Federated query engines allow data consumers to execute queries over the federation of Linked Data (LD). However, as feder-ated queries are decomposed into potentially thousands of subqueries distributed among SPARQL endpoints, data providers do not know fed-erated queries, they only know subqueries they process. Consequently, unlike warehousing approaches, LD data providers have no access to secondary data. In this paper, we propose FETA (FEderated query TrAcking), a query tracking algorithm that infers Basic Graph Patterns (BGPs) processed by a federation from a shared log maintained by data providers. Concurrent execution of thousand subqueries generated by multiple fed-erated query engines makes the query tracking process challenging and uncertain. Experiments with Anapsid show that FETA is able to extract BGPs which, even in a worst case scenario, contain BGPs of original queries.