Re: What happen with vm.v_* on stable?

On Sat, 8 Mar 2025 13:14:21 -0500
Karl Denninger <karl@denninger.net> wrote:

> > Is some commits not backported from main to stable or it some
> > refactor artefacts?
> >
> >  
> Quite a long time ago the VM system and ZFS' interaction with this 
> (specifically UMA) had some very bad effects under certain workloads
> and these three sysctl's tried to mitigate that -- and not very well
> in some cases thus I wrote up a set of patches and they were under
> active discussion for some time (they solved the problem for me but
> not for everyone.)
> 
> It only bit you under certain circumstances but if it did it was
> pretty bad (e.g. 30 second no-response pauses, etc.)
> 
> I've not seen this sort of misbehavior in the last couple of major 
> releases and am running fairly-heavy "same general type" workloads
> that used to be trouble with ZFS on both 13 and 14 without incident.
> There has been a LOT of improvement in these interactions in the
> intervening years and OpenZFS was brought in as well which also was
> involved in significant improvement for certain workloads.
> 

Probably better to remove these sysctls as they do nothing and provide no usable data?


My problem is that amdgpu start freezes in case freemem =< "v_free_min". And I have no ZFS at all.

I am looking for some sysctls or patches to mitigate freezes.

Looks like this issue is same for me:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=277476
https://github.com/freebsd/drm-kmod/issues/302