> If you log data at irregular intervals, then you'll get three effects.
> 1. Data normalisation - your samples will be modified to fit into the RRD's 
> interval.  This is common; however for small values, this can cause problems, 
> particularly if you are interested in the actual values and not the rate.
>   
This is indeed the case with some of my logging.

> 2. XFF making RRA cdps unknown.  If you have insufficient samples in an 
> interval, then an RRA may set the value for that cdp to unknown.
> 3. Heartbeat making data points unknown.  This is your biggest risk.  You 
> MUST have a sample AT LEAST as often as the heartbeat.  If the heartbeat is 
> smaller than your interval, then an interval will be set to unknown unless 
> multiple values are received during it.  If your heartbeat is exceptionally 
> large, then one sample may case several data points to be created!  This can 
> be strange.
>
> You could, for example, log the total number of bounced packets on a regular 
> basis.  This would work.  Or, every (regular) interval, log the number of 
> packets bounced since the last update.  What you should not do is to only log 
> something if and when you bounce a packet.
>   
All noted, thanks. I will then revert back to the "old" way, with a bit 
of a twist - executing the sampling at regular intervals, but via a 
special daemon and log the results (including zero/no matches) to RRD. 
That way, the database should be OK and kept in a sane state. Thanks Steve!

_______________________________________________
rrd-users mailing list
[email protected]
https://lists.oetiker.ch/cgi-bin/listinfo/rrd-users

Reply via email to