summaryrefslogtreecommitdiffstats
path: root/python.d/bind_rndc.chart.py
diff options
context:
space:
mode:
authorFederico Ceratto <federico.ceratto@gmail.com>2017-04-30 16:09:37 +0000
committerFederico Ceratto <federico.ceratto@gmail.com>2017-04-30 16:09:37 +0000
commit51f689a8e17ff3929acd2dbf39e936d2cd3ac723 (patch)
tree92e54f543171b69dcbc639be09d11221cf96ba28 /python.d/bind_rndc.chart.py
parentNew upstream version 1.5.0+dfsg (diff)
downloadnetdata-51f689a8e17ff3929acd2dbf39e936d2cd3ac723.tar.xz
netdata-51f689a8e17ff3929acd2dbf39e936d2cd3ac723.zip
New upstream version 1.6.0+dfsgupstream/1.6.0+dfsg
Diffstat (limited to 'python.d/bind_rndc.chart.py')
-rw-r--r--python.d/bind_rndc.chart.py48
1 files changed, 21 insertions, 27 deletions
diff --git a/python.d/bind_rndc.chart.py b/python.d/bind_rndc.chart.py
index e11c751f5..a4d753703 100644
--- a/python.d/bind_rndc.chart.py
+++ b/python.d/bind_rndc.chart.py
@@ -4,7 +4,7 @@
from base import SimpleService
from re import compile, findall
-from os.path import getsize, isfile, split
+from os.path import getsize, split
from os import access as is_accessible, R_OK
from subprocess import Popen
@@ -12,7 +12,6 @@ priority = 60000
retries = 60
update_every = 30
-DIRECTORIES = ['/bin/', '/usr/bin/', '/sbin/', '/usr/sbin/']
NMS = ['requests', 'responses', 'success', 'auth_answer', 'nonauth_answer', 'nxrrset', 'failure',
'nxdomain', 'recursion', 'duplicate', 'rejections']
QUERIES = ['RESERVED0', 'A', 'NS', 'CNAME', 'SOA', 'PTR', 'MX', 'TXT', 'X25', 'AAAA', 'SRV', 'NAPTR',
@@ -29,16 +28,12 @@ class Service(SimpleService):
# 'Cache DB RRsets', 'Socket I/O Statistics']
self.options = ['Name Server Statistics', 'Incoming Queries', 'Outgoing Queries']
self.regex_options = [r'(%s(?= \+\+)) \+\+([^\+]+)' % option for option in self.options]
- try:
- self.rndc = [''.join([directory, 'rndc']) for directory in DIRECTORIES
- if isfile(''.join([directory, 'rndc']))][0]
- except IndexError:
- self.rndc = False
+ self.rndc = self.find_binary('rndc')
def check(self):
# We cant start without 'rndc' command
if not self.rndc:
- self.error('Command "rndc" not found')
+ self.error('Can\'t locate \'rndc\' binary or binary is not executable by netdata')
return False
# We cant start if stats file is not exist or not readable by netdata user
@@ -55,15 +50,15 @@ class Service(SimpleService):
if not run_rndc.returncode:
# 'rndc' was found, stats file is exist and readable and we can run 'rndc stats'. Lets go!
self.create_charts()
-
+
# BIND APPEND dump on every run 'rndc stats'
# that is why stats file size can be VERY large if update_interval too small
dump_size_24hr = round(86400 / self.update_every * (int(size_after) - int(size_before)) / 1048576, 3)
-
+
# If update_every too small we should WARN user
if self.update_every < 30:
self.info('Update_every %s is NOT recommended for use. Increase the value to > 30' % self.update_every)
-
+
self.info('With current update_interval it will be + %s MB every 24hr. '
'Don\'t forget to create logrotate conf file for %s' % (dump_size_24hr, self.named_stats_path))
@@ -82,12 +77,11 @@ class Service(SimpleService):
named.stats file size
)
"""
-
try:
current_size = getsize(self.named_stats_path)
except OSError:
return None, None
-
+
run_rndc = Popen([self.rndc, 'stats'], shell=False)
run_rndc.wait()
@@ -115,23 +109,23 @@ class Service(SimpleService):
return None
rndc_stats = dict()
