feat: algo change

* fix: add adjustkeys in trace operator cte builder

* fix: more fixes

* fix: cleanup

* fix: move tests to trace operator file

* fix: address comments

* fix: lint issues

---------

Co-authored-by: Srikanth Chekuri <srikanth.chekuri92@gmail.com>

frontend/src/pages/TraceDetailsV3/TraceFlamegraph/__tests__/computeVisualLayout.test.ts

@@ -472,4 +472,98 @@ describe('computeVisualLayout', () => {
 		expect(aRow).toBeGreaterThan(1); // must NOT be at row 1
 		expect(aRow).toBe(3); // next free row after B at row 2 (A overlaps B)
 	});
+
+	// --- Wide-group fast path (> WIDE_GROUP_THRESHOLD siblings) ---
+	// Past the threshold the layout switches to exact overlap-only packing to
+	// avoid the O(N^2) connector-avoidance spiral. These lock in correctness and
+	// the no-overlap invariant at scale.
+
+	function noRowHasOverlap(
+		layout: ReturnType<typeof computeVisualLayout>,
+	): void {
+		for (const row of layout.visualRows) {
+			const sorted = [...row].sort((a, b) => a.timestamp - b.timestamp);
+			for (let i = 1; i < sorted.length; i++) {
+				const prevEnd = sorted[i - 1].timestamp + sorted[i - 1].durationNano / 1e6;
+				expect(sorted[i].timestamp).toBeGreaterThanOrEqual(prevEnd);
+			}
+		}
+	}
+
+	it('should pack thousands of sequential leaf siblings into 1 row (wide path)', () => {
+		const root = makeSpan({ spanId: 'root', timestamp: 0, durationNano: 1e12 });
+		const kids: FlamegraphSpan[] = [];
+		// 2000 strictly sequential (non-overlapping) children
+		for (let i = 0; i < 2000; i++) {
+			kids.push(
+				makeSpan({
+					spanId: `k${i}`,
+					parentSpanId: 'root',
+					timestamp: i * 10,
+					durationNano: 5e6, // 5ms, ends before next starts
+				}),
+			);
+		}
+
+		const layout = computeVisualLayout([[root], kids]);
+
+		expect(layout.spanToVisualRow['root']).toBe(0);
+		expect(layout.totalVisualRows).toBe(2); // all siblings share row 1
+		for (const k of kids) {
+			expect(layout.spanToVisualRow[k.spanId]).toBe(1);
+		}
+		noRowHasOverlap(layout);
+	});
+
+	it('should pack thousands of fully-overlapping leaf siblings without violations (wide path)', () => {
+		const root = makeSpan({ spanId: 'root', timestamp: 0, durationNano: 1e12 });
+		const kids: FlamegraphSpan[] = [];
+		// 1000 children all spanning the same window → each needs its own row
+		for (let i = 0; i < 1000; i++) {
+			kids.push(
+				makeSpan({
+					spanId: `k${i}`,
+					parentSpanId: 'root',
+					timestamp: 0,
+					durationNano: 100e6,
+				}),
+			);
+		}
+
+		const layout = computeVisualLayout([[root], kids]);
+
+		expect(layout.totalVisualRows).toBe(1001); // root + 1000 stacked rows
+		expect(Object.keys(layout.spanToVisualRow)).toHaveLength(1001);
+		noRowHasOverlap(layout);
+	});
+
+	it('should keep non-leaf subtrees adjacent within a wide mixed group (wide path)', () => {
+		const root = makeSpan({ spanId: 'root', timestamp: 0, durationNano: 1e12 });
+		const kids: FlamegraphSpan[] = [];
+		for (let i = 0; i < 1000; i++) {
+			kids.push(
+				makeSpan({
+					spanId: `k${i}`,
+					parentSpanId: 'root',
+					timestamp: i * 10,
+					durationNano: 5e6,
+				}),
+			);
+		}
+		// One of the wide siblings has a child of its own
+		const grandchild = makeSpan({
+			spanId: 'gc',
+			parentSpanId: 'k500',
+			timestamp: 5000,
+			durationNano: 2e6,
+		});
+
+		const layout = computeVisualLayout([[root], kids, [grandchild]]);
+
+		const parentRow = layout.spanToVisualRow['k500'];
+		const gcRow = layout.spanToVisualRow['gc'];
+		expect(gcRow - parentRow).toBe(1); // subtree adjacency preserved
+		expect(Object.keys(layout.spanToVisualRow)).toHaveLength(1002);
+		noRowHasOverlap(layout);
+	});
 });

