app/vmselect/promql: remove empty series before applying aggregate function

Previously empty series (e.g. series with all NaN samples) were passed to aggregate functions.
Such series must be ingored by all the aggregate functions.
So it is better from consistency PoV filtering out empty series before applying aggregate functions.

app/vmselect/promql/aggr.go

@@ -104,6 +104,9 @@ func removeGroupTags(metricName *storage.MetricName, modifier *metricsql.Modifie
 
 func aggrFuncExt(afe func(tss []*timeseries, modifier *metricsql.ModifierExpr) []*timeseries, argOrig []*timeseries,
 	modifier *metricsql.ModifierExpr, maxSeries int, keepOriginal bool) ([]*timeseries, error) {
+	// Remove empty time series, e.g. series with all NaN samples,
+	// since such series are ignored by aggregate functions.
+	argOrig = removeEmptySeries(argOrig)
 	arg := copyTimeseriesMetricNames(argOrig, keepOriginal)
 
 	// Perform grouping.

app/vmselect/promql/exec_test.go

@@ -5511,6 +5511,12 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
+	t.Run(`any(empty-series)`, func(t *testing.T) {
+		t.Parallel()
+		q := `any(label_set(time()<0, "foo", "bar"))`
+		resultExpected := []netstorage.Result{}
+		f(q, resultExpected)
+	})
 	t.Run(`group() by (test)`, func(t *testing.T) {
 		t.Parallel()
 		q := `group((

docs/CHANGELOG.md

@@ -39,13 +39,13 @@ The following tip changes can be tested by building VictoriaMetrics components f
 * BUGFIX: [vmagent](https://docs.victoriametrics.com/vmagent.html): properly parse relabel config when regex ending with escaped `$`. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3131).
 * BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): properly calculate `rate_over_sum(m[d])` as `sum_over_time(m[d])/d`. Previously the `sum_over_time(m[d])` could be improperly divided by smaller than `d` time range. See [rate_over_sum() docs](https://docs.victoriametrics.com/MetricsQL.html#rate_over_sum) and [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3045).
 * BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): properly calculate `increase(m[d])` over slow-changing counters with values smaller than 100. Previously [increase](https://docs.victoriametrics.com/MetricsQL.html#increase) could return unexpectedly big results in this case. See [the related issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/962) and [this pull request](https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3163).
+* BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): ignore empty series when applying [limit_offset](https://docs.victoriametrics.com/MetricsQL.html#limit_offset). It should improve queries with additional filters by value in expressions like `limit_offset(1,1, foo > 1)`.
 * BUGFIX: [VictoriaMetrics cluster](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html): properly calculate query results at `vmselect`. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3067). The issue has been introduced in [v1.81.0](https://docs.victoriametrics.com/CHANGELOG.html#v1810).
 * BUGFIX: [VictoriaMetrics cluster](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html): log clear error when multiple identical `-storageNode` command-line flags are passed to `vmselect` or to `vminsert`. Previously these components were crashed with cryptic panic `metric ... is already registered` in this case. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3076).
 * BUGFIX: [vmui](https://docs.victoriametrics.com/#vmui): fix `RangeError: Maximum call stack size exceeded` error when the query returns too many data points at `Table` view. See [this pull request](https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3092/files).
 * BUGFIX: [vmalert](https://docs.victoriametrics.com/vmalert.html): re-evaluate annotations per each alert evaluation. Previously, annotations were evaluated only on alert's value change. This could result in stale annotations in some cases described in [this pull request](https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3119).
 * BUGFIX: prevent from excessive CPU usage when the storage enters [read-only mode](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#readonly-mode). The previous fix in [v1.81.0](https://docs.victoriametrics.com/CHANGELOG.html#v1810) wasn't complete.
 * BUGFIX: [vmalert](https://docs.victoriametrics.com/vmalert.html): change default value for command-line flag `-datasource.queryStep` from `0s` to `5m`. Param `step` is added by vmalert to every rule evaluation request sent to datasource. Before this change, `step` was equal to group's evaluation interval by default. Param `step` for instant queries defines how far VM can look back for the last written data point. The change supposed to improve reliability of the rules evaluation when evaluation interval is lower than scraping interval.
-* BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): ignore empty series when applying `limit_offset`. It should improve queries with additional filters by value in expressions like `limit_offset(1,1, foo > 1)`. 
 
 ## [v1.81.2](https://github.com/VictoriaMetrics/VictoriaMetrics/releases/tag/v1.81.2)