[rrd-developers] problems graphing past end of RRD

[Thorsten von Eicken]

Alex van den Bogaerdt wrote:
> On Mon, Feb 25, 2008 at 07:05:06AM -0800, Thorsten von Eicken wrote:
>> Define "more data". Getting 599 data points (last data point missing) is 
>> more data in my book than getting 22 data points that happen to cover 
>> the last 20 seconds.
> 
> You aren't presented points. You are presented time and value,
> combined.

Of course these are data points! A data point can cover a time interval. 
That doesn't make it not a data point.

> You are asking for data that is not yet there. You get as little of
> it as possible. And if this means selecting the RRA at 4520 seconds
> per bucket, then so be it.

Suppose you're a sysadmin or a network admin. You're looking at the last 
day graph of a device. Data for an hour because of a problem. What do 
you expect to see? A graph that ends with the last data point (i.e. an 
hour ago) leading you to believe that everything is fine unless you 
carefully look at the X axis? Or a graph that ends "now" and shows the 
data gap at the end of the graph? I have not seen any serious network 
monitoring tool that shows the former, and I would refuse to use one, 
actually.

> It's clear that I think the decision to alter best-match is wrong.
> In my experience, when something is fixed in the wrong place this
> will eventually lead to a long series of related problems.

I've been seeing the resolution issues I described for many years. But 
in the past I used RRAs whose steps were closer to multiples of 
one-another causing the low-res graphs to clear as soon as I'd refresh. 
Also, I never saw something as dramatic as a 600 pixel graph with 22 
stairsteps. So I always shrugged my shoulders and ignored the problem 
that was clearly there. It's only now that I could fairly easily grab a 
reproducible case and track the issue down in the source code.

Thorsten