frontend/src/pages/TraceDetailsV3/TraceFlamegraph/computeVisualLayout.ts

@@ -18,6 +18,81 @@ export interface VisualLayout {
 	totalVisualRows: number;
 }
 
+// Above this many siblings under one parent, the connector-avoidance refinement
+// (Checks 2 & 3) is both visually meaningless — the row is already a dense wall —
+// and quadratic: every child deposits a connector point on each intermediate row,
+// which pushes later children even higher, which deposits more points. That
+// feedback loop inflates a layout needing ~50 rows to thousands and never
+// finishes on wide traces. Past the threshold we pack by overlap only.
+const WIDE_GROUP_THRESHOLD = 512;
+
+/**
+ * Segment tree over rows that answers "lowest row index >= `from` whose smallest
+ * span start-time is >= `end`" in O(log rows). Used to place a large group of
+ * leaf siblings by overlap only: because siblings are processed in descending
+ * start order, every already-placed span on a row starts at or after the current
+ * one, so [start, end] overlaps a row iff some span there starts before `end` —
+ * i.e. the row is free iff its minimum start >= end. Each node stores the max of
+ * its subtree's per-row minimum starts so a free row can be found by descent.
+ */
+class LowestFreeRow {
+	private readonly size: number;
+
+	private readonly tree: Float64Array;
+
+	constructor(rows: number) {
+		let size = 1;
+		while (size < rows) {
+			size *= 2;
+		}
+		this.size = size;
+		this.tree = new Float64Array(size * 2).fill(Infinity);
+	}
+
+	place(row: number, start: number): void {
+		let i = row + this.size;
+		// A row's key is the minimum start among its spans. Children are processed
+		// in descending start order so a leaf's start is the new minimum, but a
+		// non-leaf subtree's descendant can land on a row out of order — take min.
+		if (start >= this.tree[i]) {
+			return;
+		}
+		this.tree[i] = start;
+		for (i >>= 1; i >= 1; i >>= 1) {
+			const next = Math.max(this.tree[2 * i], this.tree[2 * i + 1]);
+			if (this.tree[i] === next) {
+				break;
+			}
+			this.tree[i] = next;
+		}
+	}
+
+	lowestFrom(from: number, end: number): number {
+		return this.descend(1, 0, this.size - 1, from, end);
+	}
+
+	private descend(
+		node: number,
+		lo: number,
+		hi: number,
+		from: number,
+		end: number,
+	): number {
+		if (hi < from || this.tree[node] < end) {
+			return -1;
+		}
+		if (lo === hi) {
+			return lo;
+		}
+		const mid = (lo + hi) >> 1;
+		const left = this.descend(2 * node, lo, mid, from, end);
+		if (left !== -1) {
+			return left;
+		}
+		return this.descend(2 * node + 1, mid + 1, hi, from, end);
+	}
+}
+
 /**
  * Computes an overlap-safe visual layout for flamegraph spans using DFS ordering.
  *
@@ -214,7 +289,53 @@ export function computeVisualLayout(spans: FlamegraphSpan[][]): VisualLayout {
 		arr.push(point);
 	}
 
+	// Fast path for a parent with a very large group of children: pack by overlap
+	// only (descending greedy), skipping the quadratic connector-avoidance that
+	// spirals at this scale. Leaf children — the bulk of a wide trace — are placed
+	// in O(log rows) via the segment tree; the rare non-leaf subtree falls back to
+	// findPlacement against the shared interval map. Both structures are kept in
+	// sync so each placement sees all prior occupancy. Same ShapeEntry[] contract.
+	function computeWideShape(
+		rootSpan: FlamegraphSpan,
+		children: FlamegraphSpan[],
+	): ShapeEntry[] {
+		const shape: ShapeEntry[] = [{ span: rootSpan, relativeRow: 0 }];
+		const localIntervals = new Map<number, Array<[number, number]>>();
+		// Children occupy relative rows 1..children.length in the worst case.
+		const finder = new LowestFreeRow(children.length + 2);
+
+		const occupy = (row: number, span: FlamegraphSpan): void => {
+			const s = span.timestamp;
+			const e = span.timestamp + span.durationNano / 1e6;
+			shape.push({ span, relativeRow: row });
+			addIntervalTo(localIntervals, row, s, e);
+			finder.place(row, s);
+		};
+
+		for (const child of children) {
+			if (childrenMap.has(child.spanId)) {
+				// Non-leaf: place its whole subtree shape as a unit via findPlacement.
+				const childShape = computeSubtreeShape(child);
+				const offset = findPlacement(childShape, 1, localIntervals);
+				for (const entry of childShape) {
+					occupy(entry.relativeRow + offset, entry.span);
+				}
+			} else {
+				const end = child.timestamp + child.durationNano / 1e6;
+				occupy(finder.lowestFrom(1, end), child);
+			}
+		}
+
+		return shape;
+	}
+
 	function computeSubtreeShape(rootSpan: FlamegraphSpan): ShapeEntry[] {
+		const children = childrenMap.get(rootSpan.spanId);
+
+		if (children && children.length > WIDE_GROUP_THRESHOLD) {
+			return computeWideShape(rootSpan, children);
+		}
+
 		const localIntervals = new Map<number, Array<[number, number]>>();
 		const localConnectorPoints = new Map<number, number[]>();
 		const shape: ShapeEntry[] = [];
@@ -225,7 +346,6 @@ export function computeVisualLayout(spans: FlamegraphSpan[][]): VisualLayout {
 		shape.push({ span: rootSpan, relativeRow: 0 });
 		addIntervalTo(localIntervals, 0, rootStart, rootEnd);
 
-		const children = childrenMap.get(rootSpan.spanId);
 		if (children) {
 			for (const child of children) {
 				const childShape = computeSubtreeShape(child);

frontend/src/pages/TraceDetailsV3/TraceFlamegraph/hooks/useVisualLayoutWorker.ts

@@ -94,7 +94,7 @@ export function useVisualLayoutWorker(spans: FlamegraphSpan[][]): {
 			cleanup();
 		};
 
-		// Timeout: if worker doesn't respond in 30s, terminate and error
+		// Timeout: if worker doesn't respond in 15s, terminate and error
 		const WORKER_TIMEOUT_MS = 15000;
 		const timeoutId = setTimeout(() => {
 			if (requestIdRef.current === currentId && isComputingRef.current) {

pkg/telemetrytraces/statement_builder.go

@@ -124,8 +124,10 @@ func (b *traceQueryStatementBuilder) Build(
 		-------------------------------- End of tech debt ----------------------------
 	*/
 
-	query = b.adjustKeys(ctx, keys, query, requestType)
-
+	for _, action := range adjustTraceKeys(keys, &query, requestType) {
+		// TODO: change to debug level once we are confident about the behavior
+		b.logger.InfoContext(ctx, "key adjustment action", slog.String("action", action))
+	}
 	// Create SQL builder
 	q := sqlbuilder.NewSelectBuilder()
 
@@ -193,24 +195,30 @@ func getKeySelectors(query qbtypes.QueryBuilderQuery[qbtypes.TraceAggregation])
 	return keySelectors
 }
 
