Abstract: Techniques are described for determining test cases to test a service, such as a service to manage a purchase contract in an e-commerce environment. Log data may be generated during execution of the service to process requests, the log data indicating a plurality of code segments that executed during the execution of the service. The log data may be analyzed to generate metrics data comprising a plurality of records corresponding to the processed requests. A record may include any number of counters that each indicates whether, or how many times, a particular code segment executed during processing of a request. The metrics data may be analyzed to identify a subset of records comprising representative combinations of the counters in the metrics data. The requests corresponding to one or more of the subset of records may be replayed to test the service.