Re: sole instance of a process

[Vadim Zhukov]

21 нояб. 2014 г. 10:00 пользователь "Pete Vickers"
<peter.vick...@gmail.com>
написал:
>
> Hi,
>
> I suspect this may be the wrong list for this question. However although
strictly it's a Bourne shell script query, it only seem to act up under
OpenBSD (for me).
>
> Essentially I have a job which needs to be run periodically. So I have a
shell script to do the necessary commands, and this is scheduled via
(root's) crontab.
> It is however very important that multiple instances of the job are not
run concurrently (e.g. if an previous invocation hung), and so the script
should detect this upon invocation before proceeding.
>
> I don't want a single long running job (which could e.g. sleep between
loops) for various reasons. And I also don't like PID files and other
fragile locking hacks.
>
>
> So down to business, below is the gist of my script. Most of the time it
appears to run fine. However occasionally (once every couple of days?) it
reports via email that a duplicate process is detected, but the included ps
listing shows no other instance. I don't believe that this is just due to
an old instance exiting in the small time window between the pgrep, and the
ps invocations.  So basically I guess there is an error in my script or
it's logic, or something else I'm not seeing.
>
> Any hit with the clue bat gratefully received.
>
>
>
> #!/bin/sh
> #
> #
> SHOUT="/usr/bin/logger -i -t MYPERIODICJOB"
> #
> #
> # Ensure another instance of this is not running
> #
> MYNAME=`basename $0`
> MYPID=$$
> #
> /usr/bin/pgrep -fu root $MYNAME | /usr/bin/grep -v $MYPID && \

First problem: if first shell running script has PID 75 and second has PID
5, this grep command will ignore both.

Second problem: if you have anything in system with argument containing
script's name, like text editor, or file scanner, or whatever, it will
match, too, producing the problem you see.

I'd recommend you to save output of ps in a temporary variable, and run
grep on its content.

But the better solution should be locking of some sort. You can use
dolock(1) from ports infrastructure (don't forget to remove lock file in
EXIT trap), or just switch to Perl and proper file locks.

>         {
>                 $SHOUT "HELP - duplicate process detected $?" ; \
>                 ps -axjwww | mail -s "HELP MYPERIODICJOB $MYPID $MYNAME
$PPID" m...@example.com ; \
>                 exit 1 ;
>          }
>
> #
> #
> # starting doing useful stuff here..
> #
>
>
> Disclaimer: I know my scripting is far from optimal...

--
Vadim Zhukov