Shell Programming and Scripting

I've search the various posts in these forums, but have not come up with a solution to my problem.

I have a parent process that calls a child script, runs it in the background and the parent finishes - without waiting for the child process to complete. Inside the child, a trap is issued to trap the QUIT signal and set a flag so that the child process can finish its current loop of processing, detect the flag and break out of the loop gracefully.

The problem is that the trap is not working in the orphaned child process. This only happens when I call the child from a parent and then let the parent complete without waiting for the child process to complete. As long as the parent process is still running, the child will properly trap the signal.

I would like to find a way to change this behavior so that the child - orphaned from its parent - will still trap signals sent to it.

Here's a simplistic example of how to set this up:

parent.ksh

Code:

child.ksh &

child.ksh

Code:

#!/bin/ksh
QUIT_THE_PROGRAM=N
trap "QUIT_THE_PROGRAM=Y" QUIT  #here's the trap that is not working when orphaned
while [[ ${QUIT_THE_PROGRAM} = "N" ]]
do
   echo Child PID - $$  # let me see the pid of this child process 
   sleep 5
   echo QUIT_THE_PROGRAM - ${QUIT_THE_PROGRAM}
   if [[ ${QUIT_THE_PROGRAM} = "Y" ]]
   then
      break
   fi
done

Make sure you chmod a+x child.ksh parent.ksh to make them executable.


OK - now run parent.ksh. That will call child.ksh and run it in the background and then the parent will complete and effectively orphan the child. Every 5 seconds, the child process will show you it's PID and the value of the QUIT_THE_PROGRAM variable.

Now, issue a kill -QUIT <pid> where <pid> is the PID of the child process. The child process ignores the signal - or at least the trap ignores it.

This is the behavior I want to change. I want this child process to trap the QUIT and change the value of the QUIT_THE_PROGRAM variable to Y.


Now - you can remove the & from parent.ksh and then run parent.ksh. This time the parent is waiting for the child process to complete. If you open another session on the server, you can issue a kill -QUIT <pid> where <pid> is the PID of the child process that is being printed in your first session. Now you will see that the non-orphaned child behaves properly and traps the QUIT signal and changes the value of the QUIT_THE_PROGRAM variable to Y.


Any ideas on how I can get the orphaned child process to trap the signal???

Thanks in advance for your input!

If I am getting what you are doing:

orphan processes are adopted by init, so that means you no longer own the process. [edit] init becomes the parent
[wrong]: You cannot signal a process you don't own. [/edit]

Have the parent signal the child before it exits. You do know that conventional program design in unix has the parent wait for the child, then exit - daemons excluded.

Interesting - the ps output still shows my id as running this process - though the PPID is 1.

Let's assume that you're correct (which I am sure you are despite what I see in ps).

I need a way to signal this orphaned process that is now owned by init. Is there a way?

I even tried logging in as root and issuing the kill -QUIT <orphaned_pid> and the child process would not trap the QUIT signal.

My problem is this - even if I follow proper coding practices and have the parent wait, someone or something could accidentally kill the parent and then the child will still be running. Then I have no method to gracefully end the child - which is needed. (The child process runs db table reorgs that if killed mid-process could corrupt the table.)

So the question is - how do I signal an orphaned process?

I can come up with other ways to trigger the QUIT - for example: read a control file everytime through the loop and then manually update the control file with a QUIT to trigger the QUIT_THE_PROGRAM=Y. But I like the trap

You are correct - the process owner is the original user. My mistake.

"The INT and QUIT signals for an invoked command are ignored if the command is followed by & and the monitor option is not active." -- ksh(1)

Either use the monitor option (set -m) in parent.ksh or use a different signal (perhaps SIGTERM).

Cheers,
alister

I am not a ksh expert, but I suppose that ksh is ignoring SIGQUIT for background processes. Accordingly to my system trace, your handler gets first installed, but then SIGQUIT sigaction gets reset to SIG_IGN.

If you use another signal, like SIGTERM, it seems that you can reach what you are looking for.

HTH,
Loïc.

alister wins the prize (for being first with the solution...)


Changing parent.ksh allows QUIT to be trapped by the child process.

Code:

set -m
child.ksh &

Also - even without set -m, trapping TERM also gets the desired results.

Thanks for the solutions!!!