Re: [PATCH v3] vm_swappiness=0 should still try to avoid swapping anon memory

[Johannes Weiner <hannes@cmpxchg.org>]

On Tue, Apr 19, 2022 at 07:54:52PM -0400, Nico Pache wrote:
> 
> 
> On 4/19/22 14:46, Johannes Weiner wrote:
> > Hi Nico,
> > 
> > On Tue, Apr 19, 2022 at 02:11:53PM -0400, Nico Pache wrote:
> >> I think its is important to note the issue we are seeing has greatly improved
> >> since the initial posting. However we have noticed that the issue is still
> >> present (and significantly worse) when cgroupV1 is set.
> >>
> >> We were initially testing with CgroupV1 and later found that the issue was not
> >> as bad in CgroupV2 (but was still an noticeable issue). This is also resulting
> >> in the splitting of THPs in the host kernel.
> > 
> > When swappiness is 0, cgroup limit reclaim has a fixed SCAN_FILE
> > branch, so it shouldn't ever look at anon. I'm assuming you're getting
> > global reclaim mixed in. Indeed, I think we can try harder not to swap
> > for global reclaim if the user asks for that.
> > 
> > Can you try the below patch?
> Sadly this did not fix the V1 case.
> 
> I will have to go back through my notes and over the code again to find what
> difference between the two may be causing this issue. Im just starting to focus
> on this issue again so my memory needs some refreshing of where I was last at.
> 
> The good news is that with the current state of upstream the issue of swap
> storms tearing down THPs seems to be minimized to sane amount with V2.
> 
> My swappiness=0 solution was a minimal approach to regaining the 'avoid swapping
> ANON' behavior that was previously there, but as Shakeel pointed out, there may
> be something larger at play.

So with my patch and swappiness=0 you get excessive swapping on v1 but
not on v2? And the patch to avoid DEACTIVATE_ANON fixes it?

If you haven't done so, it could be useful to litter shrink_node() and
get_scan_count() with trace_printk() to try to make sense of all the
decisions that result in it swapping.