Re: Something fishy in process timing

[Peter Stephenson <pws@xxxxxx>]

cedman@xxxxxxxxxxxxxxx wrote:
> Have a look at this paste from my xterm:

I'm looking.  (I presume that's a fish paste :-):-).)

> /tmp% xjig -file /tmp/a.gif -ts 50 &|                              7:03pm
> /tmp%                                                              7:04pm
> xjig V2.4, by Helmut Hoenig, July-24-96
> 
>   Imported from TIFF image: 09320028.tif
> X connection to :0.0 broken (explicit kill or server shutdown).
> 
> /tmp% ~/ftp                                                        8:11pm
> ~/ftp% ls                                                          8:12pm
> [ ... ]
> ls  133.68s user 37.81s system 100% cpu 0.312 total
> 
> What it looks like is that the user and system timings for the disowned
> xjig process got added to the reported timings for the ls.  That doesn't
> seem the right thing to do.

That is what happened and it isn't right.  This patch seems to fix it.
The bug is that zsh handles SIGCHLD for disowned processes
(unavoidably), but doesn't bother calculating the usage, so it doesn't
get updated into the total shell times in shtms until the next
non-disowned process is reaped.  The patch simply updates shtms even
if the process wasn't found in the table.  This is the last hunk.

The second hunk below simply does the same thing when a process
substitution process is finished.  Otherwise, it's impossible to tell
what process the times for that will tack themselves onto.

> PS:  If this really were a parallel machine in which user+system time
> legitimately exceeds wall clock time, wouldn't the right thing be to
> report more than 100% CPU usage ? 

I agree: nowadays there's less excuse than there used to be to massage
the CPU percentage --- and in fact I've relied on the accuracy of this
myself when running programmes on parallel machines.  The first hunk
below removes the limitation.  (Testing for a real parallel machine
would probably be taking things a bit far.)  I wrapped a nearby line
round as well.

By the way, my time reports look like this:

./usetime 2 1000000: 1.37s(r) 1.35s(u) 0.02s(s) (99%): %K kb, max %M kb, %F pf

I still haven't worked out what happened to %K, %M and %F.


*** Src/jobs.c.time	Mon Jul 22 22:32:14 1996
--- Src/jobs.c	Wed Jul 31 13:03:39 1996
***************
*** 725,733 ****
      elapsed_time = real->tv_sec + real->tv_usec / 1000000.0;
      user_time    = ti->ut / (double) clktck;
      system_time  = ti->st / (double) clktck;
!     percent      =  100.0 * (ti->ut + ti->st) / (clktck * real->tv_sec + clktck * real->tv_usec / 1000000.0);
!     if (percent > 100)
!         percent = 100;   /* just to make it look right */
  
      if (!(s = getsparam("TIMEFMT")))
  	s = DEFAULT_TIMEFMT;
--- 725,732 ----
      elapsed_time = real->tv_sec + real->tv_usec / 1000000.0;
      user_time    = ti->ut / (double) clktck;
      system_time  = ti->st / (double) clktck;
!     percent      =  100.0 * (ti->ut + ti->st)
! 	/ (clktck * real->tv_sec + clktck * real->tv_usec / 1000000.0);
  
      if (!(s = getsparam("TIMEFMT")))
  	s = DEFAULT_TIMEFMT;
*** Src/signals.c.time	Wed Jul 31 13:10:59 1996
--- Src/signals.c	Wed Jul 31 13:31:13 1996
***************
*** 479,484 ****
--- 479,485 ----
  			    dotrap(WTERMSIG(status));
  		    } else
  			*procsubval = WEXITSTATUS(status);
+ 		    times(&shtms);
  		    goto cont;
  		}
  		if (!es)
***************
*** 500,505 ****
--- 501,512 ----
  	    if (findproc(pid, &jn, &pn)) {
  		update_process(pn, status);
  		update_job(jn);
+ 	    } else {
+ 		/* If not found, update the shell record of time spent by
+ 		 * children in sub processes anyway:  otherwise, this
+ 		 * will get added on to the next found process that terminates.
+ 		 */
+ 		times(&shtms);
  	    }
          }
          break;


-- 
Peter Stephenson <pws@xxxxxx>       Tel: +49 33762 77366
WWW:  http://www.ifh.de/~pws/       Fax: +49 33762 77330
Deutches Electronen-Synchrotron --- Institut fuer Hochenergiephysik Zeuthen
DESY-IfH, 15735 Zeuthen, Germany.