How can a bash script ensure that not more than one copy of it is running?

Question

How can a bash script ensure that not more than one copy of it is running?

I have tried:

ps -ef| grep /user/loca/shell/sh | wc -l

This shows me 2 because of the grep command. Change it to:

ps -ef | grep /user/loca/shell/sh | grep -v 'grep' wc -l 

then its shows 1. However, if I then vim /user/loca/shell/sh and then execute:

ps -ef| grep /user/loca/shell/sh | grep -v 'grep' wc -l

that shows 2. But there is only one process, I have started.

How can the bash script check whether the process is running?

• glenn jackman: Is /user/loca/shell/sh a script or a compiled executable? I suspect that you need to search the ps output more carefully. Please show the ps output when you're both running and editing.
• joe: yes , they are running and executable ...
• joe: $ ps -ef | grep '/user/loca/shell/sh ' root 30915 30891 0 15:19 pts/3 00:00:00 vim /user/loca/shell/sh joe 30956 30927 0 15:21 pts/5 00:00:00 grep /user/loca/shell/sh
• glenn jackman: adding that to the question would make it easier to read, however you don't appear to be actually running the program in that example.
• Jonathan Leffler: Note that if I run bash /user/loca/shell/sh or bash -c /user/loca/shell/sh, then it is nigh on impossible to spot the difference between that and vim /user/loca/shell/sh or diff -c /user/loca/shell/sh /user/loca/shell/new.sh. Not to mention cp /user/loca/shell/sh xyz; ./xyz. Hence the 'lock file' approach.

Answer 1

The idiom often used in Unix Daemons is to create a file with the PID in it when they start.

Then you can check the existence and/or content of the file when you start, if it's there, you exit and leave that instance running.

At the end of the script you delete the file, ready to run next time.

The only problem comes when the script runs, but does not complete for some reason, leaving the PID file in existence. This can be taken care of by checking its timestamp, if it's too long ago, you assume that it was an aborted run and continue with the current one.

• Aaron Digulla: To delete the file use trap "rm $PID_FILE" EXIT
• jlliagre: Instead of the unreliable pid file timestamp method, better to check if the pid is in use and match the script name.
• Jonathan Leffler: Also, the daemon can check whether the process recorded in the PID file is still extant. The kill -0 PID 2>/dev/null command will report success if the process PID exists, and failure if not.
• evil otto: On most modern unix variants there is a /proc filesystem with directories or files for every currently running process. Rather than writing the pid into a file, make the pid file a symlink to the pid file in /proc, then to check if the program is still running see if the file is a broken symlink (test -h pidfile -a ! -e pidfile == broken symlink)

Answer 2

Try pgrep instead of 'ps -ef ...' in your script:

pgrep /usr/loca/shell/sh

If that won't work then I'd resort to locking by attempting to create a symbolic link from the script if one does not exist yet and bail out if it already exists, and deleting it upon exit. Creating symbolic link is an atomic operation on unix that can be used as sort of a poorman's lock for shell scripts.

Answer 3

I use a file lock:

MY_LOCK_FILE='whatever.lock'
( if (flock -nx 200); then
 # Do stuff here...
 fi
) 200>"$MY_LOCK_FILE"