Hello,
I am trying to run some distributed jobs, and unfortunately what I am running may try to display some unimportant graphics which I dont care, and thus crash if no display is present. For this reason I dispatch a job which first creates an Xvfb virtual framebuffer, and outputs everything on this display.

My problem seems to be in detecting which display is free, so that I can start an Xvfb on that display. I have come up with a bash script, which get a start display of 10000, and checks if there is a lock file for this display. If not it increments the display or else starts an Xvfb there. There is the chance that because of a race condition another process may try the same display, so I check the exit status and if it is not 0, I increment the display to use and try again.

Unfortunately in some occasions (about 5-8% of the jobs), something may go wrong, and I get the following errors
Cannot establish any listening sockets - Make sure an X server isn't already run
ning
_XSERVTransSocketINETCreateListener: ...SocketCreateListener() failed
_XSERVTransMakeAllCOTSServerListeners: server already running

Fatal server error:


The script I use to detect the display is the following
Any recommendations on my approach to detect free displays, or improvements on the script? Also any insight into what may be going wrong could help me work around the issue

Why don't you check if $disp + 1 lock file exists as well. You can't assume that if /tmp/.X1000-lock file exists that /tmp/.X1001-lock file does not.

I am not sure how your code checks that exactly.
What the current code I pasted does, is checking for lock files, if one exists increment the display number. If the lock file does not exist, then attempt to start Xvfb, and stil check if Xvfb started successfully.

So in theory I check all displays, until I find a display that has no lock file. And once I consider Xvfb to have started successfully I break the infinite while loop.

Unfortunately this method does not always work, so my criteria for successful Xvfb run is wrong. Also a completely different approach to finding free displays may exist and be far superior than mine.

My bad. I misunderstood you script. It does what mine was aiming at. Just increase the display number till a free lock file is found.
I checked some man page and the net to find a way to get a free display number but did not find anything usefull within the X family programs. What I wonder if you can start xvfb without a display number and have it create some itself. You could then get the DISPLAY variable out of the enviroment file under /proc/$pid/.

Maybe you could create a lock file after finding a free display number. Then short to starting xvfb delete it again. Or maybe just bump up the increment from 1 to say 10. Or some other alternating numbers.
Nother approach would be to monitor the /tmp directory with say inotifywait.

1 members found this post helpful.

Unfortunately Xvfb will try display :0 if no argument is given, and will crash since :0 is taken. Also since Xvfb creates its own lock files I dont see the need. The jobs are dispatched by a central machine trhough the sun grid engine. So there is no backwards communicatiton, or control of which machine each job will go to so that I can control the display. Also there is a chance that two users may start submitting jobs, thus creating another potential conflict in displays.
So most suggestions will not work unfortunately, except maybe this inotifywait. I dont even know what it is so let me take a good look and see if I can use it on this case

Thanks for the suggestions

you can use netcat to check if the port is in use: http://stackoverflow.com/questions/4...p-port-is-open. http://www.freesoft.org/CIE/Topics/113.htm

also you can use instead of:
set xvfb_pid=`echo $!`
just:
set xvfb_pid=$!

1 members found this post helpful.