chore: generate openapi specs

One request per function family, with the arguments, their values and
what the function does to the series.

Assisted-by: Claude Fable 5.1

docs/api/openapi.yml

@@ -32729,6 +32729,301 @@ paths:
                   requestType: time_series
                   schemaVersion: v1
                   start: 1640995200000
+              function_anomaly_request_rate:
+                description: 'anomaly is the one function with named arguments. seasonality
+                  says which past periods predict the current value: hourly, daily
+                  or weekly, daily when absent. z_score_threshold is how many standard
+                  deviations from the prediction count as an anomaly, 3 when absent.
+                  The response carries, next to the values, a predicted series, an
+                  upper and a lower bound series and an anomaly score series. The
+                  scoring runs in the enterprise build; the community build returns
+                  the series unchanged.'
+                summary: 'anomaly: request rate scored against its daily pattern'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - metricName: http.server.duration.count
+                          spaceAggregation: sum
+                          timeAggregation: rate
+                        functions:
+                        - args:
+                          - name: seasonality
+                            value: daily
+                          - name: z_score_threshold
+                            value: 3
+                          name: anomaly
+                        groupBy:
+                        - fieldContext: resource
+                          name: service.name
+                        name: A
+                        signal: metrics
+                        stepInterval: 300s
+                      type: builder_query
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
+              function_cumulativesum_errors_so_far:
+                description: cumulativeSum replaces each point with the sum of every
+                  point up to it, so a per-step count becomes a running total that
+                  starts at zero at the window start. It takes no arguments. The total
+                  resets with the window, so a longer time range gives a larger final
+                  value; it is a view of the window, not a counter.
+                summary: 'cumulativeSum: 5xx responses since the start of the window'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - alias: server_errors
+                          expression: count()
+                        filter:
+                          expression: http.response.status_code >= 500
+                        functions:
+                        - name: cumulativeSum
+                        groupBy:
+                        - fieldContext: resource
+                          name: service.name
+                        name: A
+                        signal: traces
+                        stepInterval: 60s
+                      type: builder_query
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
+              function_cutoffmin_hide_idle_hosts:
+                description: The four threshold functions take one number. cutOffMin
+                  and cutOffMax remove the points below or above the threshold, so
+                  an idle host disappears from the chart instead of drawing a line
+                  along the bottom. clampMin and clampMax keep the points and replace
+                  the value with the threshold, which caps spikes without hiding them.
+                  system.cpu.utilization is a ratio, so 0.05 is 5%.
+                summary: 'cutOffMin: drop hosts below 5% CPU from the chart'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - metricName: system.cpu.utilization
+                          spaceAggregation: avg
+                          timeAggregation: avg
+                        functions:
+                        - args:
+                          - value: 0.05
+                          name: cutOffMin
+                        groupBy:
+                        - fieldContext: resource
+                          name: host.name
+                        name: A
+                        signal: metrics
+                        stepInterval: 60s
+                      type: builder_query
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
+              function_ewma_smoothed_latency:
+                description: 'ewma3, ewma5 and ewma7 replace each point with an exponentially
+                  weighted average of the point and the points before it. The number
+                  in the name is the window the default weight is derived from: alpha
+                  = 2 / (n + 1), so 0.5, 0.333 and 0.25. One optional argument overrides
+                  alpha; a value close to 1 follows the raw series, a value close
+                  to 0 smooths harder. 0.3 here smooths a little more than ewma5''s
+                  default.'
+                summary: 'ewma5: smoothed p95 latency'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - metricName: http.server.duration.bucket
+                          spaceAggregation: p95
+                        functions:
+                        - args:
+                          - value: 0.3
+                          name: ewma5
+                        groupBy:
+                        - fieldContext: resource
+                          name: service.name
+                        name: A
+                        signal: metrics
+                        stepInterval: 60s
+                      type: builder_query
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
+              function_fillzero_sparse_errors:
+                description: A count over sparse events returns a point only for steps
+                  that had a match, so a chart of failed spans has gaps where nothing
+                  failed. fillZero adds a zero at every missing step. With no arguments
+                  the server fills the request window at the query's step, which is
+                  the common case. The three optional arguments are start, end and
+                  step in milliseconds, for callers that want to fill a different
+                  grid.
+                summary: 'fillZero: a zero at every step with no failed spans'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - alias: failed_spans
+                          expression: count()
+                        filter:
+                          expression: has_error = true
+                        functions:
+                        - name: fillZero
+                        groupBy:
+                        - fieldContext: resource
+                          name: service.name
+                        name: A
+                        signal: traces
+                        stepInterval: 60s
+                      type: builder_query
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
+              function_log10_network_bytes:
+                description: log2 and log10 replace each value with its logarithm,
+                  so a loopback device moving kilobytes and an uplink moving gigabytes
+                  fit on one chart. They take no arguments. Zero and negative values
+                  have no logarithm and are not drawn. absolute, also without arguments,
+                  is the usual companion for a signed series before a logarithm.
+                summary: 'log10: network throughput across devices that differ by
+                  orders of magnitude'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - metricName: system.network.io
+                          spaceAggregation: sum
+                          timeAggregation: rate
+                        filter:
+                          expression: direction = 'transmit'
+                        functions:
+                        - name: log10
+                        groupBy:
+                        - fieldContext: attribute
+                          name: device
+                        name: A
+                        signal: metrics
+                        stepInterval: 60s
+                      type: builder_query
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
+              function_median_despiked_request_rate:
+                description: median3, median5 and median7 replace each point with
+                  the median of the surrounding 3, 5 or 7 points. A spike that lasts
+                  one step vanishes, a shift that lasts longer than half the window
+                  stays. They take no arguments. Use them when the raw series is noisy
+                  and the chart is for a trend, not for catching the spike.
+                summary: 'median5: request rate without one-step spikes'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - metricName: http.server.duration.count
+                          spaceAggregation: sum
+                          timeAggregation: rate
+                        functions:
+                        - name: median5
+                        groupBy:
+                        - fieldContext: resource
+                          name: service.name
+                        name: A
+                        signal: metrics
+                        stepInterval: 60s
+                      type: builder_query
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
+              function_runningdiff_disk_growth:
+                description: 'runningDiff turns a gauge into the change between consecutive
+                  points, so a filesystem usage gauge becomes bytes written per step.
+                  It takes no arguments and drops the first point of the window, since
+                  that point has nothing to difference against; the server reads one
+                  step before the window to keep the first interval. Functions apply
+                  in order: clampMin with 0 runs on the differences and replaces the
+                  negative value a cleanup or a volume reset produces with zero.'
+                summary: 'runningDiff then clampMin: bytes written to disk per step'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - metricName: system.filesystem.usage
+                          spaceAggregation: sum
+                          timeAggregation: latest
+                        filter:
+                          expression: state = 'used'
+                        functions:
+                        - name: runningDiff
+                        - args:
+                          - value: 0
+                          name: clampMin
+                        groupBy:
+                        - fieldContext: resource
+                          name: host.name
+                        name: A
+                        signal: metrics
+                        stepInterval: 60s
+                      type: builder_query
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
+              function_timeshift_week_over_week:
+                description: timeShift moves every point of its query forward by the
+                  argument, in seconds, so a series read from the past lands on the
+                  current timestamps. B here reads the hour exactly seven days (604800
+                  s) before the requested window and comes back on this week's timestamps,
+                  so it overlays A on a chart and the formula can subtract it point
+                  by point. The argument is positive; it says how far back to read.
+                summary: 'timeShift: this week''s error logs next to last week''s'
+                value:
+                  compositeQuery:
+                    queries:
+                    - spec:
+                        aggregations:
+                        - alias: errors
+                          expression: count()
+                        filter:
+                          expression: severity_text = 'ERROR'
+                        name: A
+                        signal: logs
+                        stepInterval: 60s
+                      type: builder_query
+                    - spec:
+                        aggregations:
+                        - alias: errors
+                          expression: count()
+                        filter:
+                          expression: severity_text = 'ERROR'
+                        functions:
+                        - args:
+                          - value: 604800
+                          name: timeShift
+                        name: B
+                        signal: logs
+                        stepInterval: 60s
+                      type: builder_query
+                    - spec:
+                        expression: A - B
+                        name: F1
+                      type: builder_formula
+                  end: 1640998800000
+                  requestType: time_series
+                  schemaVersion: v1
+                  start: 1640995200000
               logs_raw:
                 summary: 'Raw: fetch raw log records'
                 value:

