An emerging Fog Computing paradigm aims at reducing overall latency and end-to-end data transfer costs between IoT devices and analytic entity as in comparison to purely Cloud based solutions. In the near future, Fog services will be offered on virtual marketplaces by Fog Node owners. To facilitate the exchange in such volatile markets, we devise a recurrent, multiunit, combinatorial auction with bidder drop and resource waste control and a two-step winner determination algorithm. Proposed solution extends on Participation Incentive Optimal Recurring Auction (PI-ORA) algorithm, which separately solves bidder drop problem for single-unit, non combinatorial auction, to include the case of multi-unit combinatorial auction which is necessary for Fog service markets. The results show stability and efficiency of the proposed algorithm.