chore: push temporary poc for testing

docs/api/openapi.yml

@@ -3064,6 +3064,79 @@ components:
       - tags
       - spec
       type: object
+    DashboardtypesHeatmapAxes:
+      properties:
+        yScale:
+          $ref: '#/components/schemas/DashboardtypesHeatmapYScale'
+      type: object
+    DashboardtypesHeatmapChartAppearance:
+      properties:
+        colors:
+          $ref: '#/components/schemas/DashboardtypesHeatmapColors'
+      type: object
+    DashboardtypesHeatmapColorMode:
+      enum:
+      - palette
+      - opacity
+      type: string
+    DashboardtypesHeatmapColorScale:
+      enum:
+      - log
+      - sqrt
+      - linear
+      type: string
+    DashboardtypesHeatmapColors:
+      properties:
+        fill:
+          type: string
+        maxCount:
+          nullable: true
+          type: number
+        minCount:
+          nullable: true
+          type: number
+        mode:
+          $ref: '#/components/schemas/DashboardtypesHeatmapColorMode'
+        palette:
+          $ref: '#/components/schemas/DashboardtypesHeatmapPalette'
+        scale:
+          $ref: '#/components/schemas/DashboardtypesHeatmapColorScale'
+        steps:
+          type: integer
+      type: object
+    DashboardtypesHeatmapPalette:
+      enum:
+      - ice
+      - moss
+      - rust
+      - graphite
+      - ember
+      - lagoon
+      - orchid
+      - verdant
+      - lava
+      - beacon
+      type: string
+    DashboardtypesHeatmapPanelSpec:
+      properties:
+        axes:
+          $ref: '#/components/schemas/DashboardtypesHeatmapAxes'
+        chartAppearance:
+          $ref: '#/components/schemas/DashboardtypesHeatmapChartAppearance'
+        formatting:
+          $ref: '#/components/schemas/DashboardtypesPanelFormatting'
+        legend:
+          $ref: '#/components/schemas/DashboardtypesLegend'
+        visualization:
+          $ref: '#/components/schemas/DashboardtypesBasicVisualization'
+      type: object
+    DashboardtypesHeatmapYScale:
+      enum:
+      - auto
+      - linear
+      - log
+      - symlog
+      type: string
     DashboardtypesHistogramBuckets:
       properties:
         bucketCount:
@@ -3416,6 +3489,7 @@ components:
       discriminator:
         mapping:
           signoz/BarChartPanel: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesBarChartPanelSpec'
+          signoz/HeatmapPanel: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesHeatmapPanelSpec'
           signoz/HistogramPanel: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesHistogramPanelSpec'
           signoz/ListPanel: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesListPanelSpec'
           signoz/NumberPanel: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesNumberPanelSpec'
@@ -3431,6 +3505,7 @@ components:
       - $ref: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesTablePanelSpec'
       - $ref: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesHistogramPanelSpec'
       - $ref: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesListPanelSpec'
+      - $ref: '#/components/schemas/DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesHeatmapPanelSpec'
       type: object
     DashboardtypesPanelPluginKind:
       enum:
@@ -3441,6 +3516,7 @@ components:
       - signoz/TablePanel
       - signoz/HistogramPanel
       - signoz/ListPanel
+      - signoz/HeatmapPanel
       type: string
     DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesBarChartPanelSpec:
       properties:
@@ -3454,6 +3530,18 @@ components:
       - kind
       - spec
       type: object
+    DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesHeatmapPanelSpec:
+      properties:
+        kind:
+          enum:
+          - signoz/HeatmapPanel
+          type: string
+        spec:
+          $ref: '#/components/schemas/DashboardtypesHeatmapPanelSpec'
+      required:
+      - kind
+      - spec
+      type: object
     DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesHistogramPanelSpec:
       properties:
         kind:
@@ -6985,10 +7073,7 @@ components:
             $ref: '#/components/schemas/Querybuildertypesv5TimeSeries'
           type: array
         meta:
-          properties:
-            unit:
-              type: string
-          type: object
+          $ref: '#/components/schemas/Querybuildertypesv5AggregationMeta'
         predictedSeries:
           items:
             $ref: '#/components/schemas/Querybuildertypesv5TimeSeries'
@@ -7003,12 +7088,51 @@ components:
             $ref: '#/components/schemas/Querybuildertypesv5TimeSeries'
           type: array
       type: object
-    Querybuildertypesv5Bucket:
+    Querybuildertypesv5AggregationMeta:
       properties:
-        step:
-          format: double
-          type: number
+        buckets:
+          items:
+            format: double
+            type: number
+          type: array
+        unit:
+          type: string
       type: object
+    Querybuildertypesv5BucketOptions:
+      discriminator:
+        mapping:
+          linear: '#/components/schemas/Querybuildertypesv5BucketOptionsLinear'
+          log: '#/components/schemas/Querybuildertypesv5BucketOptionsLog'
+        propertyName: kind
+      oneOf:
+      - $ref: '#/components/schemas/Querybuildertypesv5BucketOptionsLinear'
+      - $ref: '#/components/schemas/Querybuildertypesv5BucketOptionsLog'
+      type: object
+    Querybuildertypesv5BucketOptionsLinear:
+      properties:
+        kind:
+          $ref: '#/components/schemas/Querybuildertypesv5BucketsKind'
+        spec:
+          $ref: '#/components/schemas/Querybuildertypesv5LinearBucketsSpec'
+      required:
+      - kind
+      - spec
+      type: object
+    Querybuildertypesv5BucketOptionsLog:
+      properties:
+        kind:
+          $ref: '#/components/schemas/Querybuildertypesv5BucketsKind'
+        spec:
+          $ref: '#/components/schemas/Querybuildertypesv5LogBucketsSpec'
+      required:
+      - kind
+      - spec
+      type: object
+    Querybuildertypesv5BucketsKind:
+      enum:
+      - linear
+      - log
+      type: string
     Querybuildertypesv5BuilderQuerySpec:
       discriminator:
         mapping:
@@ -7189,6 +7313,16 @@ components:
         value:
           type: string
       type: object
+    Querybuildertypesv5LinearBucketsSpec:
+      properties:
+        maxValue:
+          format: double
+          type: number
+        numBuckets:
+          type: integer
+      required:
+      - maxValue
+      type: object
     Querybuildertypesv5LogAggregation:
       properties:
         alias:
@@ -7196,6 +7330,12 @@ components:
         expression:
           type: string
       type: object
+    Querybuildertypesv5LogBucketsSpec:
+      properties:
+        scale:
+          nullable: true
+          type: integer
+      type: object
     Querybuildertypesv5MetricAggregation:
       properties:
         comparisonSpaceAggregationParam:
@@ -7654,6 +7794,8 @@ components:
         queries (traces, logs, metrics), formulas, joins, trace operators, PromQL,
         and ClickHouse SQL queries.
       properties:
+        bucketOptions:
+          $ref: '#/components/schemas/Querybuildertypesv5BucketOptions'
         compositeQuery:
           $ref: '#/components/schemas/Querybuildertypesv5CompositeQuery'
         end:
@@ -7753,6 +7895,7 @@ components:
       - raw
       - raw_stream
       - trace
+      - heatmap
       type: string
     Querybuildertypesv5ScalarData:
       properties:
@@ -7827,8 +7970,6 @@ components:
       type: object
     Querybuildertypesv5TimeSeriesValue:
       properties:
-        bucket:
-          $ref: '#/components/schemas/Querybuildertypesv5Bucket'
         partial:
           type: boolean
         timestamp:

heatmap-poc/main.go

@@ -0,0 +1,74 @@
+package main
+
+import (
+	"embed"
+	"io/fs"
+	"log/slog"
+	"net/http"
+	"net/http/httputil"
+	"net/url"
+	"os"
+)
+
+//go:embed static
+var embedded embed.FS
+
+// diskAssets is preferred over the embedded copy when it exists, so editing the
+// UI and reloading the page needs no rebuild. It resolves against the launch
+// config's cwd, the repo root.
+const diskAssets = "heatmap-poc/static"
+
+func main() {
+	addr := envOr("HEATMAP_POC_ADDR", "localhost:8099")
+	upstream, err := url.Parse(envOr("SIGNOZ_URL", "http://localhost:8080"))
+	if err != nil {
+		slog.Error("SIGNOZ_URL is not a URL", "error", err)
+		os.Exit(1)
+	}
+
+	apiKey := os.Getenv("SIGNOZ_API_KEY")
+	if apiKey == "" {
+		slog.Warn("SIGNOZ_API_KEY is unset, every upstream call will be rejected as unauthenticated")
+	}
+
+	mux := http.NewServeMux()
+	mux.Handle("/api/", &httputil.ReverseProxy{
+		Rewrite: func(r *httputil.ProxyRequest) {
+			r.SetURL(upstream)
+			r.Out.Host = upstream.Host
+			r.Out.Header.Set("SIGNOZ-API-KEY", apiKey)
+		},
+		ErrorHandler: func(rw http.ResponseWriter, _ *http.Request, err error) {
+			slog.Error("upstream call failed", "error", err)
+			http.Error(rw, err.Error(), http.StatusBadGateway)
+		},
+	})
+	mux.Handle("/", http.FileServerFS(assets()))
+
+	slog.Info("listening", "url", "http://"+addr, "upstream", upstream.String())
+	if err := http.ListenAndServe(addr, mux); err != nil {
+		slog.Error("server stopped", "error", err)
+		os.Exit(1)
+	}
+}
+
+func assets() fs.FS {
+	if stat, err := os.Stat(diskAssets); err == nil && stat.IsDir() {
+		slog.Info("serving the UI from disk", "dir", diskAssets)
+		return os.DirFS(diskAssets)
+	}
+	slog.Info("serving the embedded UI")
+	sub, err := fs.Sub(embedded, "static")
+	if err != nil {
+		slog.Error("embedded assets are unreadable", "error", err)
+		os.Exit(1)
+	}
+	return sub
+}
+
+func envOr(key, fallback string) string {
+	if value := os.Getenv(key); value != "" {
+		return value
+	}
+	return fallback
+}

