> On Jan. 15, 2016, 6:17 a.m., Till Toenshoff wrote:
> > src/tests/environment.cpp, line 322
> > <https://reviews.apache.org/r/42197/diff/3/?file=1196925#file1196925line322>
> >
> >     This grepping for `cycles` seems dangerous assuming that we need 
> > `cpu-cycles` support (aka `cycles`) when considering that there are also 
> > things like e.g. `ref-cycles` which I believe we are not relying on. 
> >     
> >     Test-output of a box I tried:
> >     ```
> >     $ perf list hw
> >       cpu-cycles OR cycles                               [Hardware event]
> >       instructions                                       [Hardware event]
> >       cache-references                                   [Hardware event]
> >       cache-misses                                       [Hardware event]
> >       branch-instructions OR branches                    [Hardware event]
> >       branch-misses                                      [Hardware event]
> >       bus-cycles                                         [Hardware event]
> >       ref-cycles                                         [Hardware event]
> >     ```
> >     
> >     Test-output of a VM I tried:
> >     ```
> >     $ perf list hw
> >       ref-cycles                                         [Hardware event]
> >     ```
> >     
> >     More on that on 
> > https://issues.apache.org/jira/browse/MESOS-3082?focusedCommentId=15101208&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15101208

You're right! I wasn't aware of other hardware events that have `cycles` in 
their name. Should have had a better look at the documentation. Instead of 
grepping for `cycles`, let's grep for `cpu-cycles` instead, as this is a unique 
identifier in this list. The tests covered by this filter only sample this 
event, currently no other hardware events are sampled by any of the other tests.


- Jan


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/42197/#review114659
-----------------------------------------------------------


On Jan. 14, 2016, 4:28 p.m., Jan Schlicht wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/42197/
> -----------------------------------------------------------
> 
> (Updated Jan. 14, 2016, 4:28 p.m.)
> 
> 
> Review request for mesos, Benjamin Bannier and Till Toenshoff.
> 
> 
> Bugs: MESOS-3082
>     https://issues.apache.org/jira/browse/MESOS-3082
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Virtual machines may not always support 'CPU performance counters'. Tests
> trying to collect to 'cycle' value of 'perf' will fail under these
> circumstances. This will be avoided by disabling these tests if 'perf' 
> hardware
> events are not available.
> 
> 
> Diffs
> -----
> 
>   src/tests/environment.cpp 20218a086baefcefb310eb45ed9024e5425ce787 
> 
> Diff: https://reviews.apache.org/r/42197/diff/
> 
> 
> Testing
> -------
> 
> Running `sudo ./bin/mesos-tests.sh 
> --gtest_filter="*ROOT_CGROUPS_Perf:*ROOT_CGROUPS_Sample:*ROOT_CGROUPS_UserCgroup:*CGROUPS_ROOT_PerfRollForward:*ROOT_Sample"`
> 
> 1) In a Linux VM without 'CPU performance counters': This doesn't run the 
> tests and shows an error message that 'cycles' is not available, as expected
> 2) In a Linux VM without 'perf': This doesn't run the tests and shows an 
> error message that 'perf' is not installed, as expected
> 3) On OSX: This doesn't run the tests and shows an error message that 'perf' 
> is not available, as expected
> 4) NOT TESTED YET!!! On native Linux (i.e. with 'CPU performance counters'): 
> This runs the tests.
> 
> 
> Thanks,
> 
> Jan Schlicht
> 
>

Reply via email to