> On Jan. 11, 2018, 10 p.m., Greg Mann wrote:
> > src/master/master.hpp
> > Lines 881-885 (original), 881-888 (patched)
> > <https://reviews.apache.org/r/65095/diff/1/?file=1939029#file1939029line881>
> >
> >     I'm thinking that `convertResources` might be a more intuitive name for 
> > the boolean parameter here? In all cases, resources associated with the 
> > operation are recovered. In only some cases, we must first convert the 
> > consumed resources to converted resources so that we can then recover them.
> >     
> >     So, the variable behavior seems to be whether or not we are converting 
> > resources before we recover them. WDYT?
> >     
> >     
> >     I think the comment here could be tweaked a bit to better reflect this 
> > as well. Perhaps:
> >     
> >     "Transitions the operation and, if the operation becomes terminal, 
> > recovers the resources associate with the operation. If `convertResources` 
> > is false, then no conversion from consumed to converted resources is 
> > applied in the allocator prior to resource recovery."

Thank you for the suggestion, I changed the parameter name to 
`convertResources`.

I also update the comment slightly,

    // Transitions the operation, and updates and recovers resources if
    // the operation becomes terminal. If `convertResources` is `false`
    // only the consumed resources of terminal operations are recovered,
    // but no resources are converted.
    
I'd prefer a tone along these lines, so we do not just repeat the 
implementation incompletely, but instead reflect intention. The parameter 
`convertResources` e.g., does not just control resources updates in the 
allocator, but also in the master's agent state.


- Benjamin


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/65095/#review195246
-----------------------------------------------------------


On Jan. 12, 2018, 11:53 a.m., Benjamin Bannier wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/65095/
> -----------------------------------------------------------
> 
> (Updated Jan. 12, 2018, 11:53 a.m.)
> 
> 
> Review request for mesos, Gaston Kleiman, Greg Mann, and Jan Schlicht.
> 
> 
> Bugs: MESOS-8422
>     https://issues.apache.org/jira/browse/MESOS-8422
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> In certain situations it can make sense to update the state of an
> operation without also wanting to update resources. In this patch we
> modify the master's `updateOperation` function to take an additional
> parameter signifying whether resources should be updated, or whether
> we only care about updating the operation and tracking of used
> resources.
> 
> We will use this functionality in a subsequent patch to perform more
> contained updates to agent state when processing `UpdateSlaveMessages`
> which contain both resources and operations (and where any terminal
> operations were already applied to the agent's resources).
> 
> 
> Diffs
> -----
> 
>   src/master/master.hpp cdfd06ceb2a8eafa60d0af382b628b165e4ddcb5 
>   src/master/master.cpp 8921964a3ed74538b8c831a2f827d07875c1a5ab 
> 
> 
> Diff: https://reviews.apache.org/r/65095/diff/2/
> 
> 
> Testing
> -------
> 
> `make check`
> 
> 
> Thanks,
> 
> Benjamin Bannier
> 
>

Reply via email to