-func (b *traceQueryStatementBuilder) adjustKeys(ctx context.Context, keys map[string][]*telemetrytypes.TelemetryFieldKey, query qbtypes.QueryBuilderQuery[qbtypes.TraceAggregation], requestType qbtypes.RequestType) qbtypes.QueryBuilderQuery[qbtypes.TraceAggregation] {
-
-	// add deprecated fields only during statement building
-	// why?
-	// 1. to not fail filter expression that use deprecated cols
-	// 2. this could have been moved to metadata fetching itself, however, that
-	// would mean, they also show up in suggestions we we don't want to do
-	// 3. reason for not doing a simple append is to keep intrinsic/calculated field first so that it gets
-	// priority in multi_if sql expression
+// mergeDeprecatedTraceKeys prepends deprecated intrinsic/calculated trace field
+// definitions to the keys map. We do this during statement building, not at
+// metadata fetch time, because:
+//  1. Filter expressions that reference deprecated columns must continue to
+//     resolve — otherwise they fail with "key not found".
+//  2. Doing it at metadata fetch time would also surface deprecated keys in
+//     autocomplete suggestions, which we don't want.
+//  3. We prepend (not append) so the intrinsic/calculated entry wins ordering
+//     in the multi_if SQL expression.
+func mergeDeprecatedTraceKeys(keys map[string][]*telemetrytypes.TelemetryFieldKey) {
 	for fieldKeyName, fieldKey := range IntrinsicFieldsDeprecated {
 		keys[fieldKeyName] = append([]*telemetrytypes.TelemetryFieldKey{&fieldKey}, keys[fieldKeyName]...)
 	}
 	for fieldKeyName, fieldKey := range CalculatedFieldsDeprecated {
 		keys[fieldKeyName] = append([]*telemetrytypes.TelemetryFieldKey{&fieldKey}, keys[fieldKeyName]...)
 	}
+}
+
+func adjustTraceKeys(keys map[string][]*telemetrytypes.TelemetryFieldKey, query *qbtypes.QueryBuilderQuery[qbtypes.TraceAggregation], requestType qbtypes.RequestType) []string {
+
+	mergeDeprecatedTraceKeys(keys)
 
 	// Adjust keys for alias expressions in aggregations
-	actions := querybuilder.AdjustKeysForAliasExpressions(&query, requestType)
+	actions := querybuilder.AdjustKeysForAliasExpressions(query, requestType)
 
 	/*
 		Check if user is using multiple contexts or data types for same field name
@@ -228,7 +236,7 @@ func (b *traceQueryStatementBuilder) adjustKeys(ctx context.Context, keys map[st
 		and make it just http.status_code and remove the duplicate entry.
 	*/
 
-	actions = append(actions, querybuilder.AdjustDuplicateKeys(&query)...)
+	actions = append(actions, querybuilder.AdjustDuplicateKeys(query)...)
 
 	/*
 		Now adjust each key to have correct context and data type
@@ -236,24 +244,20 @@ func (b *traceQueryStatementBuilder) adjustKeys(ctx context.Context, keys map[st
 		Reason for doing this is to not create an unexpected behavior for users
 	*/
 	for idx := range query.SelectFields {
-		actions = append(actions, b.adjustKey(&query.SelectFields[idx], keys)...)
+		actions = append(actions, adjustTraceKey(&query.SelectFields[idx], keys)...)
 	}
 	for idx := range query.GroupBy {
-		actions = append(actions, b.adjustKey(&query.GroupBy[idx].TelemetryFieldKey, keys)...)
+		actions = append(actions, adjustTraceKey(&query.GroupBy[idx].TelemetryFieldKey, keys)...)
 	}
 	for idx := range query.Order {
-		actions = append(actions, b.adjustKey(&query.Order[idx].Key.TelemetryFieldKey, keys)...)
+		actions = append(actions, adjustTraceKey(&query.Order[idx].Key.TelemetryFieldKey, keys)...)
 	}
 
-	for _, action := range actions {
-		// TODO: change to debug level once we are confident about the behavior
-		b.logger.InfoContext(ctx, "key adjustment action", slog.String("action", action))
-	}
-
-	return query
+	return actions
 }
 
-func (b *traceQueryStatementBuilder) adjustKey(key *telemetrytypes.TelemetryFieldKey, keys map[string][]*telemetrytypes.TelemetryFieldKey) []string {
+// adjustTraceKey resolves a single TelemetryFieldKey against the keys map.
+func adjustTraceKey(key *telemetrytypes.TelemetryFieldKey, keys map[string][]*telemetrytypes.TelemetryFieldKey) []string {
 
 	// for recording actions taken
 	actions := []string{}

pkg/telemetrytraces/stmt_builder_test.go

@@ -1125,28 +1125,13 @@ func TestAdjustKey(t *testing.T) {
 		},
 	}
 
-	fm := NewFieldMapper()
-	cb := NewConditionBuilder(fm)
-	mockMetadataStore := telemetrytypestest.NewMockMetadataStore()
-	fl := flaggertest.New(t)
-	aggExprRewriter := querybuilder.NewAggExprRewriter(instrumentationtest.New().ToProviderSettings(), nil, fm, cb, nil, fl)
-	statementBuilder := NewTraceQueryStatementBuilder(
-		instrumentationtest.New().ToProviderSettings(),
-		mockMetadataStore,
-		fm,
-		cb,
-		aggExprRewriter,
-		nil,
-		fl,
-	)
-
 	for _, c := range cases {
 		t.Run(c.name, func(t *testing.T) {
 			// Create a copy of the input key to avoid modifying the original
 			key := c.inputKey
 
 			// Call adjustKey
-			statementBuilder.adjustKey(&key, c.keysMap)
+			adjustTraceKey(&key, c.keysMap)
 
 			// Verify the key was adjusted as expected
 			require.Equal(t, c.expectedKey.Name, key.Name, "key name should match")
@@ -1399,21 +1384,6 @@ func TestAdjustKeys(t *testing.T) {
 		},
 	}
 