heatmap-poc/static/app.js

@@ -0,0 +1,788 @@
+const TIME_AGGREGATIONS = ["latest", "sum", "avg", "min", "max", "count", "count_distinct", "rate", "increase"];
+const SPACE_AGGREGATIONS = ["sum", "avg", "min", "max", "count"];
+
+// the types whose samples reach the reader as plain values, so a bucket axis has
+// to be chosen for them
+const VALUE_TYPES = new Set(["gauge", "sum", "summary"]);
+
+// single hue, dark to light: a count's magnitude is the only thing it encodes
+const RAMP = ["#1c3557", "#22406c", "#284c82", "#2e5998", "#3668ae", "#4a80c4", "#689dd6", "#8dbbe6", "#b9d8f5"];
+const ZERO_FILL = "#0e1016";
+
+const PAD = { left: 78, right: 8, top: 8, bottom: 22 };
+const MAX_CHART_HEIGHT = 560;
+const MAX_JSON_CHARS = 300_000;
+
+const state = {
+	mode: "metric",
+	rows: [],
+	groupBy: [],
+	bucketKind: "default",
+	colorScale: "linear",
+	grid: null,
+	emptyReason: "Run a query to draw the heatmap.",
+	hidden: new Set(),
+	requestText: "",
+	responseText: "",
+};
+
+const catalogue = new Map();
+const attributesByMetric = new Map();
+let nextRowId = 0;
+
+const $ = (selector) => document.querySelector(selector);
+const esc = (value) => String(value).replace(/[&<>"]/g, (c) => ({ "&": "&amp;", "<": "&lt;", ">": "&gt;", '"': "&quot;" })[c]);
+
+function timeWindow() {
+	const end = Date.now();
+	return { start: end - Number($("#range").value) * 60_000, end };
+}
+
+function rowLetter(index) {
+	return String.fromCharCode(65 + index);
+}
+
+function typeSupport(type) {
+	if (VALUE_TYPES.has(type)) {
+		return { aggregations: true, buckets: true };
+	}
+	if (type === "histogram") {
+		return { aggregations: false, buckets: false, note: "Read with increase/sum over its own le labels. Bucket options are rejected." };
+	}
+	if (type === "exponentialhistogram") {
+		return { aggregations: false, buckets: false, bad: true, note: "Exponential histograms are not supported yet — this comes back 501." };
+	}
+	return { aggregations: false, buckets: false, bad: true, note: "No type is recorded for this metric, so no bucket axis can be chosen — this comes back 400." };
+}
+
+function scaleHint(scale) {
+	const perTwice = 2 ** scale;
+	return perTwice >= 1 ? `${perTwice} bucket${perTwice === 1 ? "" : "s"} per 2x` : `1 bucket per ${2 ** -scale}x`;
+}
+
+function formatNumber(value) {
+	if (!Number.isFinite(value)) {
+		return value > 0 ? "∞" : "-∞";
+	}
+	if (value === 0) {
+		return "0";
+	}
+	const magnitude = Math.abs(value);
+	if (magnitude >= 1e6 || magnitude < 1e-3) {
+		return value.toExponential(1).replace("e+", "e");
+	}
+	if (Number.isInteger(value)) {
+		return String(value);
+	}
+	const text = value.toPrecision(magnitude >= 1 ? 4 : 3);
+	return text.includes(".") ? text.replace(/0+$/, "").replace(/\.$/, "") : text;
+}
+
+function formatTime(ms, spanMs) {
+	const at = new Date(ms);
+	const clock = at.toTimeString().slice(0, 5);
+	return spanMs > 24 * 3600_000 ? `${String(at.getMonth() + 1).padStart(2, "0")}-${String(at.getDate()).padStart(2, "0")} ${clock}` : clock;
+}
+
+/* ---------- upstream ---------- */
+
+async function getJSON(path, params) {
+	const response = await fetch(`${path}?${new URLSearchParams(params)}`);
+	if (!response.ok) {
+		throw new Error(`${path} came back ${response.status}: ${(await response.text()).slice(0, 400)}`);
+	}
+	return response.json();
+}
+
+async function searchMetrics(searchText) {
+	const { start, end } = timeWindow();
+	const body = await getJSON("/api/v2/metrics", { start, end, limit: 60, searchText });
+	const metrics = body?.data?.metrics ?? [];
+	for (const metric of metrics) {
+		catalogue.set(metric.metricName, metric);
+	}
+	return metrics;
+}
+
+async function metricAttributes(metricName) {
+	if (attributesByMetric.has(metricName)) {
+		return attributesByMetric.get(metricName);
+	}
+	const { start, end } = timeWindow();
+	const body = await getJSON("/api/v2/metrics/attributes", { metricName, start, end });
+	const keys = (body?.data?.attributes ?? []).map((attribute) => attribute.key);
+	attributesByMetric.set(metricName, keys);
+	return keys;
+}
+
+/* ---------- metric rows ---------- */
+
+function addRow() {
+	state.rows.push({ id: nextRowId++, metric: "", type: "", timeAggregation: "max", spaceAggregation: "max", filter: "" });
+	renderRows();
+}
+
+function renderRows() {
+	const host = $("#rows");
+	host.textContent = "";
+
+	state.rows.forEach((row, index) => {
+		const node = $("#row-template").content.firstElementChild.cloneNode(true);
+		const input = node.querySelector(".metric-input");
+		const list = node.querySelector("datalist");
+		const badge = node.querySelector(".badge");
+		const aggregations = node.querySelector(".agg-fields");
+		const note = node.querySelector(".row-note");
+		const listId = `metrics-${row.id}`;
+
+		node.querySelector(".row-name").textContent = rowLetter(index);
+		list.id = listId;
+		input.setAttribute("list", listId);
+		input.value = row.metric;
+		node.querySelector(".filter-input").value = row.filter;
+		node.querySelector(".remove-row").hidden = state.mode !== "formula" || state.rows.length < 2;
+
+		const support = typeSupport(row.type);
+		if (row.metric) {
+			badge.hidden = false;
+			badge.textContent = row.type || "no type";
+			badge.classList.toggle("bad", Boolean(support.bad));
+			aggregations.hidden = !support.aggregations;
+			note.hidden = !support.note;
+			note.textContent = support.note ?? "";
+			note.classList.toggle("warn", Boolean(support.bad));
+		}
+
+		for (const [select, options, chosen] of [
+			[node.querySelector(".time-agg"), TIME_AGGREGATIONS, row.timeAggregation],
+			[node.querySelector(".space-agg"), SPACE_AGGREGATIONS, row.spaceAggregation],
+		]) {
+			select.innerHTML = options.map((option) => `<option value="${option}"${option === chosen ? " selected" : ""}>${option}</option>`).join("");
+		}
+
+		const fillOptions = async () => {
+			try {
+				const metrics = await searchMetrics(input.value.trim());
+				list.innerHTML = metrics.map((metric) => `<option value="${esc(metric.metricName)}" label="${esc(metric.type || "no type")}"></option>`).join("");
+			} catch (error) {
+				showBanner(error.message);
+			}
+		};
+
+		let searchTimer = 0;
+		input.addEventListener("input", () => {
+			clearTimeout(searchTimer);
+			searchTimer = setTimeout(fillOptions, 220);
+		});
+		input.addEventListener("focus", () => {
+			if (!list.children.length) {
+				fillOptions();
+			}
+		});
+
+		input.addEventListener("change", async () => {
+			const name = input.value.trim();
+			row.metric = name;
+			row.type = catalogue.get(name)?.type ?? "";
+			if (name && !catalogue.has(name)) {
+				try {
+					await searchMetrics(name);
+					row.type = catalogue.get(name)?.type ?? "";
+				} catch (error) {
+					showBanner(error.message);
+				}
+			}
+			renderRows();
+			refreshPanes();
+			refreshGroupOptions();
+		});
+
+		node.querySelector(".time-agg").addEventListener("change", (event) => {
+			row.timeAggregation = event.target.value;
+		});
+		node.querySelector(".space-agg").addEventListener("change", (event) => {
+			row.spaceAggregation = event.target.value;
+		});
+		node.querySelector(".filter-input").addEventListener("change", (event) => {
+			row.filter = event.target.value.trim();
+		});
+		node.querySelector(".remove-row").addEventListener("click", () => {
+			state.rows = state.rows.filter((candidate) => candidate.id !== row.id);
+			renderRows();
+			refreshPanes();
+			refreshGroupOptions();
+		});
+
+		host.append(node);
+	});
+}
+
+/* ---------- group by ---------- */
+
+function renderGroupChips() {
+	$("#group-chips").innerHTML = state.groupBy
+		.map((key) => `<span class="chip">${esc(key)}<button type="button" data-key="${esc(key)}" title="Remove">&times;</button></span>`)
+		.join("");
+}
+
+async function refreshGroupOptions() {
+	const metrics = state.rows.map((row) => row.metric).filter(Boolean);
+	const hint = $("#group-hint");
+	if (!metrics.length) {
+		$("#group-options").innerHTML = "";
+		hint.hidden = false;
+		hint.textContent = "Pick a metric to load its attributes.";
+		return;
+	}
+
+	try {
+		const keys = new Set();
+		for (const row of state.rows.filter((candidate) => candidate.metric)) {
+			for (const key of await metricAttributes(row.metric)) {
+				// the builder strips `le` from a histogram's group by and reads the
+				// bucket axis off it instead, so offering it here would do nothing
+				if (key !== "le" || row.type !== "histogram") {
+					keys.add(key);
+				}
+			}
+		}
+		const available = [...keys].filter((key) => !state.groupBy.includes(key)).sort();
+		$("#group-options").innerHTML = available.map((key) => `<option value="${esc(key)}"></option>`).join("");
+		hint.hidden = available.length > 0;
+		hint.textContent = available.length ? "" : "No further attributes on the selected metrics in this window.";
+	} catch (error) {
+		showBanner(error.message);
+	}
+}
+
+/* ---------- panes ---------- */
+
+// A formula is bucketed from its own output, so its inputs may be any type. In
+// metric mode the one metric decides, and an unpicked one keeps the pane up.
+function bucketsAllowed() {
+	if (state.mode === "promql") {
+		return false;
+	}
+	if (state.mode === "formula") {
+		return true;
+	}
+	const row = state.rows[0];
+	return !row?.metric || typeSupport(row.type).buckets;
+}
+
+function refreshPanes() {
+	const isPromql = state.mode === "promql";
+	const isFormula = state.mode === "formula";
+
+	$("#builder-pane").hidden = isPromql;
+	$("#promql-pane").hidden = !isPromql;
+	$("#add-row").hidden = !isFormula;
+	$("#formula-field").hidden = !isFormula;
+
+	const allowed = bucketsAllowed();
+	$("#bucket-pane").hidden = !allowed;
+	if (!allowed) {
+		setBucketKind("default");
+	}
+}
+
+function setBucketKind(kind) {
+	state.bucketKind = kind;
+	for (const button of $("#bucket-kinds").children) {
+		button.classList.toggle("on", button.dataset.kind === kind);
+	}
+	$("#bucket-default-hint").hidden = kind !== "default";
+	$("#scale-field").hidden = kind !== "log";
+	$("#linear-fields").hidden = kind !== "linear";
+	$("#scale-hint").textContent = scaleHint(Number($("#scale").value));
+}
+
+function setMode(mode) {
+	state.mode = mode;
+	for (const button of $("#modes").children) {
+		button.classList.toggle("on", button.dataset.mode === mode);
+	}
+	if (mode === "metric") {
+		state.rows = state.rows.slice(0, 1);
+	}
+	if (!state.rows.length) {
+		addRow();
+	}
+	if (mode === "formula" && state.rows.length < 2) {
+		addRow();
+	}
+	renderRows();
+	refreshPanes();
+	refreshGroupOptions();
+}
+
+/* ---------- request ---------- */
+
+function buildRequest() {
+	const { start, end } = timeWindow();
+	const step = Number($("#step").value) || 60;
+	const request = {
+		schemaVersion: "v1",
+		start,
+		end,
+		requestType: "heatmap",
+		compositeQuery: { queries: [] },
+		formatOptions: { formatTableResultForUI: false, fillGaps: false },
+		noCache: $("#no-cache").checked,
+	};
+
+	if (state.mode === "promql") {
+		const query = $("#promql").value.trim();
+		if (!query) {
+			throw new Error("Enter a PromQL query.");
+		}
+		request.compositeQuery.queries.push({ type: "promql", spec: { name: "A", query, step, disabled: false } });
+		return request;
+	}
+
+	const inFormula = state.mode === "formula";
+	state.rows.forEach((row, index) => {
+		if (!row.metric) {
+			throw new Error(`Query ${rowLetter(index)} has no metric selected.`);
+		}
+		const histogram = row.type === "histogram";
+		const spec = {
+			name: rowLetter(index),
+			signal: "metrics",
+			aggregations: [
+				{
+					metricName: row.metric,
+					timeAggregation: histogram ? "increase" : row.timeAggregation,
+					spaceAggregation: histogram ? "sum" : row.spaceAggregation,
+				},
+			],
+			stepInterval: step,
+			// only the enabled query renders the heatmap, so a formula's inputs ride
+			// along disabled
+			disabled: inFormula,
+		};
+		if (state.groupBy.length) {
+			spec.groupBy = state.groupBy.map((name) => ({ name }));
+		}
+		if (row.filter) {
+			spec.filter = { expression: row.filter };
+		}
+		request.compositeQuery.queries.push({ type: "builder_query", spec });
+	});
+
+	if (inFormula) {
+		const expression = $("#formula").value.trim();
+		if (!expression) {
+			throw new Error("A formula heatmap needs an expression.");
+		}
+		request.compositeQuery.queries.push({ type: "builder_formula", spec: { name: "F1", expression, disabled: false } });
+	}
+
+	if (state.bucketKind === "log") {
+		request.bucketOptions = { kind: "log", spec: { scale: Number($("#scale").value) } };
+	} else if (state.bucketKind === "linear") {
+		request.bucketOptions = { kind: "linear", spec: { maxValue: Number($("#max-value").value), numBuckets: Number($("#num-buckets").value) } };
+	}
+
+	return request;
+}
+
+async function run() {
+	let request;
+	try {
+		request = buildRequest();
+	} catch (error) {
+		showBanner(error.message);
+		return;
+	}
+
+	state.window = { start: request.start, end: request.end };
+	state.requestText = JSON.stringify(request, null, "\t");
+	renderJSON("#request", state.requestText);
+	$("#run").disabled = true;
+	$("#run").textContent = "Running…";
+	showBanner("");
+
+	try {
+		const response = await fetch("/api/v5/query_range", {
+			method: "POST",
+			headers: { "content-type": "application/json" },
+			body: JSON.stringify(request),
+		});
+		const text = await response.text();
+		let parsed = null;
+		try {
+			parsed = JSON.parse(text);
+			state.responseText = JSON.stringify(parsed, null, "\t");
+		} catch {
+			state.responseText = text;
+		}
+		renderJSON("#response", state.responseText);
+
+		if (!response.ok) {
+			const problem = parsed?.error;
+			const detail = (problem?.errors ?? []).map((entry) => entry.message ?? JSON.stringify(entry)).join("\n");
+			showBanner(`${response.status} ${problem?.code ?? ""}\n${problem?.message ?? text.slice(0, 600)}${detail ? `\n${detail}` : ""}`.trim());
+			setGrid(null, "The request was rejected — see the message above.");
+			return;
+		}
+
+		const warning = parsed?.data?.warning;
+		if (warning?.message) {
+			showBanner([warning.message, ...(warning.warnings ?? []).map((entry) => entry.message)].join("\n"));
+		}
+		const grid = buildGrid(parsed);
+		setGrid(grid, "The response carried no series, so there is nothing to draw.");
+	} catch (error) {
+		showBanner(error.message);
+		setGrid(null, "The request could not be sent — see the message above.");
+	} finally {
+		$("#run").disabled = false;
+		$("#run").textContent = "Run query";
+	}
+}
+
+/* ---------- response ---------- */
+
+function buildGrid(body) {
+	const results = body?.data?.data?.results ?? [];
+	const result = results.find((entry) => Array.isArray(entry?.aggregations) && entry.aggregations.length);
+	if (!result) {
+		return null;
+	}
+
+	const aggregation = result.aggregations[0];
+	const buckets = aggregation.meta?.buckets ?? [];
+	const rows = buckets.length + 1;
+	const partial = new Set();
+
+	const series = (aggregation.series ?? []).map((entry) => {
+		const labels = (entry.labels ?? []).map((label) => `${label.key?.name ?? "?"}=${label.value}`);
+		const byTs = new Map();
+		let total = 0;
+		for (const point of entry.values ?? []) {
+			const counts = point.values ?? [];
+			byTs.set(point.timestamp, counts);
+			total += counts.reduce((sum, count) => sum + count, 0);
+			if (point.partial) {
+				partial.add(point.timestamp);
+			}
+		}
+		return { key: labels.join(", ") || "(no labels)", byTs, total };
+	});
+
+	const timestamps = [...new Set(series.flatMap((entry) => [...entry.byTs.keys()]))].sort((a, b) => a - b);
+	series.sort((a, b) => b.total - a.total || a.key.localeCompare(b.key));
+
+	return { queryName: result.queryName, buckets, rows, timestamps, partial, series };
+}
+
+function setGrid(grid, emptyReason) {
+	state.grid = grid;
+	state.emptyReason = emptyReason;
+	state.hidden = new Set();
+	renderGroups();
+	renderChart();
+}
+
+function visibleSeries() {
+	return state.grid.series.filter((entry) => !state.hidden.has(entry.key));
+}
+
+function renderGroups() {
+	const grid = state.grid;
+	const card = $("#groups-card");
+	if (!grid || grid.series.length < 2) {
+		card.hidden = true;
+		return;
+	}
+
+	card.hidden = false;
+	$("#group-count").textContent = `${grid.series.length - state.hidden.size} of ${grid.series.length} shown`;
+	$("#groups").innerHTML = grid.series
+		.map(
+			(entry) => `<label><input type="checkbox" data-key="${esc(entry.key)}"${state.hidden.has(entry.key) ? "" : " checked"}>
+				<span class="name" title="${esc(entry.key)}">${esc(entry.key)}</span>
+				<span class="total">${formatNumber(entry.total)}</span></label>`,
+		)
+		.join("");
+}
+
+/* ---------- chart ---------- */
+
+function bucketRange(grid, row) {
+	if (row === grid.rows - 1) {
+		return grid.buckets.length ? `> ${formatNumber(grid.buckets[grid.buckets.length - 1])}` : "overflow, the response carried no bucket bounds";
+	}
+	const upper = formatNumber(grid.buckets[row]);
+	return row === 0 ? `<= ${upper}` : `(${formatNumber(grid.buckets[row - 1])}, ${upper}]`;
+}
+
+function colorFor(count, max) {
+	if (count <= 0) {
+		return ZERO_FILL;
+	}
+	const fraction = max <= 0 ? 1 : state.colorScale === "log" ? Math.log1p(count) / Math.log1p(max) : count / max;
+	return RAMP[Math.min(RAMP.length - 1, Math.max(0, Math.round(fraction * (RAMP.length - 1))))];
+}
+
+function renderChart() {
+	const host = $("#chart");
+	const legend = $("#legend");
+	const grid = state.grid;
+	legend.textContent = "";
+
+	if (!grid) {
+		host.innerHTML = `<p class="empty">${esc(state.emptyReason)}</p>`;
+		return;
+	}
+	if (!grid.timestamps.length) {
+		host.innerHTML = '<p class="empty">The query returned no columns.</p>';
+		return;
+	}
+
+	const columns = grid.timestamps.length;
+	const shown = visibleSeries();
+	const matrix = grid.timestamps.map((ts) => {
+		const column = new Array(grid.rows).fill(0);
+		for (const entry of shown) {
+			const counts = entry.byTs.get(ts);
+			if (!counts) {
+				continue;
+			}
+			for (let row = 0; row < grid.rows; row++) {
+				column[row] += counts[row] ?? 0;
+			}
+		}
+		return column;
+	});
+
+	const max = Math.max(0, ...matrix.flat());
+	// clientWidth carries the 12px padding on either side of #chart
+	const available = host.clientWidth - 24 - PAD.left - PAD.right;
+	const cellWidth = Math.max(3, available / columns);
+	const cellHeight = Math.min(22, Math.max(4, MAX_CHART_HEIGHT / grid.rows));
+	const plotWidth = cellWidth * columns;
+	const plotHeight = cellHeight * grid.rows;
+	const gap = cellWidth >= 7 && cellHeight >= 7 ? 1 : 0;
+	const span = grid.timestamps[columns - 1] - grid.timestamps[0];
+
+	const cells = [];
+	for (let column = 0; column < columns; column++) {
+		for (let row = 0; row < grid.rows; row++) {
+			const x = PAD.left + column * cellWidth;
+			const y = PAD.top + (grid.rows - 1 - row) * cellHeight;
+			cells.push(`<rect x="${x.toFixed(2)}" y="${y.toFixed(2)}" width="${(cellWidth - gap).toFixed(2)}" height="${(cellHeight - gap).toFixed(2)}" fill="${colorFor(matrix[column][row], max)}"/>`);
+		}
+	}
+
+	const rowStride = Math.max(1, Math.ceil(13 / cellHeight));
+	const rowLabels = [];
+	for (let row = 0; row < grid.rows; row++) {
+		if (row % rowStride !== 0 && row !== grid.rows - 1) {
+			continue;
+		}
+		const y = PAD.top + (grid.rows - 1 - row) * cellHeight + cellHeight / 2;
+		const text = row === grid.rows - 1 ? "∞" : formatNumber(grid.buckets[row]);
+		rowLabels.push(`<text class="axis-label" x="${PAD.left - 6}" y="${(y + 3.2).toFixed(2)}" text-anchor="end">${esc(text)}</text>`);
+	}
+
+	const columnStride = Math.max(1, Math.ceil(58 / cellWidth));
+	const columnLabels = [];
+	for (let column = 0; column < columns; column += columnStride) {
+		const ts = grid.timestamps[column];
+		const x = PAD.left + column * cellWidth;
+		columnLabels.push(`<text class="axis-label" x="${x.toFixed(2)}" y="${(PAD.top + plotHeight + 14).toFixed(2)}">${esc(formatTime(ts, span))}${grid.partial.has(ts) ? "*" : ""}</text>`);
+	}
+
+	host.innerHTML = `<svg width="${PAD.left + plotWidth + PAD.right}" height="${PAD.top + plotHeight + PAD.bottom}">
+		${cells.join("")}
+		<line class="axis-line" x1="${PAD.left}" y1="${PAD.top + plotHeight + 0.5}" x2="${PAD.left + plotWidth}" y2="${PAD.top + plotHeight + 0.5}"/>
+		${rowLabels.join("")}${columnLabels.join("")}
+		<rect class="cursor-cell" hidden/>
+	</svg>`;
+
+	// a heatmap cannot fill gaps, so a chart much shorter than the window asked
+	// for means those columns hold no data at all rather than being hidden
+	const asked = state.window ? state.window.end - state.window.start : span;
+	const covered = `${formatTime(grid.timestamps[0], asked)}–${formatTime(grid.timestamps[columns - 1], asked)}`;
+	const coverage =
+		state.window && span < 0.9 * asked
+			? `${covered}, the only columns with data in the ${formatTime(state.window.start, asked)}–${formatTime(state.window.end, asked)} requested`
+			: covered;
+
+	legend.innerHTML = `<span>0</span>
+		<div class="swatches"><div class="swatch" style="background:${ZERO_FILL};border:1px solid var(--line)"></div>${RAMP.map((color) => `<div class="swatch" style="background:${color}"></div>`).join("")}</div>
+		<span>${formatNumber(max)} per cell</span>
+		<span>· ${grid.rows} buckets × ${columns} columns · ${shown.length} of ${grid.series.length} series · ${esc(coverage)}${grid.partial.size ? " · * partial column" : ""}</span>`;
+
+	attachHover(host.querySelector("svg"), { grid, matrix, shown, columns, cellWidth, cellHeight, plotHeight, span, max });
+}
+
+function attachHover(svg, view) {
+	const tooltip = $("#tooltip");
+	const cursor = svg.querySelector(".cursor-cell");
+
+	svg.addEventListener("mouseleave", () => {
+		tooltip.hidden = true;
+		cursor.setAttribute("hidden", "");
+	});
+
+	svg.addEventListener("mousemove", (event) => {
+		const box = svg.getBoundingClientRect();
+		const column = Math.floor((event.clientX - box.left - PAD.left) / view.cellWidth);
+		const row = view.grid.rows - 1 - Math.floor((event.clientY - box.top - PAD.top) / view.cellHeight);
+		if (column < 0 || column >= view.columns || row < 0 || row >= view.grid.rows) {
+			tooltip.hidden = true;
+			cursor.setAttribute("hidden", "");
+			return;
+		}
+
+		cursor.removeAttribute("hidden");
+		cursor.setAttribute("x", PAD.left + column * view.cellWidth);
+		cursor.setAttribute("y", PAD.top + (view.grid.rows - 1 - row) * view.cellHeight);
+		cursor.setAttribute("width", view.cellWidth);
+		cursor.setAttribute("height", view.cellHeight);
+
+		const ts = view.grid.timestamps[column];
+		const breakdown = view.shown
+			.map((entry) => ({ key: entry.key, count: entry.byTs.get(ts)?.[row] ?? 0 }))
+			.filter((entry) => entry.count > 0)
+			.sort((a, b) => b.count - a.count);
+
+		tooltip.innerHTML = [
+			`<b>${esc(formatNumber(view.matrix[column][row]))}</b> in ${esc(bucketRange(view.grid, row))}`,
+			`${esc(new Date(ts).toTimeString().slice(0, 8))}${view.grid.partial.has(ts) ? "  (partial)" : ""}`,
+			...breakdown.slice(0, 6).map((entry) => `  ${esc(entry.key)}  ${esc(formatNumber(entry.count))}`),
+			breakdown.length > 6 ? `  … ${breakdown.length - 6} more` : "",
+		]
+			.filter(Boolean)
+			.join("\n");
+
+		tooltip.hidden = false;
+		const width = tooltip.offsetWidth;
+		tooltip.style.left = `${Math.min(event.clientX + 14, window.innerWidth - width - 8)}px`;
+		tooltip.style.top = `${Math.min(event.clientY + 14, window.innerHeight - tooltip.offsetHeight - 8)}px`;
+	});
+}
+
+/* ---------- chrome ---------- */
+
+function showBanner(message) {
+	const banner = $("#banner");
+	banner.textContent = message;
+	banner.hidden = !message;
+}
+
+function renderJSON(selector, text) {
+	$(selector).textContent = text.length > MAX_JSON_CHARS ? `${text.slice(0, MAX_JSON_CHARS)}\n… truncated for display, Copy takes the whole thing` : text;
+}
+
+function wire() {
+	$("#modes").addEventListener("click", (event) => {
+		if (event.target.dataset.mode) {
+			setMode(event.target.dataset.mode);
+		}
+	});
+
+	$("#bucket-kinds").addEventListener("click", (event) => {
+		if (event.target.dataset.kind) {
+			setBucketKind(event.target.dataset.kind);
+		}
+	});
+
+	$("#scale").addEventListener("input", () => {
+		$("#scale-hint").textContent = scaleHint(Number($("#scale").value));
+	});
+
+	$("#add-row").addEventListener("click", () => {
+		addRow();
+		refreshPanes();
+	});
+
+	$("#range").addEventListener("change", () => {
+		attributesByMetric.clear();
+		refreshGroupOptions();
+	});
+
+	$("#group-input").addEventListener("change", (event) => {
+		const key = event.target.value.trim();
+		if (key && !state.groupBy.includes(key)) {
+			state.groupBy.push(key);
+			renderGroupChips();
+			refreshGroupOptions();
+		}
+		event.target.value = "";
+	});
+
+	$("#group-chips").addEventListener("click", (event) => {
+		const key = event.target.dataset.key;
+		if (key) {
+			state.groupBy = state.groupBy.filter((candidate) => candidate !== key);
+			renderGroupChips();
+			refreshGroupOptions();
+		}
+	});
+
+	$("#form").addEventListener("submit", (event) => {
+		event.preventDefault();
+		run();
+	});
+
+	$("#color-scale").addEventListener("change", (event) => {
+		state.colorScale = event.target.value;
+		renderChart();
+	});
+
+	$("#groups").addEventListener("change", (event) => {
+		const key = event.target.dataset.key;
+		if (!key) {
+			return;
+		}
+		if (event.target.checked) {
+			state.hidden.delete(key);
+		} else {
+			state.hidden.add(key);
+		}
+		$("#group-count").textContent = `${state.grid.series.length - state.hidden.size} of ${state.grid.series.length} shown`;
+		renderChart();
+	});
+
+	$("#select-all").addEventListener("click", () => {
+		state.hidden.clear();
+		renderGroups();
+		renderChart();
+	});
+
+	$("#select-none").addEventListener("click", () => {
+		state.hidden = new Set(state.grid.series.map((entry) => entry.key));
+		renderGroups();
+		renderChart();
+	});
+
+	for (const button of document.querySelectorAll("[data-copy]")) {
+		button.addEventListener("click", async () => {
+			const text = button.dataset.copy === "request" ? state.requestText : state.responseText;
+			try {
+				await navigator.clipboard.writeText(text);
+				button.textContent = "Copied";
+			} catch {
+				button.textContent = "Copy failed";
+			}
+			setTimeout(() => {
+				button.textContent = "Copy";
+			}, 1200);
+		});
+	}
+
+	let resizeTimer = 0;
+	window.addEventListener("resize", () => {
+		clearTimeout(resizeTimer);
+		resizeTimer = setTimeout(renderChart, 120);
+	});
+}
+
+wire();
+setMode("metric");
+renderGroupChips();
+setBucketKind("default");

