Updated code for Tableau 9.0+ Servers

check_tableau_systeminfo

@@ -1,9 +1,9 @@
-#!/usr/bin/env ruby -w
+#!/usr/bin/ruby -w
 
 #       Usage: check_tableau_systeminfo -H <hostname>|-f [<file.xml>]
 # Description: check_tableau_systeminfo parses Tableau's systeminfo and reports Nagios status
 #              for Tableau server (http://www.tableausoftware.com/products/server)
-#              + CRITICAL: any service instance that is DOWN 
+#              + CRITICAL: any service instance that is DOWN
 #              +  WARNING: too many BUSY instances (per busy_thres)
 #      Author: gerir_at_ning_dot_com
 
@@ -12,7 +12,7 @@
    require 'net/http'
    require 'uri'
    require 'xmlsimple'
-   
+
    # busy thresholds
    # generates warning if n busy serviceinstances for service
    busy_thres = { 'W' => { 'serverwebapplication' => 12, 'vizqlserver' => 6 } }
@@ -23,10 +23,11 @@
            'C' => { 'f' => nil, 'm' => ''             , 'e' => 2, 's' => 'CRITICAL' },
            'U' => { 'f' => nil, 'm' => ''             , 'e' => 3, 's' => 'UNKNOWN'  }
          }
-
+   $host = ''
+
    # plugin output and exit code
    def nagiosMessageExit(status,message=nil)
-      unless message : message = $ns[status]['m'] end
+      message = $ns[status]['m'] unless message
       message.strip!
       puts "#{$ns[status]['s']}: #{message}"
       exit $ns[status]['e']
@@ -36,69 +37,75 @@
 # argument/options parsing and sanity checks
    begin
       options = {}
-   
+
       optparse = OptionParser.new do |opts|
-         opts.banner = "usage: check_tableau_systeminfo -H <hostname>|-f <file.xml>"
-  
+         opts.banner = "usage: check_tableau_systeminfo -H <hostname>|-f <file.xml>|-S <systemname>"
+
          options[:hostname] = nil
          opts.on( '-H', '--hostname HOSTNAME', 'Hostname' ) do |hostname|
             options[:hostname] = hostname
-         end
-
+            $host = /([^.]+)\..*/.match(hostname)
+	 end
+
          options[:filename] = nil
          opts.on( '-f', '--filename FILENAME', 'Filename' ) do |filename|
             options[:filename] = filename
+	    $host = /([^.]+)\..*/.match(filename)
          end
-         
+
          options[:verbose] = nil
          opts.on( '-v', '--verbose', 'Verbose' ) do
             options[:verbose] = true
          end
-   
+
          opts.on( '-h', '--help', 'Help' ) do
             puts opts
             exit
          end
       end
- 
+
       optparse.parse!
-      
+
       unless (options[:hostname] or options[:filename])
          nagiosMessageExit('U', "unable to parse arguments (none provided)")
       end
 
       if options[:hostname] and options[:filename]
          nagiosMessageExit('U', "--hostname and --filename are mutually exclusive options")
       end
-      
+
    rescue OptionParser::InvalidOption
       nagiosMessageExit('U', "error parsing options: #{$!}")
-      
+
    rescue RuntimeError
       nagiosMessageExit('U', "#{$!}")
-      
+
    end
 