pkg/apiserver/signozapiserver/querier.go

@@ -30,7 +30,7 @@ func (provider *provider) addQuerierRoutes(router *mux.Router) error {
 		Description:        "Execute a composite query over a time range. Supports builder queries (traces, logs, metrics), formulas, trace operators, PromQL, and ClickHouse SQL.",
 		Request:            new(qbtypes.QueryRangeRequest),
 		RequestContentType: "application/json",
-		RequestExamples: []handler.OpenAPIExample{
+		RequestExamples: append([]handler.OpenAPIExample{
 			{
 				Name:    "traces_time_series",
 				Summary: "Time series: count spans grouped by service",
@@ -598,7 +598,7 @@ func (provider *provider) addQuerierRoutes(router *mux.Router) error {
 					},
 				},
 			},
-		},
+		}, queryRangeFunctionExamples()...),
 		Response:            new(qbtypes.QueryRangeResponse),
 		ResponseContentType: "application/json",
 		SuccessStatusCode:   http.StatusOK,

pkg/apiserver/signozapiserver/querier_examples.go

@@ -0,0 +1,202 @@
+package signozapiserver
+
+import "github.com/SigNoz/signoz/pkg/http/handler"
+
+// timeSeriesExample wraps builder query specs in a one hour time series
+// request, the shape every function example shares.
+func timeSeriesExample(name, summary, description string, queries ...map[string]any) handler.OpenAPIExample {
+	list := make([]any, 0, len(queries))
+	for _, query := range queries {
+		list = append(list, query)
+	}
+	return handler.OpenAPIExample{
+		Name:        name,
+		Summary:     summary,
+		Description: description,
+		Value: map[string]any{
+			"schemaVersion":  "v1",
+			"start":          1640995200000,
+			"end":            1640998800000,
+			"requestType":    "time_series",
+			"compositeQuery": map[string]any{"queries": list},
+		},
+	}
+}
+
+func builderQueryEnvelope(spec map[string]any) map[string]any {
+	return map[string]any{"type": "builder_query", "spec": spec}
+}
+
+// function builds a function entry with positional arguments; anomaly is the
+// one function that takes named arguments and is written out in full.
+func function(name string, args ...any) map[string]any {
+	list := make([]any, 0, len(args))
+	for _, arg := range args {
+		list = append(list, map[string]any{"value": arg})
+	}
+	entry := map[string]any{"name": name}
+	if len(list) > 0 {
+		entry["args"] = list
+	}
+	return entry
+}
+
+// queryRangeFunctionExamples shows every function a builder query can carry,
+// what its arguments mean and the values they take. Functions run in the
+// order listed, after the aggregation and before formulas.
+func queryRangeFunctionExamples() []handler.OpenAPIExample {
+	return []handler.OpenAPIExample{
+		timeSeriesExample(
+			"function_timeshift_week_over_week",
+			"timeShift: this week's error logs next to last week's",
+			"timeShift moves every point of its query forward by the argument, in seconds, so a series read from the past lands on the current timestamps. B here reads the hour exactly seven days (604800 s) before the requested window and comes back on this week's timestamps, so it overlays A on a chart and the formula can subtract it point by point. The argument is positive; it says how far back to read.",
+			builderQueryEnvelope(map[string]any{
+				"name":         "A",
+				"signal":       "logs",
+				"aggregations": []any{map[string]any{"expression": "count()", "alias": "errors"}},
+				"stepInterval": "60s",
+				"filter":       map[string]any{"expression": "severity_text = 'ERROR'"},
+			}),
+			builderQueryEnvelope(map[string]any{
+				"name":         "B",
+				"signal":       "logs",
+				"aggregations": []any{map[string]any{"expression": "count()", "alias": "errors"}},
+				"stepInterval": "60s",
+				"filter":       map[string]any{"expression": "severity_text = 'ERROR'"},
+				"functions":    []any{function("timeShift", 604800)},
+			}),
+			map[string]any{
+				"type": "builder_formula",
+				"spec": map[string]any{"name": "F1", "expression": "A - B"},
+			},
+		),
+		timeSeriesExample(
+			"function_fillzero_sparse_errors",
+			"fillZero: a zero at every step with no failed spans",
+			"A count over sparse events returns a point only for steps that had a match, so a chart of failed spans has gaps where nothing failed. fillZero adds a zero at every missing step. With no arguments the server fills the request window at the query's step, which is the common case. The three optional arguments are start, end and step in milliseconds, for callers that want to fill a different grid.",
+			builderQueryEnvelope(map[string]any{
+				"name":         "A",
+				"signal":       "traces",
+				"aggregations": []any{map[string]any{"expression": "count()", "alias": "failed_spans"}},
+				"stepInterval": "60s",
+				"filter":       map[string]any{"expression": "has_error = true"},
+				"groupBy":      []any{map[string]any{"name": "service.name", "fieldContext": "resource"}},
+				"functions":    []any{function("fillZero")},
+			}),
+		),
+		timeSeriesExample(
+			"function_cutoffmin_hide_idle_hosts",
+			"cutOffMin: drop hosts below 5% CPU from the chart",
+			"The four threshold functions take one number. cutOffMin and cutOffMax remove the points below or above the threshold, so an idle host disappears from the chart instead of drawing a line along the bottom. clampMin and clampMax keep the points and replace the value with the threshold, which caps spikes without hiding them. system.cpu.utilization is a ratio, so 0.05 is 5%.",
+			builderQueryEnvelope(map[string]any{
+				"name":   "A",
+				"signal": "metrics",
+				"aggregations": []any{
+					map[string]any{"metricName": "system.cpu.utilization", "timeAggregation": "avg", "spaceAggregation": "avg"},
+				},
+				"stepInterval": "60s",
+				"groupBy":      []any{map[string]any{"name": "host.name", "fieldContext": "resource"}},
+				"functions":    []any{function("cutOffMin", 0.05)},
+			}),
+		),
+		timeSeriesExample(
+			"function_runningdiff_disk_growth",
+			"runningDiff then clampMin: bytes written to disk per step",
+			"runningDiff turns a gauge into the change between consecutive points, so a filesystem usage gauge becomes bytes written per step. It takes no arguments and drops the first point of the window, since that point has nothing to difference against; the server reads one step before the window to keep the first interval. Functions apply in order: clampMin with 0 runs on the differences and replaces the negative value a cleanup or a volume reset produces with zero.",
