[rrd-users] Cleaning a messed up rrdtool data file
david at landgren.net
Fri Mar 4 13:09:58 MET 2005
I have an rrdtool database that polls my main router to the net. From
time to time there was an ugly 70Mb spike in the graph, and I didn't pay
too much attention, thinking that it was just some sort of counter
rollover or something. It was never a sufficient annoyance to bother
looking into it more closely.
This week, the line was upgraded from 2Mb to 6Mb and as I was puttering
around with my graphing tools I noticed that I was running two copies of
my polling script that was updating the rrd file. And so it makes sense
to me that that was the source of the spikes. Now I'd like to try and
clean up the historical data.
I've generated two graphs, one clamped rigidly to the maximum expected y
value, and the other with y autoconfigured:
At first I was thinking I could use something like
but then I'll clamp new legitimate values that'll be coming down the
wire. If I clamp to the new higher value I'll have spikes in the old
traffic values. I'd really like to have clean data so that I can point
and say "look, before... and after".
Therefore, is there a way of saying:
"if date before d0, then clamp to lim1 else clamp to lim2"
The second approach I considered would be to go into the rrd file and
iterate over the values, looking for peaks greater than 2Mb, and
replacing them by a linear interpolation between to the two points
either side of the peak?
The maths doesn't pose much of a problem, I'm wondering about the
mechanics of going through the datapoints in the rrd file.
Thanks for any pointers I can use to either approach (or if there's a
Unsubscribe mailto:rrd-users-request at list.ee.ethz.ch?subject=unsubscribe
Help mailto:rrd-users-request at list.ee.ethz.ch?subject=help
More information about the rrd-users