Re: [PATCH mmotm] vmscan: fix may_swap handling for memcg

[KOSAKI Motohiro]

> On Mon,  8 Jun 2009 12:20:54 +0900 (JST), KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx> wrote:
> > Hi
> > 
> Hi, thank you for your comment.
> 
> > > From: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx>
> > > 
> > > Commit 2e2e425989080cc534fc0fca154cae515f971cf5 ("vmscan,memcg: reintroduce
> > > sc->may_swap) add may_swap flag and handle it at get_scan_ratio().
> > > 
> > > But the result of get_scan_ratio() is ignored when priority == 0, and this
> > > means, when memcg hits the mem+swap limit, anon pages can be swapped
> > > just in vain. Especially when memcg causes oom by mem+swap limit,
> > > we can see many and many pages are swapped out.
> > > 
> > > Instead of not scanning anon lru completely when priority == 0, this patch adds
> > > a hook to handle may_swap flag in shrink_page_list() to avoid using useless swaps,
> > > and calls try_to_free_swap() if needed because it can reduce
> > > both mem.usage and memsw.usage if the page(SwapCache) is unused anymore.
> > > 
> > > Such unused-but-managed-under-memcg SwapCache can be made in some paths,
> > > for example trylock_page() failure in free_swap_cache().
> > > 
> > > Signed-off-by: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx>
> > 
> > I think root cause is following branch, right?
> yes.
> 
> > if so, Why can't we handle this issue on shrink_zone()?
> > 
> Just because priority==0 means oom is about to happen and I don't
> want to see oom if possible.
> So I thought it would be better to reclaim as much pages(memsw.usage) as possible
> in this case.

hmmm..

In general, adding new branch to shrink_page_list() is not good idea.
it can cause performance degression.

Plus, it is not big problem at all. it happen only when priority==0.
Definitely, priority==0 don't occur normally.
and, too many recliaming pages is not only memcg issue. I don't think this
patch provide generic solution.


Why your test environment makes oom so frequently?




--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/