histogram_quantile and histogram_fraction return values when both classic and native histograms are present #16551
Description
What did you do?
I ran histogram_quantile and histogram_fraction over a selector that returned both classic and native histograms that would be mapped to the same output series from these functions.
In PromQL test format:
load 1m
series{host="a"} {{schema:0 sum:5 count:4 buckets:[9 2 1]}}
series{host="a", le="0.1"} 2
series{host="a", le="1"} 3
series{host="a", le="10"} 5
series{host="a", le="100"} 6
series{host="a", le="1000"} 8
series{host="a", le="+Inf"} 9
eval instant at 0 histogram_quantile(0.8, series)
expect no_info
expect warn msg: PromQL warning: vector contains a mix of classic and native histograms for metric name "series"
# Should return no results, but has series {host="a"} with value 2.297 derived from native histogram series
eval instant at 0 histogram_fraction(-Inf, 1, series)
expect no_info
expect warn msg: PromQL warning: vector contains a mix of classic and native histograms for metric name "series"
# Should return no results, but has series {host="a"} with value 1 derived from native histogram series
I suspect this is a regression introduced by #16095.
What did you expect to see?
I expected both queries to return no results and a vector contains a mix of classic and native histograms for metric name "series" warning.
What did you see instead? Under which circumstances?
Both queries return results in addition to the vector contains a mix of classic and native histograms for metric name "series" warning.
The results appear to be derived from the native histogram series.
Prometheus version
Prometheus at 0ce6022