-
+
# Result: dict.
# topic = Cache DB RRsets; body = A 178303 NS 86790 ... ; desc = A; value = 178303
# {'Cache DB RRsets': [('A', 178303), ('NS', 286790), ...],
# {Incoming Queries': [('RESERVED0', 8), ('A', 4557317680), ...],
# ......
for regex in self.regex_options:
- rndc_stats.update({topic: [(desc, int(value)) for value, desc in self.regex_values.findall(body)]
- for topic, body in findall(regex, raw_data)})
-
+ rndc_stats.update(dict([(topic, [(desc, int(value)) for value, desc in self.regex_values.findall(body)])
+ for topic, body in findall(regex, raw_data)]))
+
nms = dict(rndc_stats.get('Name Server Statistics', []))
- inc_queries = {'i' + k: 0 for k in QUERIES}
- inc_queries.update({'i' + k: v for k, v in rndc_stats.get('Incoming Queries', [])})
- out_queries = {'o' + k: 0 for k in QUERIES}
- out_queries.update({'o' + k: v for k, v in rndc_stats.get('Outgoing Queries', [])})
-
+ inc_queries = dict([('i' + k, 0) for k in QUERIES])
+ inc_queries.update(dict([('i' + k, v) for k, v in rndc_stats.get('Incoming Queries', [])]))
+ out_queries = dict([('o' + k, 0) for k in QUERIES])
+ out_queries.update(dict([('o' + k, v) for k, v in rndc_stats.get('Outgoing Queries', [])]))
+
to_netdata = dict()
to_netdata['requests'] = sum([v for k, v in nms.items() if 'request' in k and 'received' in k])
to_netdata['responses'] = sum([v for k, v in nms.items() if 'responses' in k and 'sent' in k])
@@ -145,7 +139,7 @@ class Service(SimpleService):
to_netdata['duplicate'] = nms.get('duplicate queries received', 0)
to_netdata['rejections'] = nms.get('recursive queries rejected', 0)
to_netdata['stats_size'] = size
-
+
to_netdata.update(inc_queries)
to_netdata.update(out_queries)
return to_netdata
@@ -154,20 +148,20 @@ class Service(SimpleService):
self.order = ['stats_size', 'bind_stats', 'incoming_q', 'outgoing_q']
self.definitions = {
'bind_stats': {
- 'options': [None, 'Name Server Statistics', 'stats', 'Name Server Statistics', 'bind_rndc.stats', 'line'],
+ 'options': [None, 'Name Server Statistics', 'stats', 'name server statistics', 'bind_rndc.stats', 'line'],
'lines': [
]},
'incoming_q': {
- 'options': [None, 'Incoming queries', 'queries','Incoming queries', 'bind_rndc.incq', 'line'],
+ 'options': [None, 'Incoming queries', 'queries','incoming queries', 'bind_rndc.incq', 'line'],
'lines': [
]},
'outgoing_q': {
- 'options': [None, 'Outgoing queries', 'queries','Outgoing queries', 'bind_rndc.outq', 'line'],
+ 'options': [None, 'Outgoing queries', 'queries','outgoing queries', 'bind_rndc.outq', 'line'],
'lines': [
]},
'stats_size': {
'options': [None, '%s file size' % split(self.named_stats_path)[1].capitalize(), 'megabytes',
- '%s size' % split(self.named_stats_path)[1].capitalize(), 'bind_rndc.size', 'line'],
+ '%s size' % split(self.named_stats_path)[1], 'bind_rndc.size', 'line'],
'lines': [
["stats_size", None, "absolute", 1, 1048576]
]}