dtrace discuss - Dec 2005 - truncating aggregation output only

Hello dtrace-discuss,

  Sometimes I want to run a script for some time and every n second
  output N top entries. trunc() isn''t suitable here as it also removed
  keys/values. I want it ''coz over time if I use sum() entries which
  are normally truncated can actually get to top over a time.

  Maybe printa() extension, something like: printa(@b[10]) - to output
  top 10?

-- 
Best regards,
 Robert                          mailto:rmilkowski at task.gda.pl

Robert Milkowski wrote:
> Hello dtrace-discuss,
> 
>   Sometimes I want to run a script for some time and every n second
>   output N top entries. trunc() isn''t suitable here as it also
removed
>   keys/values. I want it ''coz over time if I use sum() entries
which
>   are normally truncated can actually get to top over a time.
> 
>   Maybe printa() extension, something like: printa(@b[10]) - to output
>   top 10?
> 
Robert, that''s a really good idea.

I worry about changing the syntax and overloading [] though. Perhaps a 
better idea would be to go the same way that we are doing column 
sorting? Bryan?

alan.
-- 
Alan Hargreaves - http://blogs.sun.com/tpenta
Staff Engineer (Kernel/VOSJEC/Performance)
Product Technical Support (APAC)
Sun Microsystems

On Fri, Dec 23, 2005 at 09:11:54AM +1100, Alan Hargreaves
wrote:
> Robert Milkowski wrote:
> >Hello dtrace-discuss,
> >
> >  Sometimes I want to run a script for some time and every n second
> >  output N top entries. trunc() isn''t suitable here as it also
removed
> >  keys/values. I want it ''coz over time if I use sum() entries
which
> >  are normally truncated can actually get to top over a time.
> >
> >  Maybe printa() extension, something like: printa(@b[10]) - to output
> >  top 10?
> >
> 
> Robert, that''s a really good idea.
> 
> I worry about changing the syntax and overloading [] though. Perhaps a 
> better idea would be to go the same way that we are doing column 
> sorting? Bryan?
Yeah, the way to do this would be via a new option.  Something like 
"aggtop" or something?  So your code would look like this:

	...

	setopt("aggtop", "10");
	printf("The top 10:\n");
	printa(@);

	setopt("aggtop", "false");   /* or something... */
	printf("\nThe whole lot:\n");
	printa(@);

(This would also allow you to, e.g., see the bottom ten by setting both 
"aggsortrev" and "aggtop".)

How does that sit?

	- Bryan

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

}Yeah, the way to do this would be via a new option.  Something like 
}"aggtop" or something?  So your code would look like this:
}
}	...
}
}	setopt("aggtop", "10");
}	printf("The top 10:\n");
}	printa(@);
}
}	setopt("aggtop", "false");   /* or something... */
}	printf("\nThe whole lot:\n");
}	printa(@);
}
}(This would also allow you to, e.g., see the bottom ten by setting both 
}"aggsortrev" and "aggtop".)
}
}How does that sit?

  How about a new function that produces an aggregation with at most <n> 
elements, where <n> is the second argument:

    printa(agghead(@, 10));

  ?

 --S

On Thu, Dec 22, 2005 at 02:27:47PM -0800, Seth Goldberg
wrote:
> 
> }Yeah, the way to do this would be via a new option.  Something like 
> }"aggtop" or something?  So your code would look like this:
> }
> }	...
> }
> }	setopt("aggtop", "10");
> }	printf("The top 10:\n");
> }	printa(@);
> }
> }	setopt("aggtop", "false");   /* or something... */
> }	printf("\nThe whole lot:\n");
> }	printa(@);
> }
> }(This would also allow you to, e.g., see the bottom ten by setting both 
> }"aggsortrev" and "aggtop".)
> }
> }How does that sit?
> 
>   How about a new function that produces an aggregation with at most
<n>
> elements, where <n> is the second argument:
> 
>     printa(agghead(@, 10));
That''s what trunc does, effectively.  (It just does it without creating
a new aggregation.)  Going beyond that would require aggregation assignment
(i.e. "@a = @b"), which opens up a Pandora''s Box that we
would prefer to
keep welded shut.

	- Bryan

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

On 12/23/05, Bryan Cantrill <bmc at eng.sun.com>
wrote:
> Yeah, the way to do this would be via a new option.  Something like
> "aggtop" or something?  So your code would look like this:
>
>         ...
>
>         setopt("aggtop", "10");
>         printf("The top 10:\n");
>         printa(@);
>
>         setopt("aggtop", "false");   /* or something...
*/
>         printf("\nThe whole lot:\n");
>         printa(@);
>
> (This would also allow you to, e.g., see the bottom ten by setting both
> "aggsortrev" and "aggtop".)
>
> How does that sit?
"aggtop" is something that is pretty close to the specific printa and
will likely change for each printa''s.  Some parameter setting within
printa seems more appropriate than a global option.

I''d prefer a syntax like:
  printa("%10a", @)

or even
  printa(10@)

Either way it is implemented, such a feature is certainly useful.


--
Just me,
Wire ...

