def _init_probe(self):
if self.options.host == 'localhost':
self.options.host = socket.getfqdn()
self.hostname = self.options.host
(username, password) = open(self.options.config,
'r').readline().rstrip('\n').split(':')
self.cdh_api = get_root_resource(self.options.host, self.options.port,
username, password,
self.options.use_tls,
self.CM_API_VERSION)
return {'username': username, 'password': password}
def run(self):
(self.options, args) = self._parse_args()
if self.options.host == 'localhost':
hostname = socket.getfqdn()
else:
hostname = self.options.host
# Step 1: init container
try:
zbx_container = self._init_container()
except:
return 1
# Step 2: get data
try:
(self.options.username, self.options.password) = \
open(self.options.passfile, 'r').readline().rstrip('\n').split(':')
cdh_api = get_root_resource(
self.options.host,
self.options.port,
self.options.username,
self.options.password,
self.options.use_tls,
self.CM_API_VERSION
)
data = {}
if self.options.mode == "update_items":
zbx_container.set_type("items")
data = self._get_metrics(cdh_api, hostname)
elif self.options.mode == "discovery":
zbx_container.set_type("lld")
data = self._get_discovery(cdh_api, hostname)
except:
return 2
# Step 3: format & load data into container
try:
zbx_container.add(data)
zbx_container.add_item(hostname, "hadoop.cm.zbx_version", self.__version__)
except:
return 3
# Step 4: send container data to Zabbix server
try:
zbx_container.send(zbx_container)
except protobix.SenderException as zbx_e:
if self.options.debug:
print self.ZBX_CONN_ERR % zbx_e.err_text
return 4
# Everything went fine. Let's return 0 and exit
return 0
def main():
(options, args) = parse_args()
try:
(options.username,
options.password) = open(options.passfile,
'r').readline().rstrip('\n').split(':')
except:
print >> sys.stderr, "Unable to read username and password from file '%s'. "
"Make sure the file is readable and contains a single line of "
"the form \"<username>:<password>\"" % options.passfile
if options.host == 'localhost':
hostname = platform.node()
else:
hostname = options.host
try:
cdh_api = get_root_resource(options.host, options.port,
options.username, options.password,
options.use_tls, CM_API_VERSION)
except:
return 1
zbx_container = protobix.DataContainer()
if options.mode == "update_items":
zbx_container.set_type("items")
data = get_metrics(cdh_api, hostname)
elif options.mode == "discovery":
zbx_container.set_type("lld")
data = get_discovery(cdh_api, hostname)
zbx_container.add(data)
zbx_container.add_item(hostname, "hadoop.cm.zbx_version", __version__)
zbx_container.set_host(options.zabbix_server)
zbx_container.set_port(int(options.zabbix_port))
zbx_container.set_debug(options.debug)
zbx_container.set_verbosity(options.verbose)
zbx_container.set_dryrun(options.dry)
try:
zbxret = zbx_container.send(zbx_container)
except protobix.SenderException as zbx_e:
if options.debug:
print ZBX_CONN_ERR % zbx_e.err_text
return 2
else:
return 0
def main():
(options, args) = parse_args()
try:
(options.username, options.password) = open(options.passfile, "r").readline().rstrip("\n").split(":")
except:
print >> sys.stderr, "Unable to read username and password from file '%s'. "
"Make sure the file is readable and contains a single line of "
'the form "<username>:<password>"' % options.passfile
if options.host == "localhost":
hostname = socket.getfqdn()
else:
hostname = options.host
try:
cdh_api = get_root_resource(
options.host, options.port, options.username, options.password, options.use_tls, CM_API_VERSION
)
except:
return 1
zbx_container = protobix.DataContainer()
if options.mode == "update_items":
zbx_container.set_type("items")
data = get_metrics(cdh_api, hostname)
elif options.mode == "discovery":
zbx_container.set_type("lld")
data = get_discovery(cdh_api, hostname)
zbx_container.add(data)
zbx_container.add_item(hostname, "hadoop.cm.zbx_version", __version__)
zbx_container.set_host(options.zabbix_server)
zbx_container.set_port(int(options.zabbix_port))
zbx_container.set_debug(options.debug)
zbx_container.set_verbosity(options.verbose)
zbx_container.set_dryrun(options.dry)
try:
zbxret = zbx_container.send(zbx_container)
except protobix.SenderException as zbx_e:
if options.debug:
print ZBX_CONN_ERR % zbx_e.err_text
return 2
else:
return 0
def main():
(options, args) = parse_args()
try:
(username, password) = open(options.passfile, 'r').readline().rstrip('\n').split(':')
except:
print >> sys.stderr, "Unable to read username and password from file '%s'. "
"Make sure the file is readable and contains a single line of "
"the form \"<username>:<password>\"" % options.passfile
root = get_root_resource(options.host, options.port, username,
password, options.use_tls, CM_API_VERSION)
hosts_map = get_host_map(root)
if options.mode == "update_status":
''' The way in which the status is reported depends on whether or not the
-n/--use-send-nsca option was set. By default, the status updates are
simply appended to the external command file. If nsca is used, the
status updates are piped via stdin to send_nsca.
