modified lastruncheck to run under v2.7.x, see github.com/cafuego/check_puppetmaster

1 files changed, 100 insertions, 223 deletions

diff --git a/files/master/lastruncheck b/files/master/lastruncheck
index d59e489..72c0eb5 100644
--- a/files/master/lastruncheck
+++ b/files/master/lastruncheck
@@ -1,224 +1,101 @@
-#!/usr/bin/env ruby
-require 'puppet/application'
-
-module Puppet::Lastcheck
-  module Puppet::Lastcheck::Tests
-    def self.included(klass)
-      klass.extend ClassMethods
-    end
-    def self.tests
-      @tests ||= {}
-    end
-    module ClassMethods
-      def add_test(name, options={})
-        include Puppet::Lastcheck::Tests.const_get(name.to_s.split('_').collect{|s| s.capitalize }.join(''))
-        Puppet::Lastcheck::Tests.tests[name] = options
-        attr_accessor "ignore_#{name}".to_sym
-        option("--ignore-#{name.to_s.gsub(/_/,'-')}") do
-          self.send("ignore_#{name}=", true)
-        end
-      end
-    end
-    module Util
-      def facts_hosts
-        return @facts_hosts if @facts_hosts
-        require 'puppet/indirector/facts/yaml'
-        @facts_hosts = Puppet::Node::Facts.indirection.search("*").collect do |fqdn|
-          if node = Puppet::Node::Facts.indirection.find(fqdn)
-            { :hostname => node.name, :expired => node.expired?, :timestamp => node.values[:_timestamp], :expiration => node.expiration }
-          end
-        end.compact
-      end
-    end
-  end
-  module Puppet::Lastcheck::Reports
-    def self.included(klass)
-      klass.extend ClassMethods
-    end
-    def ordered_reports
-        @ordered_reports ||= Puppet::Lastcheck::Reports.reports.keys.sort{|a,b| Puppet::Lastcheck::Reports.reports[a][:priority] <=> Puppet::Lastcheck::Reports.reports[b][:priority] }
-    end
-
-    def self.reports
-      @reports ||= {}
-    end
-    module ClassMethods
-      def add_report(name, options={})
-        include Puppet::Lastcheck::Reports.const_get(name.to_s.split('_').collect{|s| s.capitalize }.join(''))
-        Puppet::Lastcheck::Reports.reports[name] = options
-        Puppet::Lastcheck::Reports.reports[name][:priority] ||= 100
-        attr_accessor "report_to_#{name}".to_sym
-        option("--report-to-#{name.to_s.gsub(/_/,'-')}") do
-          self.send("report_to_#{name}=", true)
-        end
-      end
-    end
-  end
-end
-
-module Puppet::Lastcheck::Tests::NoFacts
-  def analyze_no_facts
-    signed_hosts.each{|host| add_failed_host(host,"No facts available") unless facts_hosts.any?{|fhost| fhost[:hostname] == host } }
-  end
-  def setup_no_facts
-    Puppet::SSL::Host.ca_location = :only
-  end
-
-  private
-  def signed_hosts
-    ca.list
-  end
-
-  def ca
-    @ca ||= Puppet::SSL::CertificateAuthority.new
-  end
-end
-
-module Puppet::Lastcheck::Tests::ExpiredFacts
-  include Puppet::Lastcheck::Tests::Util
-  def analyze_expired_facts
-      facts_hosts.each{|host| add_failed_host(host[:hostname],"Expired at #{host[:expiration]}") if host[:expired] }
-  end
-end
-module Puppet::Lastcheck::Tests::TimedOutFacts
-  include Puppet::Lastcheck::Tests::Util
-  def analyze_timed_out_facts
-    require 'time'
-    facts_hosts.each{|host| add_failed_host(host[:hostname], "Last facts save at #{host[:timestamp]}") if Time.parse(host[:timestamp].to_s) < (Time.now - @timeout) }
-  end
-
-  def setup_timed_out_facts
-    if @timeout
-      ignore_expired_facts ||= true
-    end
-  end
-end
-module Puppet::Lastcheck::Tests::Storedconfigs
-  def analyze_storedconfigs
-    storedconfigs_hosts.each do |host|
-      if !facts_hosts.any?{|fact_host| fact_host[:hostname] == host.name }
-        add_failed_host(host.name, "In storedconfigs but no facts available!")
