GPT on Question 14, Sep 24

A Bash process cannot clean up after it has died. The usual approach is:

• 1.Handle normal exit and catchable signals with trap.
• 2.Put each child workload in its own process group/session.
• 3.Kill the whole group during cleanup.

#!/usr/bin/env bash

children=()

run() {
    setsid -- "$@" &
    children+=("$!")       # also the new process-group ID
}

cleanup() {
    status=$?
    trap - EXIT HUP INT TERM

    # Ask each child process group to terminate.
    for pgid in "${children[@]}"; do
        kill -TERM -- "-$pgid" 2>/dev/null || :
    done

    sleep 1

    # Forcefully remove anything still alive.
    for pgid in "${children[@]}"; do
        kill -KILL -- "-$pgid" 2>/dev/null || :
    done

    wait 2>/dev/null || :
    exit "$status"
}

trap cleanup EXIT
trap 'exit 129' HUP
trap 'exit 130' INT
trap 'exit 143' TERM

run command1 arg1
run command2 arg2

wait

The negative PID in kill -- "-$pgid" means “signal the entire process group,” so grandchildren are included unless they deliberately create another session/process group.

Important limitations:

• •Traps do not run for SIGKILL, kernel crashes, or abrupt power loss.
• •pkill -P $$ and jobs -p only reliably find direct children and are race-prone.
• •Avoid kill 0 unless the script is guaranteed to have its own process group; otherwise it may kill the invoking shell too.

For hard guarantees, run the script under a supervisor using a cgroup—e.g. systemd with its default KillMode=control-group. On Linux, another option is making each child set prctl(PR_SET_PDEATHSIG, ...), though Bash cannot do that natively.