Re: [PATCH V3 0/4] record non-slab shrinkers for non-root memcgs when kmem is disabled

[Andrew Morton]

On Wed, 11 Mar 2026 11:02:31 +0800 Haifeng Xu <haifeng.xu@xxxxxxxxxx> wrote:

> When registering new shrinkers, all memcgs need to expand shrinker info
> if new allocated id exceeds shrinker_nr_max. But if kmem is disabled,
> only non-slab shrinkers is useful in memcg slab shrink. So in this case,
> it is enough to allocate non-slab shrinker info for non-root memcg. This
> can save a bit of memory and reduce the holding time of shrinker lock.

Are you able to estimate how much memory it will save?

> With this optimization, the finish time of pod creation in our internal
> test is reduced from 150 seconds to 69 seconds. We test it based on
> stable kernel 6.6.102.

Is that internal test some crazy thing or is it representative of a
real-world workload?

IOW, what I'd like to see in your [0/N] is some indication of how much
time and memory we can expect real-world Linux users to see in
real-world workloads.

So please have a think about that, add some words then resend the
patchset.

Thanks.