[rrd-users] Debugging graph smoothing or bad data samples? interpolation bug?

[Tobias Oetiker]

Hi Marc,

Yesterday Marc MERLIN wrote:

> On Mon, Nov 16, 2009 at 11:21:10PM +0100, Tobias Oetiker wrote:
> > > The graph definitions are in the links and the rrd was created like so:
> > > /usr/bin/rrdtool create /var/lib/cacti/rra/housepower_21.rrd --step 120  \
> > > DS:HouseNoAC:DERIVE:600:-1000000000:1000000000 DS:House:DERIVE:600:-1000000000:1000000000 \
> > > DS:PGE:DERIVE:600:-1000000000:1000000000 DS:AC:DERIVE:600:-1000000000:1000000000 \
> > > DS:PV:DERIVE:600:-1000000000:1000000000 \
> > > RRA:AVERAGE:0.5:1:2628000 RRA:AVERAGE:0.5:1440:10950  RRA:AVERAGE:0.5:5:3153600 \
> > > RRA:AVERAGE:0.5:30:525600 RRA:AVERAGE:0.5:120:131400 \
> > > RRA:MAX:0.5:1440:10950 RRA:MAX:0.5:30:525600 RRA:MAX:0.5:120:131400 \
> > >
> > > I'm just trying to understand what could make these data problems possible.
> > > My original data on disk (before it was fed to the rrd) is correct.
> > > Even if there was a time delay in when the data was given, rrdtool knows how to interpolate
> > > data to make things right, so I don't get it.
> > >
> > > Any ideas?
> >
> > you are talking about 'original' data, are you feeding the data to
> > rrdtool INCLUDING the collection time, or are you just using N: the
> > graph looks as if there was some bad case of jitter at work ...
>
> Sorry, I should have said that: the data is fed through cacti which does not
> feed the sample time into rrdtool. So rrdtool is left with just the sample
> and time at which it received it.
> Samples are every 2mn so even if I had a few seconds of jitter, it should
> not have made such a big difference.
>
> > two samples being 'faked' into one slot ...
>
> yes, that's pretty much what I'm seeing.
>
> If cacti/cron/the system were overloaded though, I would have expected to
> have an empty sample and then a double sample to catch up.
> I'm very perplexed here since the graphs clearly show a double sample first
> and then an empty sample, which is the wrong way around.
>
> Does that make any sense to you?

not entirely, I can only say that irregular data arrival times are
rrdtools forte ... have you tried to write the data into a logfile
(including the timestamp) and then recreate the phaenomenon by hand ?

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