promql: Fix collision error with delayed name removal for non-overlapping series

Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>

promql/engine.go

@@ -1175,7 +1175,7 @@ func (ev *evaluator) Eval(ctx context.Context, expr parser.Expr) (v parser.Value
 
 	v, ws = ev.eval(ctx, expr)
 	if ev.enableDelayedNameRemoval {
-		ev.cleanupMetricLabels(v)
+		v = ev.cleanupMetricLabels(v)
 	}
 	return v, ws, nil
 }
@@ -3832,7 +3832,7 @@ func (*evaluator) aggregationCountValues(e *parser.AggregateExpr, grouping []str
 	return enh.Out, nil
 }
 
-func (ev *evaluator) cleanupMetricLabels(v parser.Value) {
+func (ev *evaluator) cleanupMetricLabels(v parser.Value) parser.Value {
 	if v.Type() == parser.ValueTypeMatrix {
 		mat := v.(Matrix)
 		for i := range mat {
@@ -3840,9 +3840,7 @@ func (ev *evaluator) cleanupMetricLabels(v parser.Value) {
 				mat[i].Metric = mat[i].Metric.DropReserved(schema.IsMetadataLabel)
 			}
 		}
-		if mat.ContainsSameLabelset() {
-			ev.errorf("vector cannot contain metrics with the same labelset")
-		}
+		return ev.mergeSeriesWithSameLabelset(mat)
 	} else if v.Type() == parser.ValueTypeVector {
 		vec := v.(Vector)
 		for i := range vec {
@@ -3853,7 +3851,98 @@ func (ev *evaluator) cleanupMetricLabels(v parser.Value) {
 		if vec.ContainsSameLabelset() {
 			ev.errorf("vector cannot contain metrics with the same labelset")
 		}
+		return vec
+	}
+	return v
+}
+
+// mergeSeriesWithSameLabelset merges series in a matrix that have the same labelset
+// after __name__ label removal. This happens when delayed name removal is enabled and
+// operations like OR combine series that originally had different names but end up
+// with the same labelset after dropping the name. If series with the same labelset
+// have overlapping timestamps, an error is returned.
+func (ev *evaluator) mergeSeriesWithSameLabelset(mat Matrix) Matrix {
+	if len(mat) <= 1 {
+		return mat
+	}
+
+	// Group series by their labelset hash.
+	seriesByHash := make(map[uint64][]int)
+	for i := range mat {
+		hash := mat[i].Metric.Hash()
+		seriesByHash[hash] = append(seriesByHash[hash], i)
+	}
+
+	// Check if any merging is needed.
+	needsMerge := false
+	for _, indices := range seriesByHash {
+		if len(indices) > 1 {
+			needsMerge = true
+			break
+		}
+	}
+
+	if !needsMerge {
+		return mat
+	}
+
+	// Merge series with the same labelset.
+	merged := make(Matrix, 0, len(seriesByHash))
+	for _, indices := range seriesByHash {
+		base := mat[indices[0]]
+
+		if len(indices) == 1 {
+			// No collision, add as-is.
+			merged = append(merged, base)
+			continue
+		}
+
+		// Multiple series with the same labelset - check for overlaps and merge.
+		// Build a set of timestamps to detect overlaps.
+		timestamps := make(map[int64]struct{}, len(base.Floats)+len(base.Histograms))
+		for _, p := range base.Floats {
+			timestamps[p.T] = struct{}{}
+		}
+		for _, p := range base.Histograms {
+			timestamps[p.T] = struct{}{}
+		}
+
+		// Merge remaining series, checking for timestamp overlaps.
+		for _, idx := range indices[1:] {
+			series := mat[idx]
+
+			// Check floats for overlaps.
+			for _, p := range series.Floats {
+				if _, exists := timestamps[p.T]; exists {
+					ev.errorf("vector cannot contain metrics with the same labelset")
+				}
+				timestamps[p.T] = struct{}{}
+			}
+			// Check histograms for overlaps.
+			for _, p := range series.Histograms {
+				if _, exists := timestamps[p.T]; exists {
+					ev.errorf("vector cannot contain metrics with the same labelset")
+				}
+				timestamps[p.T] = struct{}{}
+			}
+
+			// No overlaps, merge the samples.
+			base.Floats = append(base.Floats, series.Floats...)
+			base.Histograms = append(base.Histograms, series.Histograms...)
+		}
+
+		// Sort merged samples by timestamp.
+		sort.Slice(base.Floats, func(i, j int) bool {
+			return base.Floats[i].T < base.Floats[j].T
+		})
+		sort.Slice(base.Histograms, func(i, j int) bool {
+			return base.Histograms[i].T < base.Histograms[j].T
+		})
+
+		merged = append(merged, base)
 	}
+
+	return merged
 }
 
 func addToSeries(ss *Series, ts int64, f float64, h *histogram.FloatHistogram, numSteps int) {

promql/promqltest/testdata/name_label_dropping.test

@@ -126,3 +126,13 @@ eval instant at 10m sum by (__name__) (metric_total{env="3"} or rate(metric_tota
 # Same as above, but with reversed order.
 eval instant at 10m sum by (__name__) (rate(metric_total{env="3"}[5m]) or metric_total{env="1"})
   metric_total 10
+
+clear
+
+# Test delayed name removal with range queries and OR operator.
+load 10m
+  metric_a 1 _
+  metric_b 3 4
+
+eval range from 0 to 20m step 10m -metric_a or -metric_b
+  {} -1 -4 _