+			builderQueryEnvelope(map[string]any{
+				"name":   "A",
+				"signal": "metrics",
+				"aggregations": []any{
+					map[string]any{"metricName": "system.filesystem.usage", "timeAggregation": "latest", "spaceAggregation": "sum"},
+				},
+				"stepInterval": "60s",
+				"filter":       map[string]any{"expression": "state = 'used'"},
+				"groupBy":      []any{map[string]any{"name": "host.name", "fieldContext": "resource"}},
+				"functions":    []any{function("runningDiff"), function("clampMin", 0)},
+			}),
+		),
+		timeSeriesExample(
+			"function_ewma_smoothed_latency",
+			"ewma5: smoothed p95 latency",
+			"ewma3, ewma5 and ewma7 replace each point with an exponentially weighted average of the point and the points before it. The number in the name is the window the default weight is derived from: alpha = 2 / (n + 1), so 0.5, 0.333 and 0.25. One optional argument overrides alpha; a value close to 1 follows the raw series, a value close to 0 smooths harder. 0.3 here smooths a little more than ewma5's default.",
+			builderQueryEnvelope(map[string]any{
+				"name":   "A",
+				"signal": "metrics",
+				"aggregations": []any{
+					map[string]any{"metricName": "http.server.duration.bucket", "spaceAggregation": "p95"},
+				},
+				"stepInterval": "60s",
+				"groupBy":      []any{map[string]any{"name": "service.name", "fieldContext": "resource"}},
+				"functions":    []any{function("ewma5", 0.3)},
+			}),
+		),
+		timeSeriesExample(
+			"function_median_despiked_request_rate",
+			"median5: request rate without one-step spikes",
+			"median3, median5 and median7 replace each point with the median of the surrounding 3, 5 or 7 points. A spike that lasts one step vanishes, a shift that lasts longer than half the window stays. They take no arguments. Use them when the raw series is noisy and the chart is for a trend, not for catching the spike.",
+			builderQueryEnvelope(map[string]any{
+				"name":   "A",
+				"signal": "metrics",
+				"aggregations": []any{
+					map[string]any{"metricName": "http.server.duration.count", "timeAggregation": "rate", "spaceAggregation": "sum"},
+				},
+				"stepInterval": "60s",
+				"groupBy":      []any{map[string]any{"name": "service.name", "fieldContext": "resource"}},
+				"functions":    []any{function("median5")},
+			}),
+		),
+		timeSeriesExample(
+			"function_cumulativesum_errors_so_far",
+			"cumulativeSum: 5xx responses since the start of the window",
+			"cumulativeSum replaces each point with the sum of every point up to it, so a per-step count becomes a running total that starts at zero at the window start. It takes no arguments. The total resets with the window, so a longer time range gives a larger final value; it is a view of the window, not a counter.",
+			builderQueryEnvelope(map[string]any{
+				"name":         "A",
+				"signal":       "traces",
+				"aggregations": []any{map[string]any{"expression": "count()", "alias": "server_errors"}},
+				"stepInterval": "60s",
+				"filter":       map[string]any{"expression": "http.response.status_code >= 500"},
+				"groupBy":      []any{map[string]any{"name": "service.name", "fieldContext": "resource"}},
+				"functions":    []any{function("cumulativeSum")},
+			}),
+		),
+		timeSeriesExample(
+			"function_log10_network_bytes",
+			"log10: network throughput across devices that differ by orders of magnitude",
+			"log2 and log10 replace each value with its logarithm, so a loopback device moving kilobytes and an uplink moving gigabytes fit on one chart. They take no arguments. Zero and negative values have no logarithm and are not drawn. absolute, also without arguments, is the usual companion for a signed series before a logarithm.",
+			builderQueryEnvelope(map[string]any{
+				"name":   "A",
+				"signal": "metrics",
+				"aggregations": []any{
+					map[string]any{"metricName": "system.network.io", "timeAggregation": "rate", "spaceAggregation": "sum"},
+				},
+				"stepInterval": "60s",
+				"filter":       map[string]any{"expression": "direction = 'transmit'"},
+				"groupBy":      []any{map[string]any{"name": "device", "fieldContext": "attribute"}},
+				"functions":    []any{function("log10")},
+			}),
+		),
+		timeSeriesExample(
+			"function_anomaly_request_rate",
+			"anomaly: request rate scored against its daily pattern",
+			"anomaly is the one function with named arguments. seasonality says which past periods predict the current value: hourly, daily or weekly, daily when absent. z_score_threshold is how many standard deviations from the prediction count as an anomaly, 3 when absent. The response carries, next to the values, a predicted series, an upper and a lower bound series and an anomaly score series. The scoring runs in the enterprise build; the community build returns the series unchanged.",
+			builderQueryEnvelope(map[string]any{
+				"name":   "A",
+				"signal": "metrics",
+				"aggregations": []any{
+					map[string]any{"metricName": "http.server.duration.count", "timeAggregation": "rate", "spaceAggregation": "sum"},
+				},
+				"stepInterval": "300s",
+				"groupBy":      []any{map[string]any{"name": "service.name", "fieldContext": "resource"}},
+				"functions": []any{
+					map[string]any{
+						"name": "anomaly",
+						"args": []any{
+							map[string]any{"name": "seasonality", "value": "daily"},
+							map[string]any{"name": "z_score_threshold", "value": 3},
+						},
+					},
+				},
+			}),
+		),
+	}
+}

pkg/apiserver/signozapiserver/querier_examples_test.go

@@ -0,0 +1,21 @@
+package signozapiserver
+
+import (
+	"encoding/json"
+	"testing"
+
+	qbtypes "github.com/SigNoz/signoz/pkg/types/querybuildertypes/querybuildertypesv5"
+	"github.com/stretchr/testify/require"
+)
+
+func TestQueryRangeFunctionExamplesAreValidRequests(t *testing.T) {
+	for _, ex := range queryRangeFunctionExamples() {
+		t.Run(ex.Name, func(t *testing.T) {
+			raw, err := json.Marshal(ex.Value)
+			require.NoError(t, err)
+			var req qbtypes.QueryRangeRequest
+			require.NoError(t, json.Unmarshal(raw, &req))
+			require.NoError(t, req.Validate())
+		})
+	}
+}

