Dataset definition for CMS operations and physics analyses

Data recorded at the CMS experiment are funnelled into streams, integrated in the HLT menu, and further organised in a hierarchical structure of primary datasets, secondary datasets, and dedicated skims. Datasets are defined according to the final-state particles reconstructed by the high level trigger, the data format and the use case (physics analysis, alignment and calibration, performance studies). During the first LHC run, new workflows have been added to this canonical scheme, to exploit at best the flexibility of the CMS trigger and data acquisition systems. The concept of data parking and data scouting have been introduced to extend the physics reach of CMS, offering the opportunity of defining physics triggers with extremely loose selections (e.g. dijet resonance trigger collecting data at a 1 kHz). In this presentation, we review the evolution of the dataset definition during the first run, and we discuss the plans for the second LHC run.