[rrd-developers] DS idea

[Alex van den Bogaerdt]

----- Original Message ----- 
From: "Tobias Oetiker" <tobi at oetiker.ch>
To: "Alex van den Bogaerdt" <alex at vandenbogaerdt.nl>
Cc: <rrd-developers at lists.oetiker.ch>
Sent: Monday, October 11, 2010 6:55 AM
Subject: Re: [rrd-developers] DS idea


> Hi Alex,
>
> Today Alex van den Bogaerdt wrote:
>
>> >>   GAUGELATEST
>> >
>> > Whatever it's called, I'd certainly like to see a DS type which is like
>> > GAUGE but without the normalisation to the interval.  There are many
>> > applications for which this would be useful.
>> >
>>
>>
>> I'm probably missing something here.
>>
>> What can you do with that, which cannot be achieved with step==1,
>> steps=={whatever},heartbeat>={whatever} and CF LAST (or, as in this 
>> thread,
>> MAX) ?
>
> the use case for this is that you have a mixed set of datasource
> types in a single RRD file ... some get the normal treatment others
> get the GAUGELAST treatment. This can not be achieved today. You
> have to use multipl RRD files or maybe do some very cool stuff in
> all graph commands.


I just posted another reply. It does not use different data source types, 
only different step and steps.

Indeed, each RRA would use the same step (because that is a database 
property, not an archive property) so you cannot mix old style and new style 
max. But why would one do that?

I can see only one problem, and that has to do with sub-second precision. I 
do not have my mind made up yet, but two ideas are fighting in the back of 
my head:  one idea is that when subsecond precision was introduced, the step 
parameter should also allow for fractions, the other idea is that the whole 
step parameter is a wrong idea to begin with, that "steps" really should 
have been "seconds", and "step" just disappears.