heatmap-poc/static/index.html

@@ -0,0 +1,168 @@
+<!doctype html>
+<html lang="en">
+<head>
+	<meta charset="utf-8">
+	<meta name="viewport" content="width=device-width, initial-scale=1">
+	<title>Heatmap POC</title>
+	<link rel="stylesheet" href="style.css">
+</head>
+<body>
+	<header>
+		<h1>Heatmap POC</h1>
+		<p class="sub"><code>POST /api/v5/query_range</code> with <code>requestType: "heatmap"</code></p>
+	</header>
+
+	<div class="layout">
+		<form id="form" class="card form">
+			<div class="segmented" id="modes">
+				<button type="button" data-mode="metric" class="on">Metric</button>
+				<button type="button" data-mode="formula">Formula</button>
+				<button type="button" data-mode="promql">PromQL</button>
+			</div>
+
+			<div class="grid-2">
+				<label>Time range
+					<select id="range">
+						<option value="15">Last 15 minutes</option>
+						<option value="60" selected>Last 1 hour</option>
+						<option value="180">Last 3 hours</option>
+						<option value="360">Last 6 hours</option>
+						<option value="1440">Last 24 hours</option>
+						<option value="10080">Last 7 days</option>
+					</select>
+				</label>
+				<label>Step (seconds)
+					<input id="step" type="number" min="1" step="1" value="60">
+				</label>
+			</div>
+
+			<section id="builder-pane">
+				<div id="rows"></div>
+				<button type="button" id="add-row" class="ghost" hidden>+ Add metric</button>
+
+				<label id="formula-field" hidden>Formula
+					<input id="formula" placeholder="A / B" autocomplete="off">
+				</label>
+
+				<div class="chips-field">
+					<span class="chips-label">Group by</span>
+					<div id="group-chips" class="chips"></div>
+					<input id="group-input" list="group-options" placeholder="Add an attribute…" autocomplete="off">
+					<datalist id="group-options"></datalist>
+					<p class="hint" id="group-hint" hidden></p>
+				</div>
+			</section>
+
+			<section id="promql-pane" hidden>
+				<label>PromQL
+					<textarea id="promql" rows="4" spellcheck="false" placeholder='sum by (le) (increase({__name__="http.server.duration.bucket"}[5m]))'></textarea>
+				</label>
+				<p class="hint">The bucket axis comes from the <code>le</code> labels the query returns, so bucket options are rejected here. A dotted metric name needs the <code>{__name__="…"}</code> form.</p>
+			</section>
+
+			<fieldset id="bucket-pane">
+				<legend>Bucket options</legend>
+				<div class="segmented small" id="bucket-kinds">
+					<button type="button" data-kind="default" class="on">Default</button>
+					<button type="button" data-kind="log">Log</button>
+					<button type="button" data-kind="linear">Linear</button>
+				</div>
+				<p class="hint" id="bucket-default-hint">Omitted from the request. The backend falls back to a log axis at scale 4.</p>
+				<label id="scale-field" hidden>Scale
+					<input id="scale" type="number" min="-4" max="4" step="1" value="4">
+					<span class="hint" id="scale-hint"></span>
+				</label>
+				<div class="grid-2" id="linear-fields" hidden>
+					<label>Max value
+						<input id="max-value" type="number" min="0" step="any" value="1000">
+					</label>
+					<label>Number of buckets
+						<input id="num-buckets" type="number" min="1" max="512" step="1" value="60">
+					</label>
+				</div>
+			</fieldset>
+
+			<div class="run-row">
+				<button type="submit" id="run">Run query</button>
+				<label class="inline"><input type="checkbox" id="no-cache" checked> Bypass cache</label>
+			</div>
+		</form>
+
+		<div class="results">
+			<p id="banner" hidden></p>
+
+			<section class="card">
+				<div class="card-head">
+					<h2>Heatmap</h2>
+					<div class="head-tools">
+						<label class="inline">Colour scale
+							<select id="color-scale">
+								<option value="linear" selected>Linear</option>
+								<option value="log">Log</option>
+							</select>
+						</label>
+					</div>
+				</div>
+				<div id="chart" class="chart"><p class="empty">Run a query to draw the heatmap.</p></div>
+				<div id="legend" class="legend"></div>
+			</section>
+
+			<section class="card" id="groups-card" hidden>
+				<div class="card-head">
+					<h2>Groups <span id="group-count" class="count"></span></h2>
+					<div class="head-tools">
+						<button type="button" class="ghost" id="select-all">All</button>
+						<button type="button" class="ghost" id="select-none">None</button>
+					</div>
+				</div>
+				<div id="groups" class="groups"></div>
+			</section>
+
+			<div class="grid-2 json-grid">
+				<section class="card">
+					<div class="card-head">
+						<h2>Request</h2>
+						<button type="button" class="ghost" data-copy="request">Copy</button>
+					</div>
+					<pre id="request" class="json"></pre>
+				</section>
+				<section class="card">
+					<div class="card-head">
+						<h2>Response</h2>
+						<button type="button" class="ghost" data-copy="response">Copy</button>
+					</div>
+					<pre id="response" class="json"></pre>
+				</section>
+			</div>
+		</div>
+	</div>
+
+	<div id="tooltip" class="tooltip" hidden></div>
+
+	<template id="row-template">
+		<div class="metric-row">
+			<span class="row-name"></span>
+			<div class="row-body">
+				<div class="metric-field">
+					<input class="metric-input" placeholder="Search metrics…" autocomplete="off" spellcheck="false">
+					<datalist></datalist>
+					<span class="badge" hidden></span>
+				</div>
+				<div class="agg-fields" hidden>
+					<label>Time
+						<select class="time-agg"></select>
+					</label>
+					<label>Space
+						<select class="space-agg"></select>
+					</label>
+				</div>
+				<input class="filter-input" placeholder='Filter, e.g. service = "api"' autocomplete="off" spellcheck="false">
+				<p class="row-note hint" hidden></p>
+			</div>
+			<button type="button" class="remove-row" title="Remove this metric">&times;</button>
+		</div>
+	</template>
+
+	<script src="app.js" type="module"></script>
+</body>
+</html>

heatmap-poc/static/style.css

@@ -0,0 +1,552 @@
+:root {
+	color-scheme: dark;
+
+	--surface: #0b0d12;
+	--card: #12151d;
+	--card-head: #171b25;
+	--line: #242a38;
+	--line-soft: #1b2030;
+
+	--ink: #e6e9f0;
+	--ink-soft: #a2abbd;
+	--ink-faint: #6b7488;
+
+	--accent: #4a80c4;
+	--accent-ink: #b9d8f5;
+	--danger: #e0736b;
+	--danger-bg: #2a1618;
+
+	--cell-zero: #0e1016;
+}
+
+* {
+	box-sizing: border-box;
+}
+
+/* the display rules below would otherwise beat the UA's [hidden] rule, and SVG
+   elements never honoured the attribute on their own */
+[hidden] {
+	display: none !important;
+}
+
+body {
+	margin: 0;
+	padding: 20px;
+	background: var(--surface);
+	color: var(--ink);
+	font: 13px/1.5 ui-sans-serif, system-ui, -apple-system, "Segoe UI", sans-serif;
+}
+
+header {
+	margin-bottom: 16px;
+}
+
+h1 {
+	margin: 0;
+	font-size: 17px;
+	font-weight: 600;
+	letter-spacing: -0.01em;
+}
+
+h2 {
+	margin: 0;
+	font-size: 12px;
+	font-weight: 600;
+	text-transform: uppercase;
+	letter-spacing: 0.06em;
+	color: var(--ink-soft);
+}
+
+.sub {
+	margin: 3px 0 0;
+	color: var(--ink-faint);
+}
+
+code {
+	font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+	font-size: 0.92em;
+	color: var(--ink-soft);
+}
+
+.layout {
+	display: grid;
+	grid-template-columns: minmax(320px, 380px) minmax(0, 1fr);
+	gap: 16px;
+	align-items: start;
+}
+
+@media (max-width: 900px) {
+	.layout {
+		grid-template-columns: minmax(0, 1fr);
+	}
+}
+
+.card {
+	background: var(--card);
+	border: 1px solid var(--line);
+	border-radius: 8px;
+	overflow: hidden;
+}
+
+.card-head {
+	display: flex;
+	align-items: center;
+	justify-content: space-between;
+	gap: 10px;
+	padding: 8px 12px;
+	background: var(--card-head);
+	border-bottom: 1px solid var(--line);
+}
+
+.head-tools {
+	display: flex;
+	align-items: center;
+	gap: 8px;
+}
+
+.count {
+	color: var(--ink-faint);
+	font-weight: 400;
+	text-transform: none;
+	letter-spacing: 0;
+}
+
+/* ---------- form ---------- */
+
+.form {
+	padding: 14px;
+	display: grid;
+	gap: 14px;
+	position: sticky;
+	top: 20px;
+}
+
+label {
+	display: grid;
+	gap: 5px;
+	font-size: 12px;
+	color: var(--ink-soft);
+}
+
+label.inline {
+	display: inline-flex;
+	align-items: center;
+	gap: 6px;
+}
+
+.grid-2 {
+	display: grid;
+	grid-template-columns: 1fr 1fr;
+	gap: 10px;
+}
+
+input,
+select,
+textarea {
+	width: 100%;
+	padding: 6px 8px;
+	background: var(--surface);
+	color: var(--ink);
+	border: 1px solid var(--line);
+	border-radius: 5px;
+	font: inherit;
+}
+
+input[type="checkbox"] {
+	width: auto;
+	accent-color: var(--accent);
+}
+
+textarea {
+	font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+	resize: vertical;
+}
+
+input:focus-visible,
+select:focus-visible,
+textarea:focus-visible,
+button:focus-visible {
+	outline: 2px solid var(--accent);
+	outline-offset: 1px;
+}
+
+fieldset {
+	margin: 0;
+	padding: 10px 12px 12px;
+	border: 1px solid var(--line);
+	border-radius: 6px;
+	display: grid;
+	gap: 10px;
+}
+
+legend {
+	padding: 0 5px;
+	font-size: 11px;
+	font-weight: 600;
+	text-transform: uppercase;
+	letter-spacing: 0.06em;
+	color: var(--ink-soft);
+}
+
+button {
+	padding: 6px 12px;
+	background: var(--accent);
+	color: #fff;
+	border: 1px solid transparent;
+	border-radius: 5px;
+	font: inherit;
+	font-weight: 500;
+	cursor: pointer;
+}
+
+button:hover {
+	filter: brightness(1.12);
+}
+
+button.ghost {
+	background: transparent;
+	color: var(--ink-soft);
+	border-color: var(--line);
+	font-weight: 400;
+}
+
+button.ghost:hover {
+	color: var(--ink);
+	border-color: var(--ink-faint);
+	filter: none;
+}
+
+button[disabled] {
+	opacity: 0.5;
+	cursor: default;
+	filter: none;
+}
+
+.segmented {
+	display: flex;
+	gap: 2px;
+	padding: 2px;
+	background: var(--surface);
+	border: 1px solid var(--line);
+	border-radius: 6px;
+}
+
+.segmented button {
+	flex: 1;
+	background: transparent;
+	color: var(--ink-soft);
+	font-weight: 400;
+}
+
+.segmented button.on {
+	background: var(--line);
+	color: var(--ink);
+}
+
+.segmented button:hover {
+	filter: none;
+	color: var(--ink);
+}
+
+.segmented.small button {
+	padding: 4px 8px;
+	font-size: 12px;
+}
+
+.hint {
+	margin: 0;
+	font-size: 11px;
+	color: var(--ink-faint);
+}
+
+.hint.warn {
+	color: var(--danger);
+}
+
+.run-row {
+	display: flex;
+	align-items: center;
+	gap: 12px;
+}
+
+.run-row button {
+	flex: 1;
+}
+
+/* ---------- metric rows ---------- */
+
+#rows {
+	display: grid;
+	gap: 8px;
+}
+
+.metric-row {
+	display: grid;
+	grid-template-columns: auto minmax(0, 1fr) auto;
+	gap: 8px;
+	align-items: start;
+	padding: 8px;
+	background: var(--surface);
+	border: 1px solid var(--line-soft);
+	border-radius: 6px;
+}
+
+.row-name {
+	width: 20px;
+	padding-top: 6px;
+	font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+	font-weight: 600;
+	color: var(--accent-ink);
+	text-align: center;
+}
+
+.row-body {
+	display: grid;
+	gap: 8px;
+	min-width: 0;
+}
+
+.metric-field {
+	display: flex;
+	align-items: center;
+	gap: 6px;
+	min-width: 0;
+}
+
+.badge {
+	flex: none;
+	padding: 2px 6px;
+	background: var(--line);
+	border-radius: 4px;
+	font-size: 10px;
+	font-weight: 600;
+	text-transform: uppercase;
+	letter-spacing: 0.04em;
+	color: var(--accent-ink);
+	white-space: nowrap;
+}
+
+.badge.bad {
+	background: var(--danger-bg);
+	color: var(--danger);
+}
+
+.agg-fields {
+	display: grid;
+	grid-template-columns: 1fr 1fr;
+	gap: 8px;
+}
+
+.remove-row {
+	padding: 2px 7px;
+	background: transparent;
+	color: var(--ink-faint);
+	border-color: transparent;
+	font-size: 15px;
+	line-height: 1.2;
+}
+
+.remove-row:hover {
+	color: var(--danger);
+	filter: none;
+}
+
+/* ---------- group by chips ---------- */
+
+.chips-field {
+	display: grid;
+	gap: 6px;
+}
+
+.chips-label {
+	font-size: 12px;
+	color: var(--ink-soft);
+}
+
+.chips {
+	display: flex;
+	flex-wrap: wrap;
+	gap: 5px;
+}
+
+.chips:empty {
+	display: none;
+}
+
+.chip {
+	display: inline-flex;
+	align-items: center;
+	gap: 5px;
+	padding: 2px 4px 2px 8px;
+	background: var(--line);
+	border-radius: 11px;
+	font-size: 12px;
+}
+
+.chip button {
+	padding: 0 3px;
+	background: transparent;
+	color: var(--ink-faint);
+	border: 0;
+	font-size: 13px;
+	line-height: 1;
+}
+
+.chip button:hover {
+	color: var(--danger);
+	filter: none;
+}
+
+/* ---------- results ---------- */
+
+.results {
+	display: grid;
+	gap: 16px;
+	min-width: 0;
+}
+
+#banner {
+	margin: 0;
+	padding: 10px 12px;
+	background: var(--danger-bg);
+	border: 1px solid #4a2427;
+	border-radius: 8px;
+	color: var(--danger);
+	white-space: pre-wrap;
+}
+
+.chart {
+	padding: 12px;
+	overflow: auto;
+	max-height: 620px;
+}
+
+.chart svg {
+	display: block;
+}
+
+.empty {
+	margin: 0;
+	padding: 28px 0;
+	color: var(--ink-faint);
+	text-align: center;
+}
+
+.axis-label {
+	fill: var(--ink-faint);
+	font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+	font-size: 10px;
+}
+
+.axis-line {
+	stroke: var(--line);
+	stroke-width: 1;
+}
+
+.cursor-cell {
+	fill: none;
+	stroke: var(--accent-ink);
+	stroke-width: 1.5;
+	pointer-events: none;
+}
+
+.legend {
+	display: flex;
+	align-items: center;
+	gap: 10px;
+	padding: 0 12px 12px;
+	color: var(--ink-faint);
+	font-size: 11px;
+}
+
+.legend:empty {
+	display: none;
+}
+
+.legend .swatches {
+	display: flex;
+	gap: 2px;
+}
+
+.legend .swatch {
+	width: 20px;
+	height: 10px;
+	border-radius: 2px;
+}
+
+.groups {
+	display: grid;
+	grid-template-columns: repeat(auto-fill, minmax(220px, 1fr));
+	gap: 2px 12px;
+	padding: 10px 12px;
+	max-height: 220px;
+	overflow: auto;
+}
+
+.groups label {
+	display: flex;
+	align-items: center;
+	gap: 7px;
+	min-width: 0;
+	padding: 2px 0;
+	color: var(--ink);
+	font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+	font-size: 11px;
+	cursor: pointer;
+}
+
+.groups .name {
+	overflow: hidden;
+	text-overflow: ellipsis;
+	white-space: nowrap;
+}
+
+.groups .total {
+	margin-left: auto;
+	flex: none;
+	color: var(--ink-faint);
+}
+
+.json-grid {
+	align-items: start;
+}
+
+@media (max-width: 1200px) {
+	.json-grid {
+		grid-template-columns: minmax(0, 1fr);
+	}
+}
+
+.json {
+	margin: 0;
+	padding: 12px;
+	max-height: 420px;
+	overflow: auto;
+	font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+	font-size: 11px;
+	line-height: 1.55;
+	color: var(--ink-soft);
+	white-space: pre;
+	tab-size: 2;
+}
+
+.tooltip {
+	position: fixed;
+	z-index: 10;
+	max-width: 320px;
+	padding: 7px 9px;
+	background: #1c212e;
+	border: 1px solid var(--line);
+	border-radius: 6px;
+	box-shadow: 0 6px 18px rgb(0 0 0 / 45%);
+	font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+	font-size: 11px;
+	line-height: 1.6;
+	pointer-events: none;
+	white-space: pre;
+}
+
+.tooltip b {
+	color: var(--accent-ink);
+	font-weight: 600;
+}

pkg/querier/bucket_cache.go

@@ -451,7 +451,7 @@ func (bc *bucketCache) mergeBuckets(ctx context.Context, buckets []*qbtypes.Cach
 	// Merge values based on type
 	var mergedValue any
 	switch resultType {
-	case qbtypes.RequestTypeTimeSeries:
+	case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 		mergedValue = bc.mergeTimeSeriesValues(ctx, buckets)
 		// Raw and Scalar types are not cached, so no merge needed
 	}
@@ -476,14 +476,36 @@ func (bc *bucketCache) mergeTimeSeriesValues(ctx context.Context, buckets []*qbt
 	}
 	seriesMap := make(map[seriesKey]*qbtypes.TimeSeries, estimatedSeries)
 
+	decodedTimeSeriesData := make([]*qbtypes.TimeSeriesData, 0, len(buckets))
+
+	// Alias and Meta are taken from whichever cached bucket covers the latest
+	// range, and the buckets do not arrive in StartMs order, so keep the winner
+	// per AggregationBucket.Index alongside the StartMs that won it.
+	aggregationIndexToLatest := map[int]*qbtypes.AggregationBucket{}
+	aggregationIndexToLatestStartMs := map[int]uint64{}
+
 	for _, bucket := range buckets {
 		var tsData *qbtypes.TimeSeriesData
 		if err := json.Unmarshal(bucket.Value, &tsData); err != nil {
 			bc.logger.ErrorContext(ctx, "failed to unmarshal time series data", errors.Attr(err))
 			continue
 		}
+		decodedTimeSeriesData = append(decodedTimeSeriesData, tsData)
 
 		for _, aggBucket := range tsData.Aggregations {
+			if _, seen := aggregationIndexToLatest[aggBucket.Index]; !seen || bucket.StartMs >= aggregationIndexToLatestStartMs[aggBucket.Index] {
+				aggregationIndexToLatest[aggBucket.Index] = aggBucket
+				aggregationIndexToLatestStartMs[aggBucket.Index] = bucket.StartMs
+			}
+		}
+	}
+
+	mergedUpperBounds := qbtypes.MergeBucketUpperBounds(decodedTimeSeriesData...)
+
+	for _, tsData := range decodedTimeSeriesData {
+		for _, aggBucket := range tsData.Aggregations {
+			aggBucket.ReindexValuesToNewUpperBounds(mergedUpperBounds[aggBucket.Index])
+
 			for _, series := range aggBucket.Series {
 				// Create series key from labels
 				key := seriesKey{
@@ -556,10 +578,15 @@ func (bc *bucketCache) mergeTimeSeriesValues(ctx context.Context, buckets []*qbt
 			}
 		}
 
-		result.Aggregations = append(result.Aggregations, &qbtypes.AggregationBucket{
+		aggBucket := &qbtypes.AggregationBucket{
 			Index:  index,
 			Series: seriesList,
-		})
+		}
+		if latest, ok := aggregationIndexToLatest[index]; ok {
+			aggBucket.Alias = latest.Alias
+			aggBucket.Meta = latest.Meta
+		}
+		result.Aggregations = append(result.Aggregations, aggBucket)
 	}
 
 	return result
@@ -572,7 +599,7 @@ func (bc *bucketCache) isEmptyResult(result *qbtypes.Result) (isEmpty bool, isFi
 	}
 
 	switch result.Type {
-	case qbtypes.RequestTypeTimeSeries:
+	case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 		if tsData, ok := result.Value.(*qbtypes.TimeSeriesData); ok {
 			// No aggregations at all means truly empty
 			if len(tsData.Aggregations) == 0 {
@@ -699,14 +726,19 @@ func (bc *bucketCache) trimResultToFluxBoundary(result *qbtypes.Result, fluxBoun
 	}
 
 	switch result.Type {
-	case qbtypes.RequestTypeTimeSeries:
+	case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 		// Trim time series data
 		if tsData, ok := result.Value.(*qbtypes.TimeSeriesData); ok && tsData != nil {
 			trimmedData := &qbtypes.TimeSeriesData{}
 
 			for _, aggBucket := range tsData.Aggregations {
+				// Meta has to survive the trim: a heatmap's counts are
+				// positional against Meta.Buckets, so a cached bucket that
+				// lost its axis cannot be read back against anything.
 				trimmedBucket := &qbtypes.AggregationBucket{
 					Index: aggBucket.Index,
+					Alias: aggBucket.Alias,
+					Meta:  aggBucket.Meta,
 				}
 
 				for _, series := range aggBucket.Series {
@@ -766,7 +798,7 @@ func (bc *bucketCache) filterResultToTimeRange(result *qbtypes.Result, startMs,
 	}
 
 	switch result.Type {
-	case qbtypes.RequestTypeTimeSeries:
+	case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 		if tsData, ok := result.Value.(*qbtypes.TimeSeriesData); ok {
 			filteredData := &qbtypes.TimeSeriesData{
 				Aggregations: make([]*qbtypes.AggregationBucket, 0, len(tsData.Aggregations)),

pkg/querier/builder_query.go

@@ -92,6 +92,10 @@ func (q *builderQuery[T]) Fingerprint() string {
 	// This needs to include all fields that affect the query results
 	parts := []string{q.queryType.StringValue()}
 
+	// A heatmap and a time series query can share every spec field and still
+	// return different rows, so the request type has to separate their entries
+	parts = append(parts, fmt.Sprintf("requestType=%s", q.kind.StringValue()))
+
 	// Add signal type
 	parts = append(parts, fmt.Sprintf("signal=%s", q.spec.Signal.StringValue()))
 
@@ -130,6 +134,9 @@ func (q *builderQuery[T]) Fingerprint() string {
 					}
 					part += ":" + route
 				}
+				if a.HeatmapBucketing != nil {
+					part += ":" + fingerprintHeatmapBucketing(*a.HeatmapBucketing)
+				}
 				aggParts = append(aggParts, part)
 			}
 		}
@@ -185,6 +192,16 @@ func (q *builderQuery[T]) Fingerprint() string {
 	return strings.Join(parts, "&")
 }
 
