def execute(parameters=None, host_name=None):
"""
Returns a tuple containing the result code and a pre-formatted result label
Keyword arguments:
parameters (dictionary): a mapping of parameter key to value
host_name (string): the name of this host where the alert is running
"""
if parameters is None:
return (RESULT_CODE_UNKNOWN, ['There were no parameters supplied to the script.'])
flume_conf_directory = None
if FLUME_CONF_DIR_KEY in parameters:
flume_conf_directory = parameters[FLUME_CONF_DIR_KEY]
if flume_conf_directory is None:
return (RESULT_CODE_UNKNOWN, ['The Flume configuration directory is a required parameter.'])
if host_name is None:
host_name = socket.getfqdn()
processes = get_flume_status(flume_conf_directory, FLUME_RUN_DIR)
expected_agents = find_expected_agent_names(flume_conf_directory)
alert_label = ''
alert_state = RESULT_CODE_OK
if len(processes) == 0 and len(expected_agents) == 0:
alert_label = 'No agents defined on {0}'.format(host_name)
else:
ok = []
critical = []
text_arr = []
for process in processes:
if not process.has_key('status') or process['status'] == 'NOT_RUNNING':
critical.append(process['name'])
else:
ok.append(process['name'])
if len(critical) > 0:
text_arr.append("{0} {1} NOT running".format(", ".join(critical),
"is" if len(critical) == 1 else "are"))
if len(ok) > 0:
text_arr.append("{0} {1} running".format(", ".join(ok),
"is" if len(ok) == 1 else "are"))
plural = len(critical) > 1 or len(ok) > 1
alert_label = "Agent{0} {1} {2}".format(
"s" if plural else "",
" and ".join(text_arr),
"on " + host_name)
alert_state = RESULT_CODE_CRITICAL if len(critical) > 0 else RESULT_CODE_OK
return (alert_state, [alert_label])
def status(self, env):
import params
env.set_params(params)
processes = get_flume_status(params.flume_conf_dir, params.flume_run_dir)
expected_agents = find_expected_agent_names(params.flume_conf_dir)
json = {}
json['processes'] = processes
self.put_structured_out(json)
if len(expected_agents) > 0:
for proc in processes:
if not proc.has_key('status') or proc['status'] == 'NOT_RUNNING':
raise ComponentIsNotRunning()
elif len(expected_agents) == 0 and 'INSTALLED' == get_desired_state():
raise ComponentIsNotRunning()
def status(self, env):
import params
env.set_params(params)
processes = get_flume_status(params.flume_conf_dir, params.flume_run_dir)
expected_agents = find_expected_agent_names(params.flume_conf_dir)
json = {}
json['processes'] = processes
self.put_structured_out(json)
# only throw an exception if there are agents defined and there is a
# problem with the processes; if there are no agents defined, then
# the service should report STARTED (green) ONLY if the desired state is started. otherwise, INSTALLED (red)
if len(expected_agents) > 0:
for proc in processes:
if not proc.has_key('status') or proc['status'] == 'NOT_RUNNING':
raise ComponentIsNotRunning()
elif len(expected_agents) == 0 and 'INSTALLED' == get_desired_state():
raise ComponentIsNotRunning()
def status(self, env):
import params
env.set_params(params)
processes = get_flume_status(params.flume_conf_dir, params.flume_run_dir)
expected_agents = find_expected_agent_names(params.flume_conf_dir)
json = {}
json['processes'] = processes
self.put_structured_out(json)
# only throw an exception if there are agents defined and there is a
# problem with the processes; if there are no agents defined, then
# the service should report STARTED (green) ONLY if the desired state is started. otherwise, INSTALLED (red)
if len(expected_agents) > 0:
for proc in processes:
if not proc.has_key('status') or proc['status'] == 'NOT_RUNNING':
raise ComponentIsNotRunning()
elif len(expected_agents) == 0 and 'INSTALLED' == get_desired_state():
raise ComponentIsNotRunning()
def status(self, env):
import params
env.set_params(params)
processes = get_flume_status(params.flume_conf_dir,
params.flume_run_dir)
expected_agents = find_expected_agent_names(params.flume_conf_dir)
json = {}
json['processes'] = processes
json['alerts'] = []
alert = {}
alert['name'] = 'flume_agent'
alert['label'] = 'Flume Agent process'
if len(processes) == 0 and len(expected_agents) == 0:
alert['state'] = 'OK'
if not params.hostname is None:
alert['text'] = 'No agents defined on ' + params.hostname
else:
alert['text'] = 'No agents defined'
else:
crit = []
ok = []
for proc in processes:
if not proc.has_key(
'status') or proc['status'] == 'NOT_RUNNING':
crit.append(proc['name'])
else:
ok.append(proc['name'])
text_arr = []
if len(crit) > 0:
text_arr.append("{0} {1} NOT running".format(
", ".join(crit), "is" if len(crit) == 1 else "are"))
if len(ok) > 0:
text_arr.append("{0} {1} running".format(
", ".join(ok), "is" if len(ok) == 1 else "are"))
plural = len(crit) > 1 or len(ok) > 1
alert['text'] = "Agent{0} {1} {2}".format(
"s" if plural else "", " and ".join(text_arr),
"" if params.hostname is None else "on " +
str(params.hostname))
alert['state'] = 'CRITICAL' if len(crit) > 0 else 'OK'
json['alerts'].append(alert)
self.put_structured_out(json)
# only throw an exception if there are agents defined and there is a
# problem with the processes; if there are no agents defined, then
# the service should report STARTED (green) ONLY if the desired state is started. otherwise, INSTALLED (red)
if len(expected_agents) > 0:
for proc in processes:
if not proc.has_key(
'status') or proc['status'] == 'NOT_RUNNING':
raise ComponentIsNotRunning()
elif len(expected_agents) == 0 and 'INSTALLED' == get_desired_state():
raise ComponentIsNotRunning()