-	fm := NewFieldMapper()
-	cb := NewConditionBuilder(fm)
-	mockMetadataStore := telemetrytypestest.NewMockMetadataStore()
-	fl := flaggertest.New(t)
-	aggExprRewriter := querybuilder.NewAggExprRewriter(instrumentationtest.New().ToProviderSettings(), nil, fm, cb, nil, fl)
-	statementBuilder := NewTraceQueryStatementBuilder(
-		instrumentationtest.New().ToProviderSettings(),
-		mockMetadataStore,
-		fm,
-		cb,
-		aggExprRewriter,
-		nil,
-		fl,
-	)
-
 	for _, c := range cases {
 		t.Run(c.name, func(t *testing.T) {
 			// Create a deep copy of the keys map to avoid modifying the original
@@ -1424,7 +1394,7 @@ func TestAdjustKeys(t *testing.T) {
 			}
 
 			// Call adjustKeys
-			c.query = statementBuilder.adjustKeys(context.Background(), keysMapCopy, c.query, qbtypes.RequestTypeScalar)
+			adjustTraceKeys(keysMapCopy, &c.query, qbtypes.RequestTypeScalar)
 
 			// Verify select fields were adjusted
 			if c.expectedSelectFields != nil {

pkg/telemetrytraces/trace_operator_cte_builder.go

@@ -216,6 +216,13 @@ func (b *traceOperatorCTEBuilder) buildQueryCTE(ctx context.Context, queryName s
 	}
 	b.stmtBuilder.logger.DebugContext(ctx, "Retrieved keys for query", slog.String("query_name", queryName), slog.Int("keys_count", len(keys)))
 
+	// The CTE only selects spans matching the filter. Aggregations, group by
+	// and order by run later in buildFinalQuery, so RequestTypeRaw is fine here.
+	for _, action := range adjustTraceKeys(keys, query, qbtypes.RequestTypeRaw) {
+		// TODO: change to debug level once we are confident about the behavior
+		b.stmtBuilder.logger.InfoContext(ctx, "key adjustment action", slog.String("action", action))
+	}
+
 	// Build resource filter CTE for this specific query
 	resourceFilterCTEName := fmt.Sprintf("__resource_filter_%s", cteName)
 	resourceStmt, err := b.buildResourceFilterCTE(ctx, *query)
@@ -417,21 +424,28 @@ func (b *traceOperatorCTEBuilder) buildNotCTE(leftCTE, rightCTE string) (string,
 }
 
 func (b *traceOperatorCTEBuilder) buildFinalQuery(ctx context.Context, selectFromCTE string, requestType qbtypes.RequestType) (*qbtypes.Statement, error) {
+	keySelectors := b.getKeySelectors()
+	keys, _, err := b.stmtBuilder.metadataStore.GetKeysMulti(ctx, keySelectors)
+	if err != nil {
+		return nil, err
+	}
+	b.adjustOperatorKeys(ctx, keys, requestType)
+
 	switch requestType {
 	case qbtypes.RequestTypeRaw:
-		return b.buildListQuery(ctx, selectFromCTE)
+		return b.buildListQuery(ctx, selectFromCTE, keys)
 	case qbtypes.RequestTypeTimeSeries:
-		return b.buildTimeSeriesQuery(ctx, selectFromCTE)
+		return b.buildTimeSeriesQuery(ctx, selectFromCTE, keys)
 	case qbtypes.RequestTypeTrace:
-		return b.buildTraceQuery(ctx, selectFromCTE)
+		return b.buildTraceQuery(ctx, selectFromCTE, keys)
 	case qbtypes.RequestTypeScalar:
-		return b.buildScalarQuery(ctx, selectFromCTE)
+		return b.buildScalarQuery(ctx, selectFromCTE, keys)
 	default:
 		return nil, errors.NewInvalidInputf(errors.CodeInvalidInput, "unsupported request type: %s", requestType)
 	}
 }
 
-func (b *traceOperatorCTEBuilder) buildListQuery(ctx context.Context, selectFromCTE string) (*qbtypes.Statement, error) {
+func (b *traceOperatorCTEBuilder) buildListQuery(ctx context.Context, selectFromCTE string, keys map[string][]*telemetrytypes.TelemetryFieldKey) (*qbtypes.Statement, error) {
 	sb := sqlbuilder.NewSelectBuilder()
 
 	// Select core fields
@@ -453,22 +467,6 @@ func (b *traceOperatorCTEBuilder) buildListQuery(ctx context.Context, selectFrom
 		"parent_span_id": true,
 	}
 
-	// Get keys for selectFields
-	keySelectors := b.getKeySelectors()
-	for _, field := range b.operator.SelectFields {
-		keySelectors = append(keySelectors, &telemetrytypes.FieldKeySelector{
-			Name:          field.Name,
-			Signal:        telemetrytypes.SignalTraces,
-			FieldContext:  field.FieldContext,
-			FieldDataType: field.FieldDataType,
-		})
-	}
-
-	keys, _, err := b.stmtBuilder.metadataStore.GetKeysMulti(ctx, keySelectors)
-	if err != nil {
-		return nil, err
-	}
-
 	// Add selectFields using ColumnExpressionFor since we now have all base table columns
 	for _, field := range b.operator.SelectFields {
 		if selectedFields[field.Name] {
@@ -518,6 +516,44 @@ func (b *traceOperatorCTEBuilder) buildListQuery(ctx context.Context, selectFrom
 	}, nil
 }
 
