app/vmselect/promql: follow-up for ce4f26db02

- Document the bugfix at docs/CHANGELOG.md
- Filter out NaN values before sorting as suggested at https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5509#discussion_r1447369218
- Revert unrelated changes in lib/filestream and lib/fs
- Use simpler test at app/vmselect/promql/exec_test.go

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5509
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5506

app/vmselect/promql/aggr.go

@@ -649,18 +649,26 @@ func newAggrFuncTopK(isReverse bool) aggrFunc {
 		if err != nil {
 			return nil, err
 		}
-		lt := lessWithNaNs
-		if isReverse {
-			lt = lessWithNaNsReversed
-		}
 		afe := func(tss []*timeseries, modififer *metricsql.ModifierExpr) []*timeseries {
+			var tssNoNaNs []*timeseries
 			for n := range tss[0].Values {
-				sort.Slice(tss, func(i, j int) bool {
-					a := tss[i].Values[n]
-					b := tss[j].Values[n]
-					return lt(a, b)
+				// Drop series with NaNs at Values[n] before sorting.
+				// This is needed for https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5506
+				tssNoNaNs = tssNoNaNs[:0]
+				for _, ts := range tss {
+					if !math.IsNaN(ts.Values[n]) {
+						tssNoNaNs = append(tssNoNaNs, ts)
+					}
+				}
+				sort.Slice(tssNoNaNs, func(i, j int) bool {
+					a := tssNoNaNs[i].Values[n]
+					b := tssNoNaNs[j].Values[n]
+					if isReverse {
+						a, b = b, a
+					}
+					return a < b
 				})
-				fillNaNsAtIdx(n, ks[n], tss)
+				fillNaNsAtIdx(n, ks[n], tssNoNaNs)
 			}
 			tss = removeEmptySeries(tss)
 			reverseSeries(tss)
@@ -711,16 +719,31 @@ func getRangeTopKTimeseries(tss []*timeseries, modifier *metricsql.ModifierExpr,
 			value: value,
 		}
 	}
-	lt := lessWithNaNs
-	if isReverse {
-		lt = lessWithNaNsReversed
+	// Drop maxs with NaNs before sorting.
+	// This is needed for https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5506
+	maxsNoNaNs := make([]tsWithValue, 0, len(maxs))
+	for _, tsv := range maxs {
+		if !math.IsNaN(tsv.value) {
+			maxsNoNaNs = append(maxsNoNaNs, tsv)
+		}
 	}
-	sort.Slice(maxs, func(i, j int) bool {
-		return lt(maxs[i].value, maxs[j].value)
+	sort.Slice(maxsNoNaNs, func(i, j int) bool {
+		a := maxsNoNaNs[i].value
+		b := maxsNoNaNs[j].value
+		if isReverse {
+			a, b = b, a
+		}
+		return a < b
 	})
-	for i := range maxs {
-		tss[i] = maxs[i].ts
+	for _, tsv := range maxs {
+		if math.IsNaN(tsv.value) {
+			maxsNoNaNs = append(maxsNoNaNs, tsv)
+		}
 	}
+	for i := range maxsNoNaNs {
+		tss[i] = maxsNoNaNs[i].ts
+	}
+
 	remainingSumTS := getRemainingSumTimeseries(tss, modifier, ks, remainingSumTagName)
 	for i, k := range ks {
 		fillNaNsAtIdx(i, k, tss)
@@ -1252,20 +1275,6 @@ func newAggrQuantileFunc(phis []float64) func(tss []*timeseries, modifier *metri
 	}
 }
 
-func lessWithNaNs(a, b float64) bool {
-	if math.IsNaN(a) {
-		return !math.IsNaN(b)
-	}
-	return a < b
-}
-
-func lessWithNaNsReversed(a, b float64) bool {
-	if math.IsNaN(a) {
-		return true
-	}
-	return a > b
-}
-
 func floatToIntBounded(f float64) int {
 	if f > math.MaxInt {
 		return math.MaxInt

app/vmselect/promql/aggr_test.go

@@ -1,12 +1,8 @@
 package promql
 
 import (
-	"log"
 	"math"
-	"reflect"
 	"testing"
-
-	"github.com/VictoriaMetrics/metricsql"
 )
 