pkg/querier/bucket_cache.go

@@ -26,9 +26,8 @@ const bucketCacheCellSteps = 60
 type BucketCache interface {
 	// cached portion + list of gaps to fetch
 	GetMissRanges(ctx context.Context, orgID valuer.UUID, q qbtypes.Query, step qbtypes.Step) (cached *qbtypes.Result, missing []*qbtypes.TimeRange)
-	// store fresh buckets for future hits; only the ranges in the map were
-	// queried for fresh, the rest of it came from the cache
-	Put(ctx context.Context, orgID valuer.UUID, q qbtypes.Query, step qbtypes.Step, fresh *qbtypes.Result, execStatsForQueriedTimeRanges map[qbtypes.TimeRange]qbtypes.ExecStats)
+	// store fresh buckets for future hits
+	Put(ctx context.Context, orgID valuer.UUID, q qbtypes.Query, step qbtypes.Step, fresh *qbtypes.Result)
 }
 
 // bucketCache implements the BucketCache interface.
@@ -117,7 +116,7 @@ func (bc *bucketCache) GetMissRanges(
 	}
 
 	// Merge buckets into a single result
-	mergedResult := bc.mergeBuckets(ctx, relevantBuckets, cachedWarnings, startMs, endMs)
+	mergedResult := bc.mergeBuckets(ctx, relevantBuckets, cachedWarnings)
 
 	// Filter the merged result to only include values within the requested time range
 	mergedResult = bc.filterResultToTimeRange(mergedResult, q, startMs, endMs, stepMs)
@@ -126,7 +125,7 @@ func (bc *bucketCache) GetMissRanges(
 }
 
 // Put stores fresh query results in the cache.
-func (bc *bucketCache) Put(ctx context.Context, orgID valuer.UUID, q qbtypes.Query, step qbtypes.Step, fresh *qbtypes.Result, execStatsForQueriedTimeRanges map[qbtypes.TimeRange]qbtypes.ExecStats) {
+func (bc *bucketCache) Put(ctx context.Context, orgID valuer.UUID, q qbtypes.Query, step qbtypes.Step, fresh *qbtypes.Result) {
 	// Get query window
 	startMs, endMs := q.Window()
 
@@ -189,7 +188,7 @@ func (bc *bucketCache) Put(ctx context.Context, orgID valuer.UUID, q qbtypes.Que
 
 	// Convert trimmed result to buckets, one per grid cell it spans
 	cellWidthMs := bucketCacheCellSteps * stepMs
-	freshBuckets := bc.resultToBuckets(ctx, trimmedResult, cachableStartMs, cachableEndMs, cellWidthMs, execStatsForQueriedTimeRanges)
+	freshBuckets := bc.resultToBuckets(ctx, trimmedResult, cachableStartMs, cachableEndMs, cellWidthMs)
 
 	cellKeys := make([]string, len(freshBuckets))
 	cellEntries := make([]cachetypes.Cacheable, len(freshBuckets))
@@ -211,7 +210,7 @@ func (bc *bucketCache) Put(ctx context.Context, orgID valuer.UUID, q qbtypes.Que
 			cellEntry = cellEntries[i].(*qbtypes.CachedData)
 		}
 
-		cellBuckets, changed := bc.mergeFreshBucketIntoCell(ctx, cellEntry.Buckets, freshBucket, execStatsForQueriedTimeRanges)
+		cellBuckets, changed := bc.mergeFreshBucketIntoCell(ctx, cellEntry.Buckets, freshBucket)
 		if !changed {
 			continue
 		}
@@ -464,7 +463,7 @@ func (bc *bucketCache) filterRelevantBuckets(buckets []*qbtypes.CachedBucket, st
 }
 
 // mergeBuckets combines multiple cached buckets into a single result.
-func (bc *bucketCache) mergeBuckets(ctx context.Context, buckets []*qbtypes.CachedBucket, warnings []string, startMs, endMs uint64) *qbtypes.Result {
+func (bc *bucketCache) mergeBuckets(ctx context.Context, buckets []*qbtypes.CachedBucket, warnings []string) *qbtypes.Result {
 	if len(buckets) == 0 {
 		return &qbtypes.Result{}
 	}
@@ -472,13 +471,12 @@ func (bc *bucketCache) mergeBuckets(ctx context.Context, buckets []*qbtypes.Cach
 	// All buckets should have the same type
 	resultType := buckets[0].Type
 
-	// A bucket can reach past the window; only the part inside it is served
+	// Aggregate stats
 	var totalStats qbtypes.ExecStats
 	for _, bucket := range buckets {
-		servedStats := bucket.ProrateStatsToWindow(startMs, endMs)
-		totalStats.RowsScanned += servedStats.RowsScanned
-		totalStats.BytesScanned += servedStats.BytesScanned
-		totalStats.DurationMS += servedStats.DurationMS
+		totalStats.RowsScanned += bucket.Stats.RowsScanned
+		totalStats.BytesScanned += bucket.Stats.BytesScanned
+		totalStats.DurationMS += bucket.Stats.DurationMS
 	}
 
 	// Merge values based on type
@@ -675,7 +673,7 @@ func (bc *bucketCache) isEmptyResult(result *qbtypes.Result) (isEmpty bool, isFi
 }
 
 // resultToBuckets converts a query result into time-based buckets.
-func (bc *bucketCache) resultToBuckets(ctx context.Context, result *qbtypes.Result, startMs, endMs, cellWidthMs uint64, execStatsForQueriedTimeRanges map[qbtypes.TimeRange]qbtypes.ExecStats) []*qbtypes.CachedBucket {
+func (bc *bucketCache) resultToBuckets(ctx context.Context, result *qbtypes.Result, startMs, endMs, cellWidthMs uint64) []*qbtypes.CachedBucket {
 	// Check if result is empty
 	isEmpty, isFiltered := bc.isEmptyResult(result)
 
@@ -687,7 +685,7 @@ func (bc *bucketCache) resultToBuckets(ctx context.Context, result *qbtypes.Resu
 	}
 