+// adjustOperatorKeys runs the same key adjustments as adjustTraceKeys, but on
+// the operator's own fields. The operator has a different struct shape than
+// QueryBuilderQuery, so we copy the relevant fields into a temp query, run
+// the shared helpers, and copy the results back.
+func (b *traceOperatorCTEBuilder) adjustOperatorKeys(ctx context.Context, keys map[string][]*telemetrytypes.TelemetryFieldKey, requestType qbtypes.RequestType) {
+	mergeDeprecatedTraceKeys(keys)
+
+	tmp := qbtypes.QueryBuilderQuery[qbtypes.TraceAggregation]{
+		Aggregations: b.operator.Aggregations,
+		SelectFields: b.operator.SelectFields,
+		GroupBy:      b.operator.GroupBy,
+		Order:        b.operator.Order,
+	}
+
+	actions := querybuilder.AdjustKeysForAliasExpressions(&tmp, requestType)
+	actions = append(actions, querybuilder.AdjustDuplicateKeys(&tmp)...)
+
+	for idx := range tmp.SelectFields {
+		actions = append(actions, adjustTraceKey(&tmp.SelectFields[idx], keys)...)
+	}
+	for idx := range tmp.GroupBy {
+		actions = append(actions, adjustTraceKey(&tmp.GroupBy[idx].TelemetryFieldKey, keys)...)
+	}
+	for idx := range tmp.Order {
+		actions = append(actions, adjustTraceKey(&tmp.Order[idx].Key.TelemetryFieldKey, keys)...)
+	}
+
+	// Copy back the slices the helpers can rewrite.
+	b.operator.Aggregations = tmp.Aggregations
+	b.operator.SelectFields = tmp.SelectFields
+	b.operator.GroupBy = tmp.GroupBy
+	b.operator.Order = tmp.Order
+
+	for _, action := range actions {
+		b.stmtBuilder.logger.InfoContext(ctx, "key adjustment action", slog.String("action", action))
+	}
+}
+
 func (b *traceOperatorCTEBuilder) getKeySelectors() []*telemetrytypes.FieldKeySelector {
 	var keySelectors []*telemetrytypes.FieldKeySelector
 
@@ -545,6 +581,15 @@ func (b *traceOperatorCTEBuilder) getKeySelectors() []*telemetrytypes.FieldKeySe
 		})
 	}
 
+	for _, sf := range b.operator.SelectFields {
+		keySelectors = append(keySelectors, &telemetrytypes.FieldKeySelector{
+			Name:          sf.Name,
+			Signal:        telemetrytypes.SignalTraces,
+			FieldContext:  sf.FieldContext,
+			FieldDataType: sf.FieldDataType,
+		})
+	}
+
 	for i := range keySelectors {
 		keySelectors[i].Signal = telemetrytypes.SignalTraces
 	}
@@ -552,7 +597,7 @@ func (b *traceOperatorCTEBuilder) getKeySelectors() []*telemetrytypes.FieldKeySe
 	return keySelectors
 }
 
-func (b *traceOperatorCTEBuilder) buildTimeSeriesQuery(ctx context.Context, selectFromCTE string) (*qbtypes.Statement, error) {
+func (b *traceOperatorCTEBuilder) buildTimeSeriesQuery(ctx context.Context, selectFromCTE string, keys map[string][]*telemetrytypes.TelemetryFieldKey) (*qbtypes.Statement, error) {
 	sb := sqlbuilder.NewSelectBuilder()
 
 	sb.Select(fmt.Sprintf(
@@ -560,12 +605,6 @@ func (b *traceOperatorCTEBuilder) buildTimeSeriesQuery(ctx context.Context, sele
 		int64(b.operator.StepInterval.Seconds()),
 	))
 
-	keySelectors := b.getKeySelectors()
-	keys, _, err := b.stmtBuilder.metadataStore.GetKeysMulti(ctx, keySelectors)
-	if err != nil {
-		return nil, err
-	}
-
 	var allGroupByArgs []any
 
 	for _, gb := range b.operator.GroupBy {
@@ -644,8 +683,7 @@ func (b *traceOperatorCTEBuilder) buildTimeSeriesQuery(ctx context.Context, sele
 	combinedArgs := append(allGroupByArgs, allAggChArgs...)
 
 	// Add HAVING clause if specified
-	err = b.addHavingClause(sb)
-	if err != nil {
+	if err := b.addHavingClause(sb); err != nil {
 		return nil, err
 	}
 
@@ -672,17 +710,11 @@ func (b *traceOperatorCTEBuilder) buildTraceSummaryCTE(selectFromCTE string) {
 	b.addCTE("trace_summary", sql, args, []string{"all_spans", selectFromCTE})
 }
 
-func (b *traceOperatorCTEBuilder) buildTraceQuery(ctx context.Context, selectFromCTE string) (*qbtypes.Statement, error) {
+func (b *traceOperatorCTEBuilder) buildTraceQuery(ctx context.Context, selectFromCTE string, keys map[string][]*telemetrytypes.TelemetryFieldKey) (*qbtypes.Statement, error) {
 	b.buildTraceSummaryCTE(selectFromCTE)
 
 	sb := sqlbuilder.NewSelectBuilder()
 
-	keySelectors := b.getKeySelectors()
-	keys, _, err := b.stmtBuilder.metadataStore.GetKeysMulti(ctx, keySelectors)
-	if err != nil {
-		return nil, err
-	}
-
 	var allGroupByArgs []any
 
 	for _, gb := range b.operator.GroupBy {
@@ -764,8 +796,7 @@ func (b *traceOperatorCTEBuilder) buildTraceQuery(ctx context.Context, selectFro
 		sb.GroupBy(groupByKeys...)
 	}
 
-	err = b.addHavingClause(sb)
-	if err != nil {
+	if err := b.addHavingClause(sb); err != nil {
 		return nil, err
 	}
 
@@ -821,15 +852,9 @@ func (b *traceOperatorCTEBuilder) buildTraceQuery(ctx context.Context, selectFro
 	}, nil
 }
 
-func (b *traceOperatorCTEBuilder) buildScalarQuery(ctx context.Context, selectFromCTE string) (*qbtypes.Statement, error) {
+func (b *traceOperatorCTEBuilder) buildScalarQuery(ctx context.Context, selectFromCTE string, keys map[string][]*telemetrytypes.TelemetryFieldKey) (*qbtypes.Statement, error) {
 	sb := sqlbuilder.NewSelectBuilder()
 
-	keySelectors := b.getKeySelectors()
-	keys, _, err := b.stmtBuilder.metadataStore.GetKeysMulti(ctx, keySelectors)
-	if err != nil {
-		return nil, err
-	}
-
 	var allGroupByArgs []any
 
 	for _, gb := range b.operator.GroupBy {
@@ -911,8 +936,7 @@ func (b *traceOperatorCTEBuilder) buildScalarQuery(ctx context.Context, selectFr
 	combinedArgs := append(allGroupByArgs, allAggChArgs...)
 