+// fingerprintHeatmapBucketing captures only what changes the rows ClickHouse
+// returns, which is why LogBucketsSpec.Scale is absent: coarsening it happens in
+// postprocessing, so every scale reads one cache entry.
+func fingerprintHeatmapBucketing(b qbtypes.HeatmapBucketing) string {
+	if b.Kind == qbtypes.BucketsKindLinear {
+		return fmt.Sprintf("%s:%v:%d", b.Kind.StringValue(), b.MaxValue, b.NumBuckets)
+	}
+	return b.Kind.StringValue()
+}
+
 func fingerprintGroupByKey(gb qbtypes.GroupByKey) string {
 	return fingerprintFieldKey(gb.TelemetryFieldKey)
 }
@@ -412,7 +429,7 @@ func (q *builderQuery[T]) narrowWindowByTraceID(ctx context.Context, fromMS, toM
 func emptyResultFor(kind qbtypes.RequestType, queryName string) *qbtypes.Result {
 	var value any
 	switch kind {
-	case qbtypes.RequestTypeTimeSeries:
+	case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 		value = &qbtypes.TimeSeriesData{QueryName: queryName}
 	case qbtypes.RequestTypeScalar:
 		value = &qbtypes.ScalarData{QueryName: queryName}
@@ -465,8 +482,9 @@ func (q *builderQuery[T]) executeWithContext(ctx context.Context, query string,
 	queryWindow := &qbtypes.TimeRange{From: q.fromMS, To: q.toMS}
 
 	kind := q.kind
-	// all metric queries are time series then reduced if required
-	if q.spec.Signal == telemetrytypes.SignalMetrics {
+	// all metric queries are time series then reduced if required, except
+	// heatmaps, whose statement returns a row per bucket rather than per point
+	if q.spec.Signal == telemetrytypes.SignalMetrics && kind != qbtypes.RequestTypeHeatmap {
 		kind = qbtypes.RequestTypeTimeSeries
 	}
 

pkg/querier/builder_query_test.go

@@ -6,6 +6,7 @@ import (
 	"time"
 
 	"github.com/SigNoz/signoz/pkg/querybuilder"
+	"github.com/SigNoz/signoz/pkg/types/metrictypes"
 	qbtypes "github.com/SigNoz/signoz/pkg/types/querybuildertypes/querybuildertypesv5"
 	"github.com/SigNoz/signoz/pkg/types/telemetrytypes"
 	"github.com/stretchr/testify/assert"
@@ -120,6 +121,170 @@ func TestBuilderQueryFingerprintQueryType(t *testing.T) {
 	assert.Empty(t, ai.Fingerprint())
 }
 
+func TestBuilderQueryFingerprintHeatmapBucketing(t *testing.T) {
+	coarseLogScale := 1
+
+	testCases := []struct {
+		description   string
+		left          *builderQuery[qbtypes.MetricAggregation]
+		right         *builderQuery[qbtypes.MetricAggregation]
+		expectedEqual bool
+	}{
+		{
+			// fingerprintHeatmapBucketing leaves LogScale out, so the two are
+			// indistinguishable here by design
+			description: "a coarser logScale reads the same cache entry",
+			left: &builderQuery[qbtypes.MetricAggregation]{
+				queryType: qbtypes.QueryTypeBuilder,
+				kind:      qbtypes.RequestTypeHeatmap,
+				spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+					Signal: telemetrytypes.SignalMetrics,
+					Aggregations: []qbtypes.MetricAggregation{{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{Kind: qbtypes.BucketsKindLog, LogScale: qbtypes.MaxLogScale, NumBuckets: qbtypes.DefaultNumBuckets},
+					}},
+				},
+			},
+			right: &builderQuery[qbtypes.MetricAggregation]{
+				queryType: qbtypes.QueryTypeBuilder,
+				kind:      qbtypes.RequestTypeHeatmap,
+				spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+					Signal: telemetrytypes.SignalMetrics,
+					Aggregations: []qbtypes.MetricAggregation{{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{Kind: qbtypes.BucketsKindLog, LogScale: qbtypes.MaxLogScale, NumBuckets: qbtypes.DefaultNumBuckets},
+					}},
+				},
+			},
+			expectedEqual: true,
+		},
+		{
+			description: "linear separates on maxValue",
+			left: &builderQuery[qbtypes.MetricAggregation]{
+				queryType: qbtypes.QueryTypeBuilder,
+				kind:      qbtypes.RequestTypeHeatmap,
+				spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+					Signal: telemetrytypes.SignalMetrics,
+					Aggregations: []qbtypes.MetricAggregation{{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{Kind: qbtypes.BucketsKindLinear, MaxValue: 500, NumBuckets: 25},
+					}},
+				},
+			},
+			right: &builderQuery[qbtypes.MetricAggregation]{
+				queryType: qbtypes.QueryTypeBuilder,
+				kind:      qbtypes.RequestTypeHeatmap,
+				spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+					Signal: telemetrytypes.SignalMetrics,
+					Aggregations: []qbtypes.MetricAggregation{{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{Kind: qbtypes.BucketsKindLinear, MaxValue: 800, NumBuckets: 25},
+					}},
+				},
+			},
+			expectedEqual: false,
+		},
+		{
+			description: "linear separates on numBuckets",
+			left: &builderQuery[qbtypes.MetricAggregation]{
+				queryType: qbtypes.QueryTypeBuilder,
+				kind:      qbtypes.RequestTypeHeatmap,
+				spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+					Signal: telemetrytypes.SignalMetrics,
+					Aggregations: []qbtypes.MetricAggregation{{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{Kind: qbtypes.BucketsKindLinear, MaxValue: 500, NumBuckets: 25},
+					}},
+				},
+			},
+			right: &builderQuery[qbtypes.MetricAggregation]{
+				queryType: qbtypes.QueryTypeBuilder,
+				kind:      qbtypes.RequestTypeHeatmap,
+				spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+					Signal: telemetrytypes.SignalMetrics,
+					Aggregations: []qbtypes.MetricAggregation{{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{Kind: qbtypes.BucketsKindLinear, MaxValue: 500, NumBuckets: 40},
+					}},
+				},
+			},
+			expectedEqual: false,
+		},
+		{
+			description: "linear and log are separate entries",
+			left: &builderQuery[qbtypes.MetricAggregation]{
+				queryType: qbtypes.QueryTypeBuilder,
+				kind:      qbtypes.RequestTypeHeatmap,
+				spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+					Signal: telemetrytypes.SignalMetrics,
+					Aggregations: []qbtypes.MetricAggregation{{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{Kind: qbtypes.BucketsKindLinear, MaxValue: 500, NumBuckets: 25},
+					}},
+				},
+			},
+			right: &builderQuery[qbtypes.MetricAggregation]{
+				queryType: qbtypes.QueryTypeBuilder,
+				kind:      qbtypes.RequestTypeHeatmap,
+				spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+					Signal: telemetrytypes.SignalMetrics,
+					Aggregations: []qbtypes.MetricAggregation{{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{Kind: qbtypes.BucketsKindLog, LogScale: qbtypes.MaxLogScale, NumBuckets: qbtypes.DefaultNumBuckets},
+					}},
+				},
+			},
+			expectedEqual: false,
+		},
+	}
+
+	for _, testCase := range testCases {
+		t.Run(testCase.description, func(t *testing.T) {
+			if testCase.expectedEqual {
+				assert.Equal(t, testCase.left.Fingerprint(), testCase.right.Fingerprint())
+				return
+			}
+			assert.NotEqual(t, testCase.left.Fingerprint(), testCase.right.Fingerprint())
+		})
+	}
+
+	t.Run("a coarser scale is carried but reads the same cache entry", func(t *testing.T) {
+		finest := (&qbtypes.BucketOptions{Kind: qbtypes.BucketsKindLog, Spec: qbtypes.LogBucketsSpec{}}).ToHeatmapBucketing()
+		coarse := (&qbtypes.BucketOptions{Kind: qbtypes.BucketsKindLog, Spec: qbtypes.LogBucketsSpec{Scale: &coarseLogScale}}).ToHeatmapBucketing()
+
+		assert.NotEqual(t, finest.LogScale, coarse.LogScale)
+		assert.Equal(t, fingerprintHeatmapBucketing(finest), fingerprintHeatmapBucketing(coarse))
+	})
+
+	t.Run("a histogram folds in no bucket options at all", func(t *testing.T) {
+		// resolveHeatmapBucketing leaves histograms nil, so bucketOptions sent
+		// alongside one must not fragment its cache
+		histogram := &builderQuery[qbtypes.MetricAggregation]{
+			queryType: qbtypes.QueryTypeBuilder,
+			kind:      qbtypes.RequestTypeHeatmap,
+			spec: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+				Signal: telemetrytypes.SignalMetrics,
+				Aggregations: []qbtypes.MetricAggregation{{
+					MetricName: "signoz_latency",
+					Type:       metrictypes.HistogramType,
+				}},
+			},
+		}
+
+		fingerprint := histogram.Fingerprint()
+		assert.NotContains(t, fingerprint, qbtypes.BucketsKindLog.StringValue())
+		assert.NotContains(t, fingerprint, qbtypes.BucketsKindLinear.StringValue())
+	})
+}
+
 func TestMakeBucketsOrder(t *testing.T) {
 	// Test that makeBuckets returns buckets in reverse chronological order by default
 	// Using milliseconds as input - need > 1 hour range to get multiple buckets

pkg/querier/consume.go

@@ -31,6 +31,11 @@ var (
 	// written clickhouse query. The column alias indcate which value is
 	// to be considered as final result (or target).
 	legacyReservedColumnTargetAliases = []string{"__result", "__value", "result", "res", "value"}
+
+	// userHeatmapBucketColumn is the alias a user written clickhouse query can
+	// give its bucket upper bound column, alongside the HeatmapBucketColumn the
+	// statement builder emits.
+	userHeatmapBucketColumn = "bucket"
 )
 
 // stripKeyAlias removes the __SELECT_KEY_<n>_ / __GROUP_BY_KEY_<n>_ prefix from a result
@@ -83,6 +88,8 @@ func consume(rows driver.Rows, kind qbtypes.RequestType, queryWindow *qbtypes.Ti
 		payload, err = readAsTimeSeries(rows, queryWindow, step, queryName)
 	case qbtypes.RequestTypeScalar:
 		payload, err = readAsScalar(rows, queryName)
+	case qbtypes.RequestTypeHeatmap:
+		payload, err = readAsHeatmap(rows, queryWindow, step, queryName)
 	case qbtypes.RequestTypeRaw, qbtypes.RequestTypeTrace, qbtypes.RequestTypeRawStream:
 		payload, err = readAsRaw(rows, queryName)
 		// TODO: add support for other request types
@@ -112,35 +119,6 @@ func readAsTimeSeries(rows driver.Rows, queryWindow *qbtypes.TimeRange, step qbt
 
 	stepMs := uint64(step.Milliseconds())
 
-	// Helper function to check if a timestamp represents a partial value
-	isPartialValue := func(timestamp int64) bool {
-		if stepMs == 0 || queryWindow == nil {
-			return false
-		}
-
-		timestampMs := uint64(timestamp)
-
-		// For the first interval, check if query start is misaligned
-		// The first complete interval starts at the first timestamp >= queryWindow.From that is aligned to step
-		firstCompleteInterval := queryWindow.From
-		if queryWindow.From%stepMs != 0 {
-			// Round up to next step boundary
-			firstCompleteInterval = ((queryWindow.From / stepMs) + 1) * stepMs
-		}
-
-		// If timestamp is before the first complete interval, it's partial
-		if timestampMs < firstCompleteInterval {
-			return true
-		}
-
-		// For the last interval, check if it would extend beyond query end
-		if timestampMs+stepMs > queryWindow.To {
-			return queryWindow.To%stepMs != 0
-		}
-
-		return false
-	}
-
 	// Pre-allocate for labels based on column count
 	lblValsCapacity := len(colNames) - 1 // -1 for timestamp
 	if lblValsCapacity < 0 {
@@ -271,7 +249,7 @@ func readAsTimeSeries(rows driver.Rows, queryWindow *qbtypes.TimeRange, step qbt
 			series.Values = append(series.Values, &qbtypes.TimeSeriesValue{
 				Timestamp: ts,
 				Value:     val,
-				Partial:   isPartialValue(ts),
+				Partial:   isPartialValue(ts, queryWindow, stepMs),
 			})
 		}
 	}
@@ -315,6 +293,120 @@ func readAsTimeSeries(rows driver.Rows, queryWindow *qbtypes.TimeRange, step qbt
 	}, nil
 }
 
+func isHeatmapBucketColumn(colName string) bool {
+	name := stripKeyAlias(colName)
+	return name == qbtypes.HeatmapBucketColumn || name == userHeatmapBucketColumn
+}
+
+// readAsHeatmap folds one row per cell — (timestamp, group labels, bucket upper
+// bound, count) — into one series per group.
+func readAsHeatmap(rows driver.Rows, queryWindow *qbtypes.TimeRange, step qbtypes.Step, queryName string) (*qbtypes.TimeSeriesData, error) {
+	colTypes := rows.ColumnTypes()
+	colNames := rows.Columns()
+
+	if !slices.ContainsFunc(colNames, isHeatmapBucketColumn) {
+		// there is no heatmap bucket column so empty response is returned.
+		return &qbtypes.TimeSeriesData{QueryName: queryName}, nil
+	}
+
+	slots := make([]any, len(colTypes))
+	for i, ct := range colTypes {
+		slots[i] = reflect.New(ct.ScanType()).Interface()
+	}
+
+	stepMs := uint64(step.Milliseconds())
+
+	accumulator := newHeatmapAccumulator()
+
+	for rows.Next() {
+		if err := rows.Scan(slots...); err != nil {
+			return nil, err
+		}
+
+		var (
+			ts         int64
+			upperBound float64
+			count      float64
+			lblVals    []string
+			lblObjs    []*qbtypes.Label
+		)
+
+		for idx, ptr := range slots {
+			name := stripKeyAlias(colNames[idx])
+			value := derefValue(ptr)
+
+			if t, ok := value.(time.Time); ok {
+				ts = t.UnixMilli()
+				continue
+			}
+
+			switch name {
+			case qbtypes.HeatmapBucketColumn, userHeatmapBucketColumn:
+				upperBound = numericAsFloat(value)
+			default:
+				if aggRe.MatchString(name) || slices.Contains(legacyReservedColumnTargetAliases, name) {
+					count = numericAsFloat(value)
+					continue
+				}
+				// a nullable label column comes back as a nil any, which would
+				// otherwise key the series on the literal "<nil>"
+				if value == nil {
+					value = ""
+				}
+				lblVals = append(lblVals, fmt.Sprint(value))
+				lblObjs = append(lblObjs, &qbtypes.Label{
+					Key:   telemetrytypes.TelemetryFieldKey{Name: name},
+					Value: value,
+				})
+			}
+		}
+
+		if ts == 0 || !isValidBucketUpperBound(upperBound) || math.IsNaN(count) || math.IsInf(count, 0) {
+			continue
+		}
+		sort.Strings(lblVals)
+		labelsKey := strings.Join(lblVals, ",")
+
+		accumulator.addCell(labelsKey, lblObjs, ts, upperBound, count)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, err
+	}
+
+	return accumulator.foldSeries(queryWindow, stepMs, queryName), nil
+}
+
+// isPartialValue reports whether the step interval starting at timestamp is only
+// partly covered by the query window, which happens when the window boundaries
+// are not step-aligned.
+func isPartialValue(timestamp int64, queryWindow *qbtypes.TimeRange, stepMs uint64) bool {
+	if stepMs == 0 || queryWindow == nil {
+		return false
+	}
+
+	timestampMs := uint64(timestamp)
+
+	// For the first interval, check if query start is misaligned
+	// The first complete interval starts at the first timestamp >= queryWindow.From that is aligned to step
+	firstCompleteInterval := queryWindow.From
+	if queryWindow.From%stepMs != 0 {
+		// Round up to next step boundary
+		firstCompleteInterval = ((queryWindow.From / stepMs) + 1) * stepMs
+	}
+
+	// If timestamp is before the first complete interval, it's partial
+	if timestampMs < firstCompleteInterval {
+		return true
+	}
+
+	// For the last interval, check if it would extend beyond query end
+	if timestampMs+stepMs > queryWindow.To {
+		return queryWindow.To%stepMs != 0
+	}
+
+	return false
+}
+
 func isNumericKind(t reflect.Type) bool {
 	if t == nil {
 		return false

pkg/querier/heatmap_accumulator.go

@@ -0,0 +1,116 @@
+package querier
+
+import (
+	"math"
+	"slices"
+
+	qbtypes "github.com/SigNoz/signoz/pkg/types/querybuildertypes/querybuildertypesv5"
+)
+
+// heatmapColumn maps a bucket's upper bound to the count in it, holding one
+// timestamp's cells. Keyed rather than indexed by band because the axis is only
+// known once every cell has been seen.
+type heatmapColumn map[float64]float64
+
+func isValidBucketUpperBound(upperBound float64) bool {
+	return !math.IsNaN(upperBound) && !math.IsInf(upperBound, -1)
+}
+
+// heatmapSeries accumulates one group's columns while the rows are read.
+type heatmapSeries struct {
+	labels             []*qbtypes.Label
+	columnsByTimestamp map[int64]heatmapColumn
+}
+
+// heatmapAccumulator collects cells from either reader and folds them into one
+// series per group.
+type heatmapAccumulator struct {
+	seriesByKey map[string]*heatmapSeries
+	seriesOrder []string
+	upperBounds map[float64]struct{}
+}
+
+func newHeatmapAccumulator() *heatmapAccumulator {
+	return &heatmapAccumulator{
+		seriesByKey: map[string]*heatmapSeries{},
+		upperBounds: map[float64]struct{}{},
+	}
+}
+
+// addCell files one cell under the group labelsKey identifies, keeping the
+// labels from the first cell seen for it.
+func (a *heatmapAccumulator) addCell(labelsKey string, lbls []*qbtypes.Label, ts int64, upperBound, count float64) {
+	series, ok := a.seriesByKey[labelsKey]
+	if !ok {
+		series = &heatmapSeries{labels: lbls, columnsByTimestamp: map[int64]heatmapColumn{}}
+		a.seriesByKey[labelsKey] = series
+		a.seriesOrder = append(a.seriesOrder, labelsKey)
+	}
+	if series.columnsByTimestamp[ts] == nil {
+		series.columnsByTimestamp[ts] = heatmapColumn{}
+	}
+	series.columnsByTimestamp[ts][upperBound] += count
+	if !math.IsInf(upperBound, 1) {
+		a.upperBounds[upperBound] = struct{}{}
+	}
+}
+
+// foldSeries turns the collected cells into one series per group, in the order
+// the groups first appeared.
+func (a *heatmapAccumulator) foldSeries(queryWindow *qbtypes.TimeRange, stepMs uint64, queryName string) *qbtypes.TimeSeriesData {
+	if len(a.seriesOrder) == 0 {
+		return &qbtypes.TimeSeriesData{QueryName: queryName}
+	}
+
+	upperBounds := make([]float64, 0, len(a.upperBounds))
+	for upperBound := range a.upperBounds {
+		upperBounds = append(upperBounds, upperBound)
+	}
+	slices.Sort(upperBounds)
+
+	// the index past the last upper bound is where the +Inf overflow lands
+	upperBoundToIndex := make(map[float64]int, len(upperBounds)+1)
+	for index, upperBound := range upperBounds {
+		upperBoundToIndex[upperBound] = index
+	}
+	upperBoundToIndex[math.Inf(1)] = len(upperBounds)
+
+	bucket := &qbtypes.AggregationBucket{
+		Index:  0,
+		Alias:  "__result_0",
+		Meta:   qbtypes.AggregationMeta{Buckets: upperBounds},
+		Series: make([]*qbtypes.TimeSeries, 0, len(a.seriesOrder)),
+	}
+
+	for _, labelsKey := range a.seriesOrder {
+		accumulated := a.seriesByKey[labelsKey]
+
+		timestamps := make([]int64, 0, len(accumulated.columnsByTimestamp))
+		for ts := range accumulated.columnsByTimestamp {
+			timestamps = append(timestamps, ts)
+		}
+		slices.Sort(timestamps)
+
+		series := &qbtypes.TimeSeries{
+			Labels: accumulated.labels,
+			Values: make([]*qbtypes.TimeSeriesValue, 0, len(timestamps)),
+		}
+		for _, ts := range timestamps {
+			values := make([]float64, len(upperBounds)+1)
+			for upperBound, count := range accumulated.columnsByTimestamp[ts] {
+				values[upperBoundToIndex[upperBound]] = count
+			}
+			series.Values = append(series.Values, &qbtypes.TimeSeriesValue{
+				Timestamp: ts,
+				Values:    values,
+				Partial:   isPartialValue(ts, queryWindow, stepMs),
+			})
+		}
+		bucket.Series = append(bucket.Series, series)
+	}
+
+	return &qbtypes.TimeSeriesData{
+		QueryName:    queryName,
+		Aggregations: []*qbtypes.AggregationBucket{bucket},
+	}
+}

