i tried scaling based on cpu usage once but it just caused massive context switching spikes. i ended up moving the logic to a simple cron job that checks the
queue:monitor
output and updates the config. its much more stable than trying to react to real-time traffic bursts.