Alexandros Kosiaris has submitted this change and it was merged.
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.
Also differentiate between the runlockfile/adminlockfile introduced in
puppet 3.X and use symbols instead of magic numbers. Finally warn that
puppet is disabled unless told otherwise
Change-Id: I5a4439b18758a0915bc4ac6666f8f22435fb1689
---
M modules/base/files/monitoring/check_puppetrun
1 file changed, 54 insertions(+), 62 deletions(-)
Approvals:
Andrew Bogott: Looks good to me, but someone else must approve
Alexandros Kosiaris: Looks good to me, approved
jenkins-bot: Verified
diff --git a/modules/base/files/monitoring/check_puppetrun
b/modules/base/files/monitoring/check_puppetrun
index d16a6f3..efc159e 100755
--- a/modules/base/files/monitoring/check_puppetrun
+++ b/modules/base/files/monitoring/check_puppetrun
@@ -12,7 +12,8 @@
require 'optparse'
require 'yaml'
-lockfile = "/var/lib/puppet/state/puppetdlock"
+runlockfile = "/var/lib/puppet/state/agent_catalog_run.lock"
+adminlockfile = "/var/lib/puppet/state/agent_disabled.lock"
statefile = "/var/lib/puppet/state/state.yaml"
summaryfile = "/var/lib/puppet/state/last_run_summary.yaml"
enabled = true
@@ -23,28 +24,27 @@
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|
enabled_only = true
end
-opt.on("--lock-file [FILE]", "-l", "Location of the lock file, default
#{lockfile}") do |f|
- lockfile = f
+opt.on("--runlock-file [FILE]", "-l", "Location of the run lock file, default
#{runlockfile}") do |f|
+ runlockfile = f
+end
+
+opt.on("--adminlock-file [FILE]", "-a", "Location of the admin lock file,
default #{adminlockfile}") do |f|
+ adminlockfile = f
end
opt.on("--state-file [FILE]", "-t", "Location of the state file, default
#{statefile}") do |f|
@@ -62,12 +62,12 @@
exit 3
end
-if File.exists?(lockfile)
- if File::Stat.new(lockfile).zero?
+if File.exists?(adminlockfile)
enabled = false
- else
+end
+
+if File.exists?(runlockfile)
running = true
- end
end
lastrun = File.stat(statefile).mtime.to_i if File.exists?(statefile)
@@ -81,63 +81,55 @@
# are treated as huge failures. The yaml file will be valid but
# it wont have anything but last_run in it
unless summary.include?("events")
- failcount = 99
+ failcount = :failed
else
# and unless there are failures, the events hash just wont have
the failure count
failcount = summary["events"]["failure"] || 0
end
rescue
- failcount = 0
+ failcount = :unknown
summary = nil
end
+else
+ failcount = :nostatefile
end
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 not enabled
+ puts "WARNING: Puppet is currently disabled, last run
#{time_since_last_run} seconds ago with #{failcount} failures"
+ exit 1
+end
+
+if failcount == :failed
+ puts "CRITICAL: Complete puppet failure"
+ exit 2
+end
+
+if failcount == :unknown or failcount == :nostatefile
+ puts "UNKNOWN: Failed to check. Probably failed to read the state file"
+ exit 3
+end
+
+if failcount > 0
+ puts "CRITICAL: Puppet has ${failcount} failures"
+ exit 2
+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
+
+puts "OK: Puppet is currently enabled, last run #{time_since_last_run} seconds
ago with #{failcount} failures"
+exit 0
--
To view, visit https://gerrit.wikimedia.org/r/143332
To unsubscribe, visit https://gerrit.wikimedia.org/r/settings
Gerrit-MessageType: merged
Gerrit-Change-Id: I5a4439b18758a0915bc4ac6666f8f22435fb1689
Gerrit-PatchSet: 2
Gerrit-Project: operations/puppet
Gerrit-Branch: production
Gerrit-Owner: Andrew Bogott <[email protected]>
Gerrit-Reviewer: Alexandros Kosiaris <[email protected]>
Gerrit-Reviewer: Andrew Bogott <[email protected]>
Gerrit-Reviewer: jenkins-bot <>
_______________________________________________
MediaWiki-commits mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/mediawiki-commits
