Predictive auto-scaling and reactive auto-scaling for network accessible messaging services

Technologies are disclosed for providing predictive auto-scaling for resources of a network-accessible message processing service. Control metrics are analyzed in order to predict a future volume of messages that will be received and processed by the message processing service. Based upon the prediction, resources, in the form of computing resources, are allocated to the message processing service. Reactive auto-scaling of the resources can also be used in conjunction with predictive auto-scaling. For reactive auto-scaling, an enqueue rate at a queuing service for messages is monitored and resources are reactively auto-scaled accordingly.