mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alexander Rukletsov" <ruklet...@gmail.com>
Subject Re: Review Request 41769: Made allocator traverse all roles for quota allocation.
Date Tue, 19 Jan 2016 14:36:59 GMT


> On Jan. 19, 2016, 12:42 p.m., Klaus Ma wrote:
> > src/tests/hierarchical_allocator_tests.cpp, line 1875
> > <https://reviews.apache.org/r/41769/diff/3/?file=1199267#file1199267line1875>
> >
> >     Here's one question after reviewing this test: is there any priority between
Quota? e.g. if the first Quota did not statisfied, should we continue to offer resources to
other Quota? Should we lay aside resources for the first one? The first one of Quota is considered
to be high priority?
> 
> Guangya Liu wrote:
>     I think this is the current quota behavior. If the first quota did not satisfied,
allocator will not allocate resources for another quota based on priority with quota role
sorter.

I think your question is: "If there are not enough resources, shall we ensure we deprive quota'ed
roles evenly?" In other words, "shall we avoid cases when one quota is fully satisfied, while
the other one is starving?". My answer would be "yes", but it's not a blocker. A situation
when there are not enough resources in the cluster to satisfy all quotas should be considered
exceptional, hence being less fair is not a big deal, I would say.

Definitely let's not fix it here. Feel free to create a separate ticket though.


- Alexander


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/41769/#review115149
-----------------------------------------------------------


On Jan. 19, 2016, 12:38 p.m., Guangya Liu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/41769/
> -----------------------------------------------------------
> 
> (Updated Jan. 19, 2016, 12:38 p.m.)
> 
> 
> Review request for mesos, Alexander Rukletsov, Joris Van Remoortere, Klaus Ma, and Neil
Conway.
> 
> 
> Bugs: MESOS-4411
>     https://issues.apache.org/jira/browse/MESOS-4411
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> This patch include two parts:
> 1) If there are some non-active roles in front of active roles after quotaRoleSorter,
when the allocator encounter a non-active role, the allocator should not break but continue
to allocate Quota for other active roles to make sure other roles can get its quotaed resources.
> 2) If some role's quota reach its guaranteed value, the allocator should handle another
role but not break. Take the following case: role1 has quota 5 and got 5, role2 has quota
100 and got 50, the role1 will be put in front of role2 by the quotaRoleSorter, if allocator
break when found role1 is satisfied, then role2 will never get its quotaed resources.
> 
> 
> Diffs
> -----
> 
>   src/master/allocator/mesos/hierarchical.cpp 48acde69b1a2f305b568a7e322a58708063dd30a

>   src/tests/hierarchical_allocator_tests.cpp 9362dd306497ba01e0f387c3862456cdcac6f863

> 
> Diff: https://reviews.apache.org/r/41769/diff/
> 
> 
> Testing
> -------
> 
> make
> make check
> GLOG_v=2  ./bin/mesos-tests.sh  --gtest_filter="HierarchicalAllocatorTest.*" --verbose
--gtest_repeat=100 --gtest_shuffle
> 
> 
> Thanks,
> 
> Guangya Liu
> 
>


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