def _get_config(self, instance):
required = ['mysql_host', 'mysql_user', 'rrd_path']
for param in required:
if not instance.get(param):
raise Exception("Cacti instance missing %s. Skipping." % (param))
host = instance.get('mysql_host')
user = instance.get('mysql_user')
password = instance.get('mysql_password', '') or ''
db = instance.get('mysql_db', 'cacti')
rrd_path = instance.get('rrd_path')
whitelist = instance.get('rrd_whitelist')
field_names = instance.get('field_names', ['ifName', 'dskDevice'])
Config = namedtuple('Config', [
'host',
'user',
'password',
'db',
'rrd_path',
'whitelist',
'field_names']
)
return Config(host, user, password, db, rrd_path, whitelist, field_names)
def _parse_proc_list(self, processes):
def _compute_family(command):
if command.startswith("["):
return "kernel"
else:
return (command.split()[0]).split("/")[-1]
PSLine = namedtuple("PSLine", "user,pid,pct_cpu,pct_mem,vsz,rss,tty,stat,started,time,command")
self.start_snapshot()
for line in processes:
try:
psl = PSLine(*line)
self.add_to_snapshot(
[
psl.user,
float(psl.pct_cpu),
float(psl.pct_mem),
int(psl.vsz),
int(psl.rss),
_compute_family(psl.command),
1,
]
)
except Exception:
pass
self.end_snapshot(group_by=self.group_by_family)
def _get_config(self, instance):
required = ['mysql_host', 'mysql_user', 'rrd_path']
for param in required:
if not instance.get(param):
raise Exception("Cacti instance missing %s. Skipping." % (param))
host = instance.get('mysql_host')
user = instance.get('mysql_user')
password = instance.get('mysql_password', '') or ''
db = instance.get('mysql_db', 'cacti')
rrd_path = instance.get('rrd_path')
whitelist = instance.get('rrd_whitelist')
field_names = instance.get('field_names', ['ifName', 'dskDevice'])
Config = namedtuple('Config', [
'host',
'user',
'password',
'db',
'rrd_path',
'whitelist',
'field_names']
)
return Config(host, user, password, db, rrd_path, whitelist, field_names)
def _parse_proc_list(self, processes):
def _compute_family(command):
if command.startswith('['):
return 'kernel'
else:
return (command.split()[0]).split('/')[-1]
PSLine = namedtuple(
"PSLine",
"user,pid,pct_cpu,pct_mem,vsz,rss,tty,stat,started,time,command")
self.start_snapshot()
for line in processes:
try:
psl = PSLine(*line)
self.add_to_snapshot([
psl.user,
float(psl.pct_cpu),
float(psl.pct_mem),
int(psl.vsz),
int(psl.rss),
_compute_family(psl.command), 1
])
except:
pass
self.end_snapshot(group_by=self.group_by_family)
import time
import re
from util import namedtuple, get_hostname
from utils import TailFile
# Event types we know about but decide to ignore in the parser
IGNORE_EVENT_TYPES = []
# fields order for each event type, as named tuples
EVENT_FIELDS = {
'CURRENT HOST STATE':
namedtuple('E_CurrentHostState',
'host, event_state, event_soft_hard, return_code, payload'),
'CURRENT SERVICE STATE':
namedtuple(
'E_CurrentServiceState',
'host, check_name, event_state, event_soft_hard, return_code, payload'
),
'SERVICE ALERT':
namedtuple(
'E_ServiceAlert',
'host, check_name, event_state, event_soft_hard, return_code, payload'
),
'PASSIVE SERVICE CHECK':
namedtuple('E_PassiveServiceCheck',
'host, check_name, return_code, payload'),
'HOST ALERT':
namedtuple('E_HostAlert',
'host, event_state, event_soft_hard, return_code, payload'),
# [1305744274] SERVICE NOTIFICATION: ops;ip-10-114-237-165;Metric ETL;ACKNOWLEDGEMENT (CRITICAL);notify-service-by-email;HTTP CRITICAL: HTTP/1.1 503 Service Unavailable - 394 bytes in 0.010 second response time;datadog;alq
from util import namedtuple
url_fmt = "http://docs.google.com/spreadsheet/pub?key=%s&gid=%s&single=true&output=csv"
try:
# gid=8 for second worksheet
url = url_fmt % (remap_doc_id, 8)
result = fetch(url)
except Exception, e:
raise Exception("%s (%s)" % (e, url))
logging.info("Loading domains")
reader = csv.reader(StringIO.StringIO(result.content))
headerline = reader.next()
domains = {}
Domain = namedtuple("Domain", headerline)
for line in reader:
if not line[0]:
break
domain = Domain(*line)
data = dict(domain._asdict())
domains[domain.code] = data
logging.info(" %(code)s - %(name)s", data)
