Closing file descriptor for sshd that spawns remote threads

Greetings


I have a script that executes a remote secure shell to a server and runs a few startup scripts to launch in-house software. The scripts spawn various procs/threads so subsequently they inherit the shell's file descriptors...so the ssh session will hang as sshd is waiting for a close() it will never receive.

I can narrow down the PPID of the open fd's quite easily:
lsof -a -p `ps -ef|grep "\[priv\]" | grep -v grep | awk '{print $2}'` -d0,1,2

Unfortunately, I tried to redefine the fd inside the script and then close those before reassigning fd0,1,2 again, but it still hangs...I can always grep and | awk '{print $2}' for a string to get the PID to kill of the sshd but this could possibly return the wrong session should another sshd be open for the same grep string (which is slightly possible on the system)

Anyway, is there a good or standard way to close the fd for such an event? namely, an ssh to re remote server that executes a startup of various procs...or is there a good way to redefine the fd inside the script.

Thanks

Can you write a dummy script that any of us can run to demonstrate the problem?

It is similar, if not identical, to Bug52 in bugzilla, namely try the following:

From a host ssh to a client
Run any command that forks a process in the backgroun
Logout of client, ssh hangs

Similarly, shopt -s nohuponexit is one solution, however it doesn't work for me here. I also tried to export a PROMPT_COMMAND for "disown -a -h" but still no go. Finally, the only solution that works, which is a hack at best, is to kill -9 on the sshd that executes, but I wanted a more graceful way to get around this. Also, upgrading to a newer version of OpenSSH isn't a solution either (CM system).

I know there has been a lot of chatter on the web on these issues, but alas, no one had a solution, so figured I would give it a whirl here...

How about not running the commands from SSH but it activating an 'at' job containing those commands?