[HOWTO] RESIZE existing RRAs of existing RRDs

[gandalf]

Please find current version at http://docs.cacti.net


Preface
This tutorial is based on rrdtool, written by Tobias Oetiker. It is written by Reinhard Scheck to help you figuring out, how to resize an existing RRA of a given rrd file. This tutorial is based on Howto RESIZE existing RRAs of existing RRDs and extends this to some kind. Hope, this one will work even for windows users.

Be warned!
BACKUP ALL YOUR RRDs!
There's a good chance, that you will destroy all of your rrd files. I'm not joking! At the time of writing, rrdtool 1.2.12 is stable. Pay attention to older rrdtool-1.2.x version as they contain a bug when resizing rrd files created by rrdtool-1.0.x (see above reference for more).
SO BACKUP ALL YOUR RRDs! And check for sufficient file space!
As always: Use this information at your own risk.

Here we go!
As an attachment to the forum entry you will find a perl script resize.pl. It is necessary, to customize the /path/to/the/rrd/binary, e.g /usr/bin/rrdtool.

Help!
Put resize.pl wherever you want. There's no need to put it into the rrd working directory. But you will need some scratch space here for all rrds to be resized (due to the way rrdtool resize works). The user that runs this script must have

• write permissions to the current directory used for scratch
• read permissions on the original rrds to be resized
• write permissions to the target directory to store the resized rrds in

The script does not care about space provided. To get help, simply type

[evilzardoz]

Once I have resized my RRAs, how do I tell cacti not to consolidate?

Also, I can't seem to find RRDp.pm anywhere.

[gandalf]

[evilzardoz]

lvm - thanks for the help so far. We want to keep more detailed information for historical reasons which is why consolidation, while a good idea, isn't quite what we're looking for.

I've found some extra packages for RRD-perl stuff and they look like what I might need so I'll give those a shot.

[gandalf]

Of course you can "turn off consolidation" by only defining a single rra, consisting of LOTS of data points. That would be HUGE,of course. But for some very few data templates, I did this. Took me SOME discussions with those interpreting the results; this is a bit "funny". Try this, and you'll knwo what I mean. This effect will surely pertain to all other solutions (what do you think comes out, when you try to graph some 1000 data points when there are only about 500 pixels on the screen ...). Perhaps this would be a solution for intervals of 1h or so ...
Reinhard

[evilzardoz]

I still want the daily, monthly, yearly graphs.. but I don't want to loose granularity over time i.e. if I want to see what our network looked like three years ago.

Is this a bad idea?

[gandalf]

[gandalf]

Please define 4 new RRAs as shown (Xff is always 0.5) in picture 1. In your Data Templates, you will reference them as shown in picture 2 when creating the Data Source.
Afterwards, when the Data Sources are created (not earlier) you may change this to settings like picture 3. Now, when graphing, the commonly known 4 timespan graphs are shown.
If you add all 4 RRAs when creating the data sources, you'll get an rrd file with 4 rra's while you only want to have one. So follow the instructions carefully (perhaps a little code patch would supress generating weird rra definitions like those I mentioned).
In fact, I'd prefer to seperate the definition of graphing timespans from rra definitions. In "reality", the timespans to be defined for cacti's rra are not needed for rrdtool.
Reinhard

[time]

I only just noticed today, but some of my data sources maintain 5 minute resolution without me having done anything. It is seemingly random in that I have used the same graph and data template multiple times and in some instances it keeps all the data while for other devices it hasn't. See a couple of graphs as an example, these are taken today (28/02/2006) by zooming in on my yearly graph - notice the dates and you can see it is still data per 5 minutes.

I'm not too sure where to poke around and what to look for to work out why this is happening either, but I've checked the RRD file sizes and the ones with 5 minute resolution are ~1.5M or 3M versus ~140K. This seems to have happened for about 1500 of my 3200 data sources. The time these data sources were created isn't of any significance either by the looks of it as some of the ones that are maintaining resolution are fairly old.

Tim

[gandalf]

The size is a good indication for those rrd's who got wrong. If you're using pure cacti, I'd swear there is ABSOLUTELY NO way to change rrd definitions if the file has once been created.
So it must have occured at generation time OR was changed manually afterwards.
You may check your RRA definitions to see, whether there is actually any RRA defining 5 min intervals for infinity. Another source may be importing tenplates. In each import you'll find rra definitions that will overwrite your settings. cacti always reports "updated" whether it really changed existing rra definitions or not. To be SURE, you would have to check rra definitions EACH time after importing. No joke, this is a very useful check; you should really do this (well, it already occured to me and I claim to be a somewhat experienced cacti user).
And if you have this unwanted settings for lots of rrds, this will surely slow cactis poller performance down. Changing this is near to magic ...
Perhaps dumping, changing XML and reloading would do ...
Reinhard

[time]

I haven't changed anything manually, so perhaps things got changed with template imports as you suggested. What RRA definitions should I be checking exactly? I've looked at the rra definitions for the graphs above that have 5 minute data and they all have the Daily, Weekly, Monthly, Yearly RRAs associated with them.

It is not an unwanted setting, I actually prefer to keep all the data there as it allows me to look at old data with 5 minute accuracy. The reason I posted in this thread was because I thought maybe something happened with my installation that inadvertently caused some of my graphs to maintain 5 minute accuracy which is what a few people seem to want also.

It also doesn't seem to slow the poller down at all, my poller stats are:

[warenet]

Reinhard,

I have a question about your insructions below:

[warenet]

[gandalf]

[gandalf]