 	// Empty filtered results still become buckets, so the range is not re-queried
-	perCellBuckets, err := result.SplitIntoCacheCellBuckets(startMs, endMs, cellWidthMs, execStatsForQueriedTimeRanges)
+	perCellBuckets, err := result.SplitIntoCacheCellBuckets(startMs, endMs, cellWidthMs)
 	if err != nil {
 		bc.logger.ErrorContext(ctx, "failed to convert result to buckets", errors.Attr(err))
 		return nil
@@ -699,7 +697,7 @@ func (bc *bucketCache) resultToBuckets(ctx context.Context, result *qbtypes.Resu
 // and reports whether the cell changed: a bucket already covering the fresh
 // one leaves it as is, buckets inside the fresh one are replaced, buckets
 // overlapping or edge to edge with it are unioned, and the rest are kept apart.
-func (bc *bucketCache) mergeFreshBucketIntoCell(ctx context.Context, cellBuckets []*qbtypes.CachedBucket, freshBucket *qbtypes.CachedBucket, execStatsForQueriedTimeRanges map[qbtypes.TimeRange]qbtypes.ExecStats) ([]*qbtypes.CachedBucket, bool) {
+func (bc *bucketCache) mergeFreshBucketIntoCell(ctx context.Context, cellBuckets []*qbtypes.CachedBucket, freshBucket *qbtypes.CachedBucket) ([]*qbtypes.CachedBucket, bool) {
 	keptBuckets := make([]*qbtypes.CachedBucket, 0, len(cellBuckets)+1)
 	touchingBuckets := []*qbtypes.CachedBucket{}
 	for _, cellBucket := range cellBuckets {
@@ -712,7 +710,7 @@ func (bc *bucketCache) mergeFreshBucketIntoCell(ctx context.Context, cellBuckets
 			keptBuckets = append(keptBuckets, cellBucket)
 		}
 	}
-	keptBuckets = append(keptBuckets, bc.unionBuckets(ctx, freshBucket, touchingBuckets, execStatsForQueriedTimeRanges))
+	keptBuckets = append(keptBuckets, bc.unionBuckets(ctx, freshBucket, touchingBuckets))
 
 	slices.SortFunc(keptBuckets, func(a, b *qbtypes.CachedBucket) int {
 		return cmp.Compare(a.StartMs, b.StartMs)
@@ -723,16 +721,7 @@ func (bc *bucketCache) mergeFreshBucketIntoCell(ctx context.Context, cellBuckets
 // unionBuckets widens the fresh bucket by the buckets it touches. Buckets
 // inside it are replaced outright, so only a genuine partial overlap decodes
 // anything.
-func (bc *bucketCache) unionBuckets(ctx context.Context, freshBucket *qbtypes.CachedBucket, touchingBuckets []*qbtypes.CachedBucket, execStatsForQueriedTimeRanges map[qbtypes.TimeRange]qbtypes.ExecStats) *qbtypes.CachedBucket {
-	// Data queried again takes the new query's cost, which the fresh bucket
-	// carries; a touching bucket keeps its cost only for what was not queried again
-	for _, touchingBucket := range touchingBuckets {
-		keptStats := touchingBucket.ProrateStatsToUnqueried(execStatsForQueriedTimeRanges)
-		freshBucket.Stats.RowsScanned += keptStats.RowsScanned
-		freshBucket.Stats.BytesScanned += keptStats.BytesScanned
-		freshBucket.Stats.DurationMS += keptStats.DurationMS
-	}
-
+func (bc *bucketCache) unionBuckets(ctx context.Context, freshBucket *qbtypes.CachedBucket, touchingBuckets []*qbtypes.CachedBucket) *qbtypes.CachedBucket {
 	bucketsToMerge := []*qbtypes.CachedBucket{freshBucket}
 	for _, touchingBucket := range touchingBuckets {
 		if !freshBucket.Contains(touchingBucket) {
@@ -753,11 +742,14 @@ func (bc *bucketCache) unionBuckets(ctx context.Context, freshBucket *qbtypes.Ca
 		EndMs:   freshBucket.EndMs,
 		Type:    freshBucket.Type,
 		Value:   value,
-		Stats:   freshBucket.Stats,
+		Stats:   qbtypes.ExecStats{StepIntervals: freshBucket.Stats.StepIntervals},
 	}
 	for _, mergedBucket := range bucketsToMerge {
 		unionBucket.StartMs = min(unionBucket.StartMs, mergedBucket.StartMs)
 		unionBucket.EndMs = max(unionBucket.EndMs, mergedBucket.EndMs)
+		unionBucket.Stats.RowsScanned += mergedBucket.Stats.RowsScanned
+		unionBucket.Stats.BytesScanned += mergedBucket.Stats.BytesScanned
+		unionBucket.Stats.DurationMS += mergedBucket.Stats.DurationMS
 	}
 	return unionBucket
 }

pkg/querier/bucket_cache_bench_test.go

@@ -30,7 +30,7 @@ func BenchmarkBucketCache_GetMissRanges(b *testing.B) {
 			endMs:       uint64((i + 1) * 10000),
 		}
 		result := createBenchmarkResult(query.startMs, query.endMs, 1000)
-		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: query.startMs, To: query.endMs}: result.Stats})
+		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 	}
 
 	// Create test queries with varying cache hit patterns
@@ -121,7 +121,7 @@ func BenchmarkBucketCache_Put(b *testing.B) {
 
 			for i := 0; i < b.N; i++ {
 				for j := 0; j < tc.numQueries; j++ {
-					bc.Put(ctx, orgID, queries[j], qbtypes.Step{Duration: 1000 * time.Millisecond}, results[j], map[qbtypes.TimeRange]qbtypes.ExecStats{{From: queries[j].startMs, To: queries[j].endMs}: results[j].Stats})
+					bc.Put(ctx, orgID, queries[j], qbtypes.Step{Duration: 1000 * time.Millisecond}, results[j])
 				}
 			}
 		})
@@ -259,7 +259,7 @@ func BenchmarkBucketCache_ConcurrentOperations(b *testing.B) {
 			endMs:       uint64((i + 1) * 10000),
 		}
 		result := createBenchmarkResult(query.startMs, query.endMs, 1000)
-		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: query.startMs, To: query.endMs}: result.Stats})
+		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 	}
 
 	b.ResetTimer()
@@ -284,7 +284,7 @@ func BenchmarkBucketCache_ConcurrentOperations(b *testing.B) {
 					endMs:       uint64((i + 1) * 10000),
 				}
 				result := createBenchmarkResult(query.startMs, query.endMs, 1000)
-				bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: query.startMs, To: query.endMs}: result.Stats})
+				bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 			case 2: // Partial read
 				query := &mockQuery{
 					fingerprint: fmt.Sprintf("concurrent-query-%d", i%100),

pkg/querier/bucket_cache_step_test.go

@@ -59,7 +59,7 @@ func TestBucketCacheStepAlignment(t *testing.T) {
 	}
 
 	// Put result in cache
-	bc.Put(ctx, orgID, query, step, result, nil)
+	bc.Put(ctx, orgID, query, step, result)
 
 	// Get cached data
 	cached, missing := bc.GetMissRanges(ctx, orgID, query, step)
@@ -109,7 +109,7 @@ func TestBucketCacheNoStepInterval(t *testing.T) {
 	}
 
