使用bash脚本分发命令行

Question

I'm trying to implement trivial parallelization where lines of commands are distributed among separate processes. For that purpose I wrote this script that I named jobsel :

(only "#! /bin/bash" and help message is omitted)

slots=$1
sel=$2

[[ $slots -gt 0 ]] || die_usage
[[ $sel -lt $slots ]] || die_usage

i=0
while read line
do
        (( i % slots == sel )) && eval $line
        i=$(( i + 1 ))
done

# in case the last line does not end with EOL
if [[ $line != "" ]]; then
        (( i % slots == sel )) && eval $line
        i=$(( i + 1 ))
fi

I put eval because I couldn't use redirection or pipe in the commands without it.

When I run this like $HOME/util/jobsel 22 0 < cmds on a console emulator when cmds is a file that contains lines like echo 0 >> out with increasing numbers, it outputs, as expected, 0, 22, 44... in separate lines. Good so far.

So I put this to work. But when I ran this through secure shell, I ran it through at with backgrounding (ending each line with & ). Then there is a problem. When I entered 8 lines, 21 processes started! ps -AFH printed processes with identical command and different pIDs. All work processes were at the same level, directly under init. My program does not create child processes anyway.

Puzzled, I tried the echo 0 >> out script through at then the output contained duplicate lines. Still finding it hard to believe, and thinking simultaneous appending might have caused the anomaly, I used other methods to confirm that some lines were run multiple times.

Moreover, there was no such anomaly when everything was run in terminal or when I created separate at job for each worker process.

But how can this happen? Is something wrong with my script? Does at/atd have some bug?

Answer 1

GNU parallel is what you are trying to implement.

Hint : it works great combined to xargs