[rrd-developers] Modify step size of existing RRD

[Peter Stamfest]

Am 2014-03-08 11:34, schrieb Peter Stamfest:
> Dear List!
>
> I have just pushed a new version of my rrd modify work to
> https://github.com/stamfest/rrdtool-1.x in branch rrdmodify-master.
Which brings me to a very important point I am pondering over for some
time now: should I integrate the experimental modify command into
rrdtune? This would especially mean that all the language bindings that
use the argv[] based interface would work out of the box... This only
really has to be decided before releasing a new version, but the earlier
the better. Tobi?

peter

> This is the first "public" version to have support for a reduction of
> the basic RRD step size. Note that there are restrictions about the
> possible new step size: The current step size must be a whole-number
> multiple of the new step size. The modification works by recording the
> new step size and adjusting the pdp count of all existing RRAs. This
> also means, that reducing the step size will most likely only make sense
> if you add a new RRA with a pdp count of 1 so you can take advantage of
> the then higher resolution.
>
> Example:
>
> Given in.rra with step size 300 and initial RRA definitions of
> RRA:AVERAGE:0.5:1:100 and RRA:MAX:0.5:6:1000 the command
>
> rrdtool modify -s 60 in.rrd out.rrd RRA:AVERAGE:0.5:1:250
>
> should result in out.rrd having 3 RRAs with definitions of
> RRA:AVERAGE:0.5:1:250 (just added) and  RRA:AVERAGE:0.5:5:100 and
> RRA:MAX:0.5:30:1000. In addition, the new RRA will have been populated
> using data from the pre-existing AVERAGE RRA.
>
>
> It is not (yet?) possible to increase the step size, mostly because the
> use case is not entirely clear and additional requirements regarding the
> old and new step sizes have to be met.
>
>
> Tobi: This should be pull-able... although the travis integration still
> does not work as it should.
>
> All others: Feedback would be extremely welcome both on bugs and the
> proposed commandline syntax. What other structural modifications would
> anybody be interested in? Also, somebody might have a look at supporting
> the more esoteric CFs within rrd_modify.c
>
>
>
> Peter
>
>