[rrd-developers] Discuss regarding Ticket 263

[Tobias Oetiker]

Hi Alex,

Today Alex van den Bogaerdt wrote:

> ----- Original Message -----
> From: "Jean-Edouard Babin" <radius at gmail.com>
> To: "Alex van den Bogaerdt" <alex at vandenbogaerdt.nl>
> Cc: <rrd-developers at lists.oetiker.ch>
> Sent: Monday, May 24, 2010 7:19 PM
> Subject: Re: [rrd-developers] Discuss regarding Ticket 263
>
>
>
> > Yes, conditional (G)PRINT would be better, but rrdtool set a "false"
> > default time to the VDEF result anyway.
>
> > > Using 0 was a design choice.
>
> This is not a false timestamp, it is overloaded so to speak, 0 means false,
> unset.
>
> > > I don't think time 0 will ever be valid, so instead of adding something
> > > to
> > > this struct, why not simply if (vdef->when==0) printf("none"); else
> > > {...whatever is done now...}
>
> > Imagine someone doing statistic and importing data from 1970 to 2010,
> > with one value per month and the fewest one on 1st Jan 1970.
>
> I did back then, read on
>
> > If he wants to print the first value and use a VDEF for that, he would
> > not get it's result even if in his case 0 is the correct value.
> > From my point of view adding 'never' flag is better as it can't break
> > something
>
> Does RRDtool actually work properly when working with timestamps around
> 1970?
>
> When I programmed the (basics of) VDEF, it didn't.  If memory serves me
> well, it has been a while, so maybe the next changed, 1980-01-01 or so was
> the earliest date one could use (or reliably use).
>
> What timestamp results from local time 1970-01-01 00:00 ?  It will result in
> 0 for the UK, Iceland, and other places where during the winter there's no
> time difference with UTC. In a large part of Europe it will be 3600. In
> timezones west of Greenwich (UK) it will have to be a negative value.  Does
> that actually work? West of Greenwich includes the Americas, not really a
> small part of the world.
>
> Bottom line:
> If things have changed, perhaps 0 is no longer a suitable false value. But
> chances are it is as valid as it has been, so then use it.

I agree that 0 could serve as a value for undefined since rrdtool
curently does not let you use dates before 1980 when creating an
rrd (this is an arbitrary choice in the code and could be removed
with a very short patch). In 1.5 we shuld be able todo this at
least for platforms using 64bit dates.

In general I think it is a sub optimal solution to use inband
signaling ... here we can get proper out of band signaling without
any disadvantage ... so I welcome this.

cheers
tobi

-- 
Tobi Oetiker, OETIKER+PARTNER AG, Aarweg 15 CH-4600 Olten, Switzerland
http://it.oetiker.ch tobi at oetiker.ch ++41 62 775 9902 / sb: -9900