I have a Linux cluster that runs a script that copies a file every time a new node takes over. The copy process itself is simple.

Part of the copy routine is a simple scp process to update a file on the inactive nodes. It needs to break away from the parent script and retry until success. That could take 3 seconds or 30 seconds or 30 days. Die on reboot is the only relevant way to end because the cluster will just move to another node and start it again.

I personally don't care if the process keeps trying to do the scp steps to clean up the inactive nodes as a zombie. Is there a professional method of handling a possibly long-running process like the one I described? FWIW, I will write the script in PERL or Dash.

What I like about this method is this is, in theory, very simple and cluster specific. No syncing crontabs and scripts, cron job tests for active node status, or more complicated configuration management systems.

I would write that as a program and launch it as a daemon. In fact I wrote a few blog entries about how to create a daemon and manage it, as well as how to clean up zombies.

http://www.linuxquestions.org/questi...ocesses-35540/
http://www.linuxquestions.org/questi...zombies-35575/

To correct your perspective though. Zombies do nothing. They exist merely because the parent which forked them is still around, the former process which became a zombie terminated. In terminating, it sent a signal to the parent process. I believe this is part of the System V closure architecture. If the parent process never checks for that signal, the defunct process remains as a zombie. If the parent process itself exits, then any zombie processes attached to it will be resolved. In the end, the zombie is not an active process, it does nothing except to hold a place in the system. Ultimately they're not good to have.

My recommended architecture would be to write a daemon, run that in the system and use that to perform you timing for when you wish to launch your child process. Once the time, or conditions are correct for launching your intended process, use fork() and exec() to do this, and then have your daemon periodically check for signals from that process to clean up the zombie state.

1 members found this post helpful.

rtmistler,

Thanks for the reply. Zombie is not the right term. I just need the script to start a new process then exit leaving that process running.

I don't NEED to control the process, but a daemon the more professional way to do it.

Based on your reply, I found this handy package to do the work for me! http://packages.debian.org/wheezy/daemon Very nice features.