-      elsif host.last_compile.nil?
-        add_failed_host(host.name, "No entry in storedconfigs")
-      elsif host.last_compile < (Time.now - @timeout)
-        add_failed_host(host.name, "Last compile time in storedconfigs at #{host.last_compile}")
-      end
-    end
-  end
-
-  private
-  def storedconfigs_hosts
-    return @storedconfigs_hosts if @storedconfigs_hosts
-    Puppet::Rails.connect
-    @storedconfigs_hosts = Puppet::Rails::Host.all
-  end
-end
-module Puppet::Lastcheck::Reports::Console
-  def deliver_report_to_console(failing_hosts)
-    unless failing_hosts.empty?
-      puts 'The following hosts are out of date:'
-      puts '------------------------------------'
-      host_length = 0
-      failing_hosts.keys.each{|host| host_length = host.length if host.length > host_length }
-      failing_hosts.keys.each{ |host| puts "#{pretty_puts(host,host_length)} - Reason: #{failing_hosts[host][:reason]}" }
-      1
-    else
-      0
-    end
-  end
-end
-module Puppet::Lastcheck::Reports::Nagios
-  def deliver_report_to_nagios(failing_hosts)
-    unless failing_hosts.empty?
-      puts "PUPPETLAST CRITICAL: #{failing_hosts.size} outdated hosts: #{failing_hosts.keys.join(',')}"
-      2
-    else
-      puts "PUPPETLAST OK: No outdated hosts"
-      0
-    end
-  end
-end
-#
-# = Synopsis
-#
-# Verifiying your puppet runs. Check different places to verify
-# whether your clients actually still runs successfully.
-# Also checks for left overs of legacy hosts.
+#!/bin/bash
 #
-# = Usage
-#
-#  puppet lastcheck [-h|--help] 
-class Puppet::Application::Lastcheck < Puppet::Application
-
-  should_parse_config
-  run_mode :master
-
-  include Puppet::Lastcheck::Tests
-  add_test :no_facts
-  add_test :expired_facts, :ignore_by_default => true
-  add_test :timed_out_facts
-  add_test :storedconfigs
-
-  include Puppet::Lastcheck::Reports
-  add_report :console, :priority => 50
-  add_report :nagios
-
-  option("--timeout TIMEOUT") do |v|
-    @timeout = v.to_i
-  end
-
-  option("--ignore-hosts HOSTS") do |v|
-    @ignore_hosts = v.split(',')
-  end
-
-  def main
-
-    Puppet::Lastcheck::Tests.tests.keys.each do |test|
-      self.send("analyze_#{test}") unless self.send("ignore_#{test}")
-    end
-    exitcode = 0
-    ordered_reports.each do |report|
-      if self.send("report_to_#{report}")
-        tmpexitcode = self.send("deliver_report_to_#{report}",@failing_hosts)
-        exitcode = tmpexitcode unless exitcode > 0
-      end
-    end
-    exit(exitcode)
-  end
-
-  def setup
-    exit(Puppet.settings.print_configs ? 0 : 1) if Puppet.settings.print_configs?
-
-    #Puppet::Util::Log.newdestination :console
-    Puppet::Node::Facts.indirection.terminus_class = :yaml
-
-    Puppet::Lastcheck::Tests.tests.keys.each do |test|
-      self.send("ignore_#{test}=", Puppet::Lastcheck::Tests.tests[test][:ignore_by_default]||false) unless self.send("ignore_#{test}")
-      self.send("setup_#{test}") if self.respond_to?("setup_#{test}") and !self.send("ignore_#{test}")
-    end
-    report = nil
-    report_activated = false
-    ordered_reports.each do |report|
-      report_activated ||= self.send("report_to_#{report}")
-    end
-    self.report_to_console = true unless report_activated
-
-    @ignore_hosts = [] unless @ignore_hosts
-    @failing_hosts = {}
-    unless @timeout 
-      @timeout = Puppet[:runinterval]
-    end
-  end
-
-  private
-
-  def add_failed_host(hostname,reason)
-    @failing_hosts[hostname] = { :reason => reason } unless (@failing_hosts[hostname] || @ignore_hosts.include?(hostname))
-  end
-
-  def pretty_puts(str,length)
-    sprintf("%0-#{length}s",str)
-  end
-end  
-
-Puppet::Application.find('lastcheck').new.run
+# Nagios checking script that whines if a node hasn't checked in with the
+# master for a day or more and sets a critical status if the node hasn't
+# checked in for a week or longer. You can of course tweak those thresholds
+# if you want.
