Data duplication is a common problem on data streams processing applications that occurs due to software error or adoption of data loss prevention measures, jeopardizing real-time data analyses. This paper explores stream-based deduplication methods to identify challenges from these methods and proposes a decision method to choose the most appropriate strategy for a domain. This work investigates native solutions and auxiliary tools to provide data deduplication and fault tolerance. The experimental results show that it is necessary to use fast additional storage to persist the read keys, as long as they can appear, or to use the optimized storage, with a quick key search.