Some of my scripts kick off a number of children in background, and then wait for their completion. Currently I invoke the wait in a loop for each PID:

for pid in "${!PIDs[@]}"
do
        if wait $pid
        then
                log ${PIDs[$pid]} completed successfully
        else
                log WARNING: ${PIDs[$pid]} failed
                errors+=1
        fi
done

This works, allowing me to analyze and process failures, but the processing happens in the order, in which PIDs are listed -- not in the order, in which the processes actually complete. That is, the 5th process may finish first, but its exit-code will not be processed until the first four are done...

As far as I know, sh provides two modes for wait:

1. Bare wait will wait for all backgrounded jobs to finish, but it will always "succeed" losing the exit-codes of the backgrounded processes.
2. wait PID will wait for the specified process. This is providing the exit-code, but can only wait for that one process.

But, maybe, bash has this improved compared to the old sh? Is there a way to request bash's wait to return when any of the backgrounded processes completes -- and have it provide both the finished PID and its exit-code?

The underlying C-functions waitpid and friends can do this -- if you provide the PID of -1. I tried doing that with bash and got an error...

Some of my scripts kick off a number of children in background, and then wait for their completion. Currently I invoke the wait in a loop for each PID:

for pid in "${!PIDs[@]}"
do
        if wait $pid
        then
                log ${PIDs[$pid]} completed successfully
        else
                log WARNING: ${PIDs[$pid]} failed
                errors+=1
        fi
done

This works, allowing me to analyze and process failures, but the processing happens in the order, in which PIDs are listed -- not in the order, in which the processes actually complete. That is, the 5th process may finish first, but its exit-code will not be processed until the first four are done...

As far as I know, sh provides two modes for wait:

1. Bare wait will wait for all backgrounded jobs to finish, but it will always "succeed" losing the exit-codes of the backgrounded processes.
2. wait PID will wait for the specified process. This is providing the exit-code, but can only wait for that one process.

But, maybe, bash has this improved compared to the old sh? Is there a way to request bash's wait to return when any of the backgrounded processes completes -- and have it provide both the finished PID and its exit-code?

The underlying C-functions waitpid and friends can do this -- if you provide the PID of -1. I tried doing that with bash and got an error...

• bash
• wait
• background-process

• Did you take a look at this? stackoverflow.com/questions/356100/… – León Commented Feb 6 at 23:48
• 1 See the wait section of the Bash Reference Manual. The -n and -p options may be what you are looking for. – pjh Commented Feb 7 at 1:25
• Also see ProcessManagement - Greg's Wiki. – pjh Commented Feb 7 at 1:26

3 Answers 3

If you are restricted to a version of Bash that does not support wait -n then one possible way to do what you want is to poll with the jobs builtin to detect when background processes have completed. This Shellcheck-clean code (lightly tested with Bash 4.2) demonstrates the idea:

unwaited_pids=( "${!pids[@]}" )
declare -A is_running

while (( ${#unwaited_pids[*]} > 0 )); do
    sleep 0.1
    jobs_output=$(jobs -rp)
    is_running=()
    while IFS= read -r p; do
        [[ -n $p ]] && is_running[$p]=Y
    done <<<"$jobs_output"

    new_unwaited_pids=()

    for pid in "${unwaited_pids[@]}"; do
        if [[ -n ${is_running[$pid]-} ]]; then
            new_unwaited_pids+=( "$pid" )
        elif wait "$pid"; then
            log "${pids[$pid]} completed successfully"
        else
            log "WARNING: ${pids[$pid]} failed"
            (( errors+=1 ))
        fi
    done

    unwaited_pids=( "${new_unwaited_pids[@]}" )
done

You can adapt following script to your needs :

while test -n "${PIDs[*]}" && { wait -n -p pid "${!PIDs[@]}"; status=$?; }; do
    echo "Command [${PIDs[$pid]}] with pid:$pid exited with status:$status"
    unset "PIDs[$pid]" # Remove $pid
done

For bash 4.2, try this :

while (( ${#PIDs[@]} > 0 )); do
    sleep .1 # Sleep .1 second
    for pid in "${!PIDs[@]}"
    do  
        [ -d "/proc/$pid" ] && continue # $pid still running, check next
        if wait "$pid"
        then
            echo "${PIDs[$pid]} completed successfully"
        else
            echo "WARNING: ${PIDs[$pid]} failed"
            errors+=1
        fi
        unset "PIDs[$pid]" # Remove $pid
    done
done

Might take a few steps.

Just as a test, complete with one process killed to prove it catches error codes.

$: cat tst
#! /usr/bin/env bash
for x in 1 3 5 7 9; do sleep $x & done
declare -A rc=()
pids=($(jobs -pr))
while (( ${#pids[@]} ))
do for k in "${!pids[@]}"
   do p=${pids[$k]}
      if ps -p $p >/dev/null; then :
      else wait $p; rc+=( $p $? ); unset pids[$k]
           date +"%F %T PID $p: rc ${rc[$p]}"
      fi
   done
   ((skip++)) || kill ${pids[3]}
   sleep 1
done

$: ./tst
2025-02-07 14:55:19 PID 3036: rc 0
2025-02-07 14:55:19 PID 3039: rc 143
2025-02-07 14:55:21 PID 3037: rc 0
2025-02-07 14:55:23 PID 3038: rc 0
2025-02-07 14:55:28 PID 3040: rc 0