Usually a good trick is to run small deployments at high logging levels. Then, as soon as there are performance issues, you can dial down the logging and get t

Usually a good trick is to run small deployments at high logging levels. Then, as soon as there are performance issues, you can dial down the logging and get the hours of respite needed to actually make a bigger algorithmic improvement.