mesos-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Chi Zhang" <chzhc...@gmail.com>
Subject Re: Review Request 31276: Added cgroup memory pressure listening tests.
Date Fri, 27 Mar 2015 03:20:48 GMT


> On March 25, 2015, 9:24 p.m., Jie Yu wrote:
> > src/tests/cgroups_tests.cpp, lines 1206-1212
> > <https://reviews.apache.org/r/31276/diff/6/?file=904734#file904734line1206>
> >
> >     I don't quite get this part :(

use the guard to break out of the loop. as discussed.


> On March 25, 2015, 9:24 p.m., Jie Yu wrote:
> > src/tests/memory_test_helper.cpp, line 160
> > <https://reviews.apache.org/r/31276/diff/6/?file=904736#file904736line160>
> >
> >     Move this down after validation is done.
> >     
> >     Also, s/memory/buffer/
> 
> Chi Zhang wrote:
>     that won't change the function frame size though? Is it the size you concern, or
the style? If it's the latter, I'd move both lines down after validation?

never mind. not needed any more after writeToFile is killed.


> On March 25, 2015, 9:24 p.m., Jie Yu wrote:
> > src/tests/cgroups_tests.cpp, lines 1117-1160
> > <https://reviews.apache.org/r/31276/diff/6/?file=904734#file904734line1117>
> >
> >     This is a little unreliable and non-deterministic. Any reason why you want to
call increaseRSS multiple times?
> >     
> >     Can you increase the RSS once to a size right below 'limit' and keep reading
the counter until it shows non-zero (and set a time limit for that).

I printed out the stats when the test has pressure readings for the first time and did multiple
runs. The pressure readings show up when the rss gets high (around 90%-95%, however, the data
doesn't suggest a fixed-value threshold), which indicates the kernel starts to scan memory
to attempt to reclaim. 

But since the test only allocates unit size (a page) at a time, waits and checks the pressure
each time, we won't oom ourselves.


- Chi


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/31276/#review77779
-----------------------------------------------------------


On March 27, 2015, 3:20 a.m., Chi Zhang wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/31276/
> -----------------------------------------------------------
> 
> (Updated March 27, 2015, 3:20 a.m.)
> 
> 
> Review request for mesos, Dominic Hamon, Ian Downes, and Jie Yu.
> 
> 
> Bugs: mesos-2136
>     https://issues.apache.org/jira/browse/mesos-2136
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Added cgroup memory pressure listening tests.
> 
> 
> Diffs
> -----
> 
>   src/Makefile.am 7a06c7028eca8164b1f5fdea6a7ecd37ee6826bb 
>   src/tests/cgroups_tests.cpp 75c61aad80f894acb92a9752e8d1b6af70e5b9a6 
>   src/tests/memory_test_helper.hpp PRE-CREATION 
>   src/tests/memory_test_helper.cpp PRE-CREATION 
>   src/tests/memory_test_helper_main.cpp PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/31276/diff/
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Chi Zhang
> 
>


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