[rrd-users] Lost Data??

Steve Shipway s.shipway at auckland.ac.nz
Thu Oct 9 00:49:13 CEST 2008

> This task works fine when i am searching values in the last hour week
> month, but if i search e.g. the hour between 15:00 and 16:00 one month ago
> it give me back only three values where in my opinion has to be 12 times
> more!

My guess is that your small-granularity RRA only stretches less than 1 month; for older data, rrdtool is forced to rely on a larger granularity RRA, in this case about 30min intervals, and so you only get 3 samples.

Check the RRD file configuration and in particular the definition of the small-granularity RRA.  Multiply the number of rows by the rollup by the interval; this is as far back as you can go using this RRA.

/u01/bin/rrdtool info /u01/mrtg/data/test.rrd
step = 300
rra[0].cf = "AVERAGE"
rra[0].rows = 800
rra[0].pdp_per_row = 1
rra[0].xff = 5.0000000000e-01

Therefore, for my first RRA, I can go back (step)*(pdp_per_row)*(rows) seconds, since the step is the time interval for one pdp (5min in this case, as this is a default MRTG-generated RRD file)
  Step x pdp_per_row x rows
    = 300s x 1 x 800
    = 66 hours
    = about 2.5 days

Any further, and it will have to use rra[1], which is the same apart from a pdp_per_row of 6 (IE, it rolls up date to a 30-min average).  This is why you can make a MRTG system show a 'daily' and a '2 days' graph, but not a 'daily' graph from a week ago.


More information about the rrd-users mailing list