One solution is to have metric_daemon epoll on the /dev/chromeos-low-mem to write the timestamp to /tmp/uptime-low-mem-triggered. And later, we could read the timestamp from Tab Discarder to get the full latency for the first kill of tab/app.