HDF5 consumes too much memory (Memory Leak ?)
While profiling pallas_print -T
on a minife trace, I realized that the H5_Malloc memory seems a bit too large: here, it's 500MB for a 170MB trace, and never goes down:
I've attached the .massif file: minife_64_pallas_print.massif
As a side note: The "readCurrentLevel" memory never going down is because we never free the ThreadReaders and their ThreadTraces. It might also be because of the Savestates, which may be badly handled memory-wise.
Edited by Catherine Guelque