-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/71068/#review216816
-----------------------------------------------------------


Fix it, then Ship it!




Can you also include the additional caveats of the current approach in the 
commit description for posterity? As was done for the race caveat


src/master/quota_handler.cpp
Line 583 (original), 583-594 (patched)
<https://reviews.apache.org/r/71068/#comment304065>

    Thanks!



src/master/quota_handler.cpp
Lines 585-586 (patched)
<https://reviews.apache.org/r/71068/#comment304063>

    hm.. I think readers will wonder what compromise is being made here with 
tracking incorrect "resource state", can you clarify it here or point them to 
where they can see what this means?
    
    (I assume this is the addition overlap only right?)



src/master/quota_handler.cpp
Lines 590 (patched)
<https://reviews.apache.org/r/71068/#comment304064>

    s/churning/churn/



src/master/quota_handler.cpp
Lines 599-600 (patched)
<https://reviews.apache.org/r/71068/#comment304066>

    Maybe make this a "NOTE: " comment?


- Benjamin Mahler


On July 23, 2019, 8:38 p.m., Meng Zhu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/71068/
> -----------------------------------------------------------
> 
> (Updated July 23, 2019, 8:38 p.m.)
> 
> 
> Review request for mesos and Benjamin Mahler.
> 
> 
> Bugs: MESOS-9812
>     https://issues.apache.org/jira/browse/MESOS-9812
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Outstanding offers of a role (including offers allocated to its
> subroles) are rescinded until the sum of its consumed quota and
> outstanding offered resources are below the requested limits.
> 
> The rescind logic here, however, overlooks the race between master
> and the allocator. Namely, there might be pending offers in the master
> mailbox that have yet to be processed. Here, for simplicity, we ignore
> this race. To properly handle this, the plan is to move offer
> management logic into the allocator and rescind offers there.
> 
> Also added a test.
> 
> 
> Diffs
> -----
> 
>   src/master/quota_handler.cpp 2a3ca5640c850ec8a8aca7abed45386b5b5e1125 
>   src/tests/master_quota_tests.cpp 8105765172f17f3ea50aa09bc66fede8d21365ab 
> 
> 
> Diff: https://reviews.apache.org/r/71068/diff/5/
> 
> 
> Testing
> -------
> 
> Ran continously without failure.
> 
> 
> Thanks,
> 
> Meng Zhu
> 
>

Reply via email to