 	// Should cache the entire range when step is 0
-	bc.Put(ctx, orgID, query, step, result, nil)
+	bc.Put(ctx, orgID, query, step, result)
 
 	cached, missing := bc.GetMissRanges(ctx, orgID, query, step)
 	assert.NotNil(t, cached)

pkg/querier/bucket_cache_test.go

@@ -156,7 +156,7 @@ func TestBucketCache_Put_And_Get(t *testing.T) {
 	}
 
 	// Store in cache
-	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: query.startMs, To: query.endMs}: result.Stats})
+	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 
 	// Wait a bit for cache to be written
 	time.Sleep(10 * time.Millisecond)
@@ -189,7 +189,7 @@ func TestBucketCache_PartialHit(t *testing.T) {
 		Type:  qbtypes.RequestTypeTimeSeries,
 		Value: createTestTimeSeries("A", 1000, 3000, 1000),
 	}
-	bc.Put(context.Background(), valuer.UUID{}, query1, qbtypes.Step{Duration: 1000 * time.Millisecond}, result1, nil)
+	bc.Put(context.Background(), valuer.UUID{}, query1, qbtypes.Step{Duration: 1000 * time.Millisecond}, result1)
 
 	// Wait for cache write
 	time.Sleep(10 * time.Millisecond)
@@ -225,7 +225,7 @@ func TestBucketCache_MultipleBuckets(t *testing.T) {
 	bc.Put(context.Background(), valuer.UUID{}, query1, qbtypes.Step{Duration: 100 * time.Millisecond}, &qbtypes.Result{
 		Type:  qbtypes.RequestTypeTimeSeries,
 		Value: createTestTimeSeries("A", 1000, 2000, 100),
-	}, nil)
+	})
 
 	query2 := &mockQuery{
 		fingerprint: "test-query",
@@ -235,7 +235,7 @@ func TestBucketCache_MultipleBuckets(t *testing.T) {
 	bc.Put(context.Background(), valuer.UUID{}, query2, qbtypes.Step{Duration: 100 * time.Millisecond}, &qbtypes.Result{
 		Type:  qbtypes.RequestTypeTimeSeries,
 		Value: createTestTimeSeries("A", 3000, 4000, 100),
-	}, nil)
+	})
 
 	// Wait for cache writes
 	time.Sleep(10 * time.Millisecond)
@@ -280,7 +280,7 @@ func TestBucketCache_FluxInterval(t *testing.T) {
 	}
 
 	// This should not be cached due to flux interval
-	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 
 	// Wait a bit
 	time.Sleep(10 * time.Millisecond)
@@ -358,7 +358,7 @@ func TestBucketCache_MergeTimeSeriesResults(t *testing.T) {
 				{Series: []*qbtypes.TimeSeries{series1}},
 			},
 		},
-	}, nil)
+	})
 
 	// Cache second part
 	query2 := &mockQuery{
@@ -374,7 +374,7 @@ func TestBucketCache_MergeTimeSeriesResults(t *testing.T) {
 				{Series: []*qbtypes.TimeSeries{series2}},
 			},
 		},
-	}, nil)
+	})
 
 	// Wait for cache writes
 	time.Sleep(10 * time.Millisecond)
@@ -441,7 +441,7 @@ func TestBucketCache_RawData(t *testing.T) {
 		Value: rawData,
 	}
 
-	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 	time.Sleep(10 * time.Millisecond)
 
 	cached, missing := bc.GetMissRanges(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond})
@@ -481,7 +481,7 @@ func TestBucketCache_ScalarData(t *testing.T) {
 		Value: scalarData,
 	}
 
-	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 	time.Sleep(10 * time.Millisecond)
 
 	cached, missing := bc.GetMissRanges(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond})
@@ -509,7 +509,7 @@ func TestBucketCache_EmptyFingerprint(t *testing.T) {
 		Value: createTestTimeSeries("A", 1000, 5000, 1000),
 	}
 
-	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+	bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 	time.Sleep(10 * time.Millisecond)
 
 	// Should still be able to retrieve
@@ -564,7 +564,7 @@ func TestBucketCache_ConcurrentAccess(t *testing.T) {
 				Type:  qbtypes.RequestTypeTimeSeries,
 				Value: createTestTimeSeries(fmt.Sprintf("Q%d", id), query.startMs, query.endMs, 100),
 			}
-			bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 100 * time.Microsecond}, result, nil)
+			bc.Put(context.Background(), valuer.UUID{}, query, qbtypes.Step{Duration: 100 * time.Microsecond}, result)
 			done <- true
 		}(i)
 	}
@@ -624,7 +624,7 @@ func TestBucketCache_GetMissRanges_FluxInterval(t *testing.T) {
 		},
 	}
 
-	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, cachedResult, nil)
+	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, cachedResult)
 
 	// Get miss ranges
 	cached, missing := bc.GetMissRanges(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond})
@@ -686,7 +686,7 @@ func TestBucketCache_Put_FluxIntervalTrimming(t *testing.T) {
 	}
 
 	// Put the result
-	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: query.startMs, To: query.endMs}: result.Stats})
+	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 
 	// Retrieve cached data
 	cached, missing := bc.GetMissRanges(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond})
@@ -756,7 +756,7 @@ func TestBucketCache_Put_EntireRangeInFluxInterval(t *testing.T) {
 	}
 
 	// Put the result - should not cache anything
-	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 
 	// Try to get cached data - should have no cached data
 	cached, missing := bc.GetMissRanges(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond})
@@ -862,7 +862,7 @@ func TestBucketCache_EmptyDataHandling(t *testing.T) {
 			}
 
 			// Put the result
-			bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, tt.result, nil)
+			bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, tt.result)
 
 			// Wait a bit for cache to be written
 			time.Sleep(10 * time.Millisecond)
@@ -927,7 +927,7 @@ func TestBucketCache_PartialValues(t *testing.T) {
 	}
 
 	// Put the result
-	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: query.startMs, To: query.endMs}: result.Stats})
+	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 
 	// Wait for cache to be written
 	time.Sleep(10 * time.Millisecond)
@@ -997,7 +997,7 @@ func TestBucketCache_AllPartialValues(t *testing.T) {
 	}
 
 	// Put the result
-	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 
 	// Wait for cache to be written
 	time.Sleep(10 * time.Millisecond)
@@ -1058,7 +1058,7 @@ func TestBucketCache_FilteredCachedResults(t *testing.T) {
 	}
 
 	// Cache the wide range
-	bc.Put(ctx, orgID, query1, qbtypes.Step{Duration: 1000 * time.Millisecond}, result1, nil)
+	bc.Put(ctx, orgID, query1, qbtypes.Step{Duration: 1000 * time.Millisecond}, result1)
 	time.Sleep(10 * time.Millisecond)
 
 	// Now query for a smaller range (2000-3500ms)
@@ -1125,7 +1125,7 @@ func TestBucketCache_PromQLKeepsTheValueAtTheWindowEnd(t *testing.T) {
 				}},
 			}},
 		},
-	}, nil)
+	})
 	time.Sleep(10 * time.Millisecond)
 
 	cached, missing := bc.GetMissRanges(ctx, orgID, query, step)
@@ -1291,7 +1291,7 @@ func TestBucketCache_PartialValueDetection(t *testing.T) {
 		}
 