 func TestModeNoNaNs(t *testing.T) {
@@ -38,268 +34,3 @@ func TestModeNoNaNs(t *testing.T) {
 	f(1, []float64{2, 3, 3, 4, 4}, 3)
 	f(1, []float64{4, 3, 2, 3, 4}, 3)
 }
-
-func TestLessWithNaNs(t *testing.T) {
-	f := func(a, b float64, expectedResult bool) {
-		t.Helper()
-		result := lessWithNaNs(a, b)
-		if result != expectedResult {
-			t.Fatalf("unexpected result; got %v; want %v", result, expectedResult)
-		}
-	}
-	f(nan, nan, false)
-	f(nan, 1, true)
-	f(1, nan, false)
-	f(1, 2, true)
-	f(2, 1, false)
-	f(1, 1, false)
-}
-
-func TestLessWithNaNsReversed(t *testing.T) {
-	f := func(a, b float64, expectedResult bool) {
-		t.Helper()
-		result := lessWithNaNsReversed(a, b)
-		if result != expectedResult {
-			t.Fatalf("unexpected result; got %v; want %v", result, expectedResult)
-		}
-	}
-	f(nan, nan, true)
-	f(nan, 1, true)
-	f(1, nan, false)
-	f(1, 2, false)
-	f(2, 1, true)
-	f(1, 1, false)
-}
-
-func TestTopK(t *testing.T) {
-	f := func(all [][]*timeseries, expected []*timeseries, k int, reversed bool) {
-		t.Helper()
-		topKFunc := newAggrFuncTopK(reversed)
-		actual, err := topKFunc(&aggrFuncArg{
-			args: all,
-			ae: &metricsql.AggrFuncExpr{
-				Limit:    1,
-				Modifier: metricsql.ModifierExpr{},
-			},
-			ec: nil,
-		})
-		if err != nil {
-			log.Fatalf("failed to call topK, err=%v", err)
-		}
-		for i := range actual {
-			if !eq(expected[i], actual[i]) {
-				t.Fatalf("unexpected result: i:%v got:\n%v; want:\t%v", i, actual[i], expected[i])
-			}
-		}
-	}
-
-	f(newTestSeries(), []*timeseries{
-		{
-			Timestamps: []int64{1, 2, 3, 4, 5},
-			Values:     []float64{nan, nan, 3, 2, 1},
-		},
-		{
-			Timestamps: []int64{1, 2, 3, 4, 5},
-			Values:     []float64{1, 2, 3, 4, 5},
-		},
-		{
-			Timestamps: []int64{1, 2, 3, 4, 5},
-			Values:     []float64{2, 3, nan, nan, nan},
-		},
-	}, 2, true)
-	f(newTestSeries(), []*timeseries{
-		{
-			Timestamps: []int64{1, 2, 3, 4, 5},
-			Values:     []float64{3, 4, 5, 6, 7},
-		},
-		{
-			Timestamps: []int64{1, 2, 3, 4, 5},
-			Values:     []float64{nan, nan, 4, 5, 6},
-		},
-		{
-			Timestamps: []int64{1, 2, 3, 4, 5},
-			Values:     []float64{5, 4, nan, nan, nan},
-		},
-	}, 2, false)
-	f(newTestSeriesWithNaNsWithoutOverlap(), []*timeseries{
-		{
-			Values:     []float64{nan, nan, nan, 2, 1},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{nan, nan, 5, 6, 7},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{2, 3, 4, nan, nan},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{1, 2, nan, nan, nan},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-	}, 2, true)
-	f(newTestSeriesWithNaNsWithoutOverlap(), []*timeseries{
-		{
-			Values:     []float64{nan, nan, 5, 6, 7},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{nan, nan, 6, 2, 1},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{2, 3, nan, nan, nan},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{1, 2, nan, nan, nan},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-	}, 2, false)
-	f(newTestSeriesWithNaNsWithOverlap(), []*timeseries{
-		{
-			Values:     []float64{nan, nan, nan, 2, 1},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{nan, nan, nan, 6, 7},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{1, 2, 3, nan, nan},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{2, 3, 4, nan, nan},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-	}, 2, true)
-	f(newTestSeriesWithNaNsWithOverlap(), []*timeseries{
-		{
-			Values:     []float64{nan, nan, 5, 6, 7},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{nan, nan, 6, 2, 1},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{2, 3, nan, nan, nan},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-		{
-			Values:     []float64{1, 2, nan, nan, nan},
-			Timestamps: []int64{1, 2, 3, 4, 5},
-		},
-	}, 2, false)
-}
-
-func newTestSeries() [][]*timeseries {
-	return [][]*timeseries{
-		{
-			{
-				Values:     []float64{2, 2, 2, 2, 2},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-		},
-		{
-			{
-				Values:     []float64{1, 2, 3, 4, 5},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{2, 3, 4, 5, 6},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{5, 4, 3, 2, 1},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{3, 4, 5, 6, 7},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-		},
-	}
-}
-
-func newTestSeriesWithNaNsWithoutOverlap() [][]*timeseries {
-	return [][]*timeseries{
-		{
-			{
-				Values:     []float64{2, 2, 2, 2, 2},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-		},
-		{
-			{
-				Values:     []float64{1, 2, nan, nan, nan},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{2, 3, 4, nan, nan},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{nan, nan, 6, 2, 1},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{nan, nan, 5, 6, 7},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-		},
-	}
-}
-
-func newTestSeriesWithNaNsWithOverlap() [][]*timeseries {
-	return [][]*timeseries{
-		{
-			{
-				Values:     []float64{2, 2, 2, 2, 2},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-		},
-		{
-			{
-				Values:     []float64{1, 2, 3, nan, nan},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{2, 3, 4, nan, nan},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{nan, nan, 6, 2, 1},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-			{
-				Values:     []float64{nan, nan, 5, 6, 7},
-				Timestamps: []int64{1, 2, 3, 4, 5},
-			},
-		},
-	}
-}
-
-func eq(a, b *timeseries) bool {
-	if !reflect.DeepEqual(a.Timestamps, b.Timestamps) {
-		return false
-	}
-	for i := range a.Values {
-		if !eqWithNan(a.Values[i], b.Values[i]) {
-			return false
-		}
-	}
-	return true
-}
-
-func eqWithNan(a, b float64) bool {
-	if math.IsNaN(a) && math.IsNaN(b) {
-		return true
-	}
-	if math.IsNaN(a) || math.IsNaN(b) {
-		return false
-	}
-	return a == b
-}

