Re: [PATCH] Add rusage reporting to procstat

On Monday, January 14, 2013 5:44:42 pm Konstantin Belousov wrote:
> On Mon, Jan 14, 2013 at 04:39:17PM -0500, John Baldwin wrote:
> > This patch adds a new -r flag to dump the resource usage information (what 
you 
> > would get from getrusage() or wait()) for a given process.  Sample output:
> > 
> > % procstat -r $$
> >   PID COMM             TYPE                              VALUE        
> >  1428 tcsh             user time                     00:00:00.050182  
> >  1428 tcsh             system time                   00:00:00.040145  
> >  1428 tcsh             maximum RSS                              3328 B
> >  1428 tcsh             integral shared memory                   2844 B
> >  1428 tcsh             integral unshared data                   6372 B
> >  1428 tcsh             integral unshared stack                  1152 B
> >  1428 tcsh             page reclaims                            1306  
> >  1428 tcsh             page faults                                12  
> >  1428 tcsh             swaps                                       0  
> >  1428 tcsh             block reads                                50  
> >  1428 tcsh             block writes                                0  
> >  1428 tcsh             messages sent                             172  
> >  1428 tcsh             messages received                           0  
> >  1428 tcsh             signals received                           33  
> >  1428 tcsh             voluntary context switches               1167  
> >  1428 tcsh             involuntary context switches                1  
> > 
> > http://www.FreeBSD.org/~jhb/patches/procstat_rusage.patch
> > 
> > Any thoughts, etc.?
> 
> It looks fine, but use of the human-oriented resource values, together
> with spaces in names, makes the parsing of the output unfeasible.

Yes, but that is par for the course with procstat.  If we wanted to add a
global flag to request machine-readable output and update the various procstat
backends to honor it, I think that would be a great project.

> The patch only reports the process-cumulative rusage, and not the per-thread
> rusage, it seems.

Yes, this is true.  It operates on the process-wide kinfo that procstat has
handy.  It would not be too hard to add a thread mode (-H flag perhaps?) to
have it request the kinfo_proc structures for all threads and then output
those separately.

-- 
John Baldwin