Re: [PATCH] mm/zswap: Improve with alloc_workqueue() call

[Yosry Ahmed <yosryahmed@google.com>]

On Thu, Jan 18, 2024 at 9:39 AM Johannes Weiner <hannes@cmpxchg.org> wrote:
>
> On Thu, Jan 18, 2024 at 09:06:43AM -0800, Yosry Ahmed wrote:
> > > > > On a different note, I wonder if it would help to perform synchronous
> > > > > reclaim here instead. With our current design, the zswap store failure
> > > > > (due to global limit hit) would leave the incoming page going to swap
> > > > > instead, creating an LRU inversion. Not sure if that's ideal.
> > > >
> > > > The global shrink path keeps reclaiming until zswap can accept again
> > > > (by default, that means reclaiming 10% of the total limit). I think
> > > > this is too expensive to be done synchronously.
> > >
> > > That thresholding code is a bit weird right now.
> > >
> > > It wakes the shrinker and rejects at the same time. We're guaranteed
> > > to see rejections, even if the shrinker has no trouble flushing some
> > > entries a split second later.
> > >
> > > It would make more sense to wake the shrinker at e.g. 95% full and
> > > have it run until 90%.
> > >
> > > But with that in place we also *should* do synchronous reclaim once we
> > > hit 100%. Just enough to make room for the store. This is important to
> > > catch the case where reclaim rate exceeds swapout rate. Rejecting and
> > > going to swap means the reclaimer will be throttled down to IO rate
> > > anyway, and the app latency isn't any worse. But this way we keep the
> > > pipeline alive, and keep swapping out the oldest zswap entries,
> > > instead of rejecting and swapping what would be the hottest ones.
> >
> > I fully agree with the thresholding code being weird, and with waking
> > up the shrinker before the pool is full. What I don't understand is
> > how we can do synchronous reclaim when we hit 100% and still respect
> > the acceptance threshold :/
> >
> > Are you proposing we change the semantics of the acceptance threshold
> > to begin with?
>
> I kind of am. It's worth looking at the history of this knob.
>
> It was added in 2020 by 45190f01dd402112d3d22c0ddc4152994f9e1e55, and
> from the changelogs and the code in this patch I do not understand how
> this was supposed to work.
>
> It also *didn't* work for very basic real world applications. See
> Domenico's follow-up (e0228d590beb0d0af345c58a282f01afac5c57f3), which
> effectively reverted it to get halfway reasonable behavior.
>
> If there are no good usecases for this knob, then I think it makes
> sense to phase it out again.

I am always nervous about removing/altering user visible knobs, but if
you think it's fine then I am all for it. I think it makes more sense
to start writeback early to avoid the whole situation if possible, and
synchronously reclaim a little bit if we hit 100%. I think the
proactive writeback should reduce the amount of synchronous IO we need
to do in reclaim as well, so we may see some latency improvements.