try:
# gid=9 for 3rd worksheet
url = url_fmt % (remap_doc_id, 9)
result = fetch(url)
except Exception, e:
raise Exception("%s (%s)" % (e, url))
import time
import re
from util import namedtuple, get_hostname
from utils import TailFile
# Event types we know about but decide to ignore in the parser
IGNORE_EVENT_TYPES = []
# fields order for each event type, as named tuples
EVENT_FIELDS = {
'CURRENT HOST STATE': namedtuple('E_CurrentHostState', 'host, event_state, event_soft_hard, return_code, payload'),
'CURRENT SERVICE STATE': namedtuple('E_CurrentServiceState', 'host, check_name, event_state, event_soft_hard, return_code, payload'),
'SERVICE ALERT': namedtuple('E_ServiceAlert', 'host, check_name, event_state, event_soft_hard, return_code, payload'),
'PASSIVE SERVICE CHECK': namedtuple('E_PassiveServiceCheck', 'host, check_name, return_code, payload'),
'HOST ALERT': namedtuple('E_HostAlert', 'host, event_state, event_soft_hard, return_code, payload'),
# [1305744274] SERVICE NOTIFICATION: ops;ip-10-114-237-165;Metric ETL;ACKNOWLEDGEMENT (CRITICAL);notify-service-by-email;HTTP CRITICAL: HTTP/1.1 503 Service Unavailable - 394 bytes in 0.010 second response time;datadog;alq
'SERVICE NOTIFICATION': namedtuple('E_ServiceNotification', 'contact, host, check_name, event_state, notification_type, payload'),
# [1296509331] SERVICE FLAPPING ALERT: ip-10-114-97-27;cassandra JVM Heap;STARTED; Service appears to have started flapping (23.4% change >= 20.0% threshold)
# [1296662511] SERVICE FLAPPING ALERT: ip-10-114-97-27;cassandra JVM Heap;STOPPED; Service appears to have stopped flapping (3.8% change < 5.0% threshold)
'SERVICE FLAPPING ALERT': namedtuple('E_FlappingAlert', 'host, check_name, flap_start_stop, payload'),
# Reference for external commands: http://old.nagios.org/developerinfo/externalcommands/commandlist.php
# Command Format:
# ACKNOWLEDGE_SVC_PROBLEM;<host_name>;<service_description>;<sticky>;<notify>;<persistent>;<author>;<comment>
# [1305832665] EXTERNAL COMMAND: ACKNOWLEDGE_SVC_PROBLEM;ip-10-202-161-236;Resources ETL;2;1;0;datadog;alq checking
'ACKNOWLEDGE_SVC_PROBLEM': namedtuple('E_ServiceAck', 'host, check_name, sticky_ack, notify_ack, persistent_ack, ack_author, payload'),
# Command Format:
# ACKNOWLEDGE_HOST_PROBLEM;<host_name>;<sticky>;<notify>;<persistent>;<author>;<comment>
import time
import re
from util import namedtuple, get_hostname
from utils import TailFile
# Event types we know about but decide to ignore in the parser
IGNORE_EVENT_TYPES = []
# fields order for each event type, as named tuples
EVENT_FIELDS = {
'CURRENT HOST STATE': namedtuple('E_CurrentHostState', 'host, event_state, event_soft_hard, return_code, payload'),
'CURRENT SERVICE STATE': namedtuple('E_CurrentServiceState', 'host, check_name, event_state, event_soft_hard, return_code, payload'),
'SERVICE ALERT': namedtuple('E_ServiceAlert', 'host, check_name, event_state, event_soft_hard, return_code, payload'),
'PASSIVE SERVICE CHECK': namedtuple('E_PassiveServiceCheck', 'host, check_name, return_code, payload'),
'HOST ALERT': namedtuple('E_HostAlert', 'host, event_state, event_soft_hard, return_code, payload'),
# [1305744274] SERVICE NOTIFICATION: ops;ip-10-114-237-165;Metric ETL;ACKNOWLEDGEMENT (CRITICAL);notify-service-by-email;HTTP CRITICAL: HTTP/1.1 503 Service Unavailable - 394 bytes in 0.010 second response time;datadog;alq
'SERVICE NOTIFICATION': namedtuple('E_ServiceNotification', 'contact, host, check_name, event_state, notification_type, payload'),
# [1296509331] SERVICE FLAPPING ALERT: ip-10-114-97-27;cassandra JVM Heap;STARTED; Service appears to have started flapping (23.4% change >= 20.0% threshold)
# [1296662511] SERVICE FLAPPING ALERT: ip-10-114-97-27;cassandra JVM Heap;STOPPED; Service appears to have stopped flapping (3.8% change < 5.0% threshold)
'SERVICE FLAPPING ALERT': namedtuple('E_FlappingAlert', 'host, check_name, flap_start_stop, payload'),
