Infrastructure/ansible
1
0
Fork 0
Code Issues Pull requests 6 Projects Releases Packages Wiki Activity Actions

[nagios/datanommer] this is what happens when you have 2 files which are supposeldy the same file. You edit one in nagios_server and miss the one in nagios_client. Bad nagios. Bad

Browse source
This commit is contained in:
Stephen Smoogen 2019-05-30 16:56:26 +00:00
parent 0e2870c9c5
commit 40a819e1d5
1 changed files with 12 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

17
roles/nagios_server/files/nagios/plugins/check_datanommer_timesince.py
View file

@ -2,6 +2,8 @@
""" NRPE check for datanommer/fedmsg health. """ NRPE check for datanommer/fedmsg health.
Given a category like 'bodhi', 'buildsys', or 'git', return an error if Given a category like 'bodhi', 'buildsys', or 'git', return an error if
datanommer hasn't seen a message of that type in such and such time. datanommer hasn't seen a message of that type in such and such time.
You can alternatively provide a 'topic' which might look like
org.fedoraproject.prod.bodhi.update.comment.
Requires: python-dateutil Requires: python-dateutil
@ -19,19 +21,24 @@ import sys
import json import json
def query_timesince(category): def query_timesince(identifier):
cmd = 'datanommer-latest --category %s --timesince' % category # If it has a '.', then assume it is a topic.
if '.' in identifier:
cmd = 'datanommer-latest --topic %s --timesince' % identifier
else:
cmd = 'datanommer-latest --category %s --timesince' % identifier
sys.stderr.write("Running %r\n" % cmd) sys.stderr.write("Running %r\n" % cmd)
process = subprocess.Popen(cmd.split(), shell=False, process = subprocess.Popen(cmd.split(), shell=False,
stdout=subprocess.PIPE, stderr=subprocess.PIPE) stdout=subprocess.PIPE, stderr=subprocess.PIPE)
stdout, stderr = process.communicate() stdout, stderr = process.communicate()
prefix, stdout = stdout.split("INFO] ", 1)
data = json.loads(stdout) data = json.loads(stdout)
return float(data[0]) return float(data[0])
def main(): def main():
category, warning_threshold, critical_threshold = sys.argv[-3:] identifier, warning_threshold, critical_threshold = sys.argv[-3:]
timesince = query_timesince(category) timesince = query_timesince(identifier)
warning_threshold = int(warning_threshold) warning_threshold = int(warning_threshold)
critical_threshold = int(critical_threshold) critical_threshold = int(critical_threshold)
@ -43,7 +50,7 @@ def main():
time_strings.append("%d %s" % (value, denomination)) time_strings.append("%d %s" % (value, denomination))
string = ", ".join(time_strings) string = ", ".join(time_strings)
reason = "datanommer has not seen a %r message in %s" % (category, string) reason = "datanommer has not seen a %r message in %s" % (identifier, string)
if timesince > critical_threshold: if timesince > critical_threshold:
print "CRIT: ", reason print "CRIT: ", reason