Panel stacking should automatically consider NULL as 0 for the graphic representation

[sylr]

I know there as been a number of issues about this but using null as 0 is not a good solution because it messes up with the aggregation values (min, max, average, current) in the legend.

If you are stacking you are obliged to use null as 0 otherwise you might have surprises in your graphs but if you do so you can't trust the aggregation values in the legend because a Null in the timerange will:

• Set min to 0 even if your metric never had this value
• Set current to 0 even if the metric (/labels combinations) does not exist anymore
• And it will make the average totally bogus

So my proposal is to automatically consider Null values as being 0 (but would be super nice if nothing was drawn) for the graphic representation when stacking so that you can choose whether or not you want to consider Null as 0 for the aggregation functions.

Cheers.

[xlson]

Leaving this one to you @torkelo @Ijin08

[roidelapluie]

I agree with OP.

[sylr]

@torkelo @Ijin08 care to comment on this ?

[sylr]

@torkelo @Ijin08 ?

[onumossn]

Any plans for this?

[grandich]

+1

I'm new to Grafana and painfully stumbled upon this. One example of 'surprises' in graphs: when turning on and off Prometheus node exporter' collectors:

Node Exporter Scrape Time

With Null Value=null (incorrect)

With Null Value=null as zero

Albeit no difference in legends...

[djahandarie]

I think this is a UI/readability issue pretty much anyone who uses stacked graphs in Grafana will run into. Is there any chance this can get prioritized?

Also, it's likely this should be recategorized as a bug instead of a feature request.

[jhoos70]

Me too, missing the possibility to keep zero values down at the axis.

[tvsaru]

I am yet another person who tripped on this weird behavior (in Grafana v8.4.6)

In my example screenshot below, of stacking graphs, the yellow line is supposed to be 0, while green line was 1, but the yellow line masks the green line giving a wrong visualization of the data (makes it look like yellow was 1 for the whole time period).

[AndrewJDR]

Looking forward to a solution for this as well