# Reference for external commands: http://old.nagios.org/developerinfo/externalcommands/commandlist.php
# Command Format:
# ACKNOWLEDGE_SVC_PROBLEM;<host_name>;<service_description>;<sticky>;<notify>;<persistent>;<author>;<comment>
# [1305832665] EXTERNAL COMMAND: ACKNOWLEDGE_SVC_PROBLEM;ip-10-202-161-236;Resources ETL;2;1;0;datadog;alq checking
'ACKNOWLEDGE_SVC_PROBLEM': namedtuple('E_ServiceAck', 'host, check_name, sticky_ack, notify_ack, persistent_ack, ack_author, payload'),
# Command Format:
# ACKNOWLEDGE_HOST_PROBLEM;<host_name>;<sticky>;<notify>;<persistent>;<author>;<comment>
return 0
@staticmethod
def append(args):
l = []
for arg in args:
if type(arg) == type("") or type(arg) == type(u""):
l.extend(arg.split(","))
else:
l.append(str(arg))
return ",".join(list(set(l)))
MetricDescriptor = namedtuple('MetricDescriptor',['version','name','type','aggregator',
'temporal_aggregator','server_aggregator','server_temporal_aggregator',
'group_on','temporal_group_on'])
SnapshotDesc = namedtuple('SnapshotDesc',['version','fields'])
def SnapshotField(name,_type,aggregator=sum,temporal_aggregator=agg.avg,
server_aggregator=None,server_temporal_aggregator=None,
group_on = False, temporal_group_on = False):
if server_aggregator is None:
if _type == 'str':
server_aggregator = agg.append
else:
server_aggregator = sum
if server_temporal_aggregator is None:
if _type == 'str':
server_temporal_aggregator = agg.append
return 0
@staticmethod
def append(args):
l = []
for arg in args:
if type(arg) == type("") or type(arg) == type(u""):
l.extend(arg.split(","))
else:
l.append(str(arg))
return ",".join(list(set(l)))
MetricDescriptor = namedtuple('MetricDescriptor',['version','name','type','aggregator',
'temporal_aggregator','server_aggregator','server_temporal_aggregator',
'group_on','temporal_group_on'])
SnapshotDesc = namedtuple('SnapshotDesc',['version','fields'])
def SnapshotField(name,_type,aggregator=sum,temporal_aggregator=agg.avg,
server_aggregator=None,server_temporal_aggregator=None,
group_on = False, temporal_group_on = False):
if server_aggregator is None:
if _type == 'str':
server_aggregator = agg.append
else:
server_aggregator = sum
if server_temporal_aggregator is None:
if _type == 'str':
server_temporal_aggregator = agg.append
import time
import re
from util import namedtuple, get_hostname
from utils import TailFile
# Event types we know about but decide to ignore in the parser
IGNORE_EVENT_TYPES = []
# fields order for each event type, as named tuples
EVENT_FIELDS = {
"CURRENT HOST STATE": namedtuple("E_CurrentHostState", "host, event_state, event_soft_hard, return_code, payload"),
"CURRENT SERVICE STATE": namedtuple(
"E_CurrentServiceState", "host, check_name, event_state, event_soft_hard, return_code, payload"
),
"SERVICE ALERT": namedtuple(
"E_ServiceAlert", "host, check_name, event_state, event_soft_hard, return_code, payload"
),
"PASSIVE SERVICE CHECK": namedtuple("E_PassiveServiceCheck", "host, check_name, return_code, payload"),
"HOST ALERT": namedtuple("E_HostAlert", "host, event_state, event_soft_hard, return_code, payload"),
# [1305744274] SERVICE NOTIFICATION: ops;ip-10-114-237-165;Metric ETL;ACKNOWLEDGEMENT (CRITICAL);notify-service-by-email;HTTP CRITICAL: HTTP/1.1 503 Service Unavailable - 394 bytes in 0.010 second response time;datadog;alq
"SERVICE NOTIFICATION": namedtuple(
"E_ServiceNotification", "contact, host, check_name, event_state, notification_type, payload"
),
# [1296509331] SERVICE FLAPPING ALERT: ip-10-114-97-27;cassandra JVM Heap;STARTED; Service appears to have started flapping (23.4% change >= 20.0% threshold)
# [1296662511] SERVICE FLAPPING ALERT: ip-10-114-97-27;cassandra JVM Heap;STOPPED; Service appears to have stopped flapping (3.8% change < 5.0% threshold)
"SERVICE FLAPPING ALERT": namedtuple("E_FlappingAlert", "host, check_name, flap_start_stop, payload"),
# Reference for external commands: http://old.nagios.org/developerinfo/externalcommands/commandlist.php
# Command Format:
# ACKNOWLEDGE_SVC_PROBLEM;<host_name>;<service_description>;<sticky>;<notify>;<persistent>;<author>;<comment>
# [1305832665] EXTERNAL COMMAND: ACKNOWLEDGE_SVC_PROBLEM;ip-10-202-161-236;Resources ETL;2;1;0;datadog;alq checking
"ACKNOWLEDGE_SVC_PROBLEM": namedtuple(