'''
# <host>\t<service>\t<return_code>\t<message>\n
NSCA_STATUS_FORMAT = "{0}\t{1}\t{2}\t{3}\n"
# [<timestamp>] PROCESS_SERVICE_CHECK_RESULT;<host>;<service>;<return_code>;<message>\n
CMD_STATUS_FORMAT = "[{0}] PROCESS_SERVICE_CHECK_RESULT;{1};{2};{3};{4}\n"
try:
services_list = get_services(root, hosts_map, "full")
timestamp = long(time()) # approximate time the services were queried
tmp_status_file = tempfile.TemporaryFile()
try:
for service in services_list:
health = service["health_summary"]
if not health is None:
code = CM_STATE_CODES[health]
if options.use_send_nsca:
status_line = NSCA_STATUS_FORMAT.format(
service["hostname"], service["name"], code, service["status"])
else:
status_line = CMD_STATUS_FORMAT.format(
timestamp, service["hostname"], service["name"], code, service["status"])
tmp_status_file.write(status_line)
tmp_status_file.flush()
tmp_status_file.seek(0)
if options.use_send_nsca:
submit_status_send_nsca(options.send_nsca_path, options.nsca_host,
options.nsca_port, options.send_nsca_config,
tmp_status_file, options.verbose)
else:
submit_status_external_cmd(options.cmd_file, tmp_status_file)
finally:
tmp_status_file.close()
except SystemExit as e:
sys.exit(e)
except Exception as e:
print >> sys.stderr, "An unknown error occurred: %s" % str(e)
raise
elif options.mode == "generate_cfg":
host_cfg_file = open(join(options.cfg_dir, CM_HOST_CFG_FILENAME), 'w+')
host_cfg_file.write(PASSIVE_HOST_TEMPLATE)
for host in hosts_map.values():
host_cfg_file.write(CM_HOST_TEMPLATE %
(host["hostname"], host["hostname"], host["address"],))
host_cfg_file.close()
services_cfg_file = open(join(options.cfg_dir, CM_SERVICE_CFG_FILENAME), 'w+')
services_cfg_file.write(PASSIVE_SERVICE_TEMPLATE)
services_list = get_services(root, hosts_map)
for service in services_list:
services_cfg_file.write(CM_SERVICE_TEMPLATE %
(service["hostname"],
service["name"],
service["display_name"],
service["url"]))
services_cfg_file.close()
else:
print >> sys.stderr, "Fatal error: Unrecognized mode"
def main():
(options, args) = parse_args()
try:
(username, password) = open(options.passfile, 'r').readline().rstrip('\n').split(':')
except:
print >> sys.stderr, "Unable to read username and password from file '%s'. "
"Make sure the file is readable and contains a single line of "
"the form \"<username>:<password>\"" % options.passfile
root = get_root_resource(options.host, options.port, username,
password, options.use_tls, CM_API_VERSION)
clusters = root.get_all_clusters()
for cluster in clusters:
host_refs = cluster.list_hosts()
services = cluster.get_all_services()
print "# Cluster Summary #"
print "Name: %s" % cluster.displayName
print "Number of hosts: %d" % len(host_refs)
print "Number of services: %d" % len(services)
print "## Host Summary"
for ref in host_refs:
host = root.get_host(ref.hostId)
print "Host: %(host)s, Health Summary: %(health)s" % {
'host': host.hostname
, 'health': host.healthSummary
}
print "## Host Health Checks"
for ref in host_refs:
host = root.get_host(ref.hostId)
checks = host.healthChecks
for check in checks:
print "Host: %(host)s, Health Check: %(health)s, Status: %(status)s" % {
'host': host.hostname
, 'health': check['name']
, 'status': check['summary']
}
print "## Service Summary"
for service in services:
print "Service: %(name)s, Health Summary: %(health)s" % {
'name': service.displayName
, 'health': service.healthSummary
}
for check in service.healthChecks:
print "Service: %(name)s, Health Check: %(health)s, Status: %(status)s" % {
'name': service.displayName
, 'health': check['name']
, 'status': check['summary']
}
for role in service.get_all_roles():
print "Service: %(service_name)s, Role: %(role_name)s, Health Summary: %(health)s" % {
'service_name': service.displayName
, 'role_name': role.name
, 'health': role.healthSummary
}
for check in role.healthChecks:
print "Service: %(service_name)s, Role: %(role_name)s, Health Check: %(health)s, Status: %(status)s" % {
'service_name': service.displayName
, 'role_name': role.name
, 'health': check['name']
, 'status': check['summary']
}
sys.exit(0)
def main():
(options, args) = parse_args()
try:
(username, password) = open(options.passfile,
'r').readline().rstrip('\n').split(':')
except:
print >> sys.stderr, "Unable to read username and password from file '%s'. "
"Make sure the file is readable and contains a single line of "
"the form \"<username>:<password>\"" % options.passfile
root = get_root_resource(options.host, options.port, username, password,
