On Wed, Jan 30, 2019 at 05:54:27AM +0000, Roman Gushchin wrote:
Hi, Andrew!
I believe, that Rik's patch ( https://lkml.org/lkml/2019/1/28/1865 ) can make a difference here, and might fix the regression. I'd give it a chance, before reverting these two patches. Reverting will re-introduce the memcg-leak, which is quite bad.
Rik's change is just another hack that will still have effects on reclaim behaviour.
Indeed, the fs/inode.c change definitely needs reverting, because that is just *plain wrong* and breaks long-standing memory reclaim behaviour.
I seriously disagree with shovelling a different, largely untested and contentious change to the shrinker algorithm to try and patch over the symptoms of the original change. It leaves the underlying problem unfixed (dying memcgs need a reaper to shrink the remaining slab objects that pin that specific memcg) and instead plays "whack-a-mole" on what we alreayd know is a fundamentally broken assumption (i.e. that shrinking small slabs more agressively is side-effect free).
-Dave.