 		// Put the result
-		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 		time.Sleep(10 * time.Millisecond)
 
 		// Get cached data
@@ -1345,7 +1345,7 @@ func TestBucketCache_PartialValueDetection(t *testing.T) {
 		}
 
 		// Put the result
-		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 		time.Sleep(10 * time.Millisecond)
 
 		// Get cached data
@@ -1397,7 +1397,7 @@ func TestBucketCache_PartialValueDetection(t *testing.T) {
 		}
 
 		// Put the result
-		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+		bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 		time.Sleep(10 * time.Millisecond)
 
 		// Get cached data
@@ -1454,7 +1454,7 @@ func TestBucketCache_NoCache(t *testing.T) {
 	}
 
 	// Put the result in cache
-	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result, nil)
+	bc.Put(ctx, orgID, query, qbtypes.Step{Duration: 1000 * time.Millisecond}, result)
 	time.Sleep(10 * time.Millisecond)
 
 	// Verify data is cached
@@ -1475,12 +1475,11 @@ func TestBucketCachePutSlicesAtCellBoundaries(t *testing.T) {
 
 	// a 1s step gives 60s cells, so 30s to 150s spans three of them
 	query := &mockQuery{fingerprint: "sliced-query", startMs: 30000, endMs: 150000}
-	result := &qbtypes.Result{
+	bc.Put(ctx, orgID, query, step, &qbtypes.Result{
 		Type:  qbtypes.RequestTypeTimeSeries,
 		Value: createTestTimeSeries("A", 30000, 150000, 1000),
 		Stats: qbtypes.ExecStats{RowsScanned: 120, BytesScanned: 1200, DurationMS: 12},
-	}
-	bc.Put(ctx, orgID, query, step, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: 30000, To: 150000}: result.Stats})
+	})
 
 	type bucketShape struct {
 		cell, startMs, endMs, rowsScanned uint64
@@ -1540,7 +1539,7 @@ func TestBucketCachePutAbsorbsWritesWithinACell(t *testing.T) {
 			name:                  "OverlappingWrites_BecomeOne",
 			firstWindow:           qbtypes.TimeRange{From: 1000, To: 4000},
 			secondWindow:          qbtypes.TimeRange{From: 3000, To: 6000},
-			expectedBuckets:       []bucketShape{{1000, 6000, 5}},
+			expectedBuckets:       []bucketShape{{1000, 6000, 6}},
 			expectedMissingRanges: []*qbtypes.TimeRange{},
 			expectedValueCount:    5,
 		},
@@ -1571,13 +1570,12 @@ func TestBucketCachePutAbsorbsWritesWithinACell(t *testing.T) {
 
 			for _, window := range []qbtypes.TimeRange{testCase.firstWindow, testCase.secondWindow} {
 				query := &mockQuery{fingerprint: "absorb-query", startMs: window.From, endMs: window.To}
-				result := &qbtypes.Result{
+				bc.Put(ctx, orgID, query, step, &qbtypes.Result{
 					Type:  qbtypes.RequestTypeTimeSeries,
 					Value: createTestTimeSeries("A", window.From, window.To, 1000),
-					// one row per second, so a bucket's stats read as the seconds of data it holds
+					// one row per second, so a bucket's stats read as the seconds it was scanned for
 					Stats: qbtypes.ExecStats{RowsScanned: (window.To - window.From) / 1000},
-				}
-				bc.Put(ctx, orgID, query, step, result, map[qbtypes.TimeRange]qbtypes.ExecStats{window: result.Stats})
+				})
 			}
 
 			query := &mockQuery{
@@ -1611,7 +1609,7 @@ func TestBucketCachePutLeavesAnUnchangedCellAlone(t *testing.T) {
 		Type:     qbtypes.RequestTypeTimeSeries,
 		Value:    createTestTimeSeries("A", 1000, 10000, 1000),
 		Warnings: []string{"from the wide write"},
-	}, nil)
+	})
 
 	// a write already covered by the cell's bucket must not rewrite the entry,
 	// which shows as its warnings never arriving
@@ -1620,7 +1618,7 @@ func TestBucketCachePutLeavesAnUnchangedCellAlone(t *testing.T) {
 		Type:     qbtypes.RequestTypeTimeSeries,
 		Value:    createTestTimeSeries("A", 3000, 5000, 1000),
 		Warnings: []string{"from the inside write"},
-	}, nil)
+	})
 
 	var cellEntry qbtypes.CachedData
 	require.NoError(t, bc.cache.Get(ctx, orgID, bc.generateCacheKey(wide, 0), &cellEntry))

pkg/querier/querier.go

@@ -823,7 +823,7 @@ func (q *querier) executeWithCache(ctx context.Context, orgID valuer.UUID, query
 				return nil, err
 			}
 			// Store in cache for future use
-			q.bucketCache.Put(ctx, orgID, query, step, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: startMs, To: endMs}: result.Stats})
+			q.bucketCache.Put(ctx, orgID, query, step, result)
 			return result, nil
 		}
 	}
@@ -879,19 +879,16 @@ func (q *querier) executeWithCache(ctx context.Context, orgID valuer.UUID, query
 			if err != nil {
 				return nil, err
 			}
-			startMs, endMs := query.Window()
-			q.bucketCache.Put(ctx, orgID, query, step, result, map[qbtypes.TimeRange]qbtypes.ExecStats{{From: startMs, To: endMs}: result.Stats})
+			q.bucketCache.Put(ctx, orgID, query, step, result)
 			return result, nil
 		}
 	}
 
 	// Calculate total stats and filter out nil results
 	validResults := make([]*qbtypes.Result, 0, len(freshResults))
-	execStatsForQueriedTimeRanges := make(map[qbtypes.TimeRange]qbtypes.ExecStats, len(freshResults))
-	for i, result := range freshResults {
+	for _, result := range freshResults {
 		if result != nil {
 			validResults = append(validResults, result)
-			execStatsForQueriedTimeRanges[*missingRanges[i]] = result.Stats
 			totalStats.RowsScanned += result.Stats.RowsScanned
 			totalStats.BytesScanned += result.Stats.BytesScanned
 			totalStats.DurationMS += result.Stats.DurationMS
@@ -906,7 +903,7 @@ func (q *querier) executeWithCache(ctx context.Context, orgID valuer.UUID, query
 	mergedResult.Stats.DurationMS += totalStats.DurationMS
 
 	// Store merged result in cache
-	q.bucketCache.Put(ctx, orgID, query, step, mergedResult, execStatsForQueriedTimeRanges)
+	q.bucketCache.Put(ctx, orgID, query, step, mergedResult)
 
 	return mergedResult, nil
 }

pkg/types/querybuildertypes/querybuildertypesv5/cached.go

@@ -43,23 +43,6 @@ func (c *CachedBucket) Touches(other *CachedBucket) bool {
 	return c.StartMs <= other.EndMs && other.StartMs <= c.EndMs
 }
 
