Unexpected behavior for absent_over_time

[psalaberria002]

Describe the bug
The absent_over_time in PromQL is expected to return 0 or 1 rows at most.
Definition taken from https://prometheus.io/docs/prometheus/latest/querying/functions/#absent_over_time:

absent_over_time(v range-vector) returns an empty vector if the range vector passed to it has any elements and a 1-element vector with the value 1 if the range vector passed to it has no elements.

VictoriaMetrics returns multiple rows in some cases, where it should not have reported any.

Expected behavior
See the following chart for example. Those are metrics from pods that are probably gone during the last few minutes, but there are still thousands of up metrics being reported. The expected result is an empty response.

Version
v1.71.0 cluster mode

[valyala]

@psalaberria002 , thanks for the bug report! It should be fixed in the commit 989668b . Could you build vmselect from this commit according to these docs and try executing absent_over_time() queries over this vmselect node in order to check whether the issue has been fixed?

This commit will be included in the next release of VictoriaMetrics.

[Wilderone]

We are face this problem in process of integrating VM with kubecost.
@valyala i was build image from this commit and everything seems ok now

[valyala]

@Wilderone , thanks for the update!

The bugfix has been integrated into VictoriaMetrics v1.73.0. Closing the issue as fixed.

[psalaberria002]

Thanks for the fix! Will try it in a couple of days.