 	// Add HAVING clause if specified
-	err = b.addHavingClause(sb)
-	if err != nil {
+	if err := b.addHavingClause(sb); err != nil {
 		return nil, err
 	}
 

pkg/telemetrytraces/trace_operator_cte_builder_test.go

@@ -14,6 +14,24 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+func newTestTraceOperatorStatementBuilder(t *testing.T) *traceOperatorStatementBuilder {
+	t.Helper()
+	fm := NewFieldMapper()
+	cb := NewConditionBuilder(fm)
+	mockMetadataStore := telemetrytypestest.NewMockMetadataStore()
+	mockMetadataStore.KeysMap = buildCompleteFieldKeyMap()
+	fl := flaggertest.New(t)
+	aggExprRewriter := querybuilder.NewAggExprRewriter(instrumentationtest.New().ToProviderSettings(), nil, fm, cb, nil, fl)
+	traceStmtBuilder := NewTraceQueryStatementBuilder(
+		instrumentationtest.New().ToProviderSettings(),
+		mockMetadataStore, fm, cb, aggExprRewriter, nil, fl,
+	)
+	return NewTraceOperatorStatementBuilder(
+		instrumentationtest.New().ToProviderSettings(),
+		mockMetadataStore, fm, cb, traceStmtBuilder, aggExprRewriter, fl,
+	)
+}
+
 func TestTraceOperatorStatementBuilder(t *testing.T) {
 	cases := []struct {
 		name           string
@@ -463,32 +481,7 @@ func TestTraceOperatorStatementBuilder(t *testing.T) {
 		},
 	}
 
-	fm := NewFieldMapper()
-	cb := NewConditionBuilder(fm)
-	mockMetadataStore := telemetrytypestest.NewMockMetadataStore()
-	mockMetadataStore.KeysMap = buildCompleteFieldKeyMap()
-	fl := flaggertest.New(t)
-	aggExprRewriter := querybuilder.NewAggExprRewriter(instrumentationtest.New().ToProviderSettings(), nil, fm, cb, nil, fl)
-
-	traceStmtBuilder := NewTraceQueryStatementBuilder(
-		instrumentationtest.New().ToProviderSettings(),
-		mockMetadataStore,
-		fm,
-		cb,
-		aggExprRewriter,
-		nil,
-		fl,
-	)
-
-	statementBuilder := NewTraceOperatorStatementBuilder(
-		instrumentationtest.New().ToProviderSettings(),
-		mockMetadataStore,
-		fm,
-		cb,
-		traceStmtBuilder,
-		aggExprRewriter,
-		fl,
-	)
+	statementBuilder := newTestTraceOperatorStatementBuilder(t)
 
 	for _, c := range cases {
 		t.Run(c.name, func(t *testing.T) {
@@ -579,32 +572,7 @@ func TestTraceOperatorStatementBuilderErrors(t *testing.T) {
 		},
 	}
 
-	fm := NewFieldMapper()
-	cb := NewConditionBuilder(fm)
-	mockMetadataStore := telemetrytypestest.NewMockMetadataStore()
-	mockMetadataStore.KeysMap = buildCompleteFieldKeyMap()
-	fl := flaggertest.New(t)
-	aggExprRewriter := querybuilder.NewAggExprRewriter(instrumentationtest.New().ToProviderSettings(), nil, fm, cb, nil, fl)
-
-	traceStmtBuilder := NewTraceQueryStatementBuilder(
-		instrumentationtest.New().ToProviderSettings(),
-		mockMetadataStore,
-		fm,
-		cb,
-		aggExprRewriter,
-		nil,
-		fl,
-	)
-
-	statementBuilder := NewTraceOperatorStatementBuilder(
-		instrumentationtest.New().ToProviderSettings(),
-		mockMetadataStore,
-		fm,
-		cb,
-		traceStmtBuilder,
-		aggExprRewriter,
-		fl,
-	)
+	statementBuilder := newTestTraceOperatorStatementBuilder(t)
 
