> On April 2, 2015, 5:24 p.m., Ian Downes wrote:
> > src/linux/perf.cpp, line 188
> > <https://reviews.apache.org/r/32699/diff/1/?file=911392#file911392line188>
> >
> >     I thought about this again, and it won't be sufficient for perf: we use 
> > perf with a child sleep process to control the sample duration, e.g., "perf 
> > stat -a --event cycles -- sleep 123", and killing perf does not kill the 
> > sleep process. I tested this in a shell so it may not be identical 
> > behavior, but DEATHSIG is not preserved across a fork so it likely is the 
> > same.
> >     
> >     {noformat}
> >     [idownes@hostname ~]$ sudo perf stat -a --event cycles -- sleep 123
> >     ... different shell ...
> >     [idownes@hostname ~]$ pgrep sleep | xargs ps
> >       PID TTY      STAT   TIME COMMAND
> >     57461 pts/0    S+     0:00 sleep 123
> >     [idownes@hostname ~]$ sudo pkill perf
> >     [idownes@hostname ~]$ pgrep sleep | xargs ps
> >       PID TTY      STAT   TIME COMMAND
> >     57461 pts/0    S      0:00 sleep 123
> >     {noformat}
> >     
> >     This does not apply to the `du` usecase which is a single process.
> >     
> >     What *may* work for perf, but is not clearly documented in the prctl 
> > manpage, is to run the processes in a process group and use -SIGKILL as the 
> > DEATHSIG. Depending on how this was implemented, this might signal the 
> > entire process group?

I checked the kernel code. This won't work. ::kill(-pid, SIGKILL) is the one 
that kill a process group (not ::kill(pid, -SIGKILL) :))


- Jie


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/32699/#review78678
-----------------------------------------------------------


On March 31, 2015, 8:14 p.m., Jie Yu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/32699/
> -----------------------------------------------------------
> 
> (Updated March 31, 2015, 8:14 p.m.)
> 
> 
> Review request for mesos, Ben Mahler, Ian Downes, and Vinod Kone.
> 
> 
> Bugs: MESOS-2462
>     https://issues.apache.org/jira/browse/MESOS-2462
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Set death signal for the perf subprocess.
> 
> 
> Diffs
> -----
> 
>   src/linux/perf.cpp 863aa4a972289a59f57e93cd06ba2bf9df949fe2 
> 
> Diff: https://reviews.apache.org/r/32699/diff/
> 
> 
> Testing
> -------
> 
> sudo make check
> 
> 
> Thanks,
> 
> Jie Yu
> 
>

Reply via email to