Re: [RFC 0/2] opportunistic memory reclaim of a killed process

[Michal Hocko]

On Thu 11-04-19 12:56:32, Suren Baghdasaryan wrote:
> On Thu, Apr 11, 2019 at 11:19 AM Michal Hocko <mhocko@xxxxxxxxxx> wrote:
> >
> > On Thu 11-04-19 09:47:31, Suren Baghdasaryan wrote:
> > [...]
> > > > I would question whether we really need this at all? Relying on the exit
> > > > speed sounds like a fundamental design problem of anything that relies
> > > > on it.
> > >
> > > Relying on it is wrong, I agree. There are protections like allocation
> > > throttling that we can fall back to stop memory depletion. However
> > > having a way to free up resources that are not needed by a dying
> > > process quickly would help to avoid throttling which hurts user
> > > experience.
> >
> > I am not opposing speeding up the exit time in general. That is a good
> > thing. Especially for a very large processes (e.g. a DB). But I do not
> > really think we want to expose an API to control this specific aspect.
> 
> Great! Thanks for confirming that the intent is not worthless.
> There were a number of ideas floating both internally and in the 2/2
> of this patchset. I would like to get some input on which
> implementation would be preferable. From your answer sounds like you
> think it should be a generic feature, should not require any new APIs
> or hints from the userspace and should be conducted for all kills
> unconditionally (irrespective of memory pressure, who is waiting for
> victim's death, etc.). Do I understand correctly that this would be
> the preferred solution?

Yes, I think the general tear down solution is much more preferable than
a questionable API. How that solution should look like is an open
question. I am not sure myself to be honest.
-- 
Michal Hocko
SUSE Labs