Shell Programming and Scripting

That is exactly what your code does, but killing the subshell leaves each process in the tail|awk|tee pipeline alive and orphaned.

On a more general note, I can't think of any reason to ever use the following pipeline to count the lines of a single file:

Code:

wc -l file |awk '{print $1}'

wc will not print a pathname if you redirect stdin, wc -l < file. Alternatively, you can use AWK alone and print NR from an END block.

After running your script, you may want to consult with ps and, if necessary, rework your solution.

Regards,
Alister

Good point. (I did say it was untested! )

It's not hard to correct though. Something like:

Code:

childpids=$(ps -ef | awk -vPPID=$tailpid -vORS=' ' '$3==PPID {print $2}')
kill $childpids
 kill $tailpid

It would be simpler to just place the monitored and monitoring code in the same process group and use kill 0 to terminate the entire group.

Code:

set -m
exec 3>&2 2>/dev/null

sh 2>&3 <<'EOF'
    tail ... | awk ... | tee ... &

    while :; do
        monitoring-commands
    done

    cleanup-commands
    kill 0
EOF

The highlighted redirections can be omitted if the job control messages aren't a bother.

However, in my opinion, a fifo-based solution is a superior approach. Since reading on a fifo blocks, there's no need for a polling while-loop and explicit sleeps.

Code:

mkfifo data-pipe

tail -f file > data-pipe &
awk 'awk-script' data-pipe
kill $!

When the awk script exits (after three matches, presumably), tail is killed.

If portability isn't a concern, (ksh or bash) coprocesses can be used instead of mkfifo.

Regards,
Alister