GPT on Question 14, Sep 15

The idiomatic approach is to have the parent clean up via trap, rather than having children poll $PPID.

#!/usr/bin/env bash
set -Eeuo pipefail

children=()

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

    if ((${#children[@]})); then
        kill -TERM -- "${children[@]}" 2>/dev/null || true
        wait "${children[@]}" 2>/dev/null || true
    fi

    exit "$status"
}
trap cleanup EXIT INT TERM HUP

some_command &
children+=("$!")

another_command &
children+=("$!")

wait

This handles normal exit and catchable signals. It cannot handle the parent being killed with SIGKILL, crashing before cleanup, or the machine losing power.

For entire descendant trees, use a dedicated process group/session or, preferably, a supervisor/cgroup:

• •systemd service: KillMode=control-group ensures all service processes are terminated.
• •Containers/cgroups: terminate the whole cgroup.
• •Linux child-side guarantee: use a small wrapper that calls prctl(PR_SET_PDEATHSIG, SIGTERM) before exec. Bash has no built-in equivalent.

Avoid periodically checking $PPID == 1; orphaned processes may be reparented to a subreaper rather than PID 1, and such polling is inherently racy.