Re: Trap not being run for all CHLD signals, 4.3

[Chet Ramey]

On 9/3/14, 10:08 AM, crispusfairbairn@gmail.com wrote:
> $ cat parallel-test.bash
> function process_job {
>     sleep 1
> }
> 
> function main {
>     typeset -i index=0 cur_jobs=0 max_jobs=6
>     trap '((cur_jobs--))' CHLD
>     set -m
> 
>     while ((index++ < 30)); do
>         echo -n "index: $index, cur_jobs: $cur_jobs"
>         set +m
>         childs=$(pgrep -P $$ | wc -w)
>         (( childs < cur_jobs )) && echo -n ", actual childs: $childs"
>         echo
>         set -m
>         process_job &
>         ((++cur_jobs >= max_jobs)) && POSIXLY_CORRECT= wait;
>     done
> 
>     echo 'finished, waiting for remaining jobs...'
>     wait
> }
> 
> main
> echo "done"
> 
> This works on:
> GNU bash, version 4.1.2(1)-release (x86_64-redhat-linux-gnu)
> 
> But on:
> GNU bash, version 4.3.11(1)-release (x86_64-pc-linux-gnu)
> and
> GNU bash, version 4.3.24(1)-release (x86_64-unknown-linux-gnu)
> 
> it will around "index: 9" start missing traps (not decrementing cur_jobs):
> 
> $ bash-4.3.24/bin/bash parallel-test.bash
> index: 1, cur_jobs: 0
> index: 2, cur_jobs: 1
> index: 3, cur_jobs: 2
> index: 4, cur_jobs: 3
> index: 5, cur_jobs: 4
> index: 6, cur_jobs: 5
> index: 7, cur_jobs: 5
> index: 8, cur_jobs: 5
> index: 9, cur_jobs: 5, actual childs: 4
> index: 10, cur_jobs: 5, actual childs: 3
> index: 11, cur_jobs: 5, actual childs: 3
> ...
> 
> 
> If the sleep is changed to be random, it might work correctly for the whole 
> 30 iterations, which points to a race condition somewhere?

The problem is running the wait builtin in posixly-correct mode.  That
causes the first SIGCHLD to interrupt wait (as Posix requires) and results
in timing issues.

I will look at making SIGCHLD traps more reliable in the face of the Posix
requirements.

Chet
-- 
``The lyf so short, the craft so long to lerne.'' - Chaucer
                 ``Ars longa, vita brevis'' - Hippocrates
Chet Ramey, ITS, CWRU    chet@case.edu    http://cnswww.cns.cwru.edu/~chet/