pkg/querier/heatmap_test.go

@@ -0,0 +1,101 @@
+package querier
+
+import (
+	"testing"
+
+	"github.com/SigNoz/signoz/pkg/instrumentation/instrumentationtest"
+	qbtypes "github.com/SigNoz/signoz/pkg/types/querybuildertypes/querybuildertypesv5"
+	"github.com/SigNoz/signoz/pkg/types/telemetrytypes"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestMergeTimeSeriesResultsUnionsHeatmapAxes(t *testing.T) {
+	// a log axis holds whichever bands the data reached, so a wide cached range
+	// and a narrow fresh one routinely disagree on which bands exist
+	cached := &qbtypes.TimeSeriesData{
+		QueryName: "A",
+		Aggregations: []*qbtypes.AggregationBucket{{
+			Index: 0,
+			Meta:  qbtypes.AggregationMeta{Buckets: []float64{1, 4, 16}},
+			Series: []*qbtypes.TimeSeries{{
+				Labels: []*qbtypes.Label{{Key: telemetrytypes.TelemetryFieldKey{Name: "host.name"}, Value: "node-1"}},
+				Values: []*qbtypes.TimeSeriesValue{{Timestamp: 1710000000000, Values: []float64{1, 2, 3, 4}}},
+			}},
+		}},
+	}
+	fresh := []*qbtypes.Result{{
+		Value: &qbtypes.TimeSeriesData{
+			QueryName: "A",
+			Aggregations: []*qbtypes.AggregationBucket{{
+				Index: 0,
+				Meta:  qbtypes.AggregationMeta{Buckets: []float64{2, 4}},
+				Series: []*qbtypes.TimeSeries{{
+					Labels: []*qbtypes.Label{{Key: telemetrytypes.TelemetryFieldKey{Name: "host.name"}, Value: "node-1"}},
+					Values: []*qbtypes.TimeSeriesValue{{Timestamp: 1710000060000, Values: []float64{5, 6, 7}}},
+				}},
+			}},
+		},
+	}}
+
+	merged := (&querier{}).mergeTimeSeriesResults(cached, fresh)
+
+	require.Len(t, merged.Aggregations, 1)
+	aggBucket := merged.Aggregations[0]
+	assert.Equal(t, []float64{1, 2, 4, 16}, aggBucket.Meta.Buckets)
+
+	require.Len(t, aggBucket.Series, 1)
+	require.Len(t, aggBucket.Series[0].Values, 2)
+	// the cached 16 band survives even though the fresh range never reached it
+	assert.Equal(t, []float64{1, 0, 2, 3, 4}, aggBucket.Series[0].Values[0].Values)
+	// and the fresh 2 band survives even though the cached range never had it
+	assert.Equal(t, []float64{0, 5, 6, 0, 7}, aggBucket.Series[0].Values[1].Values)
+}
+
+func TestTrimResultToFluxBoundaryKeepsTheHeatmapAxis(t *testing.T) {
+	cache := &bucketCache{logger: instrumentationtest.New().Logger()}
+
+	result := &qbtypes.Result{
+		Type: qbtypes.RequestTypeHeatmap,
+		Value: &qbtypes.TimeSeriesData{
+			Aggregations: []*qbtypes.AggregationBucket{{
+				Index: 0,
+				Alias: "__result_0",
+				Meta:  qbtypes.AggregationMeta{Unit: "By", Buckets: []float64{1, 2, 4}},
+				Series: []*qbtypes.TimeSeries{{
+					Values: []*qbtypes.TimeSeriesValue{
+						{Timestamp: 1710000000000, Values: []float64{1, 2, 3, 4}},
+					},
+				}},
+			}},
+		},
+	}
+
+	trimmed := cache.trimResultToFluxBoundary(result, 1710000060000)
+
+	tsData, ok := trimmed.Value.(*qbtypes.TimeSeriesData)
+	require.True(t, ok)
+	require.Len(t, tsData.Aggregations, 1)
+
+	// the counts are positional against the axis, so a cached bucket that lost
+	// Meta.Buckets would be realigned from an empty axis and collapse into the
+	// overflow slot on the way back out
+	aggBucket := tsData.Aggregations[0]
+	assert.Equal(t, []float64{1, 2, 4}, aggBucket.Meta.Buckets)
+	assert.Equal(t, "By", aggBucket.Meta.Unit)
+	assert.Equal(t, "__result_0", aggBucket.Alias)
+}
+
+func TestRealignFromAnEmptyAxisCollapsesIntoTheOverflow(t *testing.T) {
+	// pins the behaviour the trim bug exposed: with no axis to read the counts
+	// against, everything lands in the overflow slot
+	aggBucket := &qbtypes.AggregationBucket{
+		Series: []*qbtypes.TimeSeries{{
+			Values: []*qbtypes.TimeSeriesValue{{Timestamp: 1710000000000, Values: []float64{7, 8, 9, 10}}},
+		}},
+	}
+
+	aggBucket.ReindexValuesToNewUpperBounds([]float64{1, 2, 4})
+
+	assert.Equal(t, []float64{0, 0, 0, 7}, aggBucket.Series[0].Values[0].Values)
+}

pkg/querier/postprocess.go

@@ -195,6 +195,16 @@ func postProcessBuilderQuery[T any](
 	return result
 }
 
+// resolveHeatmapBucketAxis brings the bucket axis to the resolution the caller
+// asked for. Downscaling runs first so AddHeatmapBucketsWithNoCounts adds them
+// at that resolution rather than the finer one ClickHouse bucketed at.
+func resolveHeatmapBucketAxis(tsData *qbtypes.TimeSeriesData, bucketing qbtypes.HeatmapBucketing) {
+	if bucketing.Kind == qbtypes.BucketsKindLog {
+		qbtypes.DownscaleHeatmapResolution(tsData, bucketing.LogScale)
+	}
+	qbtypes.AddHeatmapBucketsWithNoCounts(tsData, bucketing)
+}
+
 // postProcessMetricQuery applies postprocessing to a metric query result.
 func postProcessMetricQuery(
 	q *querier,
@@ -216,6 +226,12 @@ func postProcessMetricQuery(
 		}
 	}
 
+	if req.RequestType == qbtypes.RequestTypeHeatmap && config.HeatmapBucketing != nil {
+		if tsData, ok := result.Value.(*qbtypes.TimeSeriesData); ok {
+			resolveHeatmapBucketAxis(tsData, *config.HeatmapBucketing)
+		}
+	}
+
 	result = q.applySeriesLimit(result, query.Limit, query.Order)
 
 	if len(query.Functions) > 0 {
@@ -342,6 +358,19 @@ func (q *querier) applyFormulas(ctx context.Context, results map[string]*qbtypes
 				result = q.applySeriesLimit(result, formula.Limit, formula.Order)
 				results[name] = result
 			}
+		case qbtypes.RequestTypeHeatmap:
+			// The queries a formula reads were run as time series, so what
+			// arrives here is one value per group per timestamp.
+			result := q.processTimeSeriesFormula(ctx, results, formula, req)
+			if result != nil {
+				if tsData, ok := result.Value.(*qbtypes.TimeSeriesData); ok {
+					bucketing := req.BucketOptions.ToHeatmapBucketing()
+					bucketFormulaOutputAsHeatmap(tsData, bucketing)
+					resolveHeatmapBucketAxis(tsData, bucketing)
+				}
+				result = q.applySeriesLimit(result, formula.Limit, formula.Order)
+				results[name] = result
+			}
 		case qbtypes.RequestTypeScalar:
 			result := q.processScalarFormula(ctx, results, formula, req)
 			// For scalar results, apply limit by processScalarFormula itself since it needs to be applied before converting back to scalar format
@@ -410,6 +439,89 @@ func (q *querier) processTimeSeriesFormula(
 	return result
 }
 
+func bucketFormulaOutputAsHeatmap(tsData *qbtypes.TimeSeriesData, bucketing qbtypes.HeatmapBucketing) {
+	// A formula is one expression, so processTimeSeriesFormula gives it one
+	// aggregation.
+	if tsData == nil || len(tsData.Aggregations) == 0 || tsData.Aggregations[0] == nil {
+		return
+	}
+	aggBucket := tsData.Aggregations[0]
+
+	calculateUpperBound := calculateLogValueUpperBound
+	if bucketing.Kind == qbtypes.BucketsKindLinear {
+		calculateUpperBound = func(value float64) float64 {
+			return calculateLinearValueUpperBound(bucketing, value)
+		}
+	}
+
+	// +Inf is the open-above overflow rather than an upper bound of its own, and
+	// a NaN value has no bucket at all, so neither goes on the axis.
+	upperBounds := []float64{}
+	for _, series := range aggBucket.Series {
+		for _, point := range series.Values {
+			upperBound := calculateUpperBound(point.Value)
+			if !math.IsNaN(upperBound) && !math.IsInf(upperBound, 0) {
+				upperBounds = append(upperBounds, upperBound)
+			}
+		}
+	}
+	slices.Sort(upperBounds)
+	upperBounds = slices.Compact(upperBounds)
+
+	upperBoundToIndex := make(map[float64]int, len(upperBounds))
+	for index, upperBound := range upperBounds {
+		upperBoundToIndex[upperBound] = index
+	}
+
+	overflowIndex := len(upperBounds)
+	for _, series := range aggBucket.Series {
+		for _, point := range series.Values {
+			upperBound := calculateUpperBound(point.Value)
+			point.Values = make([]float64, overflowIndex+1)
+			point.Value = 0
+			switch {
+			case math.IsNaN(upperBound):
+			case math.IsInf(upperBound, 1):
+				point.Values[overflowIndex] = 1
+			default:
+				point.Values[upperBoundToIndex[upperBound]] = 1
+			}
+		}
+	}
+
+	aggBucket.Meta.Buckets = upperBounds
+}
+
+// calculateLinearValueUpperBound and calculateLogValueUpperBound are the Go side
+// of what renderLinearUpperBoundExpr and renderLogUpperBoundExpr emit, and have
+// to stay identical to them: a formula heatmap and a metric heatmap that
+// disagreed here would put their counts in different buckets.
+func calculateLinearValueUpperBound(bucketing qbtypes.HeatmapBucketing, value float64) float64 {
+	if value > bucketing.MaxValue {
+		return math.Inf(1)
+	}
+	numBuckets := float64(bucketing.NumBuckets)
+	index := math.Min(math.Max(math.Ceil(value*numBuckets/bucketing.MaxValue), 1), numBuckets)
+	return index * bucketing.MaxValue / numBuckets
+}
+
+// Like renderLogUpperBoundExpr, this reads MaxLogScale rather than the requested
+// scale: ClickHouse buckets at the finest resolution and resolveHeatmapBucketAxis
+// folds the axis down afterwards.
+func calculateLogValueUpperBound(value float64) float64 {
+	if value <= 0 {
+		return 0
+	}
+	if value <= qbtypes.MinLogUpperBound {
+		return qbtypes.MinLogUpperBound
+	}
+	if value > qbtypes.MaxLogUpperBound {
+		return math.Inf(1)
+	}
+	bucketsPerDoubling := math.Exp2(qbtypes.MaxLogScale)
+	return math.Exp2(math.Ceil(math.Log2(value)*bucketsPerDoubling) / bucketsPerDoubling)
+}
+
 func (q *querier) processScalarFormula(
 	ctx context.Context,
 	results map[string]*qbtypes.Result,
@@ -494,7 +606,7 @@ func (q *querier) processScalarFormula(
 				bucket := &qbtypes.AggregationBucket{
 					Index:  aggIdx,
 					Alias:  scalarData.Columns[colIdx].Name,
-					Meta:   scalarData.Columns[colIdx].Meta,
+					Meta:   qbtypes.AggregationMeta{Unit: scalarData.Columns[colIdx].Meta.Unit},
 					Series: make([]*qbtypes.TimeSeries, 0),
 				}
 
@@ -667,13 +779,14 @@ func convertTimeSeriesDataToScalar(tsData *qbtypes.TimeSeriesData, queryName str
 		if name == "" {
 			name = fmt.Sprintf("__result_%d", agg.Index)
 		}
-		columns = append(columns, &qbtypes.ColumnDescriptor{
+		column := &qbtypes.ColumnDescriptor{
 			TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{Name: name},
 			QueryName:         queryName,
 			AggregationIndex:  int64(agg.Index),
-			Meta:              agg.Meta,
 			Type:              qbtypes.ColumnTypeAggregation,
-		})
+		}
+		column.Meta.Unit = agg.Meta.Unit
+		columns = append(columns, column)
 	}
 
 	// Build rows.

pkg/querier/preview.go

@@ -50,7 +50,7 @@ func (q *querier) QueryRangePreview(
 		env := []qbtypes.QueryEnvelope{req.CompositeQuery.Queries[idx]}
 		ps.Warnings = append(ps.Warnings, q.adjustStepInterval(env, req.Start, req.End)...)
 
-		missingMetricQueries, metricWarnings, mErr := q.resolveMetricMetadata(ctx, orgID, env, req.Start, req.End)
+		missingMetricQueries, metricWarnings, mErr := q.resolveMetricMetadata(ctx, orgID, env, req.Start, req.End, req.RequestType, req.BucketOptions)
 		if mErr != nil {
 			// Report this query's error but keep previewing the rest.
 			ps.Error = mErr

pkg/querier/promql_heatmap.go

@@ -0,0 +1,131 @@
+package querier
+
+import (
+	"fmt"
+	"math"
+	"slices"
+	"sort"
+	"strconv"
+	"strings"
+
+	"github.com/prometheus/prometheus/model/labels"
+
+	"github.com/prometheus/prometheus/promql"
+
+	qbv5 "github.com/SigNoz/signoz/pkg/types/querybuildertypes/querybuildertypesv5"
+	"github.com/SigNoz/signoz/pkg/types/telemetrytypes"
+)
+
+// promHistogramBucketLabel is the label a classic histogram carries its
+// cumulative upper bound on.
+const promHistogramBucketLabel = "le"
+
+// cumulativeColumn maps a bucket's upper bound to the cumulative count at it.
+// Differencing turns it into the per-band counts a heatmapColumn holds.
+type cumulativeColumn map[float64]float64
+
+// promHeatmapGroup assembles one group across the several matrix series its `le`
+// values arrive as, since differencing needs all of them.
+type promHeatmapGroup struct {
+	labels     []*qbv5.Label
+	labelsKey  string
+	cumulative map[int64]cumulativeColumn
+}
+
+// foldMatrixAsHeatmap folds a matrix of one cumulative series per (group, `le`)
+// into one series per group whose points hold a count per band.
+func foldMatrixAsHeatmap(matrix promql.Matrix, queryWindow *qbv5.TimeRange, stepMs uint64, queryName string) *qbv5.TimeSeriesData {
+	groups, groupOrder := collectCumulativeGroups(matrix)
+
+	accumulator := newHeatmapAccumulator()
+	for _, labelsKey := range groupOrder {
+		groups[labelsKey].addDifferencedCells(accumulator)
+	}
+
+	return accumulator.foldSeries(queryWindow, stepMs, queryName)
+}
+
+// collectCumulativeGroups reads the matrix into one group per label set. A series
+// without `le` has no band to sit in, so an expression that dropped the label
+// draws nothing.
+func collectCumulativeGroups(matrix promql.Matrix) (groups map[string]*promHeatmapGroup, groupOrder []string) {
+	groups = map[string]*promHeatmapGroup{}
+
+	for _, promSeries := range matrix {
+		upperBound, ok := extractBucketUpperBound(promSeries.Metric)
+		if !ok {
+			continue
+		}
+
+		lbls, labelsKey := extractHeatmapGroup(promSeries.Metric)
+		group, ok := groups[labelsKey]
+		if !ok {
+			group = &promHeatmapGroup{labels: lbls, labelsKey: labelsKey, cumulative: map[int64]cumulativeColumn{}}
+			groups[labelsKey] = group
+			groupOrder = append(groupOrder, labelsKey)
+		}
+
+		for _, point := range promSeries.Floats {
+			// skipping widens the band above onto the next upper bound that has
+			// a count, which is what lagInFrame does with an absent row
+			if math.IsNaN(point.F) || math.IsInf(point.F, 0) {
+				continue
+			}
+			if group.cumulative[point.T] == nil {
+				group.cumulative[point.T] = cumulativeColumn{}
+			}
+			group.cumulative[point.T][upperBound] = point.F
+		}
+	}
+
+	return groups, groupOrder
+}
+
+func extractBucketUpperBound(metric labels.Labels) (float64, bool) {
+	raw := metric.Get(promHistogramBucketLabel)
+	if raw == "" {
+		return 0, false
+	}
+	upperBound, err := strconv.ParseFloat(raw, 64)
+	if err != nil || !isValidBucketUpperBound(upperBound) {
+		return 0, false
+	}
+	return upperBound, true
+}
+
+// extractHeatmapGroup returns a series' group labels — everything but `le`.
+func extractHeatmapGroup(metric labels.Labels) ([]*qbv5.Label, string) {
+	lbls := make([]*qbv5.Label, 0, metric.Len())
+	pairs := make([]string, 0, metric.Len())
+
+	metric.Range(func(l labels.Label) {
+		if l.Name == promHistogramBucketLabel || excludePromLabel(l.Name) {
+			return
+		}
+		lbls = append(lbls, &qbv5.Label{
+			Key:   telemetrytypes.TelemetryFieldKey{Name: l.Name},
+			Value: l.Value,
+		})
+		pairs = append(pairs, fmt.Sprintf("%s=%s", l.Name, l.Value))
+	})
+
+	sort.Strings(pairs)
+	return lbls, strings.Join(pairs, ",")
+}
+
+// each cell is its upper bound's cumulative count minus the one below it.
+func (g *promHeatmapGroup) addDifferencedCells(accumulator *heatmapAccumulator) {
+	for ts, cumulative := range g.cumulative {
+		upperBounds := make([]float64, 0, len(cumulative))
+		for upperBound := range cumulative {
+			upperBounds = append(upperBounds, upperBound)
+		}
+		slices.Sort(upperBounds)
+
+		previous := float64(0)
+		for _, upperBound := range upperBounds {
+			accumulator.addCell(g.labelsKey, g.labels, ts, upperBound, math.Max(cumulative[upperBound]-previous, 0))
+			previous = cumulative[upperBound]
+		}
+	}
+}

pkg/querier/promql_heatmap_test.go

@@ -0,0 +1,86 @@
+package querier
+
+import (
+	"log/slog"
+	"math"
+	"testing"
+	"time"
+
+	"github.com/prometheus/prometheus/model/labels"
+	"github.com/prometheus/prometheus/promql"
+
+	qbv5 "github.com/SigNoz/signoz/pkg/types/querybuildertypes/querybuildertypesv5"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// The cache key is the fingerprint alone, so two request types over one
+// expression must not produce the same one — a time series payload served to a
+// heatmap request has no axis and reads back as a single collapsed band.
+func TestFingerprintSeparatesHeatmapFromTimeSeries(t *testing.T) {
+	fingerprintFor := func(requestType qbv5.RequestType) string {
+		q := &promqlQuery{
+			logger:      slog.New(slog.DiscardHandler),
+			query:       qbv5.PromQuery{Name: "A", Query: "sum by (le) (increase(signoz_latency_bucket[5m]))", Step: qbv5.Step{Duration: time.Minute}},
+			tr:          qbv5.TimeRange{From: 1710000000000, To: 1710003600000},
+			requestType: requestType,
+		}
+		return q.Fingerprint()
+	}
+
+	heatmap := fingerprintFor(qbv5.RequestTypeHeatmap)
+	timeSeries := fingerprintFor(qbv5.RequestTypeTimeSeries)
+
+	assert.NotEmpty(t, heatmap, "a heatmap decomposes into time buckets like a time series")
+	assert.NotEqual(t, timeSeries, heatmap)
+	assert.Empty(t, fingerprintFor(qbv5.RequestTypeScalar), "a scalar result is its window's last point")
+}
+
+func TestFoldMatrixAsHeatmapClampsADecreasingCumulativeCount(t *testing.T) {
+	at := int64(1710000000000)
+
+	matrix := promql.Matrix{
+		{
+			Metric: labels.FromStrings("le", "5"),
+			Floats: []promql.FPoint{{T: at, F: 10}},
+		},
+		{
+			Metric: labels.FromStrings("le", "10"),
+			Floats: []promql.FPoint{{T: at, F: 4}},
+		},
+	}
+
+	data := foldMatrixAsHeatmap(matrix, &qbv5.TimeRange{From: 1710000000000, To: 1710000060000}, uint64(time.Minute.Milliseconds()), "A")
+	require.Len(t, data.Aggregations, 1)
+
+	// a cumulative count that went backwards would difference to -6
+	assert.Equal(t, []float64{10, 0, 0}, data.Aggregations[0].Series[0].Values[0].Values)
+}
+
+func TestFoldMatrixAsHeatmapWidensTheBandOverAMissingUpperBound(t *testing.T) {
+	at := int64(1710000000000)
+
+	matrix := promql.Matrix{
+		{
+			Metric: labels.FromStrings("le", "5"),
+			Floats: []promql.FPoint{{T: at, F: 3}},
+		},
+		{
+			Metric: labels.FromStrings("le", "10"),
+			Floats: []promql.FPoint{{T: at, F: math.NaN()}},
+		},
+		{
+			Metric: labels.FromStrings("le", "20"),
+			Floats: []promql.FPoint{{T: at, F: 30}},
+		},
+	}
+
+	data := foldMatrixAsHeatmap(matrix, &qbv5.TimeRange{From: 1710000000000, To: 1710000060000}, uint64(time.Minute.Milliseconds()), "A")
+	require.Len(t, data.Aggregations, 1)
+
+	aggregation := data.Aggregations[0]
+	// 10 carried nothing to difference against, so it is not on the axis at all
+	// and 20 differences against 5, holding what (5,10] and (10,20] would split
+	assert.Equal(t, []float64{5, 20}, aggregation.Meta.Buckets)
+	assert.Equal(t, []float64{3, 27, 0}, aggregation.Series[0].Values[0].Values)
+}

pkg/querier/promql_query.go

@@ -155,7 +155,10 @@ func (q *promqlQuery) Fingerprint() string {
 	if q.opts.serve != nil {
 		return ""
 	}
-	if q.requestType != qbv5.RequestTypeTimeSeries {
+
+	switch q.requestType {
+	case qbv5.RequestTypeTimeSeries, qbv5.RequestTypeHeatmap:
+	default:
 		return ""
 	}
 
