I started up my larger glusterfs (16 disks on one node mirrored to 16
other disks on another node, all unified), and noticed that stat-prefetch
still has some issues.
This filesystem had a bunch of /usr copies and copies of my home
directory. A du was not fully successful (it would die prematurely,saying
that several directories were not found, even though they were clearly
there). Also, when I started multiple 'rm's on both nodes, the
glusterfs processes would die within a few minutes. I remounted and
tried again many times.
When I removed stat-prefetch from my configuration, the removes ran to
completion, so it seems solid without stat-prefetch.
It looks like stat-prefetch dies under heavy meta-data stress...
Thanks,
Brent