 # fetch systeminfo
  # filename (useful for debugging)
    if options[:filename]
-      if options[:verbose] : puts "reading file #{options[:filename]}" end
-      begin    
+      puts "reading file #{options[:filename]}" if options[:verbose]
+      begin
+	 puts "#{$host[1]}" if options[:verbose]
          systeminfo = XmlSimple.xml_in(options[:filename], { 'KeyAttr' => 'name' })
       rescue ArgumentError
          nagiosMessageExit('U', "#{$!}")
       end
  # hostname
    elsif options[:hostname]
-      if options[:verbose] : puts "contacting host #{options[:hostname]}" end
+      puts "contacting host #{options[:hostname]}" if options[:verbose]
+#      host = /([^.]+)\..*/.match(#{options[:hostname]})
       begin
-         url = URI.parse("http://#{options[:hostname]}/admin/systeminfo")
+         url = URI.parse("http://#{options[:hostname]}/admin/systeminfo.xml")
          req = Net::HTTP::Get.new(url.path)
          res = Net::HTTP.start(url.host, url.port) {|http|
             http.request(req)
          }
-         if options[:verbose] : puts "parsing systeminfo\n#{res.body}" end
+         puts "parsing systeminfo\n#{res.body}" if options[:verbose]
          systeminfo = XmlSimple.xml_in(res.body, { 'KeyAttr' => 'name' })
+      rescue Timeout::Error
+         nagiosMessageExit('C', "#{$!}")
       rescue Errno::ECONNREFUSED
          nagiosMessageExit('C', "#{$!}")
       rescue REXML::ParseException
@@ -108,21 +115,31 @@
       end
    end
 
-# analyze systeminfo   
+#      systeminfo["machines"][0]["machine"]["Primary"].each do |service|
+# analyze systeminfo
    if systeminfo["machines"]
-      systeminfo["machines"][0]["machine"]["Primary"].each do |service|
+      systeminfo["machines"][0]["machine"]["#{$host[1]}"].each do |service|
          busy_c = 0
          service[1].each do |serviceinstance|
-            if (serviceinstance["status"].casecmp "down") == 0
-               $ns['C']['m'] += "#{service[0]} #{serviceinstance["worker"]} is DOWN! "
-               $ns['C']['f'] = true
-               break
-            elsif (serviceinstance["status"].casecmp "busy") == 0
-               busy_c += 1
+	    puts "service #{service[0]} #{serviceinstance["status"]}" if options[:verbose]
+            if (service[0] == "filestore")
+              if (serviceinstance["failedTransfers"].to_i >0)
+                $ns['C']['m'] += "filestore failedTransfers >0 "
+                $ns['C']['f'] = true
+              end
+            end
+            if (serviceinstance["status"].casecmp("Active") < 0)
+              if (serviceinstance["status"].casecmp("busy")) == 0
+                busy_c += 1
+              else
+                $ns['C']['m'] += "#{service[0]} is DOWN! "
+                $ns['C']['f'] = true
+                break
+              end
             end
          end
          break if $ns['C']['f']
-      
+
          if busy_thres['W'][service[0]]
             if busy_c >= busy_thres['W'][service[0]]
                $ns['W']['f'] = true
@@ -134,7 +151,7 @@
       nagiosMessageExit('U', "unable to parse systeminfo")
    end
 
-# report status   
+# report status
    if $ns['C']['f']
       nagiosMessageExit('C')
    elsif $ns['W']['f']