+
+# Spot of configuration. Basically just specify where the yaml files live.
+YAMLPATH=/var/lib/puppet/yaml/node
+STATUS_OK=0
+STATUS_WARNING=1
+STATUS_CRITICAL=2
+STATUS_UNKNOWN=3
+INTERVAL_WARNING=$((60 * 60 * 2))
+INTERVAL_CRITICAL=$((60 * 60 * 24))
+PARAMS="-la"
+
+# A space separated list of hostnames to ignore. These might for instance be
+# laptops that just don't get used every day and thus don't sync.
+IGNORE_HOSTS=""
+
+# Override settings from a config file if one exists.
+if [ -f /etc/default/check_puppetmaster ]; then
+  . /etc/default/check_puppetmaster
+fi
+
+# Early exit if no read access to the yaml files.
+if [ ! -r ${YAMLPATH} ]; then
+  echo "UNKNOWN: Cannot access ${YAMLPATH}"
+  exit ${STATUS_UNKNOWN}
+fi
+
+# Bunch of internal vars used for status info output.
+status="OK"
+ret=${STATUS_OK}
+i_count=0
+o_count=0
+w_count=0
+w_string=""
+e_count=0
+e_string=""
+
+# Current time.
+NOW=$(date +"%s")
+
+# The meat.
+
+# Get all hostnames associated with active certificates, and check the time
+# each of these last checked in with the server. Do this by converting the
+# yaml file expiration datestamp to epoch format and subtracting it from now.
+for node in $(/usr/sbin/puppetca ${PARAMS} | awk '/^\+/ {print $2}' | tr -d '"'); do
+
+  EXPIRATION=$(grep expiration ${YAMLPATH}/$node.yaml | awk '{printf("%s %s", $2, $3);}')
+  typeset -i CHECKIN=$(date +"%s" -d "${EXPIRATION}")
+  DIFFERENCE=$((${NOW} - ${CHECKIN}))
+
+  # Count hosts and generate some output strings based on the status.
+  if [ ${DIFFERENCE} -lt ${INTERVAL_WARNING} ]; then
+    o_count=$((${o_count} + 1));
+  else
+    # If there is an issue, first check if we can ignore this host.
+    if [ -n "${IGNORE_HOSTS}" ]; then
+       if [[ ${IGNORE_HOSTS} =~ ${node} ]]; then
+         i_count=$((${i_count} + 1))
+         continue
+       fi
+    fi
+    if [ ${DIFFERENCE} -gt ${INTERVAL_CRITICAL} ]; then
+      e_count=$((${e_count} + 1))
+      e_string="${e_string} ${node}"
+    else
+      w_count=$((${w_count} + 1))
+      w_string="${w_string} ${node}"
+    fi
+  fi
+done
+
+# Generate a status string for user display.
+if [ -n "${e_string}" ]; then
+  s_string="${s_string} ${e_count} critical (${e_string## });"
+fi
+if [ -n "${w_string}" ]; then
+  s_string="${s_string} ${w_count} warning (${w_string## });"
+fi
+if [ ${i_count} -gt 0 ]; then
+  s_string="${s_string} ${i_count} ignored;"
+fi
+s_string="${s_string} ${o_count} ok."
+
+# Create a return value and status string.
+if [ ${e_count} -gt 0 ]; then
+  status="CRITICAL"
+  ret=${STATUS_CRITICAL}
+elif [ ${w_count} -gt 0 ]; then
+  status="WARNING"
+  ret=${STATUS_WARNING}
+fi
+
+# Output the status and inform the user about which hosts are lagging.
+echo -n "${status}:${s_string}"
+exit $ret