rollup func problems (such as avg_over_time)

[davidsome]

Describe the bug
when i use sum(sum_over_time(credis.exceptions.count_sum))，everything is ok。

when i use avg(avg_over_time(credis.exceptions.count_sum))，I found something strange.

I found the raw data matches sum_over_time, why avg_over_time add so many none-exists datapoints ?

$ ./victoria-metrics-prod --version
victoria-metrics-v1.37.3

[valyala]

@davidsome , thanks for the bug report!

The bug should be fixed in the following commits:

• Single-node version: b1f6843
• Cluster version: 88e8bed

Could you build VictoriaMetrics from the commit b1f6843 according to these docs and verify whether avg_over_time works as expected now?

[valyala]

The bugfix has been included in v1.39.0. Closing the issue as fixed.

@davidsome , feel free re-opening the issue if v1.39.0 doesn't work as expected for your case.

[davidsome]

@valyala the newest version I found sum_over_time behaves strange，

when use sum_over_time, null values becomes zero. avg_over_time is ok .is this a bug？

[valyala]

@davidsome , I cannot reproduce this issue on the latest release of VictoriaMetrics (v1.57.1) - both sum_over_time() and count_over_time() return empty values on time ranges without raw samples.

Try exporting raw samples for these time series on the selected time range via /api/v1/export and inspecting their contents - probably they contain some unusual values there.