The earthquake swarm that took place in 2012–2015 in the Upper Ubaye Valley (the most active seismic zone in the French Alps) is peculiar for two reasons: (1) it occurred a few kilometers from a previous swarm (active in 2003–2004); (2) it was initiated by an ML 4.3 shock and reactivated, more than two years later, by another ML 4.8 shock with an identical epicenter but a deeper focus. We present here the corresponding data set of ∼13,000 events, of which ∼3000 were relocated using a double‐difference algorithm. The swarm extends north‐northwest–south‐southeast (N165°E) over a distance of 11 km, but daily snapshots along a 2.5 yr period allow us to identify transverse faults whose activity was often ephemeral. Focal mechanisms for 13 ML≥3 events confirm the complexity of the swarm geometry, although the fault plane for the two “mainshocks” is very consistent (N156°E–N160°E strike, 52°–55°SW dip), with clear normal faulting and a slight dextral strike‐slip component. Most foci were located in the 4–11‐km depth range, within the crystalline basement. Taking into account the source sizes for the two “mainshocks”, the hydraulic diffusivity of 0.05 m2 s−1 found for the 2003–2004 swarm is shown to fit reasonably well the 2012 data, but not the 2014 reactivation sequence. Throughout the article, we discuss the difficult issue of the identification of “foreshocks” and “aftershocks” within such a sequence, even though a swarm, per se, makes this terminology inadequate. As previously suggested by other authors, between a foreshock–mainshock–aftershock sequence and an earthquake swarm also exists a whole gamut of seismic activity which makes this dichotomy much more complex than anticipated.