Hi.
For the clarity, that pr is basically started based on the idea that
currently the memory used by the prewarm pool is not being taken into
account while calculating the resource of invokers.
So we can't have a big enough prewarm pool as it will reduce the size of
"user memory". Otherwise, a bigger prewarm pool may cause an excess of
memory in the invoker machine.
If we can keep memory used by the prewarm pool as a part of the invoker
user memory, we can safely increase the size of the prewarm pool.
But still, as Markus mentioned, I think it is the key to this PR to define
a good eviction process for the prewarm pool.

We can easily imagine that if only actions whose kinds are not in the
prewarm pool are invoked and all user memory is consumed, an invoker would
not create more containers even if there are a bunch of free containers in
the prewarm pool.
So it is required to have a process to deal with such situations.

I haven't think of this in-depth yet, I have no good idea yet.
But I just wanted to make the intention of this PR clear.

Regards
Dominic.




2020년 3월 30일 (월) 오후 5:28, Markus Thömmes <markusthoem...@apache.org>님이 작성:

> Hi,
>
> this does mean, that the prewarm pool's capacity cannot be reclaimed by
> other actions should the need arise, right? I think that's problematic if
> that notion isn't carried through into loadbalancing itself as AFAIK the
> controller currently does not take the prewarm pool into account at all.
>
> Which kind of a workload pattern would that benefit? Got any benchmark
> and/or just data showcasing that?
>
> Cheers,
> Markus
>
> Am Mo., 30. März 2020 um 10:13 Uhr schrieb 甯尤刚 <ning.youg...@navercorp.com
> >:
>
> > Hi, guys
> > ​
> >     I wrote a pr here: https://github.com/apache/openwhisk/pull/4870
> >     this pr's target: make prewarmedPool as `already consumed memory
> > resources`, this pr has an benefit that if consider prewarmedPool as
> > consumed memory resource, can aovid to do  much `cold start a container`
> > when the time of `space is consumed finished` is coming
> > ​
> >     Have any suggestion?
> >
>

Reply via email to