To start with the last:

> I hope this better addresses your original query :-}

Yes, it does. Thanks.

> After thinking about this for a while, it seems that your preference
> could prove workable for STACK graphs.
> 
> > My preference would be to skip the unknown entirely (as with normal
> > area and/or lines) so that #3 gets stacked on top of #1.
> 
> I would not like to see this become the default behavior for all types
> of aggregated graphs - just the STACK type.  I can think of contexts
> for area and line plots where I do want RRD to stop graphing when
> unknown data points come into play and I like the idea that I have
> control over this behavior.
> 
Agree 100% but I don't see stacked graphs as being aggregated. This
may be the problem in my explanation and your understanding of it.
I have no doubts at all for things like val1+val2+val3 being unknown.
This is correct behaviour and only when you decide it is okay to have
a zero value there, you do so yourself.

> The question I am left with however, is whether it would be possible
> to have some sort of switch that does allow us to avoid painting
> altogether when one of the values is unknown.  I note from the

Correct. I can think of one way. If it is important to know when
data was unknown, I would like to be able to show this. Preferably
using a red line from top to bottom, no matter what those actual values
are. For this we would need the 100% value. It would be drawn over
other values thereby masking the wrong values. I'm not sure but I think
when one would use the background color for this line, it would be the
same as deleting the values. (the grid is put in last, isn't it?)

This could be the resulting line (only missing operator is "full")

   CDEF:switch=val1,UN,val2,UN,val3,UN,+,+,0,GT,full,UNKN

This was a logical OR. If any of val1,val2,val3 is unknown, the result
is 1. Then, the result is used to return either unknown or full (being
the 100% value). If this cdef is drawn, it will only be drawn if there
was an unknown value.
(sidenote: it works, but "AND", "OR" and friends *are* welcome)

Above approach will work for stacked and for non-stacked areas/lines.

> examples we exchanged earlier today that this is not really the
> current behavior - since you always have to start with either an AREA
> or LINE plot for the first data point, you will always have at least
> that one part of the graph assuming that the first data point is not
> *unknown*.  We are currently sort of straddling the fence on this
> item.

So if it is desired behaviour to not graph at all, we would need to
do some calculations anyway. Put in the switch as described above and
this may be all there is to it.

Thanks for the input. Next please ?!?   :-)

Alex

--
* To unsubscribe from the rrd-developers mailing list, send a message with the
  subject: unsubscribe to rrd-developers-request@list.ee.ethz.ch