-// ProrateStatsToWindow returns the share of the stats for the part of the
-// bucket inside [startMs, endMs).
-func (c *CachedBucket) ProrateStatsToWindow(startMs, endMs uint64) ExecStats {
-	return c.Stats.prorate(overlapMs(c.StartMs, c.EndMs, startMs, endMs), c.EndMs-c.StartMs)
-}
-
-// ProrateStatsToUnqueried returns the share of the stats for the part of the
-// bucket no queried range covers.
-func (c *CachedBucket) ProrateStatsToUnqueried(execStatsForQueriedTimeRanges map[TimeRange]ExecStats) ExecStats {
-	widthMs := c.EndMs - c.StartMs
-	queriedMs := uint64(0)
-	for queried := range execStatsForQueriedTimeRanges {
-		queriedMs += overlapMs(c.StartMs, c.EndMs, queried.From, queried.To)
-	}
-	return c.Stats.prorate(widthMs-min(queriedMs, widthMs), widthMs)
-}
-
 // CachedData represents the full cached data for a query.
 type CachedData struct {
 	Buckets  []*CachedBucket `json:"buckets"`
@@ -105,7 +88,7 @@ func (c *CachedData) Cost() int64 {
 	return size
 }
 
-func (r *Result) SplitIntoCacheCellBuckets(startMs, endMs, cellWidthMs uint64, execStatsForQueriedTimeRanges map[TimeRange]ExecStats) ([]*CachedBucket, error) {
+func (r *Result) SplitIntoCacheCellBuckets(startMs, endMs, cellWidthMs uint64) ([]*CachedBucket, error) {
 	tsData, ok := r.Value.(*TimeSeriesData)
 	if !ok || tsData == nil {
 		return nil, errors.NewInternalf(errors.CodeInternal, "result of type %s cannot be bucketed", r.Type.StringValue())
@@ -128,39 +111,18 @@ func (r *Result) SplitIntoCacheCellBuckets(startMs, endMs, cellWidthMs uint64, e
 		if err != nil {
 			return nil, err
 		}
-		// A slice costs what was queried inside it; the rest of the result came
-		// out of the cache and is costed there already
-		sliceStats := ExecStats{StepIntervals: maps.Clone(r.Stats.StepIntervals)}
-		for queried, stats := range execStatsForQueriedTimeRanges {
-			queriedMs := overlapMs(from, to, queried.From, queried.To)
-			if queriedMs == 0 {
-				continue
-			}
-			share := stats.prorate(queriedMs, queried.To-queried.From)
-			sliceStats.RowsScanned += share.RowsScanned
-			sliceStats.BytesScanned += share.BytesScanned
-			sliceStats.DurationMS += share.DurationMS
-		}
 		cellBuckets = append(cellBuckets, &CachedBucket{
 			StartMs: from,
 			EndMs:   to,
 			Type:    r.Type,
 			Value:   value,
-			Stats:   sliceStats,
+			Stats:   r.Stats.prorate(to-from, endMs-startMs),
 		})
 		from = to
 	}
 	return cellBuckets, nil
 }
 
-// overlapMs is the length of [aFromMs, aToMs) ∩ [bFromMs, bToMs).
-func overlapMs(aFromMs, aToMs, bFromMs, bToMs uint64) uint64 {
-	if fromMs, toMs := max(aFromMs, bFromMs), min(aToMs, bToMs); fromMs < toMs {
-		return toMs - fromMs
-	}
-	return 0
-}
-
 // prorate returns the share of the stats that partMs out of wholeMs of the
 // scanned window accounts for.
 func (e ExecStats) prorate(partMs, wholeMs uint64) ExecStats {

tests/integration/tests/queriermetrics/17_cache.py

@@ -442,55 +442,3 @@ def test_builder_refreshing_a_sliding_time_range(
     assert uncached.status_code == HTTPStatus.OK, uncached.text
 
     assert_results_equal(from_cache.json(), uncached.json(), "A", "the twentieth refresh")
-
-
-def test_builder_refreshing_does_not_inflate_rows_scanned(
-    signoz: types.SigNoz,
-    create_user_admin: None,  # pylint: disable=unused-argument
-    get_token: Callable[[str, str], str],
-    insert_metrics: Callable[[list[Metrics]], None],
-) -> None:
-    metric_name = f"cache_rows_scanned_{uuid4().hex[:8]}"
-
-    # 90 minutes back so even the last refresh closes clear of the flux
-    # interval, which holds recent data out of the cache
-    start_time = datetime.fromtimestamp(int((datetime.now(tz=UTC) - timedelta(minutes=90)).timestamp()) // 60 * 60, tz=UTC)
-    start_time_ms = int(start_time.timestamp() * 1000)
-
-    query = [build_builder_query("A", metric_name, "max", "max")]
-
-    insert_metrics(
-        [
-            Metrics(
-                metric_name=metric_name,
-                labels={"service": "api"},
-                timestamp=start_time + timedelta(minutes=minute),
-                value=minute,
-                type_="Gauge",
-                is_monotonic=False,
-            )
-            for minute in range(70)
-        ]
-    )
-
-    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
-
-    # a dashboard left open on a thirty minute range, re-running a minute later
-    # each time. rowsScanned is a cost indicator, so a refresh served mostly out
-    # of the cache cannot claim more than every read made so far: the first
-    # full read and then one new minute per refresh, measured with the cache off
-    uncached_rows_scanned_so_far = 0
-    for refresh in range(10):
-        refresh_start_ms = start_time_ms + refresh * MINUTE_MS
-        refresh_end_ms = refresh_start_ms + 30 * MINUTE_MS
-
-        uncached_start_ms = refresh_start_ms if refresh == 0 else refresh_end_ms - MINUTE_MS
-        uncached_response = make_query_request(signoz, token, uncached_start_ms, refresh_end_ms, query, no_cache=True)
-        assert uncached_response.status_code == HTTPStatus.OK, f"refresh {refresh}: uncached read failed: {uncached_response.text}"
-        uncached_rows_scanned_so_far += uncached_response.json()["data"]["meta"]["rowsScanned"]
-        assert uncached_rows_scanned_so_far > 0, f"refresh {refresh}: uncached read reported no rows scanned"
-
-        cached_response = make_query_request(signoz, token, refresh_start_ms, refresh_end_ms, query, no_cache=False)
-        assert cached_response.status_code == HTTPStatus.OK, f"refresh {refresh}: cached read failed: {cached_response.text}"
-        cached_rows_scanned = cached_response.json()["data"]["meta"]["rowsScanned"]
-        assert cached_rows_scanned <= uncached_rows_scanned_so_far, f"refresh {refresh}: cached read reported {cached_rows_scanned} rows scanned, more than the {uncached_rows_scanned_so_far} rows all uncached reads scanned so far"