@@ -166,6 +169,8 @@ func (q *promqlQuery) Fingerprint() string {
 	}
 	parts := []string{
 		"promql",
+		// one expression returns a different shape per request type
+		fmt.Sprintf("requestType=%s", q.requestType.StringValue()),
 		query,
 		q.query.Step.String(),
 	}
@@ -449,25 +454,52 @@ func (q *promqlQuery) Execute(ctx context.Context) (*qbv5.Result, error) {
 	return q.toResult(matrix, warnings, began, &statsMu, &rowsScanned, &bytesScanned), nil
 }
 
-// toResult converts an evaluated matrix into the v5 result shape, attaching
-// the ClickHouse scan stats accumulated during evaluation.
-func (q *promqlQuery) toResult(matrix promql.Matrix, warnings []string, began time.Time, statsMu *sync.Mutex, rowsScanned, bytesScanned *uint64) *qbv5.Result {
-	// Hide only known SigNoz storage keys: label names are user data and may
-	// legitimately start with "__" (e.g. __address__), so a blanket dunder
-	// strip mangles user labelsets. The __scope./__resource. prefixes cover
-	// every exporter version's keys.
-	excludeLabel := func(labelName string) bool {
-		return labelName == "__temporality__" ||
-			strings.HasPrefix(labelName, "__scope.") ||
-			strings.HasPrefix(labelName, "__resource.")
-	}
+// excludePromLabel hides only known SigNoz storage keys: label names are user
+// data and may legitimately start with "__" (e.g. __address__), so a blanket
+// dunder strip mangles user labelsets. The __scope./__resource. prefixes cover
+// every exporter version's keys.
+func excludePromLabel(labelName string) bool {
+	return labelName == "__temporality__" ||
+		strings.HasPrefix(labelName, "__scope.") ||
+		strings.HasPrefix(labelName, "__resource.")
+}
 
+// collectExecStats snapshots the scan counters a query accumulated. Callers take
+// it at the point they are done with the matrix, so the duration covers the
+// shaping they did.
+func collectExecStats(began time.Time, statsMu *sync.Mutex, rowsScanned, bytesScanned *uint64) qbv5.ExecStats {
+	statsMu.Lock()
+	defer statsMu.Unlock()
+	return qbv5.ExecStats{
+		RowsScanned:  *rowsScanned,
+		BytesScanned: *bytesScanned,
+		DurationMS:   uint64(time.Since(began).Milliseconds()),
+	}
+}
+
+func (q *promqlQuery) toResult(matrix promql.Matrix, warnings []string, began time.Time, statsMu *sync.Mutex, rowsScanned, bytesScanned *uint64) *qbv5.Result {
+	if q.requestType == qbv5.RequestTypeHeatmap {
+		return q.toResultForHeatmap(matrix, warnings, began, statsMu, rowsScanned, bytesScanned)
+	}
+	return q.toResultForTimeSeriesAndScalar(matrix, warnings, began, statsMu, rowsScanned, bytesScanned)
+}
+
+func (q *promqlQuery) toResultForHeatmap(matrix promql.Matrix, warnings []string, began time.Time, statsMu *sync.Mutex, rowsScanned, bytesScanned *uint64) *qbv5.Result {
+	return &qbv5.Result{
+		Type:     q.requestType,
+		Value:    foldMatrixAsHeatmap(matrix, &q.tr, uint64(q.query.Step.Milliseconds()), q.query.Name),
+		Warnings: warnings,
+		Stats:    collectExecStats(began, statsMu, rowsScanned, bytesScanned),
+	}
+}
+
+func (q *promqlQuery) toResultForTimeSeriesAndScalar(matrix promql.Matrix, warnings []string, began time.Time, statsMu *sync.Mutex, rowsScanned, bytesScanned *uint64) *qbv5.Result {
 	var series []*qbv5.TimeSeries
 	for _, v := range matrix {
 		var s qbv5.TimeSeries
 		lbls := make([]*qbv5.Label, 0, v.Metric.Len())
 		v.Metric.Range(func(l labels.Label) {
-			if excludeLabel(l.Name) {
+			if excludePromLabel(l.Name) {
 				return
 			}
 			lbls = append(lbls, &qbv5.Label{
@@ -495,13 +527,7 @@ func (q *promqlQuery) toResult(matrix promql.Matrix, warnings []string, began ti
 		series = append(series, &s)
 	}
 
-	statsMu.Lock()
-	stats := qbv5.ExecStats{
-		RowsScanned:  *rowsScanned,
-		BytesScanned: *bytesScanned,
-		DurationMS:   uint64(time.Since(began).Milliseconds()),
-	}
-	statsMu.Unlock()
+	stats := collectExecStats(began, statsMu, rowsScanned, bytesScanned)
 
 	tsData := &qbv5.TimeSeriesData{QueryName: q.query.Name}
 	// No bucket at all when nothing survived: a bucket holding no series reads

pkg/querier/querier.go

@@ -156,7 +156,7 @@ func (q *querier) QueryRange(ctx context.Context, orgID valuer.UUID, req *qbtype
 	// We need to set if it is unspecified or adjust it if value is not within recommended range
 	intervalWarnings := q.adjustStepInterval(req.CompositeQuery.Queries, req.Start, req.End)
 
-	missingMetricQueries, metricWarnings, err := q.resolveMetricMetadata(ctx, orgID, req.CompositeQuery.Queries, req.Start, req.End)
+	missingMetricQueries, metricWarnings, err := q.resolveMetricMetadata(ctx, orgID, req.CompositeQuery.Queries, req.Start, req.End, req.RequestType, req.BucketOptions)
 	if err != nil {
 		return nil, err
 	}
@@ -177,7 +177,7 @@ func (q *querier) QueryRange(ctx context.Context, orgID valuer.UUID, req *qbtype
 	preseededResults := make(map[string]any)
 	for _, name := range missingMetricQueries {
 		switch req.RequestType {
-		case qbtypes.RequestTypeTimeSeries:
+		case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 			preseededResults[name] = &qbtypes.TimeSeriesData{QueryName: name}
 		case qbtypes.RequestTypeScalar:
 			preseededResults[name] = &qbtypes.ScalarData{QueryName: name}
@@ -334,15 +334,22 @@ func (q *querier) buildQueries(
 				if missingMetricQuerySet[spec.Name] {
 					continue
 				}
+				requestType := req.RequestType
+				if requestType == qbtypes.RequestTypeHeatmap && spec.Disabled {
+					// A disabled query in a heatmap request feeds a formula, and the
+					// formula converts time series into heatmap data, so its inputs
+					// run as time series queries.
+					requestType = qbtypes.RequestTypeTimeSeries
+				}
 				spec.ShiftBy = extractShiftFromBuilderQuery(spec)
-				timeRange := adjustTimeRangeForShift(spec, qbtypes.TimeRange{From: req.Start, To: req.End}, req.RequestType)
+				timeRange := adjustTimeRangeForShift(spec, qbtypes.TimeRange{From: req.Start, To: req.End}, requestType)
 				var bq *builderQuery[qbtypes.MetricAggregation]
 
 				if spec.Source == telemetrytypes.SourceMeter {
 					event.Source = telemetrytypes.SourceMeter.StringValue()
-					bq = newBuilderQuery(q.logger, q.telemetryStore, orgID, q.meterStmtBuilder, query.Type, spec, timeRange, req.RequestType, tmplVars, builderConfig{})
+					bq = newBuilderQuery(q.logger, q.telemetryStore, orgID, q.meterStmtBuilder, query.Type, spec, timeRange, requestType, tmplVars, builderConfig{})
 				} else {
-					bq = newBuilderQuery(q.logger, q.telemetryStore, orgID, q.metricStmtBuilder, query.Type, spec, timeRange, req.RequestType, tmplVars, builderConfig{})
+					bq = newBuilderQuery(q.logger, q.telemetryStore, orgID, q.metricStmtBuilder, query.Type, spec, timeRange, requestType, tmplVars, builderConfig{})
 				}
 
 				queries[spec.Name] = bq
@@ -415,7 +422,7 @@ func (q *querier) populateQBEvent(event *qbtypes.QBEvent, queries []qbtypes.Quer
 //     resolved: never-seen metrics and dormant metrics (seen but no data in
 //     the query window).
 //   - err: Internal when a metadata fetch fails.
-func (q *querier) resolveMetricMetadata(ctx context.Context, orgID valuer.UUID, queries []qbtypes.QueryEnvelope, start, end uint64) (missingMetricQueries []string, metricWarnings []string, err error) {
+func (q *querier) resolveMetricMetadata(ctx context.Context, orgID valuer.UUID, queries []qbtypes.QueryEnvelope, start, end uint64, requestType qbtypes.RequestType, bucketOptions *qbtypes.BucketOptions) (missingMetricQueries []string, metricWarnings []string, err error) {
 	metricNames := make([]string, 0)
 	for idx := range queries {
 		if queries[idx].Type != qbtypes.QueryTypeBuilder {
@@ -465,6 +472,13 @@ func (q *querier) resolveMetricMetadata(ctx context.Context, orgID valuer.UUID,
 					spec.Aggregations[i].Type = foundMetricType
 				}
 			}
+			// Only the enabled query is used to render the heatmap, so bucket
+			// options are only applied to the enabled query.
+			if requestType == qbtypes.RequestTypeHeatmap && !spec.Disabled {
+				if err := spec.Aggregations[i].VerifyAndApplyBucketOptions(bucketOptions); err != nil {
+					return nil, nil, err
+				}
+			}
 			if spec.Aggregations[i].Type == metrictypes.UnspecifiedType {
 				missingMetrics = append(missingMetrics, spec.Aggregations[i].MetricName)
 				continue
@@ -679,7 +693,7 @@ func (q *querier) run(
 			if val, ok := result.Value.(*qbtypes.RawData); ok && val != nil {
 				return len(val.Rows) != 0
 			}
-		case qbtypes.RequestTypeTimeSeries:
+		case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 			if val, ok := result.Value.(*qbtypes.TimeSeriesData); ok && val != nil {
 				if len(val.Aggregations) != 0 {
 					anyNonEmpty := false
@@ -1000,7 +1014,7 @@ func (q *querier) mergeResults(cached *qbtypes.Result, fresh []*qbtypes.Result)
 
 		// Merge all fresh results including the first one
 		switch merged.Type {
-		case qbtypes.RequestTypeTimeSeries:
+		case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 			// Pass nil as cached value to ensure proper merging of all fresh results
 			merged.Value = q.mergeTimeSeriesResults(nil, fresh)
 		}
@@ -1023,7 +1037,7 @@ func (q *querier) mergeResults(cached *qbtypes.Result, fresh []*qbtypes.Result)
 	}
 
 	switch merged.Type {
-	case qbtypes.RequestTypeTimeSeries:
+	case qbtypes.RequestTypeTimeSeries, qbtypes.RequestTypeHeatmap:
 		merged.Value = q.mergeTimeSeriesResults(cached.Value.(*qbtypes.TimeSeriesData), fresh)
 	}
 
@@ -1044,6 +1058,16 @@ func (q *querier) mergeResults(cached *qbtypes.Result, fresh []*qbtypes.Result)
 	return merged
 }
 
+func mergeBucketUpperBounds(cachedValue *qbtypes.TimeSeriesData, freshResults []*qbtypes.Result) map[int][]float64 {
+	upperBoundSources := make([]*qbtypes.TimeSeriesData, 0, len(freshResults)+1)
+	upperBoundSources = append(upperBoundSources, cachedValue)
+	for _, result := range freshResults {
+		freshTS, _ := result.Value.(*qbtypes.TimeSeriesData)
+		upperBoundSources = append(upperBoundSources, freshTS)
+	}
+	return qbtypes.MergeBucketUpperBounds(upperBoundSources...)
+}
+
 // mergeTimeSeriesResults merges time series data.
 func (q *querier) mergeTimeSeriesResults(cachedValue *qbtypes.TimeSeriesData, freshResults []*qbtypes.Result) *qbtypes.TimeSeriesData {
 
@@ -1052,12 +1076,15 @@ func (q *querier) mergeTimeSeriesResults(cachedValue *qbtypes.TimeSeriesData, fr
 	// Map to store aggregation bucket metadata
 	bucketMetadata := make(map[int]*qbtypes.AggregationBucket)
 
+	mergedUpperBounds := mergeBucketUpperBounds(cachedValue, freshResults)
+
 	// Process cached data if available
 	if cachedValue != nil && cachedValue.Aggregations != nil {
 		for _, aggBucket := range cachedValue.Aggregations {
 			if seriesMap[aggBucket.Index] == nil {
 				seriesMap[aggBucket.Index] = make(map[string]*qbtypes.TimeSeries)
 			}
+			aggBucket.ReindexValuesToNewUpperBounds(mergedUpperBounds[aggBucket.Index])
 			if bucketMetadata[aggBucket.Index] == nil {
 				bucketMetadata[aggBucket.Index] = aggBucket
 			}
@@ -1109,6 +1136,7 @@ func (q *querier) mergeTimeSeriesResults(cachedValue *qbtypes.TimeSeriesData, fr
 		}
 
 		for _, aggBucket := range freshTS.Aggregations {
+			aggBucket.ReindexValuesToNewUpperBounds(mergedUpperBounds[aggBucket.Index])
 			for _, series := range aggBucket.Series {
 				key := qbtypes.GetUniqueSeriesKey(series.Labels)
 

pkg/statementbuilder/meterstatementbuilder/statement_builder.go

@@ -129,7 +129,7 @@ func (b *meterQueryStatementBuilder) buildPipelineStatement(
 	}
 
 	// final SELECT
-	return b.metricsStatementBuilder.BuildFinalSelect(cteFragments, cteArgs, query)
+	return b.metricsStatementBuilder.BuildFinalSelect(cteFragments, cteArgs, qbtypes.RequestTypeTimeSeries, query)
 }
 
 func (b *meterQueryStatementBuilder) buildTemporalAggDeltaFastPath(

pkg/statementbuilder/metricsstatementbuilder/statement_builder.go

@@ -4,9 +4,13 @@ import (
 	"context"
 	"fmt"
 	"log/slog"
+	"math"
 	"slices"
+	"strconv"
+	"strings"
 	"time"
 
+	"github.com/SigNoz/signoz/pkg/errors"
 	"github.com/SigNoz/signoz/pkg/factory"
 	"github.com/SigNoz/signoz/pkg/flagger"
 	"github.com/SigNoz/signoz/pkg/querybuilder"
@@ -113,7 +117,7 @@ func (b *StatementBuilder) Build(
 	orgID valuer.UUID,
 	start uint64,
 	end uint64,
-	_ qbtypes.RequestType,
+	requestType qbtypes.RequestType,
 	query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation],
 	variables map[string]qbtypes.VariableItem,
 ) (*qbtypes.Statement, error) {
@@ -125,13 +129,14 @@ func (b *StatementBuilder) Build(
 
 	start, end = querybuilder.AdjustedMetricTimeRange(start, end, uint64(query.StepInterval.Seconds()), query)
 
-	return b.buildPipelineStatement(ctx, orgID, start, end, query, keys, variables)
+	return b.buildPipelineStatement(ctx, orgID, start, end, requestType, query, keys, variables)
 }
 
 func (b *StatementBuilder) buildPipelineStatement(
 	ctx context.Context,
 	orgID valuer.UUID,
 	start, end uint64,
+	requestType qbtypes.RequestType,
 	query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation],
 	keys map[string][]*telemetrytypes.TelemetryFieldKey,
 	variables map[string]qbtypes.VariableItem,
@@ -144,7 +149,7 @@ func (b *StatementBuilder) buildPipelineStatement(
 	cteQuery := query
 	if query.Aggregations[0].Type == metrictypes.HistogramType {
 		query.GroupBy = slices.DeleteFunc(slices.Clone(query.GroupBy), isHistogramBucket)
-		cteQuery = histogramCTEQuery(query)
+		cteQuery = rewriteQueryForHistogramCTE(requestType, query)
 	}
 
 	agg := cteQuery.Aggregations[0]
@@ -216,7 +221,7 @@ func (b *StatementBuilder) buildPipelineStatement(
 		}
 	}
 
-	mainStmt, err := b.BuildFinalSelect(cteFragments, cteArgs, query)
+	mainStmt, err := b.BuildFinalSelect(cteFragments, cteArgs, requestType, query)
 	if err != nil {
 		return nil, err
 	}
@@ -224,13 +229,29 @@ func (b *StatementBuilder) buildPipelineStatement(
 	if reducedFragments == nil {
 		return mainStmt, nil
 	}
-	reducedStmt, err := b.BuildFinalSelect(reducedFragments, reducedArgs, query)
+	reducedStmt, err := b.BuildFinalSelect(reducedFragments, reducedArgs, requestType, query)
 	if err != nil {
 		return nil, err
 	}
 	return unionStatements(mainStmt, reducedStmt, query)
 }
 
+func rewriteQueryForHistogramCTE(requestType qbtypes.RequestType, query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]) qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation] {
+	query.GroupBy = append(slices.Clone(query.GroupBy), qbtypes.GroupByKey{
+		TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{Name: histogramBucketKey},
+	})
+
+	query.Aggregations = slices.Clone(query.Aggregations)
+	if query.Aggregations[0].SpaceAggregation.IsPercentile() && requestType != qbtypes.RequestTypeHeatmap {
+		query.Aggregations[0].TimeAggregation = metrictypes.TimeAggregationRate
+	} else {
+		query.Aggregations[0].TimeAggregation = metrictypes.TimeAggregationIncrease
+	}
+	query.Aggregations[0].SpaceAggregation = metrictypes.SpaceAggregationSum
+
+	return query
+}
+
 func unionStatements(main, reduced *qbtypes.Statement, query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]) (*qbtypes.Statement, error) {
 	orderBy := "ts"
 	for i, g := range query.GroupBy {
@@ -758,11 +779,9 @@ func (b *StatementBuilder) buildSpatialAggregationCTE(
 func (b *StatementBuilder) BuildFinalSelect(
 	cteFragments []string,
 	cteArgs [][]any,
+	requestType qbtypes.RequestType,
 	query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation],
 ) (*qbtypes.Statement, error) {
-	metricType := query.Aggregations[0].Type
-	spaceAgg := query.Aggregations[0].SpaceAggregation
-
 	combined := querybuilder.CombineCTEs(cteFragments)
 
 	var args []any
@@ -770,6 +789,22 @@ func (b *StatementBuilder) BuildFinalSelect(
 		args = append(args, a...)
 	}
 
+	if requestType == qbtypes.RequestTypeHeatmap {
+		return buildHeatmapFinalSelect(combined, args, query)
+	}
+	return buildAggregationFinalSelect(combined, args, query)
+}
+
+// buildAggregationFinalSelect reads __spatial_aggregation_cte as one value per
+// (group, timestamp), which is what every request type but heatmap wants.
+func buildAggregationFinalSelect(
+	combined string,
+	args []any,
+	query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation],
+) (*qbtypes.Statement, error) {
+	metricType := query.Aggregations[0].Type
+	spaceAgg := query.Aggregations[0].SpaceAggregation
+
 	sb := sqlbuilder.NewSelectBuilder()
 
 	if metricType == metrictypes.HistogramType && spaceAgg.IsPercentile() {
@@ -842,24 +877,136 @@ func (b *StatementBuilder) BuildFinalSelect(
 	return &qbtypes.Statement{Query: combined + q, Args: append(args, a...)}, nil
 }
 
-const histogramBucketKey = "le"
+const (
+	histogramBucketKey = "le"
+
+	heatmapValueAlias = "__result_0"
+	heatmapWindow     = "__heatmap_window"
+)
 
 func isHistogramBucket(k qbtypes.GroupByKey) bool { return k.Name == histogramBucketKey }
 
-func histogramCTEQuery(query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]) qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation] {
-	query.GroupBy = append(slices.Clone(query.GroupBy), qbtypes.GroupByKey{
-		TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{Name: histogramBucketKey},
-	})
-
-	query.Aggregations = slices.Clone(query.Aggregations)
-	if query.Aggregations[0].SpaceAggregation.IsPercentile() {
-		query.Aggregations[0].TimeAggregation = metrictypes.TimeAggregationRate
-	} else {
-		query.Aggregations[0].TimeAggregation = metrictypes.TimeAggregationIncrease
+// buildHeatmapFinalSelect turns __spatial_aggregation_cte into one row per
+// heatmap cell: (ts, group labels..., bucket upper bound, count).
+func buildHeatmapFinalSelect(
+	combined string,
+	args []any,
+	query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation],
+) (*qbtypes.Statement, error) {
+	if query.Aggregations[0].Type == metrictypes.HistogramType {
+		return buildHistogramHeatmapFinalSelect(combined, args, query)
 	}
-	query.Aggregations[0].SpaceAggregation = metrictypes.SpaceAggregationSum
+	return buildValueHeatmapFinalSelect(combined, args, query)
+}
 
