For children started by the Bash script, the usual pattern is to record their PIDs and clean them up with an `EXIT` trap:
```bash
#!/usr/bin/env bash
children=()
cleanup() {
rc=$?
trap - EXIT INT TERM
if ((${#children[@]})); then
kill -TERM -- "${children[@]}" 2>/dev/null || true
wait "${children[@]}" 2>/dev/null || true
fi
exit "$rc"
}
trap cleanup EXIT
trap 'exit 130' INT
trap 'exit 143' TERM
some_command &
children+=("$!")
another_command &
children+=("$!")
wait
```
Important limitations:
- `EXIT` traps do **not** run if the parent receives `SIGKILL`, crashes severely, or the machine loses power.
- This only handles recorded direct children, not necessarily grandchildren or processes that daemonize.
- Avoid blindly using `kill 0` or `kill -- -$$`; those signal an entire process group and may include unrelated processes or the shell itself.
If children must die even when the Bash parent is killed with `SIGKILL`, use an external mechanism:
- **systemd service/cgroup** with `KillMode=control-group`—generally the most robust option.
- On Linux, have each child set `prctl(PR_SET_PDEATHSIG, SIGTERM)`.
- Run the workload in a dedicated process group/session managed by a surviving supervisor.
So: use an `EXIT` trap for normal script cleanup; use a supervisor/cgroup or parent-death signal for a hard guarantee.