Add nagios-check-owfs-temp

git-svn-id: svn+ssh://asteria.noreply.org/svn/weaselutils/trunk@253 bc3d92e2-beff-0310-a7cd-cc87d7ac0ede

1 files changed, 117 insertions, 0 deletions

diff --git a/nagios-check-owfs-temp b/nagios-check-owfs-temp
new file mode 100755
index 0000000..65cf851
--- /dev/null
+++ b/nagios-check-owfs-temp
@@ -0,0 +1,117 @@
+#!/usr/bin/ruby
+#
+# Copyright (c) 2006,2007 Peter Palfrader
+
+require 'optparse'
+
+NAGIOS_STATUS = { :OK => 0, :WARNING => 1, :CRITICAL => 2, :UNKNOWN => -1 };
+
+$owfs_path = '/var/lib/owfs'
+$owfs_path = ENV['OWFSPATH'] if ENV['OWFSPATH']
+
+def bail_out(m)
+	STDERR.puts "#{$0}: #{m}"
+	exit 1
+end
+
+def runcmd(command, input)
+	rdin , wrin  = IO.pipe
+	rdout, wrout = IO.pipe
+	rderr, wrerr = IO.pipe
+
+	pid = fork
+	unless pid
+		# child
+		wrin.close
+		rdout.close
+		rderr.close
+		STDIN.reopen rdin
+		STDOUT.reopen wrout
+		STDERR.reopen wrerr
+		exec(*command)
+		throw("fell through exec(). WTF.")
+	end
+	rdin.close
+	wrout.close
+	wrerr.close
+
+	out = []
+	err = []
+	tin  = Thread.new { wrin.print input; wrin.close }
+	tout = Thread.new { out = rdout.readlines }
+	terr = Thread.new { err = rderr.readlines }
+	tin.join
+	tout.join
+	terr.join
+	Process.wait pid
+
+	exitstatus = $?.exitstatus
+
+	[exitstatus, out, err]
+end
+
+def report(device)
+	Process.gid=0
+	Process.egid=0
+
+	# fuse does weird checks.  this fails:
+	# File.new($owfs_path+'/'+device+'/temperature', "r") ---> FAILS: in `initialize': Permission denied - /var/lib/owfs/10.D234EE000800/temperature (Errno::EACCES)
+
+	# Sometimes the directory in OWFS does not exist, and cat will say
+	# cat: /var/lib/owfs/10.D234EE000800/temperature: Invalid argument
+	# so retry it a few times, since ls'ing the directory helps on the command line at least
+	5.times do
+		exitstatus, out, err = runcmd(['cat', $owfs_path+'/'+device+'/temperature'], '')
+		if exitstatus != 0
+			STDERR.puts "Child exited with non-zero exit code(%d): %s"%[exitstatus >> 8, err.join]
+			runcmd(['ls', $owfs_path], '')
+			sleep 1
+		else
+			STDERR.puts "command succeeded but returned something on stderr: #{err.join}" if err.size > 0
+			return out.join.to_f
+		end
+	end
+	puts "Could not get data for #{device}."
+	exit NAGIOS_STATUS[:UNKNOWN];
+end
+
+
+@lowcrit = 10;
+@lowwarn = 15;
+@highwarn = 25;
+@highcrit = 30;
+
+def show_help(parser, code=0, io=STDOUT)
+  program_name = File.basename($0, '.*')
+  io.puts "Usage: #{program_name} [options]"
+  io.puts parser.summarize
+  exit(code)
+end
+ARGV.options do |opts|
+	opts.on_tail("-h", "--help" , "Display this help screen")                { show_help(opts) }
+	opts.on("-d", "--device=<ow device>"  , String, "Device ID")   { |@device| }
+	opts.on("-n", "--name=<name>"         , String, "Human readable device specifier")   { |@name| }
+	opts.on("-w", "--high-warn=integer"  , Integer, "Upper warning limit")   { |@highwarn| }
+	opts.on("-c", "--high-crit=integer"  , Integer, "Upper critical limit")  { |@highcrit| }
+	opts.on("-W", "--low-warn=integer"  , Integer, "Lower warning limit")    { |@lowwarn| }
+	opts.on("-C", "--low-crit=integer"  , Integer, "Lower critical limit")   { |@lowcrit| }
+	opts.parse!
+end
+show_help(ARGV.options, 1, STDERR) if ARGV.length != 0
+show_help(ARGV.options, 1, STDERR) unless @device
+
+@name = @device unless @name;
+
+temp = report @device
+
+
+if temp > @highcrit  or   temp < @lowcrit
+	puts "CRITICAL: device #{@name} is at #{temp}"
+	exit NAGIOS_STATUS[:CRITICAL]
+elsif temp > @highwarn  or  temp < @lowwarn
+	puts "WARN: device #{@name} is at #{temp}"
+	exit NAGIOS_STATUS[:WARNING]
+else
+	puts "OK: device #{@name} is at #{temp}"
+	exit NAGIOS_STATUS[:OK]
+end