On Wed, Aug 08, 2018 at 05:07:08PM -0700, Matthew Wilcox wrote: > On Thu, Aug 09, 2018 at 07:31:25AM +1000, Dave Chinner wrote: > > IMO, we've had enough recent bugs to deal with from shrinkers being > > called before the filesystem is set up and from trying to handle > > allocation errors during setup. Do we really want to make shrinker > > shutdown just as prone to mismanagement and subtle, hard to hit > > bugs? I don't think we do - unmount is simply not a critical > > performance path. > > It's never been performance critical for me, but I'm not so sure that > there aren't container workloads which unmount filesystems multiple > times per second. What? Why would they do that? Who cares about tear-down speeds? Start up speeds I can kind of understand...
