In #59 it was reported that an application crashed with SIGBUS if /dev/shm was configured with insufficient space. On the mailing list it was suggested that this case could be detected and reported more gracefully if we replace the use of ftruncate() with fallocate(), although that would sacrifice portability since fallocate() is Linux-specific. @adammoody also pointed out that we'd need to examine how memory pages are assigned to NUMA banks with fallocate().
In #59 it was reported that an application crashed with SIGBUS if /dev/shm was configured with insufficient space. On the mailing list it was suggested that this case could be detected and reported more gracefully if we replace the use of
ftruncate()withfallocate(), although that would sacrifice portability sincefallocate()is Linux-specific. @adammoody also pointed out that we'd need to examine how memory pages are assigned to NUMA banks withfallocate().