User code deployment across compute resource partitions

Systems for processing requests to execute a program code of a user use a deployment model to select one of multiple virtual computing environments, each implemented on a plurality of server computers, which will produce the optimal program code execution, according to metrics such as latency, cost, and resource availability. The system receives the requests in the form of event messages associated with triggering events occurring on networks across the environments. The system feeds network usage data and event message metadata describing the event, event source, other target resources, and the like, into the deployment model to identify a candidate environment. The system enables the candidate environment to execute the program code, and then routes a subset of the event messages to the candidate environment, monitoring associated performance data. If the request processing is improved, the system continues routing some or all of the event messages to the candidate environment.