Andrew Bogott has uploaded a new change for review.
https://gerrit.wikimedia.org/r/143332
Change subject: Simplify check_puppetrun.
......................................................................
Simplify check_puppetrun.
Previously it had different modes; now it just checks anything.
Now:
It reports staleness first. If fresh, it reports compile failures.
If compile is working, it reports errors. If no errors, it reports
time since last run.
Change-Id: I5a4439b18758a0915bc4ac6666f8f22435fb1689
---
M modules/base/files/monitoring/check_puppetrun
1 file changed, 33 insertions(+), 53 deletions(-)
git pull ssh://gerrit.wikimedia.org:29418/operations/puppet
refs/changes/32/143332/1
diff --git a/modules/base/files/monitoring/check_puppetrun
b/modules/base/files/monitoring/check_puppetrun
index d16a6f3..52606bf 100755
--- a/modules/base/files/monitoring/check_puppetrun
+++ b/modules/base/files/monitoring/check_puppetrun
@@ -23,20 +23,15 @@
warn = 0
crit = 0
enabled_only = false
-failures = false
opt = OptionParser.new
-opt.on("--critical [CRIT]", "-c", Integer, "Critical threshold, time or failed
resources") do |f|
+opt.on("--critical [CRIT]", "-c", Integer, "Critical staleness threshold, time
in seconds") do |f|
crit = f.to_i
end
-opt.on("--warn [WARN]", "-w", Integer, "Warning thresold, time of failed
resources") do |f|
+opt.on("--warn [WARN]", "-w", Integer, "Warning staleness threshold, time in
seconds") do |f|
warn = f.to_i
-end
-
-opt.on("--check-failures", "-f", "Check for failed resources instead of time
since run") do |f|
- failures = true
end
opt.on("--only-enabled", "-e", "Only alert if Puppet is enabled") do |f|
@@ -94,50 +89,35 @@
time_since_last_run = Time.now.to_i - lastrun
-unless failures
- if enabled_only && enabled == false
- puts "OK: Puppet is currently disabled, not alerting. Last run
#{time_since_last_run} seconds ago with #{failcount} failures"
- exit 0
- end
-
- if time_since_last_run >= crit
- puts "CRITICAL: Puppet last ran #{time_since_last_run} seconds ago,
expected < #{crit}"
- exit 2
-
- elsif time_since_last_run >= warn
- puts "WARNING: Puppet last ran #{time_since_last_run} seconds ago,
expected < #{warn}"
- exit 1
-
- else
- if enabled
- puts "OK: Puppet is currently enabled, last run
#{time_since_last_run} seconds ago with #{failcount} failures"
- else
- puts "OK: Puppet is currently disabled, last run
#{time_since_last_run} seconds ago with #{failcount} failures"
- end
-
- exit 0
- end
-else
- if enabled_only && enabled == false
- puts "OK: Puppet is currently disabled, not alerting. Last run
#{time_since_last_run} seconds ago with #{failcount} failures"
- exit 0
- end
-
- if failcount >= crit
- puts "CRITICAL: Puppet last ran had #{failcount} failures, expected <
#{crit}"
- exit 2
-
- elsif failcount >= warn
- puts "WARNING: Puppet last ran had #{failcount} failures, expected <
#{warn}"
- exit 1
-
- else
- if enabled
- puts "OK: Puppet is currently enabled, last run
#{time_since_last_run} seconds ago with #{failcount} failures"
- else
- puts "OK: Puppet is currently disabled, last run
#{time_since_last_run} seconds ago with #{failcount} failures"
- end
-
- exit 0
- end
+if enabled_only && enabled == false
+ puts "OK: Puppet is currently disabled, not alerting. Last run
#{time_since_last_run} seconds ago with #{failcount} failures"
+ exit 0
end
+
+if time_since_last_run >= crit
+ puts "CRITICAL: Puppet last ran #{time_since_last_run} seconds ago,
expected < #{crit}"
+ exit 2
+end
+
+if time_since_last_run >= warn
+ puts "WARNING: Puppet last ran #{time_since_last_run} seconds ago,
expected < #{warn}"
+ exit 1
+end
+
+if failcount == 99
+ puts "CRITICAL: Complete puppet failure."
+ exit 2
+end
+
+if failcount > 0
+ puts "CRITICAL: Puppet has ${failcount} failures."
+ exit 2
+end
+
+if enabled
+ puts "OK: Puppet is currently enabled, last run #{time_since_last_run}
seconds ago with #{failcount} failures"
+else
+ puts "OK: Puppet is currently disabled, last run #{time_since_last_run}
seconds ago with #{failcount} failures"
+end
+
+exit 0
--
To view, visit https://gerrit.wikimedia.org/r/143332
To unsubscribe, visit https://gerrit.wikimedia.org/r/settings
Gerrit-MessageType: newchange
Gerrit-Change-Id: I5a4439b18758a0915bc4ac6666f8f22435fb1689
Gerrit-PatchSet: 1
Gerrit-Project: operations/puppet
Gerrit-Branch: production
Gerrit-Owner: Andrew Bogott <[email protected]>
_______________________________________________
MediaWiki-commits mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/mediawiki-commits
