A method that involves receiving budget information of a containerized application deployed with a set of containers to a set of machine instances; receiving pricing information of a list of machine instance types; receiving performance information of the set of containers; receiving an alert generated based on the performance information by comparing the performance information to a set of thresholds; generating, after receiving the alert, an output vector from a machine learning model, wherein the machine learning model uses the performance information; and adjusting a resource of the set of containers by updating a parameter based on the output vector in response to the alert, wherein the resource is controlled by the parameter, and wherein the parameter is identified in the alert.