCode.OK:
bricks_ok += 1
elif brick_detail[1] == utils.PluginStatusCode.CRITICAL:
bricks_critical += 1
# get the critical brick's host uuid if not present
# int the list
custom_vars = json.loads(livestatus.readLiveStatusAsJSON(
"GET hosts\n"
"Columns: custom_variables\n"
"Filter: groups >= %s\n"
"Filter: name = %s\n"
% (hostgroup, brick_detail[3])))
brick_dict = {}
brick_dict['brick'] = brick_detail[2] + ":" + \
brick_detail[0][brick_detail[0].find("/"):]
brick_dict['uuid'] = custom_vars[0][0]['HOST_UUID']
brick_list_critical.append(brick_dict)
# Get volume details
nrpeStatus, nrpeOut = _executeRandomHost(
hostgroup, _getVolDetailNRPECommand(volume))
if nrpeStatus != utils.PluginStatusCode.OK:
return utils.PluginStatusCode.UNKNOWN, nrpeOut
volInfo = json.loads(nrpeOut)
# Get the volume type
vol_type = volInfo[volume]['type']
if bricks_ok == 0 and bricks_critical > 0:
status = utils.PluginStatusCode.CRITICAL
output = "CRITICAL: Volume : %s type - All bricks " \
"are down " % (vol_type)
elif bricks_ok > 0 and bricks_critical == 0:
status = utils.PluginStatusCode.OK
output = "OK: Volume : %s type - All bricks " \
"are Up " % (vol_type)
elif bricks_critical > 0:
if (vol_type == "DISTRIBUTE"):
status = utils.PluginStatusCode.CRITICAL
output = "CRITICAL: Volume : %s type \n Brick(s) - <%s> " \
"is|are down " % \
(vol_type, ', '.join(dict['brick']for dict in
brick_list_critical))
elif (vol_type == "DISTRIBUTED_REPLICATE" or
vol_type == "REPLICATE" or
vol_type == "DISTRIBUTED_DISPERSE" or
vol_type == "DISPERSE"):
status, output = _getDisperseOrReplicaStatus(
vol_type,
brick_list_critical,
volInfo, volume)
else:
output = "WARNING: Volume : %s type \n Brick(s) - <%s> " \
"is|are down" % (vol_type,
', '.join(dict['brick']
for dict in
brick_list_critical))
status = utils.PluginStatusCode.WARNING
return status, output
return status, output