work around ipmitool aborts

git-svn-id: svn+ssh://asteria.noreply.org/svn/weaselutils/trunk@126 bc3d92e2-beff-0310-a7cd-cc87d7ac0ede

1 files changed, 56 insertions, 28 deletions

diff --git a/munin/ipmi_sensor_ b/munin/ipmi_sensor_
index f364fc7..c2e0876 100755
--- a/munin/ipmi_sensor_
+++ b/munin/ipmi_sensor_
@@ -22,6 +22,7 @@ VALID_UNITS = %w{volts degrees_c rpm amps watts}
 CACHEDIR = "/var/lib/munin/plugin-state"
 CACHEFILE = "plugin-ipmi_sensor.cache"
 CACHE_AGE = 275
+IPMITOOL = '/usr/bin/ipmitool'
 
 def bail_out(m)
 	STDERR.puts "#{$0}: #{m}"
@@ -64,36 +65,63 @@ def get_sensor_data
 
 	data = []
 	names = {}
-	IO.popen("ipmitool -I open sensor") do |f|
-		f.each do |line|
-			line.strip!
-			# cpu1.vtt-s3      | 1.200      | Volts      | ok    | na        | 1.080     | na        | na        | 1.320     | na        
-			# CPU0 IERR        | 0x0        | discrete   | 0x0100| na        | na        | na        | na        | na        | na········
-			items = line.split(/\s*\|\s*/)
-			unless items.size == 10
-				bail_out "Got weird number of fields in ipmitool output.  Expected 10 but got #{items.size} for line '#{line}'."
-			end
-			(name, value, unit, status, lower_non_recov, lower_crit, lower_non_crit, upper_non_crit, upper_crit, upper_non_recv) = items
-			nname = normalize_sensor name;
-			if names.has_key?(nname)
-				bail_out "Sensor name #{name} (normalized to #{nname}) appears more than once in ipmitool output."
-			end
-			names[nname] = 1
-			data << {
-				:name => name,
-				:value => value,
-				:unit => unit,
-				:status => status,
-				:lower_non_recov => lower_non_recov,
-				:lower_crit => lower_crit,
-				:lower_non_crit => lower_non_crit,
-				:upper_non_crit => upper_non_crit,
-				:upper_crit => upper_crit,
-				:upper_non_recv => upper_non_recv,
-				:line => line
-			}
+
+	rdout, wrout = IO.pipe
+	rderr, wrerr = IO.pipe
+
+	pid = fork
+	unless pid
+		# child
+		rdout.close
+		rderr.close
+		STDOUT.reopen wrout
+		STDERR.reopen wrerr
+		exec(IPMITOOL, '-I', 'open', 'sensor')
+		bail_out("fell through exec(). WTF.")
+	end
+	wrout.close
+	wrerr.close
+
+
+	out = []
+	err = []
+	tout = Thread.new { out = rdout.readlines }
+	terr = Thread.new { err = rderr.readlines }
+	tout.join
+	terr.join
+	Process.wait pid
+
+	bail_out("ipmitool printed stuff on stdout: "+err.join('')) if (err.length > 0)
+
+	out.each do |line|
+		line.strip!
+		# cpu1.vtt-s3      | 1.200      | Volts      | ok    | na        | 1.080     | na        | na        | 1.320     | na        
+		# CPU0 IERR        | 0x0        | discrete   | 0x0100| na        | na        | na        | na        | na        | na········
+		items = line.split(/\s*\|\s*/)
+		unless items.size == 10
+			bail_out "Got weird number of fields in ipmitool output.  Expected 10 but got #{items.size} for line '#{line}'."
+		end
+		(name, value, unit, status, lower_non_recov, lower_crit, lower_non_crit, upper_non_crit, upper_crit, upper_non_recv) = items
+		nname = normalize_sensor name;
+		if names.has_key?(nname)
+			bail_out "Sensor name #{name} (normalized to #{nname}) appears more than once in ipmitool output."
 		end
+		names[nname] = 1
+		data << {
+			:name => name,
+			:value => value,
+			:unit => unit,
+			:status => status,
+			:lower_non_recov => lower_non_recov,
+			:lower_crit => lower_crit,
+			:lower_non_crit => lower_non_crit,
+			:upper_non_crit => upper_non_crit,
+			:upper_crit => upper_crit,
+			:upper_non_recv => upper_non_recv,
+			:line => line
+		}
 	end
+
 	write_cache data
 	data
 end