mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Benjamin Mahler <bmah...@apache.org>
Subject Re: Review Request 64467: Rewrote the quota headroom enforcement logic in the allocator.
Date Wed, 13 Dec 2017 21:08:16 GMT


> On Dec. 12, 2017, 5:03 a.m., Benjamin Mahler wrote:
> > Can you also describe how it worked before in the description? I think that would
be helpful for the reader and for posterity.
> 
> Meng Zhu wrote:
>     Done.

Hm.. did you forget to add it?


> On Dec. 12, 2017, 5:03 a.m., Benjamin Mahler wrote:
> > src/master/allocator/mesos/hierarchical.cpp
> > Lines 1861 (patched)
> > <https://reviews.apache.org/r/64467/diff/3/?file=1912390#file1912390line1901>
> >
> >     Why are you removing shared?
> 
> Meng Zhu wrote:
>     OK, that is redundant. But just so we are in the same page, shared resources should
not be hold back for headroom.

Can you elaborate? Not that clear to me.


> On Dec. 12, 2017, 5:03 a.m., Benjamin Mahler wrote:
> > src/master/allocator/mesos/hierarchical.cpp
> > Lines 1931-1939 (original), 1926-1932 (patched)
> > <https://reviews.apache.org/r/64467/diff/3/?file=1912390#file1912390line1972>
> >
> >     Hm.. doesn't the ancestor comment still apply? Your change is altering the sematics?
> >     
> >     It should be the case that "eng/frontend" would get offered a reservation for
"eng", but that won't happen if the headroom is not satisfied? It would only go to "eng/frontend"
directly in that case?
> 
> Meng Zhu wrote:
>     OK, fixed and kept the original semantics.
>     
>     hm.. looks like resources reserved by my ancestor, while allocatable to me, is not
"reserved" by me. I wonder what's the problem of making the notion of reservation heritable,
seems to be easier to understand? -- my parents' reservations are also my reservations.

> looks like resources reserved by my ancestor, while allocatable to me, is not "reserved"
by me

Yes, a reservation to "eng" could be allocated to "eng", "eng/frontend", "eng/backend", "eng/c/d/e/f/g",
etc.

> I wonder what's the problem of making the notion of reservation heritable, seems to be
easier to understand? -- my parents' reservations are also my reservations.

Not sure I follow, but here's what I'm wondering about:

Let's say I have a reservation for "eng". When we have hierarchical quota, let's say we "eng/frontend"
that has reached its quota limit and "eng/backend" which has not. I would expect that the
"eng" reservation would only be given to "eng/backend". Or if both "eng/frontend" and "eng/backend"
have quota satisfied, then I would expect neither of them to get the "eng" reservation. Make
sense?


- Benjamin


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/64467/#review193497
-----------------------------------------------------------


On Dec. 13, 2017, 6:54 a.m., Meng Zhu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/64467/
> -----------------------------------------------------------
> 
> (Updated Dec. 13, 2017, 6:54 a.m.)
> 
> 
> Review request for mesos, Benjamin Mahler and Michael Park.
> 
> 
> Bugs: MESOS-8293
>     https://issues.apache.org/jira/browse/MESOS-8293
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Now before offering unreserved resources to frameworks, the
> resources are holdout for the quota headroom until the headroom
> is met (reserved resources are offered unaffected).
> 
> 
> Diffs
> -----
> 
>   src/master/allocator/mesos/hierarchical.cpp 2b2d1fd2802203eba482be2992a5f2756d100cbf

>   src/tests/hierarchical_allocator_tests.cpp 862f4683da04d37d9fe9f471d6ec9cd7751f39ec

> 
> 
> Diff: https://reviews.apache.org/r/64467/diff/4/
> 
> 
> Testing
> -------
> 
> make check and a dediated test in #64465
> 
> 
> Thanks,
> 
> Meng Zhu
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message