Values in graph legends don't match the actual graph

[colkinz]

Recently I realized that on the weekly/monthly/yearly graphs, the graph legends don't match the actual graph anymore.
Also when zooming into for example the graph one can see that it is using the 5 minute averages and not the 1 day averages as it should use.

A few weeks ago I have changed the number of rows to a higher value in the Cacti RRA settings as well as in the actual RRD files. Could this cause the problem? Why is Cacti not using the appropriate RRAs for it's graphs?

I attached all the relevant configuration and "rrdtool info" output for the RRD file. Maybe someone can give me a hint what the reason might be....

[colkinz]

Here is the rrdtool info output...

[colkinz]

Here is the rrdtool info output...

[gandalf]

A few weeks ago I have changed the number of rows to a higher value in the Cacti RRA settings as well as in the actual RRD files. Could this cause the problem?

Yes, that's most likely what has happened

Why is Cacti not using the appropriate RRAs for it's graphs?

Cacti in fact does NOT specify the consolidation level to be used. It relies on the rrdtool automatism to select the "best matching" consolidation level. After resizing, consolidation levels will have to be filled first, so the automatism may fail. There's no way to force that with cacti.

Reinhard

[colkinz]

Hello Reinhard

Thank you for your prompt answer.

I guess you are saying that Cacti/rrdtool is using always the most granular RRA available for a certain period of time to display a graph.
So depending on how many rows I have configured for my RRAs the weekly/monthly/yearly graphs will always contain 5 minute/30 minute/1 day average values in one graph. As a result the monthly graph for example will contain far to granular 5 minute average data which doesn't show up on the graph (only when you zoom in) but will influence the max/min values of the legend...

That I now increased the rows only worsened the problem but even with the default RRA configuration (5 mins - 2 days retention, 30 mins - 14.58 days retention, 1 day - 64.58 days retention ...) the graphs and it's legends won't show what they are supposed to show.

Is the above true and is there no way to minimize the effect of this "phenomenon" than reducing the rows as much as possible or did you mean sth different?

[gandalf]

Hello Reinhard

Thank you for your prompt answer.

I guess you are saying that Cacti/rrdtool is using always the most granular RRA available for a certain period of time to display a graph.
So depending on how many rows I have configured for my RRAs the weekly/monthly/yearly graphs will always contain 5 minute/30 minute/1 day average values in one graph. As a result the monthly graph for example will contain far to granular 5 minute average data which doesn't show up on the graph (only when you zoom in) but will influence the max/min values of the legend...

Correct. I call this "graphical consolidation". In fact, what shall rrdtool do if you request it to plot some thousands data pints and you give it only 500 pixels graph width? By default, rrdtool will use the same consolidation method for graphical consolidation as the one given in the DEF (which is AVERAGE in most cases, unfortunately). On cli, you may instruct rrdtool to use a different consolidation method for "graphical consolidation", but that is not (yet) supported by cacti.

Is the above true and is there no way to minimize the effect of this "phenomenon" than reducing the rows as much as possible or did you mean sth different?

My workaround for this is to graph AVERAGE and MAX consolidation function in the same graph. See e.g. 4th link of my sig, the new Traffic Graph Template
Reinhard