mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Benno Evers <bev...@mesosphere.com>
Subject Re: Review Request 68131: Added MasterActorResponsiveness_BENCHMARK_Test.
Date Mon, 19 Nov 2018 12:22:36 GMT


> On Nov. 6, 2018, 12:19 p.m., Benno Evers wrote:
> > src/tests/master_benchmarks.cpp
> > Lines 616 (patched)
> > <https://reviews.apache.org/r/68131/diff/3-4/?file=2070789#file2070789line620>
> >
> >     Last time we tried running this benchmark, we discovered a dead-lock caused
by the interaction between the use of libprocess worker threads in both our test code and
in the mesos code. 
> >     
> >     Ideally we wouldn't use libprocess at all here, but to avoid another big change
late in the review cycle, adding a check like this should be sufficient:
> >     ```
> >     if (numClients >= LIBPROCESS_NUM_WORKER_THREADS - 3) {
> >       cerr << "Not enough worker threads for the desired number of clients.";
> >       exit(1);
> >     }
> >     ```
> 
> Alexander Rukletsov wrote:
>     Though I'm in favour of the idea, I'm not sure we can reliably get `LIBPROCESS_NUM_WORKER_THREADS`
in tests. Do you think we should try find some substitute or punt on this and leave a comment?

I think it's fine to punt for now, assuming that the comment WILL BE VERY NOTICABLE. I've
created https://issues.apache.org/jira/browse/MESOS-9400 for this purpose.


> On Nov. 6, 2018, 12:19 p.m., Benno Evers wrote:
> > src/tests/master_benchmarks.cpp
> > Line 646 (original), 702 (patched)
> > <https://reviews.apache.org/r/68131/diff/3-4/?file=2070789#file2070789line707>
> >
> >     Given that we're capturing and printing baseline statistics anyways, would it
make sense to print the relative slowdown between baseline performance and performance under
load here, to get a metric that is comparable across Mesos versions?
> 
> Alexander Rukletsov wrote:
>     I'm not sure what a formual for a standard relative slowdown would be : ). Do you
have a suggestion? I'm inclining to leave it as is.

I'm itching to sit down with a statistics textbook to figure this out, but for efficiency's
sake let's leave it as is for now :)


- Benno


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/68131/#review210347
-----------------------------------------------------------


On Nov. 18, 2018, 8:10 p.m., Alexander Rukletsov wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/68131/
> -----------------------------------------------------------
> 
> (Updated Nov. 18, 2018, 8:10 p.m.)
> 
> 
> Review request for mesos, Benno Evers and Benjamin Mahler.
> 
> 
> Bugs: MESOS-8975
>     https://issues.apache.org/jira/browse/MESOS-8975
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> See summary.
> 
> 
> Diffs
> -----
> 
>   src/tests/master_benchmarks.cpp fbfffb69930c30b038f74e0b831fc0ae41c820f0 
> 
> 
> Diff: https://reviews.apache.org/r/68131/diff/5/
> 
> 
> Testing
> -------
> 
> See https://reviews.apache.org/r/68132/
> 
> 
> Thanks,
> 
> Alexander Rukletsov
> 
>


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