 	for _, c := range cases {
 		t.Run(c.name, func(t *testing.T) {
@@ -626,3 +594,142 @@ func TestTraceOperatorStatementBuilderErrors(t *testing.T) {
 		})
 	}
 }
+
+func TestTraceOperatorStatementBuilderAdjustsKeys(t *testing.T) {
+	cases := []struct {
+		name          string
+		requestType   qbtypes.RequestType
+		operator      qbtypes.QueryBuilderTraceOperator
+		builderFilter string
+		wantSQL       string
+		wantArgs      []any
+	}{
+		{
+			name:        "deprecated duration filter in referenced builder query",
+			requestType: qbtypes.RequestTypeRaw,
+			operator: qbtypes.QueryBuilderTraceOperator{
+				Expression: "A",
+				Limit:      10,
+			},
+			builderFilter: "durationNano = '3s'",
+			wantSQL:       "duration_nano = ?",
+			wantArgs:      []any{int64(3000000000)},
+		},
+		{
+			name:        "context-prefixed aggregation alias in order by",
+			requestType: qbtypes.RequestTypeScalar,
+			operator: qbtypes.QueryBuilderTraceOperator{
+				Expression: "A",
+				Aggregations: []qbtypes.TraceAggregation{
+					{
+						Expression: "count()",
+						Alias:      "span.count_",
+					},
+				},
+				Order: []qbtypes.OrderBy{
+					{
+						Key: qbtypes.OrderByKey{
+							TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+								Name:         "count_",
+								FieldContext: telemetrytypes.FieldContextSpan,
+							},
+						},
+						Direction: qbtypes.OrderDirectionDesc,
+					},
+				},
+			},
+			wantSQL: "ORDER BY __result_0 desc",
+		},
+	}
+
+	statementBuilder := newTestTraceOperatorStatementBuilder(t)
+
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			err := c.operator.ParseExpression()
+			require.NoError(t, err)
+
+			filter := c.builderFilter
+			if filter == "" {
+				filter = "service.name = 'frontend'"
+			}
+
+			q, err := statementBuilder.Build(
+				context.Background(),
+				1747947419000,
+				1747983448000,
+				c.requestType,
+				c.operator,
+				&qbtypes.CompositeQuery{
+					Queries: []qbtypes.QueryEnvelope{
+						{
+							Type: qbtypes.QueryTypeBuilder,
+							Spec: qbtypes.QueryBuilderQuery[qbtypes.TraceAggregation]{
+								Name:   "A",
+								Signal: telemetrytypes.SignalTraces,
+								Filter: &qbtypes.Filter{Expression: filter},
+							},
+						},
+					},
+				},
+			)
+
+			require.NoError(t, err)
+			require.Contains(t, q.Query, c.wantSQL)
+			for _, arg := range c.wantArgs {
+				require.Contains(t, q.Args, arg)
+			}
+		})
+	}
+}
+
+// TestTraceOperatorStatementBuilderDeduplicatesKeys checks that a trace
+// operator with the same field name listed twice in GroupBy (once with a
+// context, once without) ends up with a single column in the outer SELECT
+// and a single entry in GROUP BY.
+func TestTraceOperatorStatementBuilderDeduplicatesKeys(t *testing.T) {
+	statementBuilder := newTestTraceOperatorStatementBuilder(t)
+
+	operator := qbtypes.QueryBuilderTraceOperator{
+		Expression: "A",
+		Aggregations: []qbtypes.TraceAggregation{
+			{Expression: "count()"},
+		},
+		GroupBy: []qbtypes.GroupByKey{
+			{TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+				Name:         "http.method",
+				FieldContext: telemetrytypes.FieldContextAttribute,
+			}},
+			// Same name, no context — should be merged with the entry above.
+			{TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+				Name: "http.method",
+			}},
+		},
+	}
+	require.NoError(t, operator.ParseExpression())
+
+	q, err := statementBuilder.Build(
+		context.Background(),
+		1747947419000,
+		1747983448000,
+		qbtypes.RequestTypeScalar,
+		operator,
+		&qbtypes.CompositeQuery{
+			Queries: []qbtypes.QueryEnvelope{
+				{
+					Type: qbtypes.QueryTypeBuilder,
+					Spec: qbtypes.QueryBuilderQuery[qbtypes.TraceAggregation]{
+						Name:   "A",
+						Signal: telemetrytypes.SignalTraces,
+						Filter: &qbtypes.Filter{Expression: "service.name = 'frontend'"},
+					},
+				},
+			},
+		},
+	)
+
+	require.NoError(t, err)
+
+	require.Contains(t, q.Query,
+		"SELECT toString(multiIf(mapContains(attributes_string, 'http.method') = ?, attributes_string['http.method'], NULL)) AS `http.method`, count() AS __result_0 FROM A GROUP BY `http.method` ORDER BY __result_0 DESC")
+}

tests/fixtures/querier.py

@@ -459,6 +459,57 @@ def find_named_result(
     )
 
 
+def assert_scalar_value(
+    response: requests.Response,
+    name: str,
+    expected: Any,
+    *,
+    row: int = 0,
+    col: int = 0,
+) -> None:
+    """Assert that the named scalar result has `expected` at data[row][col]."""
+    result = find_named_result(response.json()["data"]["data"]["results"], name)
+    assert result is not None, f"no result for query {name}"
+    assert result["data"][row][col] == expected, f"expected {expected} at [{row}][{col}], got {result['data'][row][col]}"
+
+
+def assert_grouped_scalar(
+    response: requests.Response,
+    name: str,
+    *,
+    expected_groups: int,
+    expected_columns: int,
+    last_col_value: Any | None = None,
+) -> None:
+    """Assert grouped scalar result has the expected column count and group count.
+    If `last_col_value` is set and there is exactly one group, also assert the
+    last column of that single row equals it (a common aggregation-value check)."""
+    result = find_named_result(response.json()["data"]["data"]["results"], name)
+    assert result is not None, f"no result for query {name}"
+    columns = result["columns"]
+    rows = result["data"]
+    assert len(columns) == expected_columns, f"expected {expected_columns} columns, got {len(columns)}: {columns}"
+    assert len(rows) == expected_groups, f"expected {expected_groups} groups, got {len(rows)}: {rows}"
+    if last_col_value is not None and expected_groups == 1:
+        assert rows[0][-1] == last_col_value, f"expected last col {last_col_value}, got row {rows[0]}"
+
+
+def assert_raw_row_subset(
+    response: requests.Response,
+    name: str,
+    expected: dict[str, Any],
+    *,
+    row: int = 0,
+) -> None:
+    """Assert that the named raw result's rows[row]['data'] is a superset of `expected`."""
+    result = find_named_result(response.json()["data"]["data"]["results"], name)
+    assert result is not None, f"no result for query {name}"
+    rows = result["rows"]
+    assert rows is not None, f"no rows for query {name}"
+    data = rows[row]["data"]
+    assert expected.items() <= data.items(), f"expected subset {expected}, got data {data}"
+
+
 def build_scalar_query(
     name: str,
     signal: str,

tests/integration/tests/querier/15_trace_operator.py

@@ -25,13 +25,22 @@ returnSpansFrom="A"
 from collections.abc import Callable
 from datetime import UTC, datetime, timedelta
 from http import HTTPStatus
+from typing import Any
 
 import pytest
 import requests
 
 from fixtures import types
 from fixtures.auth import USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD
-from fixtures.querier import get_rows
+from fixtures.querier import (
+    assert_grouped_scalar,
+    assert_raw_row_subset,
+    assert_scalar_value,
+    format_timestamp,
+    generate_traces_with_corrupt_metadata,
+    get_rows,
+    make_query_request,
+)
 from fixtures.traces import TraceIdGenerator, Traces, TracesKind, TracesStatusCode
 
