Re: [PATCH v3 15/16] memcg/sl[au]b: shrink dead caches

[Tejun Heo <tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

Hello,

On Mon, Sep 24, 2012 at 12:25:00PM +0400, Glauber Costa wrote:
> > This is kinda nasty.  Do we really need to do this?  How long would a
> > dead cache stick around?
> 
> Without targeted shrinking, until all objects are manually freed, which
> may need to wait global reclaim to kick in.
> 
> In general, if we agree with duplicating the caches, the problem that
> they may stick around for some time will not be avoidable. If you have
> any suggestions about alternative ways for it, I'm all ears.

I don't have much problem with caches sticking around waiting to be
reaped.  I'm just wondering whether renaming trick is really
necessary.

> > Reaping dead caches doesn't exactly sound like a high priority thing
> > and adding a branch to hot path for that might not be the best way to
> > do it.  Why not schedule an extremely lazy deferrable delayed_work
> > which polls for emptiness, say, every miniute or whatever?
> > 
> 
> Because this branch is marked as unlikely, I would expect it not to be a
> big problem. It will be not taken most of the time, and becomes a very
> cheap branch. I considered this to be simpler than a deferred work
> mechanism.
> 
> If even then, you guys believe this is still too high, I can resort to that.

It's still an otherwise unnecessary branch on a very hot path.  If you
can remove it, there's no reason not to.

Thanks.

-- 
tejun