options.use_tls, CM_API_VERSION)
clusters = root.get_all_clusters()
for cluster in clusters:
host_refs = cluster.list_hosts()
services = cluster.get_all_services()
print "# Cluster Summary #"
print "Name: %s" % cluster.displayName
print "Number of hosts: %d" % len(host_refs)
print "Number of services: %d" % len(services)
print "## Host Summary"
for ref in host_refs:
host = root.get_host(ref.hostId)
print "Host: %(host)s, Health Summary: %(health)s" % {
'host': host.hostname,
'health': host.healthSummary
}
print "## Host Health Checks"
for ref in host_refs:
host = root.get_host(ref.hostId)
checks = host.healthChecks
for check in checks:
print "Host: %(host)s, Health Check: %(health)s, Status: %(status)s" % {
'host': host.hostname,
'health': check['name'],
'status': check['summary']
}
print "## Service Summary"
for service in services:
print "Service: %(name)s, Health Summary: %(health)s" % {
'name': service.displayName,
'health': service.healthSummary
}
for check in service.healthChecks:
print "Service: %(name)s, Health Check: %(health)s, Status: %(status)s" % {
'name': service.displayName,
'health': check['name'],
'status': check['summary']
}
for role in service.get_all_roles():
print "Service: %(service_name)s, Role: %(role_name)s, Health Summary: %(health)s" % {
'service_name': service.displayName,
'role_name': role.name,
'health': role.healthSummary
}
for check in role.healthChecks:
print "Service: %(service_name)s, Role: %(role_name)s, Health Check: %(health)s, Status: %(status)s" % {
'service_name': service.displayName,
'role_name': role.name,
'health': check['name'],
'status': check['summary']
}
sys.exit(0)
def main():
(options, args) = parse_args()
try:
(username, password) = open(options.passfile,
'r').readline().rstrip('\n').split(':')
except:
print >> sys.stderr, "Unable to read username and password from file '%s'. "
"Make sure the file is readable and contains a single line of "
"the form \"<username>:<password>\"" % options.passfile
root = get_root_resource(options.host, options.port, username, password,
options.use_tls, CM_API_VERSION)
hosts_map = get_host_map(root)
if options.mode == "update_status":
''' The way in which the status is reported depends on whether or not the
-n/--use-send-nsca option was set. By default, the status updates are
simply appended to the external command file. If nsca is used, the
status updates are piped via stdin to send_nsca.
'''
# <host>\t<service>\t<return_code>\t<message>\n
NSCA_STATUS_FORMAT = "{0}\t{1}\t{2}\t{3}\n"
# [<timestamp>] PROCESS_SERVICE_CHECK_RESULT;<host>;<service>;<return_code>;<message>\n
CMD_STATUS_FORMAT = "[{0}] PROCESS_SERVICE_CHECK_RESULT;{1};{2};{3};{4}\n"
try:
services_list = get_services(root, hosts_map, "full")
timestamp = long(
time()) # approximate time the services were queried
tmp_status_file = tempfile.TemporaryFile()
try:
for service in services_list:
health = service["health_summary"]
if not health is None:
code = CM_STATE_CODES[health]
if options.use_send_nsca:
status_line = NSCA_STATUS_FORMAT.format(
service["hostname"], service["name"], code,
service["status"])
else:
status_line = CMD_STATUS_FORMAT.format(
timestamp, service["hostname"],
service["name"], code, service["status"])
tmp_status_file.write(status_line)
tmp_status_file.flush()
tmp_status_file.seek(0)
if options.use_send_nsca:
submit_status_send_nsca(options.send_nsca_path,
options.nsca_host,
options.nsca_port,
options.send_nsca_config,
tmp_status_file, options.verbose)
else:
submit_status_external_cmd(options.cmd_file,
tmp_status_file)
finally:
tmp_status_file.close()
except SystemExit as e:
sys.exit(e)
except Exception as e:
print >> sys.stderr, "An unknown error occurred: %s" % str(e)
raise
elif options.mode == "generate_cfg":
host_cfg_file = open(join(options.cfg_dir, CM_HOST_CFG_FILENAME), 'w+')
host_cfg_file.write(PASSIVE_HOST_TEMPLATE)
for host in hosts_map.values():
host_cfg_file.write(CM_HOST_TEMPLATE % (
host["hostname"],
host["hostname"],
host["address"],
))
host_cfg_file.close()
services_cfg_file = open(
join(options.cfg_dir, CM_SERVICE_CFG_FILENAME), 'w+')
services_cfg_file.write(PASSIVE_SERVICE_TEMPLATE)
services_list = get_services(root, hosts_map)
for service in services_list:
services_cfg_file.write(CM_SERVICE_TEMPLATE %
(service["hostname"], service["name"],
service["display_name"], service["url"]))
services_cfg_file.close()
else:
print >> sys.stderr, "Fatal error: Unrecognized mode"
