mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jie Yu" <yujie....@gmail.com>
Subject Re: Review Request 33865: Added RevocableInfo message to Resource protobuf.
Date Thu, 07 May 2015 00:38:27 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/33865/#review82763
-----------------------------------------------------------



include/mesos/mesos.proto
<https://reviews.apache.org/r/33865/#comment133596>

    Chatted with Vinod offline (PS: Vinod is going to send out a summary of the discussion).
    
    In short, OVERSUBSCRIBED = (Allocated but unused) + (Unallocated)
    
    So, the type `OVERSUBSCRIBED` here is too general and ambiguate.
    
    You description of this review is interesting. Resources reserved for a different role
but unused (or unallocated) can be revokable. The question is to whom should we send this
revokable resources. For instance, if a framework under role A reserved some resources on
the slave (resources' role == A) and not using any of them yet. Should we send revokable offers
to framework under role A, or all frameworks. It's not clear yet and I think that's a policy
issue. It would be better to let a modulized component (e.g., resources estimator) to control
that so that we can potentially use different policies. If we decide to send revokable offer
for those reserved resources to all frameworks, we could strip the role in the revokable resources
(i.e., make them * resources).
    
    However, I don't know what type should we set for the RevokableInfo for the above case?
How can we distinguish that with the case where the resources are from unreserved resources
(i.e., anyone can use that)?
    
    As a result, I think maybe we should punt on the `type` here right now and just leave
an empty RevokableInfo since it's not strictly needed?



include/mesos/mesos.proto
<https://reviews.apache.org/r/33865/#comment133576>

    2 spaces indent please.



src/common/resources.cpp
<https://reviews.apache.org/r/33865/#comment133597>

    Should we put these checks to master validation? As I discussed with Mpark week ago, the
validations in Resources object should be minimal checks to ensure all methods in Resources
are well behaved.
    
    Clearly, the check you have here is a bit high level and probably should be put in master
validation?


- Jie Yu


On May 5, 2015, 9:13 p.m., Vinod Kone wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/33865/
> -----------------------------------------------------------
> 
> (Updated May 5, 2015, 9:13 p.m.)
> 
> 
> Review request for mesos, Jie Yu, Joris Van Remoortere, and Niklas Nielsen.
> 
> 
> Bugs: MESOS-2691
>     https://issues.apache.org/jira/browse/MESOS-2691
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> RevocableInfo currently supports OVERSUBSCRIBED resources. In the future it can be easily
extended to use other types of revocable reosurces (e.g., resources allocated to other roles).
> 
> Disabled the ability to use revocable resources for reservation or persistence because
the semantics seem weird. We can enable it in the future if there is a use case for that.
> 
> 
> Diffs
> -----
> 
>   include/mesos/mesos.proto db4fc8c001dd68bc3b9ca83650170c4f26db18c7 
>   src/common/resources.cpp 235930ff2dbb3ea49a3a0696dc070f2bd56fba4b 
>   src/tests/resources_tests.cpp a7ec59ea217ad71f7d1e93ca6039d5b2491b3237 
> 
> Diff: https://reviews.apache.org/r/33865/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> 
> Thanks,
> 
> Vinod Kone
> 
>


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