dtrace discuss - Aug 2006 - Printing aggregations, multiple values per line?

Is it possible, when printing aggregations, to get multiple values per 
line, or somehow link different aggregations that use the same keys to 
extract the values?

An example might explain better.  I''m using DTrace to instrument a 
function in an application.  When a particular probe is hit I start 
counting all the system calls that the app makes, the total time for 
these system calls, and the average time, in three aggregations.

I stop doing this when a second probe is hit.

The D looks something like this (and is, I think, fairly normal for this 
sort of thing).

sendmail*:::queue-createentry-start
{ self->trace = 1;  }

sendmail*:::queue-createentry-end
/self->trace/
{ self->trace = 0;  }

syscall:::entry
/self->trace/
{
   self->start = timestamp;
}

syscall:::return
/self->trace/
{
   this->time               = timestamp - self->start;
   @func_count[probefunc]   = count();
   @func_time[probefunc]    = sum(this->time);
   @func_avgtime[probefunc] = avg(this->time);
}

END
{
   printf("%20s\t%20s\n", "SYSCALL", "COUNT");
   printa("%20s\t%@20d\n", @func_count);
   printf("\n");
   printf("%20s\t%20s\n", "SYSCALL", "ELAPSED
(ns)");
   printa("%20s\t%@20d\n", @func_time);
   printf("\n");
   printf("%20s\t%20s\n", "SYSCALL", "AVERAGE
(ns)");
   printa("%20s\t%@20d\n", @func_avgtime);
}

This gives me some reasonably nicely formatted tables.

   SYSCALL               COUNT
   brk                     120
   close                 30000
   ...

   SYSCALL        ELAPSED (ns)
   brk                  868871
   close              62427883
   ...

   SYSCALL        AVERAGE (ns)
   fxstat                 1040
   fcntl                  1498
   ...

That''s great if I want to review the data on screen, not so great if I 
want to import it into a spreadsheet, or do any other automated munging 
with it -- I need to construct an awk or Perl pipeline to munge the 
format in to something a little more regular.  That''s not a horrendous 
task, but it''s a minor annoyance.

I''d like to be able to produce CSV, or thereabouts.  So it would look 
something like this:

   SYSCALL,COUNT,ELAPSED,AVERAGE
   brk,120,868871,7240
   close,30000,137680912,4589
   ...

Since the three aggregations all have the same key (the syscall name) 
I''m hoping there''s some way to ''link'' them
to produce this sort of output.

N

Hey Nik,

On Sun, Aug 06, 2006 at 06:32:06PM +0100, Nik Clayton
wrote:
> Is it possible, when printing aggregations, to get multiple values per 
> line, or somehow link different aggregations that use the same keys to 
> extract the values?
Yes, it is -- but the work was done after we wrote the doc, I''m afraid,
so the documentation is a little...informal.  In particular, details are
here:

  http://mail.opensolaris.org/pipermail/dtrace-discuss/2005-November/000602.html

Sorry to have that be in such an odd-ball place.  Maybe someone could
update the FAQ when they get the chance?

  http://www.genunix.org/wiki/index.php/DTrace_FAQ

	- Bryan

--------------------------------------------------------------------------
Bryan Cantrill, Solaris Kernel Development.       http://blogs.sun.com/bmc

Bryan,

Bryan Cantrill wrote:
> On Sun, Aug 06, 2006 at 06:32:06PM +0100, Nik Clayton wrote:
>> Is it possible, when printing aggregations, to get multiple values per 
>> line, or somehow link different aggregations that use the same keys to 
>> extract the values?
> 
> Yes, it is -- but the work was done after we wrote the doc, I''m
afraid,
> so the documentation is a little...informal.  In particular, details are
> here:
> 
>  
http://mail.opensolaris.org/pipermail/dtrace-discuss/2005-November/000602.html
> 
> Sorry to have that be in such an odd-ball place.
Ah, perfect.  In particular, the bit that I was missing was that 
printa() could take multiple aggregations as arguments.

Is the DTrace Guide (http://docs.sun.com/app/docs/doc/817-6223) in a 
repository anywhere public?  It strikes me as the document to which 
patches should be applied.

N