The introduction of WebRTC has opened a new playground for large-scale distributed applications consisting of large numbers of directly-communicating web browsers. In this context, gossip-based peer-sampling protocols appear as a particularly promising tool thanks to their inherent ability to build overlay networks that can cope with network dynamics. However, the dynamic nature of browser-to-browser communication combined with the connection establishment procedures that characterize WebRTC make current peer sampling solutions inefficient or simply unreliable. In this paper, we address the limitations of current peer-sampling approaches by introducing Spray, a novel peer-sampling protocol designed to avoid the constraints introduced by WebRTC. Unlike most recent peer-sampling approaches, Spray has the ability to adapt its operation to networks that can grow or shrink very rapidly. Moreover, by using only neighbor-to-neighbor interactions, it limits the impact of the threeway connection establishment process that characterizes WebRTC. Our experiments demonstrate the ability of Spray to adapt to dynamic networks and highlight its efficiency improvements with respect to existing protocols.