Catch SIGINT/SIGTERM/SIGHUP in start-runner.sh so the VM always powers off

[ophirt]

Problem

start-runner.sh powers the VM off when the runner exits, so Tartelet can recreate a fresh ephemeral VM:

function onexit {
  sudo shutdown -h now
}
trap onexit EXIT

The script is launched inside a GUI Terminal session (open -a Terminal ~/start-runner.sh). If that session is torn down — or the runner process is otherwise terminated by a signal — the shell receives SIGHUP/SIGTERM, and a zsh EXIT trap does not run when the shell is killed by a signal. So onexit never fires, shutdown -h now never runs, and the VM stays powered on with no runner.

Because Tartelet only provisions a replacement VM after it observes the current one stop, such a VM becomes a permanent "offline zombie": it shows offline in GitHub's runners page indefinitely and is never recycled, silently shrinking the pool while healthy VMs absorb all jobs.

We hit this in production: a runner died ~3s into boot with an I/O error on its console (the Terminal session went away before it reached "Listening for Jobs"), start-runner.sh got SIGTERM, the EXIT trap was skipped, and the VM never powered off.

Fix

Trap the termination signals too, so onexit (and therefore shutdown -h now) runs whether the shell exits normally or is killed by a signal:

-trap onexit EXIT
+trap onexit EXIT INT TERM HUP

onexit only runs sudo shutdown -h now, so invoking it on a signal (and again on the subsequent EXIT) is idempotent — the machine is halting either way.

Notes

• One-line change inside the generated start-runner.sh heredoc; no Swift API or type changes.

🤖 Generated with Claude Code