Comment (by Gabriele Bozzola):

I think I have another instance of this bug with a completely different parameter file (but same machine/compiler). I am performing a simulation of head-on collision with moving grids. Initially Carpet says that the required memory is 235 GB, and this goes down to 180 GB as time goes on (as I would expect, since the grids are colliding). Then, the simulation is killed due to the lack of memory. The job IDs on blue waters are 10105916.bw and 10106827.bw (the second restart, when the simulation crashes).

--
Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2194/memory-increase-during-regridding