app/vmselect/promql/exec_test.go

@@ -6590,7 +6590,7 @@ func TestExecSuccess(t *testing.T) {
 	})
 	t.Run(`bottomk(1)`, func(t *testing.T) {
 		t.Parallel()
-		q := `bottomk(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss"))`
+		q := `bottomk(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss") or label_set(time()<100, "a", "b"))`
 		r1 := netstorage.Result{
 			MetricName: metricNameExpected,
 			Values:     []float64{nan, nan, nan, 10, 10, 10},

docs/CHANGELOG.md

@@ -51,6 +51,7 @@ The sandbox cluster installation is running under the constant load generated by
 * BUGFIX: [vmalert](https://docs.victoriametrics.com/vmalert.html): check `-external.url` schema when starting vmalert, must be `http` or `https`. Before, alertmanager could reject alert notifications if `-external.url` contained no or wrong schema.
 * BUGFIX: [vmalert](https://docs.victoriametrics.com/vmalert.html): automatically add `exported_` prefix for original evaluation result label if it's conflicted with external or reserved one, previously it was overridden. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5161).
 * BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): consistently sort results for `q1 or q2` query, so they do not change colors with each refresh in Grafana. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5393).
+* BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): properly return results from [bottomk](https://docs.victoriametrics.com/MetricsQL.html#bottomk) and `bottomk_*()` functions when some of these results contain NaN values. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5506). Thanks to @xiaozongyang for [the fix](https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5509).
 * BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): properly handle queries, which wrap [rollup functions](https://docs.victoriametrics.com/MetricsQL.html#rollup-functions) with multiple arguments without explicitly specified lookbehind window in square brackets into [aggregate functions](https://docs.victoriametrics.com/MetricsQL.html#aggregate-functions). For example, `sum(quantile_over_time(0.5, process_resident_memory_bytes))` was resulting to `expecting at least 2 args to ...; got 1 args` error. Thanks to @atykhyy for [the pull request](https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5414).
 * BUGFIX: [vmctl](https://docs.victoriametrics.com/vmctl.html): retry on import errors in `vm-native` mode. Before, retries happened only on writes into a network connection between source and destination. But errors returned by server after all the data was transmitted were logged, but not retried.
 * BUGFIX: [vmagent](https://docs.victoriametrics.com/vmagent.html): properly assume role with [AWS IRSA authorization](https://docs.aws.amazon.com/eks/latest/userguide/iam-roles-for-service-accounts.html). Previously role chaining was not supported. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3822) for details.

lib/filestream/filestream.go

@@ -334,6 +334,6 @@ var bwPool sync.Pool
 
 type streamTracker struct {
 	fd     uintptr
-	offset uint64 // nolint
-	length uint64 // nolint
+	offset uint64
+	length uint64
 }

lib/filestream/filestream_darwin.go

@@ -1,6 +1,6 @@
 package filestream
 
-func (st *streamTracker) adviseDontNeed(n int, fdatasync bool) error { // nolint
+func (st *streamTracker) adviseDontNeed(n int, fdatasync bool) error {
 	return nil
 }
 

lib/fs/fadvise_darwin.go

@@ -4,7 +4,7 @@ import (
 	"os"
 )
 
-func fadviseSequentialRead(f *os.File, prefetch bool) error { // nolint
+func fadviseSequentialRead(f *os.File, prefetch bool) error {
 	// TODO: implement this properly
 	return nil
 }