-	return query
+// buildHistogramHeatmapFinalSelect differences the cumulative per-`le` counts in
+// __spatial_aggregation_cte into a count per band. The upper bound reported is the
+// `le` itself, so the `le=+Inf` row reaches the reader as an infinite upper bound
+// for it to fold into the overflow band.
+func buildHistogramHeatmapFinalSelect(
+	combined string,
+	args []any,
+	query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation],
+) (*qbtypes.Statement, error) {
+	groupAliases := GroupByAliases(query.GroupBy)
+	partitionBy := append(append([]string{}, groupAliases...), "ts")
+
+	sb := sqlbuilder.NewSelectBuilder()
+	sb.Select("ts")
+	sb.SelectMore(groupAliases...)
+	sb.SelectMore(fmt.Sprintf("toFloat64(%s) AS %s", histogramBucketKey, qbtypes.HeatmapBucketColumn))
+	// a partial scrape can break monotonicity across `le`, and a negative cell
+	// count has no meaning
+	sb.SelectMore(fmt.Sprintf(
+		"greatest(value - lagInFrame(value, 1, 0) OVER %s, 0) AS %s",
+		heatmapWindow, heatmapValueAlias,
+	))
+	// sqlbuilder has no WINDOW clause; appending it to FROM lands it between FROM
+	// and ORDER BY, since these statements carry no WHERE or GROUP BY
+	sb.From(fmt.Sprintf(
+		"__spatial_aggregation_cte WINDOW %s AS (PARTITION BY %s ORDER BY toFloat64(%s))",
+		heatmapWindow, strings.Join(partitionBy, ", "), histogramBucketKey,
+	))
+	sb.OrderBy(groupAliases...)
+	sb.OrderBy("ts", fmt.Sprintf("toFloat64(%s)", histogramBucketKey))
+
+	q, a := sb.BuildWithFlavor(sqlbuilder.ClickHouse)
+	return &qbtypes.Statement{Query: combined + q, Args: append(args, a...)}, nil
+}
+
+// buildValueHeatmapFinalSelect places each spatially aggregated value in a band
+// of the requested axis. __spatial_aggregation_cte holds one row per (group,
+// timestamp), so every cell counts exactly one.
+func buildValueHeatmapFinalSelect(
+	combined string,
+	args []any,
+	query qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation],
+) (*qbtypes.Statement, error) {
+	bucketing := query.Aggregations[0].HeatmapBucketing
+	if bucketing == nil {
+		return nil, errors.NewInternalf(errors.CodeInternal,
+			"heatmap over a %s metric reached the statement builder without a resolved bucket axis",
+			query.Aggregations[0].Type.StringValue())
+	}
+
+	upperBound, err := renderHeatmapUpperBoundExpr(*bucketing)
+	if err != nil {
+		return nil, err
+	}
+
+	groupAliases := GroupByAliases(query.GroupBy)
+
+	sb := sqlbuilder.NewSelectBuilder()
+	sb.Select("ts")
+	sb.SelectMore(groupAliases...)
+	sb.SelectMore(fmt.Sprintf("%s AS %s", upperBound, qbtypes.HeatmapBucketColumn))
+	sb.SelectMore(fmt.Sprintf("toFloat64(1) AS %s", heatmapValueAlias))
+	sb.From("__spatial_aggregation_cte")
+	sb.OrderBy(groupAliases...)
+	sb.OrderBy("ts", qbtypes.HeatmapBucketColumn)
+
+	q, a := sb.BuildWithFlavor(sqlbuilder.ClickHouse)
+	return &qbtypes.Statement{Query: combined + q, Args: append(args, a...)}, nil
+}
+
+// renderHeatmapUpperBoundExpr renders the upper bound of the band `value` falls in.
+func renderHeatmapUpperBoundExpr(bucketing qbtypes.HeatmapBucketing) (string, error) {
+	switch bucketing.Kind {
+	case qbtypes.BucketsKindLinear:
+		return renderLinearUpperBoundExpr(bucketing), nil
+	case qbtypes.BucketsKindLog:
+		return renderLogUpperBoundExpr(), nil
+	default:
+		return "", errors.NewInvalidInputf(errors.CodeInvalidInput,
+			"unsupported bucketsScaling %q for heatmap requests", bucketing.Kind.StringValue())
+	}
+}
+
+func renderLinearUpperBoundExpr(bucketing qbtypes.HeatmapBucketing) string {
+	maxValue := formatFloat(bucketing.MaxValue)
+	numBuckets := strconv.Itoa(bucketing.NumBuckets)
+	return fmt.Sprintf(
+		"multiIf(value > %s, toFloat64('+Inf'), least(greatest(ceil(value * %s / %s), 1), %s) * %s / %s)",
+		maxValue, numBuckets, maxValue, numBuckets, maxValue, numBuckets,
+	)
+}
+
+// ClickHouse buckets at MaxLogScale whatever HeatmapBucketing.LogScale asks for;
+// postprocessing folds the axis down afterwards.
+func renderLogUpperBoundExpr() string {
+	bandsPerDoubling := formatFloat(math.Exp2(qbtypes.MaxLogScale))
+	lowest := formatFloat(qbtypes.MinLogUpperBound)
+	highest := formatFloat(qbtypes.MaxLogUpperBound)
+	return fmt.Sprintf(
+		"multiIf(value <= 0, toFloat64(0), value <= %s, %s, value > %s, toFloat64('+Inf'), pow(2, ceil(log2(value) * %s) / %s))",
+		lowest, lowest, highest, bandsPerDoubling, bandsPerDoubling,
+	)
+}
+
+// formatFloat renders a float64 as the shortest literal that reads back as the
+// same value, so an upper bound computed from it is identical on every row.
+func formatFloat(v float64) string {
+	return strconv.FormatFloat(v, 'g', -1, 64)
 }
 
 func GroupByColumnAlias(i int, name string) string {

pkg/statementbuilder/metricsstatementbuilder/stmt_builder_test.go

@@ -284,6 +284,199 @@ func TestStatementBuilder(t *testing.T) {
 			},
 			expectedErr: nil,
 		},
+		{
+			name:        "test_histogram_heatmap_sum",
+			requestType: qbtypes.RequestTypeHeatmap,
+			query: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+				Signal:       telemetrytypes.SignalMetrics,
+				StepInterval: qbtypes.Step{Duration: 60 * time.Second},
+				Aggregations: []qbtypes.MetricAggregation{
+					{
+						MetricName:       "signoz_latency",
+						Type:             metrictypes.HistogramType,
+						Temporality:      metrictypes.Delta,
+						TimeAggregation:  metrictypes.TimeAggregationIncrease,
+						SpaceAggregation: metrictypes.SpaceAggregationSum,
+					},
+				},
+				GroupBy: []qbtypes.GroupByKey{
+					{
+						TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+							Name: "service.name",
+						},
+					},
+				},
+			},
+			expected: qbtypes.Statement{
+				Query: "WITH __spatial_aggregation_cte AS (SELECT toStartOfInterval(toDateTime(intDiv(unix_milli, 1000)), toIntervalSecond(60)) AS ts, `__GROUP_BY_KEY_0_service.name`, `le`, sum(value) AS value FROM signoz_metrics.distributed_samples_v4 AS points INNER JOIN (SELECT fingerprint, JSONExtractString(labels, 'service.name') AS `__GROUP_BY_KEY_0_service.name`, JSONExtractString(labels, 'le') AS `le` FROM signoz_metrics.time_series_v4_6hrs WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli <= ? AND LOWER(temporality) LIKE LOWER(?) GROUP BY fingerprint, `__GROUP_BY_KEY_0_service.name`, `le`) AS filtered_time_series ON points.fingerprint = filtered_time_series.fingerprint WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli < ? GROUP BY ts, `__GROUP_BY_KEY_0_service.name`, `le`) SELECT ts, `__GROUP_BY_KEY_0_service.name`, toFloat64(le) AS __bucket, greatest(value - lagInFrame(value, 1, 0) OVER __heatmap_window, 0) AS __result_0 FROM __spatial_aggregation_cte WINDOW __heatmap_window AS (PARTITION BY `__GROUP_BY_KEY_0_service.name`, ts ORDER BY toFloat64(le)) ORDER BY `__GROUP_BY_KEY_0_service.name`, ts, toFloat64(le)",
+				Args:  []any{"signoz_latency", uint64(1747936800000), uint64(1747983420000), "delta", "signoz_latency", uint64(1747947360000), uint64(1747983420000)},
+			},
+			expectedErr: nil,
+		},
+		{
+			name:        "test_histogram_heatmap_percentile",
+			requestType: qbtypes.RequestTypeHeatmap,
+			query: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+				Signal:       telemetrytypes.SignalMetrics,
+				StepInterval: qbtypes.Step{Duration: 60 * time.Second},
+				Aggregations: []qbtypes.MetricAggregation{
+					{
+						MetricName:       "signoz_latency",
+						Type:             metrictypes.HistogramType,
+						Temporality:      metrictypes.Delta,
+						TimeAggregation:  metrictypes.TimeAggregationRate,
+						SpaceAggregation: metrictypes.SpaceAggregationPercentile95,
+					},
+				},
+				GroupBy: []qbtypes.GroupByKey{
+					{
+						TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+							Name: "service.name",
+						},
+					},
+				},
+			},
+			expected: qbtypes.Statement{
+				Query: "WITH __spatial_aggregation_cte AS (SELECT toStartOfInterval(toDateTime(intDiv(unix_milli, 1000)), toIntervalSecond(60)) AS ts, `__GROUP_BY_KEY_0_service.name`, `le`, sum(value) AS value FROM signoz_metrics.distributed_samples_v4 AS points INNER JOIN (SELECT fingerprint, JSONExtractString(labels, 'service.name') AS `__GROUP_BY_KEY_0_service.name`, JSONExtractString(labels, 'le') AS `le` FROM signoz_metrics.time_series_v4_6hrs WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli <= ? AND LOWER(temporality) LIKE LOWER(?) GROUP BY fingerprint, `__GROUP_BY_KEY_0_service.name`, `le`) AS filtered_time_series ON points.fingerprint = filtered_time_series.fingerprint WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli < ? GROUP BY ts, `__GROUP_BY_KEY_0_service.name`, `le`) SELECT ts, `__GROUP_BY_KEY_0_service.name`, toFloat64(le) AS __bucket, greatest(value - lagInFrame(value, 1, 0) OVER __heatmap_window, 0) AS __result_0 FROM __spatial_aggregation_cte WINDOW __heatmap_window AS (PARTITION BY `__GROUP_BY_KEY_0_service.name`, ts ORDER BY toFloat64(le)) ORDER BY `__GROUP_BY_KEY_0_service.name`, ts, toFloat64(le)",
+				Args:  []any{"signoz_latency", uint64(1747936800000), uint64(1747983420000), "delta", "signoz_latency", uint64(1747947360000), uint64(1747983420000)},
+			},
+			expectedErr: nil,
+		},
+		{
+			name:        "test_gauge_heatmap_log",
+			requestType: qbtypes.RequestTypeHeatmap,
+			query: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+				Signal:       telemetrytypes.SignalMetrics,
+				StepInterval: qbtypes.Step{Duration: 60 * time.Second},
+				Aggregations: []qbtypes.MetricAggregation{
+					{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						Temporality:      metrictypes.Unspecified,
+						TimeAggregation:  metrictypes.TimeAggregationAvg,
+						SpaceAggregation: metrictypes.SpaceAggregationSum,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{
+							Kind:       qbtypes.BucketsKindLog,
+							LogScale:   qbtypes.MaxLogScale,
+							NumBuckets: qbtypes.DefaultNumBuckets,
+						},
+					},
+				},
+				GroupBy: []qbtypes.GroupByKey{
+					{
+						TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+							Name: "host.name",
+						},
+					},
+				},
+			},
+			expected: qbtypes.Statement{
+				Query: "WITH __temporal_aggregation_cte AS (SELECT fingerprint, toStartOfInterval(toDateTime(intDiv(unix_milli, 1000)), toIntervalSecond(60)) AS ts, `__GROUP_BY_KEY_0_host.name`, avg(value) AS per_series_value FROM signoz_metrics.distributed_samples_v4 AS points INNER JOIN (SELECT fingerprint, JSONExtractString(labels, 'host.name') AS `__GROUP_BY_KEY_0_host.name` FROM signoz_metrics.time_series_v4_6hrs WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli <= ? AND LOWER(temporality) LIKE LOWER(?) GROUP BY fingerprint, `__GROUP_BY_KEY_0_host.name`) AS filtered_time_series ON points.fingerprint = filtered_time_series.fingerprint WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli < ? GROUP BY fingerprint, ts, `__GROUP_BY_KEY_0_host.name` ORDER BY fingerprint, ts), __spatial_aggregation_cte AS (SELECT ts, `__GROUP_BY_KEY_0_host.name`, sum(per_series_value) AS value FROM __temporal_aggregation_cte WHERE isNaN(per_series_value) = ? GROUP BY ts, `__GROUP_BY_KEY_0_host.name`) SELECT ts, `__GROUP_BY_KEY_0_host.name`, multiIf(value <= 0, toFloat64(0), value <= 2.3283064365386963e-10, 2.3283064365386963e-10, value > 1.8446744073709552e+19, toFloat64('+Inf'), pow(2, ceil(log2(value) * 16) / 16)) AS __bucket, toFloat64(1) AS __result_0 FROM __spatial_aggregation_cte ORDER BY `__GROUP_BY_KEY_0_host.name`, ts, __bucket",
+				Args:  []any{"system.memory.usage", uint64(1747936800000), uint64(1747983420000), "unspecified", "system.memory.usage", uint64(1747947360000), uint64(1747983420000), 0},
+			},
+			expectedErr: nil,
+		},
+		{
+			name:        "test_gauge_heatmap_linear",
+			requestType: qbtypes.RequestTypeHeatmap,
+			query: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+				Signal:       telemetrytypes.SignalMetrics,
+				StepInterval: qbtypes.Step{Duration: 60 * time.Second},
+				Aggregations: []qbtypes.MetricAggregation{
+					{
+						MetricName:       "system.memory.usage",
+						Type:             metrictypes.GaugeType,
+						Temporality:      metrictypes.Unspecified,
+						TimeAggregation:  metrictypes.TimeAggregationAvg,
+						SpaceAggregation: metrictypes.SpaceAggregationSum,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{
+							Kind:       qbtypes.BucketsKindLinear,
+							LogScale:   qbtypes.MaxLogScale,
+							MaxValue:   500,
+							NumBuckets: 25,
+						},
+					},
+				},
+				GroupBy: []qbtypes.GroupByKey{
+					{
+						TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+							Name: "host.name",
+						},
+					},
+				},
+			},
+			expected: qbtypes.Statement{
+				Query: "WITH __temporal_aggregation_cte AS (SELECT fingerprint, toStartOfInterval(toDateTime(intDiv(unix_milli, 1000)), toIntervalSecond(60)) AS ts, `__GROUP_BY_KEY_0_host.name`, avg(value) AS per_series_value FROM signoz_metrics.distributed_samples_v4 AS points INNER JOIN (SELECT fingerprint, JSONExtractString(labels, 'host.name') AS `__GROUP_BY_KEY_0_host.name` FROM signoz_metrics.time_series_v4_6hrs WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli <= ? AND LOWER(temporality) LIKE LOWER(?) GROUP BY fingerprint, `__GROUP_BY_KEY_0_host.name`) AS filtered_time_series ON points.fingerprint = filtered_time_series.fingerprint WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli < ? GROUP BY fingerprint, ts, `__GROUP_BY_KEY_0_host.name` ORDER BY fingerprint, ts), __spatial_aggregation_cte AS (SELECT ts, `__GROUP_BY_KEY_0_host.name`, sum(per_series_value) AS value FROM __temporal_aggregation_cte WHERE isNaN(per_series_value) = ? GROUP BY ts, `__GROUP_BY_KEY_0_host.name`) SELECT ts, `__GROUP_BY_KEY_0_host.name`, multiIf(value > 500, toFloat64('+Inf'), least(greatest(ceil(value * 25 / 500), 1), 25) * 500 / 25) AS __bucket, toFloat64(1) AS __result_0 FROM __spatial_aggregation_cte ORDER BY `__GROUP_BY_KEY_0_host.name`, ts, __bucket",
+				Args:  []any{"system.memory.usage", uint64(1747936800000), uint64(1747983420000), "unspecified", "system.memory.usage", uint64(1747947360000), uint64(1747983420000), 0},
+			},
+			expectedErr: nil,
+		},
+		{
+			// cumulative keeps CanShortCircuitDelta false, so the counts reach the
+			// bucket differencing through the temporal CTE rather than the delta
+			// fast path
+			name:        "test_histogram_heatmap_cumulative",
+			requestType: qbtypes.RequestTypeHeatmap,
+			query: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+				Signal:       telemetrytypes.SignalMetrics,
+				StepInterval: qbtypes.Step{Duration: 60 * time.Second},
+				Aggregations: []qbtypes.MetricAggregation{
+					{
+						MetricName:       "http_server_duration_bucket",
+						Type:             metrictypes.HistogramType,
+						Temporality:      metrictypes.Cumulative,
+						TimeAggregation:  metrictypes.TimeAggregationRate,
+						SpaceAggregation: metrictypes.SpaceAggregationPercentile95,
+					},
+				},
+				GroupBy: []qbtypes.GroupByKey{
+					{
+						TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+							Name: "service.name",
+						},
+					},
+				},
+			},
+			expected: qbtypes.Statement{
+				Query: "WITH __temporal_aggregation_cte AS (SELECT ts, `__GROUP_BY_KEY_0_service.name`, `le`, multiIf(row_number() OVER rate_window = 1, nan, (per_series_value - lagInFrame(per_series_value, 1) OVER rate_window) < 0, per_series_value, per_series_value - lagInFrame(per_series_value, 1) OVER rate_window) AS per_series_value FROM (SELECT fingerprint, toStartOfInterval(toDateTime(intDiv(unix_milli, 1000)), toIntervalSecond(60)) AS ts, `__GROUP_BY_KEY_0_service.name`, `le`, max(value) AS per_series_value FROM signoz_metrics.distributed_samples_v4 AS points INNER JOIN (SELECT fingerprint, JSONExtractString(labels, 'service.name') AS `__GROUP_BY_KEY_0_service.name`, JSONExtractString(labels, 'le') AS `le` FROM signoz_metrics.time_series_v4_6hrs WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli <= ? AND LOWER(temporality) LIKE LOWER(?) GROUP BY fingerprint, `__GROUP_BY_KEY_0_service.name`, `le`) AS filtered_time_series ON points.fingerprint = filtered_time_series.fingerprint WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli < ? GROUP BY fingerprint, ts, `__GROUP_BY_KEY_0_service.name`, `le` ORDER BY fingerprint, ts) WINDOW rate_window AS (PARTITION BY fingerprint ORDER BY fingerprint, ts)), __spatial_aggregation_cte AS (SELECT ts, `__GROUP_BY_KEY_0_service.name`, `le`, sum(per_series_value) AS value FROM __temporal_aggregation_cte WHERE isNaN(per_series_value) = ? GROUP BY ts, `__GROUP_BY_KEY_0_service.name`, `le`) SELECT ts, `__GROUP_BY_KEY_0_service.name`, toFloat64(le) AS __bucket, greatest(value - lagInFrame(value, 1, 0) OVER __heatmap_window, 0) AS __result_0 FROM __spatial_aggregation_cte WINDOW __heatmap_window AS (PARTITION BY `__GROUP_BY_KEY_0_service.name`, ts ORDER BY toFloat64(le)) ORDER BY `__GROUP_BY_KEY_0_service.name`, ts, toFloat64(le)",
+				Args:  []any{"http_server_duration_bucket", uint64(1747936800000), uint64(1747983420000), "cumulative", "http_server_duration_bucket", uint64(1747947300000), uint64(1747983420000), 0},
+			},
+			expectedErr: nil,
+		},
+		{
+			name:        "test_sum_heatmap_cumulative",
+			requestType: qbtypes.RequestTypeHeatmap,
+			query: qbtypes.QueryBuilderQuery[qbtypes.MetricAggregation]{
+				Signal:       telemetrytypes.SignalMetrics,
+				StepInterval: qbtypes.Step{Duration: 60 * time.Second},
+				Aggregations: []qbtypes.MetricAggregation{
+					{
+						MetricName:       "signoz_calls_total",
+						Type:             metrictypes.SumType,
+						Temporality:      metrictypes.Cumulative,
+						TimeAggregation:  metrictypes.TimeAggregationIncrease,
+						SpaceAggregation: metrictypes.SpaceAggregationSum,
+						HeatmapBucketing: &qbtypes.HeatmapBucketing{
+							Kind:       qbtypes.BucketsKindLog,
+							LogScale:   qbtypes.MaxLogScale,
+							NumBuckets: qbtypes.DefaultNumBuckets,
+						},
+					},
+				},
+				GroupBy: []qbtypes.GroupByKey{
+					{
+						TelemetryFieldKey: telemetrytypes.TelemetryFieldKey{
+							Name: "service.name",
+						},
+					},
+				},
+			},
+			expected: qbtypes.Statement{
+				Query: "WITH __temporal_aggregation_cte AS (SELECT ts, `__GROUP_BY_KEY_0_service.name`, multiIf(row_number() OVER rate_window = 1, nan, (per_series_value - lagInFrame(per_series_value, 1) OVER rate_window) < 0, per_series_value, per_series_value - lagInFrame(per_series_value, 1) OVER rate_window) AS per_series_value FROM (SELECT fingerprint, toStartOfInterval(toDateTime(intDiv(unix_milli, 1000)), toIntervalSecond(60)) AS ts, `__GROUP_BY_KEY_0_service.name`, max(value) AS per_series_value FROM signoz_metrics.distributed_samples_v4 AS points INNER JOIN (SELECT fingerprint, JSONExtractString(labels, 'service.name') AS `__GROUP_BY_KEY_0_service.name` FROM signoz_metrics.time_series_v4_6hrs WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli <= ? AND LOWER(temporality) LIKE LOWER(?) GROUP BY fingerprint, `__GROUP_BY_KEY_0_service.name`) AS filtered_time_series ON points.fingerprint = filtered_time_series.fingerprint WHERE metric_name IN (?) AND unix_milli >= ? AND unix_milli < ? GROUP BY fingerprint, ts, `__GROUP_BY_KEY_0_service.name` ORDER BY fingerprint, ts) WINDOW rate_window AS (PARTITION BY fingerprint ORDER BY fingerprint, ts)), __spatial_aggregation_cte AS (SELECT ts, `__GROUP_BY_KEY_0_service.name`, sum(per_series_value) AS value FROM __temporal_aggregation_cte WHERE isNaN(per_series_value) = ? GROUP BY ts, `__GROUP_BY_KEY_0_service.name`) SELECT ts, `__GROUP_BY_KEY_0_service.name`, multiIf(value <= 0, toFloat64(0), value <= 2.3283064365386963e-10, 2.3283064365386963e-10, value > 1.8446744073709552e+19, toFloat64('+Inf'), pow(2, ceil(log2(value) * 16) / 16)) AS __bucket, toFloat64(1) AS __result_0 FROM __spatial_aggregation_cte ORDER BY `__GROUP_BY_KEY_0_service.name`, ts, __bucket",
+				Args:  []any{"signoz_calls_total", uint64(1747936800000), uint64(1747983420000), "cumulative", "signoz_calls_total", uint64(1747947300000), uint64(1747983420000), 0},
+			},
+			expectedErr: nil,
+		},
 		{
 			name:        "test_gauge_avg_sum",
 			requestType: qbtypes.RequestTypeTimeSeries,