On Fri, Dec 23, 2005 at 09:53:54AM +0800, Wee Yeh Tan
wrote:
> On 12/23/05, Bryan Cantrill <bmc at eng.sun.com> wrote:
> > Yeah, the way to do this would be via a new option.  Something like
> > "aggtop" or something?  So your code would look like this:
> >
> >         ...
> >
> >         setopt("aggtop", "10");
> >         printf("The top 10:\n");
> >         printa(@);
> >
> >         setopt("aggtop", "false");   /* or
something... */
> >         printf("\nThe whole lot:\n");
> >         printa(@);
> >
> > (This would also allow you to, e.g., see the bottom ten by setting
both
> > "aggsortrev" and "aggtop".)
> >
> > How does that sit?
> 
> "aggtop" is something that is pretty close to the specific printa
and
> will likely change for each printa''s.  Some parameter setting
within
> printa seems more appropriate than a global option.
Then use setopt() to change the option between each printa().  The precedent
here is the aggregation sorting and printing options that I just putback:
aggsortrev, aggsortkey, aggsortpos, aggsortkeypos, etc.
> I''d prefer a syntax like:
>   printa("%10a", @)
That''s actually already valid D.  If we had T-shirts, mugs, or any
DTrace
junk(*), I would offer it to the first person who can reply explaining what
the above does, and using it in a valid script.  (Note: valid, not
useful.)

	- Bryan

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

(*) Trying (and failing) to resist the urge to point out the inverse
    correlation between quantity of such junk and timeliness of putback.

On 12/23/05, Bryan Cantrill <bmc at eng.sun.com>
wrote:
> Then use setopt() to change the option between each printa().  The
precedent
> here is the aggregation sorting and printing options that I just putback:
> aggsortrev, aggsortkey, aggsortpos, aggsortkeypos, etc.
Yeah, the problem here is "aggtop" likely change a lot more often than
the other "agg*" that it might make sense to tie it with printa().
> > I''d prefer a syntax like:
> >   printa("%10a", @)
>
> That''s actually already valid D.  If we had T-shirts, mugs, or any
DTrace
> junk(*), I would offer it to the first person who can reply explaining what
> the above does, and using it in a valid script.  (Note: valid, not
> useful.)
*Smack face into the keyboard* That''s the width specifier.

dtrace -n ''profile-1001{@[arg0]=count()}'' -n
''tick-5s{printa("%40a", @)}''

I had a tingle that something was off when I suggested that.


--
Just me,
Wire ...

On Fri, Dec 23, 2005 at 10:26:50AM +0800, Wee Yeh Tan
wrote:
> On 12/23/05, Bryan Cantrill <bmc at eng.sun.com> wrote:
> > Then use setopt() to change the option between each printa().  The
precedent
> > here is the aggregation sorting and printing options that I just
putback:
> > aggsortrev, aggsortkey, aggsortpos, aggsortkeypos, etc.
> 
> Yeah, the problem here is "aggtop" likely change a lot more often
than
> the other "agg*" that it might make sense to tie it with
printa().
We really don''t want to sully-up printa() with additional options;
that''s
what the option mechanism is for.  (And I''m not even sure that I buy 
that one would want to change "aggtop" more than, say,
"aggsortkey" or
"aggsortrev" or whatever.)  Or put it this way:  we''ll
implement "aggtop"
as an option for now, and if you find you have zillions of scripts littered
with "setopt("aggtop", "whatever")",
we''ll talk about some syntactic
sweetener...
> > > I''d prefer a syntax like:
> > >   printa("%10a", @)
> >
> > That''s actually already valid D.  If we had T-shirts, mugs,
or any DTrace
> > junk(*), I would offer it to the first person who can reply explaining
what
> > the above does, and using it in a valid script.  (Note: valid, not
> > useful.)
> 
> *Smack face into the keyboard* 
Ouch!  Careful... ;)
> That''s the width specifier.
Indeed it is.
> dtrace -n ''profile-1001{@[arg0]=count()}'' -n
''tick-5s{printa("%40a", @)}''
Right again.  You and Seth Goldberg are our two winners.  Mike:  tell
them what they win!  (Oh right, nothing.  Still, if we ever have something,
you''ll win it.  Honest.)

	- Bryan

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

Hello Bryan,

Thursday, December 22, 2005, 11:19:29 PM, you wrote:

BC> On Fri, Dec 23, 2005 at 09:11:54AM +1100, Alan Hargreaves
wrote:
>> Robert Milkowski wrote:
>> >Hello dtrace-discuss,
>> >
>> >  Sometimes I want to run a script for some time and every n second
>> >  output N top entries. trunc() isn''t suitable here as it
also removed
>> >  keys/values. I want it ''coz over time if I use sum()
entries which
>> >  are normally truncated can actually get to top over a time.
>> >
>> >  Maybe printa() extension, something like: printa(@b[10]) - to
output
>> >  top 10?
>> >
>> 
>> Robert, that''s a really good idea.
>> 
>> I worry about changing the syntax and overloading [] though. Perhaps a 
>> better idea would be to go the same way that we are doing column 
>> sorting? Bryan?
BC> Yeah, the way to do this would be via a new option.  Something like 
BC> "aggtop" or something?  So your code would look like this:

BC>         ...

BC>         setopt("aggtop", "10");
BC>         printf("The top 10:\n");
BC>         printa(@);

BC>         setopt("aggtop", "false");   /* or
something... */
BC>         printf("\nThe whole lot:\n");
BC>         printa(@);

BC> (This would also allow you to, e.g., see the bottom ten by setting both 
BC> "aggsortrev" and "aggtop".)

BC> How does that sit?


This is good enough :)
Additional option to printa would be easier/quicker to use but above
example also solves the problem.

I take it I do not have to fill RFE for it :)


-- 
Best regards,
 Robert                            mailto:rmilkowski at task.gda.pl