check_tableau_systeminfo_9_0

@@ -0,0 +1,140 @@
+#!/usr/bin/env ruby
+
+#       Usage: check_tableau_systeminfo -H <hostname>|-f [<file.xml>]
+# Description: check_tableau_systeminfo parses Tableau's systeminfo and reports Nagios status
+#              for Tableau server (http://www.tableausoftware.com/products/server)
+#              + CRITICAL: any service instance that is DOWN
+#              +  WARNING: too many BUSY instances (per busy_thres)
+#      Author: gerir_at_ning_dot_com
+
+   require 'rubygems'
+   require 'optparse'
+   require 'net/http'
+   require 'uri'
+   require 'xmlsimple'
+
+   # busy thresholds
+   # generates warning if n busy serviceinstances for service
+   busy_thres = { 'W' => { 'serverwebapplication' => 12, 'vizqlserver' => 6 } }
+
+   # global indeed; apologies
+   $ns = { 'O' => { 'f' => nil, 'm' => 'systeminfo OK', 'e' => 0, 's' => 'OK'       },
+           'W' => { 'f' => nil, 'm' => ''             , 'e' => 1, 's' => 'WARNING'  },
+           'C' => { 'f' => nil, 'm' => ''             , 'e' => 2, 's' => 'CRITICAL' },
+           'U' => { 'f' => nil, 'm' => ''             , 'e' => 3, 's' => 'UNKNOWN'  }
+         }
+
+   # plugin output and exit code
+   def nagiosMessageExit(status,message=nil)
+      unless message : message = $ns[status]['m'] end
+      message.strip!
+      puts "#{$ns[status]['s']}: #{message}"
+      exit $ns[status]['e']
+   end
+
+
+# argument/options parsing and sanity checks
+   begin
+      options = {}
+
+      optparse = OptionParser.new do |opts|
+         opts.banner = "usage: check_tableau_systeminfo -H <hostname>|-f <file.xml>"
+
+         options[:hostname] = nil
+         opts.on( '-H', '--hostname HOSTNAME', 'Hostname' ) do |hostname|
+            options[:hostname] = hostname
+         end
+
+         options[:filename] = nil
+         opts.on( '-f', '--filename FILENAME', 'Filename' ) do |filename|
+            options[:filename] = filename
+         end
+
+         options[:verbose] = nil
+         opts.on( '-v', '--verbose', 'Verbose' ) do
+            options[:verbose] = true
+         end
+
+         opts.on( '-h', '--help', 'Help' ) do
+            puts opts
+            exit
+         end
+      end
+
+      optparse.parse!
+
+      unless (options[:hostname] or options[:filename])
+         nagiosMessageExit('U', "unable to parse arguments (none provided)")
+      end
+
+      if options[:hostname] and options[:filename]
+         nagiosMessageExit('U', "--hostname and --filename are mutually exclusive options")
+      end
+
+   rescue OptionParser::InvalidOption
+      nagiosMessageExit('U', "error parsing options: #{$!}")
+
+   rescue RuntimeError
+      nagiosMessageExit('U', "#{$!}")
+
+   end
+
+# fetch systeminfo
+ # filename (useful for debugging)
+   if options[:filename]
+      if options[:verbose] : puts "reading file #{options[:filename]}" end
+      begin
+         systeminfo = XmlSimple.xml_in(options[:filename], { 'KeyAttr' => 'name' })
+      rescue ArgumentError
+         nagiosMessageExit('U', "#{$!}")
+      end
+ # hostname
+   elsif options[:hostname]
+      if options[:verbose] : puts "contacting host #{options[:hostname]}" end
+      begin
+         url = URI.parse("http://#{options[:hostname]}/admin/systeminfo.xml")
+         req = Net::HTTP::Get.new(url.path)
+         res = Net::HTTP.start(url.host, url.port) {|http|
+            http.request(req)
+         }
+         if options[:verbose] : puts "parsing systeminfo\n#{res.body}" end
+         systeminfo = XmlSimple.xml_in(res.body, { 'KeyAttr' => 'name' })
+      rescue Errno::ECONNREFUSED
+         nagiosMessageExit('C', "#{$!}")
+      rescue REXML::ParseException
+         nagiosMessageExit('U', "unable to parse systeminfo")
+      rescue
+         nagiosMessageExit('U', "#{$!}")
+      end
+   end
+#parse XML
+    begin
+        sysinfo = systeminfo["machines"][0]["machine"]
+        serinfo = systeminfo["service"]
+        for system in sysinfo.keys()
+            for app in sysinfo[system].keys()
+                for status in sysinfo[system][app]
+                    if status.keys().include?("status")
+                        if not (status["status"] == "Active" or status["status"]  == "Standby")
+                            nagiosMessageExit('C',"Non OK or Standby status in: #{system}:#{app}:#{status}")
+                        end
+                    end
+                end
+            end
+        end
+    rescue
+        nagiosMessageExit('C', "Error parsing tableau state.  systeminfo is: #{systeminfo}")
+    end
+    # parse service
+    if not serinfo[0]["status"] == "Active"
+        nagiosMessageExit('C',"Service status not Active: #{serinfo[0]["status"]}")
+    end
+
+# report status
+   if $ns['C']['f']
+      nagiosMessageExit('C')
+   elsif $ns['W']['f']
+      nagiosMessageExit('W')
+   else
+      nagiosMessageExit('O')
+   end