Re: [PATCH] mm: vmscan: split khugepaged stats from direct reclaim stats

[Yosry Ahmed]

On Thu, Oct 27, 2022 at 7:15 AM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
>
> On Wed, Oct 26, 2022 at 07:41:21PM -0700, Yosry Ahmed wrote:
> > On Wed, Oct 26, 2022 at 1:51 PM Yang Shi <shy828301@xxxxxxxxx> wrote:
> > >
> > > On Wed, Oct 26, 2022 at 10:32 AM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
> > > >
> > > > On Tue, Oct 25, 2022 at 02:53:01PM -0700, Yang Shi wrote:
> > > > > On Tue, Oct 25, 2022 at 1:54 PM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Tue, Oct 25, 2022 at 12:40:15PM -0700, Yang Shi wrote:
> > > > > > > On Tue, Oct 25, 2022 at 10:05 AM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
> > > > > > > >
> > > > > > > > Direct reclaim stats are useful for identifying a potential source for
> > > > > > > > application latency, as well as spotting issues with kswapd. However,
> > > > > > > > khugepaged currently distorts the picture: as a kernel thread it
> > > > > > > > doesn't impose allocation latencies on userspace, and it explicitly
> > > > > > > > opts out of kswapd reclaim. Its activity showing up in the direct
> > > > > > > > reclaim stats is misleading. Counting it as kswapd reclaim could also
> > > > > > > > cause confusion when trying to understand actual kswapd behavior.
> > > > > > > >
> > > > > > > > Break out khugepaged from the direct reclaim counters into new
> > > > > > > > pgsteal_khugepaged, pgdemote_khugepaged, pgscan_khugepaged counters.
> > > > > > > >
> > > > > > > > Test with a huge executable (CONFIG_READ_ONLY_THP_FOR_FS):
> > > > > > > >
> > > > > > > > pgsteal_kswapd 1342185
> > > > > > > > pgsteal_direct 0
> > > > > > > > pgsteal_khugepaged 3623
> > > > > > > > pgscan_kswapd 1345025
> > > > > > > > pgscan_direct 0
> > > > > > > > pgscan_khugepaged 3623
> > > > > > >
> > > > > > > There are other kernel threads or works may allocate memory then
> > > > > > > trigger memory reclaim, there may be similar problems for them and
> > > > > > > someone may try to add a new stat. So how's about we make the stats
> > > > > > > more general, for example, call it "pg{steal|scan}_kthread"?
> > > > > >
> > > > > > I'm not convinved that's a good idea.
> > > > > >
> > > > > > Can you generally say that userspace isn't indirectly waiting for one
> > > > > > of those allocating threads? With khugepaged, we know.
> > > > >
> > > > > AFAIK, ksm may do slab allocation with __GFP_DIRECT_RECLAIM.
> > > >
> > > > Right, but ksm also uses __GFP_KSWAPD_RECLAIM. So while userspace
> > > > isn't directly waiting for ksm, when ksm enters direct reclaim it's
> > > > because kswapd failed. This is of interest to kernel developers.
> > > > Userspace will likely see direct reclaim in that scenario as well, so
> > > > the ksm direct reclaim counts aren't liable to confuse users.
> > > >
> > > > Khugepaged on the other hand will *always* reclaim directly, even if
> > > > there is no memory pressure or kswapd failure. The direct reclaim
> > > > counts there are misleading to both developers and users.
> > > >
> > > > What it really should be is pgscan_nokswapd_nouserprocesswaiting, but
> > > > that just seems kind of long ;-)
> > > >
> > > > I'm also not sure anybody but khugepaged is doing direct reclaim
> > > > without kswapd reclaim. It seems unlikely we'll get more of those.
> > >
> > > IIUC you actually don't care about how many direct reclaim are
> > > triggered by khugepaged, but you would like to separate the direct
> > > reclaim stats between that are triggered directly by userspace
> > > actions, which may stall userspace, and that aren't, which don't stall
> > > userspace. If so it doesn't sound that important to distinguish
> > > whether the direct reclaim are triggered by khugepaged or other kernel
> > > threads even though other kthreads are not liable to confuse users
> > > IMHO.
>
> I feel like I've sufficiently explained my reason for wanting to
> separate out the __GFP_KSWAPD_RECLAIM special case from other sites.
>
> > My 2c, if we care about direct reclaim as in reclaim that may stall
> > user space application allocations, then there are other reclaim
> > contexts that may pollute the direct reclaim stats. For instance,
> > proactive reclaim, or reclaim done by writing a limit lower than the
> > current usage to memory.max or memory.high, as they are not done in
> > the context of the application allocating memory.
> >
> > At Google, we have some internal direct reclaim memcg statistics, and
> > the way we handle this is by passing a flag from such contexts to
> > try_to_free_mem_cgroup_pages() in the reclaim_options arg. This flag
> > is echod into a scan_struct bit, which we then use to filter out
> > direct reclaim operations that actually cause latencies in user space
> > allocations.
> >
> > Perhaps something similar might be more generic here? I am not sure
> > what context khugepaged reclaims memory from, but I think it's not a
> > memcg context, so maybe we want to generalize the reclaim_options arg
> > to try_to_free_pages() or whatever interface khugepaged uses to free
> > memory.
>
> So at the /proc/vmstat level, I'm not sure it matters much because it
> doesn't count any cgroup_reclaim() activity.
>
> But at the cgroup level, it sure would be nice to split out proactive
> reclaim churn. Both in terms of not polluting direct reclaim counts,
> but also for *knowing* how much proactive reclaim is doing.
>
> Do you have separate counters for this?

Not yet. Currently we only have the first part, not polluting direct
reclaim counts.

We basically exclude reclaim coming from memory.reclaim, setting
memory.max/memory.limit_in_bytes, memory.high (on write, not hitting
the high limit), and memory.force_empty from direct reclaim stats.

As for having a separate counter for proactive reclaim, do you think
it should be limited to reclaim coming from memory.reclaim (and
potentially memory.force_empty), or should it include reclaim coming
from limit-setting as well?