-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/44379/#review129245
-----------------------------------------------------------




src/linux/perf.cpp (line 346)
<https://reviews.apache.org/r/44379/#comment192695>

    Hi, @fan.du. I believe @bmahler would like to elaborate more details here. 
Just my idea, feel free to correct it because I am not a native speaker.
    
    ```
        // NOTE: perf statistics csv output format for cgroups could possibly 
be one
        // of the following formats:
        //   * value,event,cgroup (since 2.6.38)
        //   * value,unit,event,cgroup (since Linux kernel 3.14)
        //   * value,unit,event,cgroup,running,ratio (since Linux kernel 4.1)
        // Howerver, we could not match them exactly by check Linux kernel 
version
        // because Linux distributions backport the related patches without 
change
        // kernel version number. So we match the `perf stat` output by token 
size
        // here.
        if (tokens.size() == 3) {
          return Sample({tokens[0], internal::normalize(tokens[1]), tokens[2]});
        } else if ((tokens.size() == 4) || (tokens.size() == 6)) {
          return Sample({tokens[0], internal::normalize(tokens[2]), tokens[3]});
        }
    ```


- haosdent huang


On April 11, 2016, 7:01 a.m., fan du wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/44379/
> -----------------------------------------------------------
> 
> (Updated April 11, 2016, 7:01 a.m.)
> 
> 
> Review request for mesos and Ben Mahler.
> 
> 
> Bugs: MESOS-4705
>     https://issues.apache.org/jira/browse/MESOS-4705
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Use tokens size to parse perf stat format.
> 
> Co-authored with haosdent.
> 
> 
> Diffs
> -----
> 
>   src/linux/perf.cpp 749e676aaf2ce639dd976f2b23e323300c6114c5 
> 
> Diff: https://reviews.apache.org/r/44379/diff/
> 
> 
> Testing
> -------
> 
> 1. {Found and Test} with Serenity, ema filter could get perf event statistics 
> correctly as expected.
> 2. ./bin/mesos-tests.sh --gtest_filter=PerfEventIsolatorTest* 
> --log_dir=/tmp/mesos/
> 
> 
> Thanks,
> 
> fan du
> 
>

Reply via email to