 
@@ -434,3 +443,173 @@ def test_trace_operator(
     )
     assert response.status_code == HTTPStatus.OK, f"HTTP {response.status_code}: {response.text}"
     assert case["validate"](response), f"validation failed: {response.json()}"
+
+
+def _expected_trace_subset(trace: Traces) -> dict[str, Any]:
+    return {
+        "duration_nano": trace.duration_nano,
+        "name": trace.name,
+        "parent_span_id": trace.parent_span_id,
+        "span_id": trace.span_id,
+        "timestamp": format_timestamp(trace.timestamp),
+        "trace_id": trace.trace_id,
+    }
+
+
+@pytest.mark.parametrize(
+    "payload_factory,request_type,assert_result",
+    [
+        # Case 1: CTE filter uses the deprecated intrinsic field `durationNano`.
+        pytest.param(
+            lambda traces: [
+                {
+                    "type": "builder_query",
+                    "spec": {
+                        "name": "A",
+                        "signal": "traces",
+                        "filter": {"expression": 'durationNano = "3s"'},
+                    },
+                },
+                {
+                    "type": "builder_query",
+                    "spec": {
+                        "name": "B",
+                        "signal": "traces",
+                        "filter": {"expression": 'durationNano = "5s"'},
+                    },
+                },
+                {
+                    "type": "builder_trace_operator",
+                    "spec": {
+                        "name": "C",
+                        "expression": "A => B",
+                        "limit": 1,
+                    },
+                },
+            ],
+            "raw",
+            lambda response, traces: assert_raw_row_subset(response, "C", _expected_trace_subset(traces[0])),
+            id="deprecated-intrinsic-filter",
+        ),
+        # Case 2: CTE filter uses the deprecated calculated field `responseStatusCode`.
+        pytest.param(
+            lambda traces: [
+                {
+                    "type": "builder_query",
+                    "spec": {
+                        "name": "A",
+                        "signal": "traces",
+                        "filter": {"expression": 'responseStatusCode = "200"'},
+                    },
+                },
+                {
+                    "type": "builder_query",
+                    "spec": {
+                        "name": "B",
+                        "signal": "traces",
+                        "filter": {"expression": 'durationNano = "5s"'},
+                    },
+                },
+                {
+                    "type": "builder_trace_operator",
+                    "spec": {
+                        "name": "C",
+                        "expression": "A => B",
+                        "limit": 1,
+                    },
+                },
+            ],
+            "raw",
+            lambda response, traces: assert_raw_row_subset(response, "C", _expected_trace_subset(traces[0])),
+            id="deprecated-calculated-filter",
+        ),
+        # Case 3: order by uses `count_` with fieldContext `span`, which has
+        # to be rewritten to the aggregation alias `span.count_`.
+        pytest.param(
+            lambda traces: [
+                {
+                    "type": "builder_query",
+                    "spec": {
+                        "name": "A",
+                        "signal": "traces",
+                        "aggregations": [{"expression": "count()"}],
+                    },
+                },
+                {
+                    "type": "builder_trace_operator",
+                    "spec": {
+                        "name": "C",
+                        "expression": "A",
+                        "aggregations": [{"expression": "count()", "alias": "span.count_"}],
+                        "order": [{"key": {"name": "count_", "fieldContext": "span"}, "direction": "desc"}],
+                    },
+                },
+            ],
+            "scalar",
+            lambda response, traces: assert_scalar_value(response, "C", len(traces)),
+            id="context-prefixed-aggregation-alias-order",
+        ),
+        # Case 4: group by lists `cloud.provider` twice (once with a resource
+        # context, once without).
+        pytest.param(
+            lambda traces: [
+                {
+                    "type": "builder_query",
+                    "spec": {
+                        "name": "A",
+                        "signal": "traces",
+                        "disabled": True,
+                        "aggregations": [{"expression": "count()"}],
+                    },
+                },
+                {
+                    "type": "builder_trace_operator",
+                    "spec": {
+                        "name": "C",
+                        "expression": "A",
+                        "aggregations": [{"expression": "count()"}],
+                        "groupBy": [
+                            {"name": "cloud.provider", "fieldContext": "resource"},
+                            {"name": "cloud.provider"},
+                        ],
+                    },
+                },
+            ],
+            "scalar",
+            lambda response, traces: assert_grouped_scalar(response, "C", expected_groups=1, expected_columns=2, last_col_value=len(traces)),
+            id="duplicate-group-by-deduplicated",
+        ),
+    ],
+)
+def test_trace_operator_with_adjusted_keys(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_traces: Callable[[list[Traces]], None],
+    payload_factory: Callable[[list[Traces]], list[dict[str, Any]]],
+    request_type: str,
+    assert_result: Callable[[requests.Response, list[Traces]], None],
+) -> None:
+    """
+    Trace operators build a CTE per referenced builder query and an outer
+    query on top. Both layers need the same key adjustment as regular trace
+    queries, otherwise deprecated keys and context-prefixed aliases don't
+    resolve.
+    """
+    traces = generate_traces_with_corrupt_metadata()
+    insert_traces(traces)
+    payload = payload_factory(traces)
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms=int((datetime.now(tz=UTC) - timedelta(minutes=5)).timestamp() * 1000),
+        end_ms=int(datetime.now(tz=UTC).timestamp() * 1000),
+        request_type=request_type,
+        queries=payload,
+    )
+
+    assert response.status_code == HTTPStatus.OK, response.text
+    assert_result(response, traces)