Retry gathering data

git-svn-id: svn+ssh://asteria.noreply.org/svn/weaselutils/trunk@147 bc3d92e2-beff-0310-a7cd-cc87d7ac0ede

1 files changed, 53 insertions, 17 deletions

diff --git a/munin/owfs_temperature_ b/munin/owfs_temperature_
index f797b21..8713990 100755
--- a/munin/owfs_temperature_
+++ b/munin/owfs_temperature_
@@ -74,32 +74,68 @@ def config
 	puts "#{n}.label #{label}"
 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 = query_device
 	Process.gid=0
 	Process.egid=0
-	temp = nil
 
 	# 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)
-	IO.popen("-") do |f|
-		unless f  # child
-			begin
-				exec('cat', $owfs_path+'/'+device+'/temperature')
-			rescue => e
-				puts "Cannot exec cat: "+e.message
-				exit 1
-			end
-		end
-		temp = f.readlines
-	end
 
-	if $? != 0
-		STDERR.puts "Child exited with non-zero exit code(%d): %s"%[$? >> 8, temp.join('')]
-	else
-		n = normalize_sensor device
-		puts "#{n}.value #{temp.join('')}"
+	# 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
+			n = normalize_sensor device
+			puts "#{n}.value #{out.join}"
+			return
+		end
 	end
+	STDERR.puts "Could not get data."
 end