pkg/types/dashboardtypes/perses_dashboard_test.go

@@ -524,6 +524,96 @@ func TestInvalidateUnknownPluginKind(t *testing.T) {
 	}
 }
 
+// TestHeatmapPanelQueryKinds pins the panel allowlist to what validateHeatmap
+// accepts in querybuildertypesv5: everything but a trace operator.
+func TestHeatmapPanelQueryKinds(t *testing.T) {
+	testCases := []struct {
+		description     string
+		queryPluginKind string
+		queryPluginSpec string
+		expectedAllowed bool
+	}{
+		{
+			description:     "a metrics builder query is allowed",
+			queryPluginKind: "signoz/BuilderQuery",
+			queryPluginSpec: `{"name": "A", "signal": "metrics", "aggregations": [
+				{"metricName": "http.server.request.duration", "timeAggregation": "increase", "spaceAggregation": "sum"}
+			]}`,
+			expectedAllowed: true,
+		},
+		{
+			description:     "a promql query is allowed",
+			queryPluginKind: "signoz/PromQLQuery",
+			queryPluginSpec: `{"name": "A", "query": "sum by (le) (increase(signoz_latency_bucket[5m]))"}`,
+			expectedAllowed: true,
+		},
+		{
+			description:     "a clickhouse query is allowed",
+			queryPluginKind: "signoz/ClickHouseSQL",
+			queryPluginSpec: `{"name": "A", "query": "SELECT ts, bucket, value FROM cells"}`,
+			expectedAllowed: true,
+		},
+		{
+			description:     "a formula is allowed",
+			queryPluginKind: "signoz/Formula",
+			queryPluginSpec: `{"name": "F1", "expression": "A / B"}`,
+			expectedAllowed: true,
+		},
+		{
+			description:     "a composite query is allowed, since a formula needs its disabled inputs alongside it",
+			queryPluginKind: "signoz/CompositeQuery",
+			queryPluginSpec: `{"queries": [
+				{"type": "builder_query", "spec": {"name": "A", "signal": "metrics", "disabled": true, "aggregations": [
+					{"metricName": "http.server.request.duration", "timeAggregation": "increase", "spaceAggregation": "sum"}
+				]}},
+				{"type": "builder_formula", "spec": {"name": "F1", "expression": "A * 2"}}
+			]}`,
+			expectedAllowed: true,
+		},
+		{
+			description:     "a trace operator is refused",
+			queryPluginKind: "signoz/TraceOperator",
+			queryPluginSpec: `{"name": "T1", "expression": "A => B"}`,
+			expectedAllowed: false,
+		},
+	}
+
+	for _, testCase := range testCases {
+		t.Run(testCase.description, func(t *testing.T) {
+			data := fmt.Sprintf(`{
+				"variables": [],
+				"panels": {
+					"p1": {
+						"kind": "Panel",
+						"spec": {
+							"links": [],
+							"plugin": {"kind": "signoz/HeatmapPanel", "spec": {}},
+							"queries": [{
+								"kind": "heatmap",
+								"spec": {
+									"plugin": {"kind": %q, "spec": %s}
+								}
+							}]
+						}
+					}
+				},
+				"links": [],
+				"layouts": []
+			}`, testCase.queryPluginKind, testCase.queryPluginSpec)
+
+			_, err := unmarshalDashboard([]byte(data))
+
+			if testCase.expectedAllowed {
+				require.NoError(t, err)
+				return
+			}
+
+			require.Error(t, err)
+			assert.Contains(t, err.Error(), "is not supported by panel kind")
+		})
+	}
+}
+
 func TestInvalidateOneInvalidPanel(t *testing.T) {
 	data := []byte(`{
 		"variables": [],

pkg/types/dashboardtypes/perses_plugin_wrappers.go

@@ -35,6 +35,7 @@ func (PanelPlugin) PrepareJSONSchema(s *jsonschema.Schema) error {
 		string(PanelKindTable):      schemaRef("DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesTablePanelSpec"),
 		string(PanelKindHistogram):  schemaRef("DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesHistogramPanelSpec"),
 		string(PanelKindList):       schemaRef("DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesListPanelSpec"),
+		string(PanelKindHeatmap):    schemaRef("DashboardtypesPanelPluginVariantGithubComSigNozSignozPkgTypesDashboardtypesHeatmapPanelSpec"),
 	})
 }
 
@@ -65,6 +66,7 @@ func (PanelPlugin) JSONSchemaOneOf() []any {
 		PanelPluginVariant[TablePanelSpec]{Kind: string(PanelKindTable)},
 		PanelPluginVariant[HistogramPanelSpec]{Kind: string(PanelKindHistogram)},
 		PanelPluginVariant[ListPanelSpec]{Kind: string(PanelKindList)},
+		PanelPluginVariant[HeatmapPanelSpec]{Kind: string(PanelKindHeatmap)},
 	}
 }
 
@@ -228,6 +230,7 @@ var (
 		PanelKindTable:      func() any { return new(TablePanelSpec) },
 		PanelKindHistogram:  func() any { return new(HistogramPanelSpec) },
 		PanelKindList:       func() any { return new(ListPanelSpec) },
+		PanelKindHeatmap:    func() any { return new(HeatmapPanelSpec) },
 	}
 	queryPluginSpecs = map[QueryPluginKind]func() any{
 		QueryKindBuilder:       func() any { return new(BuilderQuerySpec) },
@@ -250,6 +253,7 @@ var (
 		PanelKindPieChart:   {QueryKindBuilder, QueryKindComposite, QueryKindFormula, QueryKindTraceOperator, QueryKindClickHouseSQL},
 		PanelKindTable:      {QueryKindBuilder, QueryKindComposite, QueryKindFormula, QueryKindTraceOperator, QueryKindClickHouseSQL},
 		PanelKindList:       {QueryKindBuilder},
+		PanelKindHeatmap:    {QueryKindBuilder, QueryKindComposite, QueryKindFormula, QueryKindPromQL, QueryKindClickHouseSQL},
 	}
 )
 

pkg/types/dashboardtypes/perses_signoz_plugins.go

@@ -173,10 +173,11 @@ const (
 	PanelKindTable      PanelPluginKind = "signoz/TablePanel"
 	PanelKindHistogram  PanelPluginKind = "signoz/HistogramPanel"
 	PanelKindList       PanelPluginKind = "signoz/ListPanel"
+	PanelKindHeatmap    PanelPluginKind = "signoz/HeatmapPanel"
 )
 
 func (PanelPluginKind) Enum() []any {
-	return []any{PanelKindTimeSeries, PanelKindBarChart, PanelKindNumber, PanelKindPieChart, PanelKindTable, PanelKindHistogram, PanelKindList}
+	return []any{PanelKindTimeSeries, PanelKindBarChart, PanelKindNumber, PanelKindPieChart, PanelKindTable, PanelKindHistogram, PanelKindList, PanelKindHeatmap}
 }
 
 type TimeSeriesPanelSpec struct {
@@ -237,6 +238,56 @@ type ListPanelSpec struct {
 	SelectFields []telemetrytypes.TelemetryFieldKey `json:"selectFields,omitzero" validate:"dive"`
 }
 
+type HeatmapPanelSpec struct {
+	Visualization   BasicVisualization     `json:"visualization"`
+	Formatting      PanelFormatting        `json:"formatting"`
+	Axes            HeatmapAxes            `json:"axes"`
+	Legend          Legend                 `json:"legend"`
+	ChartAppearance HeatmapChartAppearance `json:"chartAppearance"`
+}
+
+// HeatmapAxes carries only the Y scale. The shared Axes type models a value
+// axis with soft bounds, where a heatmap's Y axis is the bucket boundaries the
+// response already fixed.
+type HeatmapAxes struct {
+	YScale HeatmapYScale `json:"yScale"`
+}
+
+type HeatmapChartAppearance struct {
+	Colors HeatmapColors `json:"colors"`
+}
+
+type HeatmapColors struct {
+	Mode    HeatmapColorMode  `json:"mode"`
+	Palette HeatmapPalette    `json:"palette"`
+	Scale   HeatmapColorScale `json:"scale"`
+	Steps   int               `json:"steps" validate:"omitempty,min=2,max=128"`
+	// MinCount and MaxCount clamp the colour scale; nil derives them from the
+	// grid, 0 and the highest count in it.
+	MinCount *float64 `json:"minCount"`
+	MaxCount *float64 `json:"maxCount"`
+	// Fill applies in opacity mode; empty means the selected group's legend colour.
+	Fill string `json:"fill"`
+}
+
+func (c *HeatmapColors) UnmarshalJSON(data []byte) error {
+	type alias HeatmapColors
+	var tmp alias
+	if err := json.Unmarshal(data, &tmp); err != nil {
+		return errors.WrapInvalidInputf(err, ErrCodeDashboardInvalidInput, "invalid heatmap colors")
+	}
+	*c = HeatmapColors(tmp)
+	return c.validate()
+}
+
+func (c HeatmapColors) validate() error {
+	if c.MinCount != nil && c.MaxCount != nil && *c.MinCount > *c.MaxCount {
+		return errors.NewInvalidInputf(ErrCodeDashboardInvalidInput,
+			"heatmap colors.minCount %v is greater than colors.maxCount %v", *c.MinCount, *c.MaxCount)
+	}
+	return nil
+}
+
 // ══════════════════════════════════════════════
 // Panel common types
 // ══════════════════════════════════════════════
@@ -709,3 +760,168 @@ func (p *PrecisionOption) UnmarshalJSON(data []byte) error {
 		return errors.NewInvalidInputf(ErrCodeDashboardInvalidInput, "invalid precision option %q: must be `0`, `1`, `2`, `3`, `4`, or `full`", v)
 	}
 }
+
+type HeatmapColorMode struct{ valuer.String }
+
+var (
+	HeatmapColorModePalette = HeatmapColorMode{valuer.NewString("palette")} // default
+	HeatmapColorModeOpacity = HeatmapColorMode{valuer.NewString("opacity")}
+)
+
+func (HeatmapColorMode) Enum() []any {
+	return []any{HeatmapColorModePalette, HeatmapColorModeOpacity}
+}
+
+func (m HeatmapColorMode) ValueOrDefault() string {
+	if m.IsZero() {
+		return HeatmapColorModePalette.StringValue()
+	}
+	return m.StringValue()
+}
+
+func (m HeatmapColorMode) MarshalJSON() ([]byte, error) {
+	return json.Marshal(m.ValueOrDefault())
+}
+
+func (m *HeatmapColorMode) UnmarshalJSON(data []byte) error {
+	var v string
+	if err := json.Unmarshal(data, &v); err != nil {
+		return errors.WrapInvalidInputf(err, ErrCodeDashboardInvalidInput, "invalid heatmap color mode: must be a string, one of `palette` or `opacity`")
+	}
+	mode := HeatmapColorMode{valuer.NewString(v)}
+	switch mode {
+	case HeatmapColorModePalette, HeatmapColorModeOpacity:
+		*m = mode
+		return nil
+	default:
+		return errors.NewInvalidInputf(ErrCodeDashboardInvalidInput, "invalid heatmap color mode %q: must be `palette` or `opacity`", v)
+	}
+}
+
+type HeatmapPalette struct{ valuer.String }
+
+var (
+	HeatmapPaletteIce      = HeatmapPalette{valuer.NewString("ice")}
+	HeatmapPaletteMoss     = HeatmapPalette{valuer.NewString("moss")}
+	HeatmapPaletteRust     = HeatmapPalette{valuer.NewString("rust")}
+	HeatmapPaletteGraphite = HeatmapPalette{valuer.NewString("graphite")}
+	HeatmapPaletteEmber    = HeatmapPalette{valuer.NewString("ember")}
+	HeatmapPaletteLagoon   = HeatmapPalette{valuer.NewString("lagoon")}
+	HeatmapPaletteOrchid   = HeatmapPalette{valuer.NewString("orchid")}
+	HeatmapPaletteVerdant  = HeatmapPalette{valuer.NewString("verdant")}
+	HeatmapPaletteLava     = HeatmapPalette{valuer.NewString("lava")} // default
+	HeatmapPaletteBeacon   = HeatmapPalette{valuer.NewString("beacon")}
+)
+
+func (HeatmapPalette) Enum() []any {
+	return []any{
+		HeatmapPaletteIce, HeatmapPaletteMoss, HeatmapPaletteRust, HeatmapPaletteGraphite,
+		HeatmapPaletteEmber, HeatmapPaletteLagoon, HeatmapPaletteOrchid, HeatmapPaletteVerdant,
+		HeatmapPaletteLava, HeatmapPaletteBeacon,
+	}
+}
+
+func (p HeatmapPalette) ValueOrDefault() string {
+	if p.IsZero() {
+		return HeatmapPaletteLava.StringValue()
+	}
+	return p.StringValue()
+}
+
+func (p HeatmapPalette) MarshalJSON() ([]byte, error) {
+	return json.Marshal(p.ValueOrDefault())
+}
+
+func (p *HeatmapPalette) UnmarshalJSON(data []byte) error {
+	var v string
+	if err := json.Unmarshal(data, &v); err != nil {
+		return errors.WrapInvalidInputf(err, ErrCodeDashboardInvalidInput, "invalid heatmap palette: must be a string, one of `ice`, `moss`, `rust`, `graphite`, `ember`, `lagoon`, `orchid`, `verdant`, `lava`, or `beacon`")
+	}
+	palette := HeatmapPalette{valuer.NewString(v)}
+	switch palette {
+	case HeatmapPaletteIce, HeatmapPaletteMoss, HeatmapPaletteRust, HeatmapPaletteGraphite,
+		HeatmapPaletteEmber, HeatmapPaletteLagoon, HeatmapPaletteOrchid, HeatmapPaletteVerdant,
+		HeatmapPaletteLava, HeatmapPaletteBeacon:
+		*p = palette
+		return nil
+	default:
+		return errors.NewInvalidInputf(ErrCodeDashboardInvalidInput, "invalid heatmap palette %q: must be `ice`, `moss`, `rust`, `graphite`, `ember`, `lagoon`, `orchid`, `verdant`, `lava`, or `beacon`", v)
+	}
+}
+
+type HeatmapYScale struct{ valuer.String }
+
+var (
+	HeatmapYScaleAuto   = HeatmapYScale{valuer.NewString("auto")} // default
+	HeatmapYScaleLinear = HeatmapYScale{valuer.NewString("linear")}
+	HeatmapYScaleLog    = HeatmapYScale{valuer.NewString("log")}
+	HeatmapYScaleSymlog = HeatmapYScale{valuer.NewString("symlog")}
+)
+
+func (HeatmapYScale) Enum() []any {
+	return []any{HeatmapYScaleAuto, HeatmapYScaleLinear, HeatmapYScaleLog, HeatmapYScaleSymlog}
+}
+
+func (s HeatmapYScale) ValueOrDefault() string {
+	if s.IsZero() {
+		return HeatmapYScaleAuto.StringValue()
+	}
+	return s.StringValue()
+}
+
+func (s HeatmapYScale) MarshalJSON() ([]byte, error) {
+	return json.Marshal(s.ValueOrDefault())
+}
+
+func (s *HeatmapYScale) UnmarshalJSON(data []byte) error {
+	var v string
+	if err := json.Unmarshal(data, &v); err != nil {
+		return errors.WrapInvalidInputf(err, ErrCodeDashboardInvalidInput, "invalid heatmap y scale: must be a string, one of `auto`, `linear`, `log`, or `symlog`")
+	}
+	scale := HeatmapYScale{valuer.NewString(v)}
+	switch scale {
+	case HeatmapYScaleAuto, HeatmapYScaleLinear, HeatmapYScaleLog, HeatmapYScaleSymlog:
+		*s = scale
+		return nil
+	default:
+		return errors.NewInvalidInputf(ErrCodeDashboardInvalidInput, "invalid heatmap y scale %q: must be `auto`, `linear`, `log`, or `symlog`", v)
+	}
+}
+
+type HeatmapColorScale struct{ valuer.String }
+
+var (
+	HeatmapColorScaleLog    = HeatmapColorScale{valuer.NewString("log")} // default
+	HeatmapColorScaleSqrt   = HeatmapColorScale{valuer.NewString("sqrt")}
+	HeatmapColorScaleLinear = HeatmapColorScale{valuer.NewString("linear")}
+)
+
+func (HeatmapColorScale) Enum() []any {
+	return []any{HeatmapColorScaleLog, HeatmapColorScaleSqrt, HeatmapColorScaleLinear}
+}
+
+func (s HeatmapColorScale) ValueOrDefault() string {
+	if s.IsZero() {
+		return HeatmapColorScaleLog.StringValue()
+	}
+	return s.StringValue()
+}
+
+func (s HeatmapColorScale) MarshalJSON() ([]byte, error) {
+	return json.Marshal(s.ValueOrDefault())
+}
+
+func (s *HeatmapColorScale) UnmarshalJSON(data []byte) error {
+	var v string
+	if err := json.Unmarshal(data, &v); err != nil {
+		return errors.WrapInvalidInputf(err, ErrCodeDashboardInvalidInput, "invalid heatmap color scale: must be a string, one of `log`, `sqrt`, or `linear`")
+	}
+	scale := HeatmapColorScale{valuer.NewString(v)}
+	switch scale {
+	case HeatmapColorScaleLog, HeatmapColorScaleSqrt, HeatmapColorScaleLinear:
+		*s = scale
+		return nil
+	default:
+		return errors.NewInvalidInputf(ErrCodeDashboardInvalidInput, "invalid heatmap color scale %q: must be `log`, `sqrt`, or `linear`", v)
+	}
+}

pkg/types/querybuildertypes/querybuildertypesv5/builder_elements.go

@@ -540,6 +540,8 @@ type MetricAggregation struct {
 	// reduce to operator for metric scalar requests
 	ReduceTo ReduceTo `json:"reduceTo,omitzero"`
 
+	HeatmapBucketing *HeatmapBucketing `json:"-"`
+
 	Reduced bool `json:"-"`
 }
 
@@ -554,6 +556,10 @@ func (m MetricAggregation) Copy() MetricAggregation {
 		valueFilterCopy := *m.ValueFilter
 		c.ValueFilter = &valueFilterCopy
 	}
+	if m.HeatmapBucketing != nil {
+		bucketingCopy := *m.HeatmapBucketing
+		c.HeatmapBucketing = &bucketingCopy
+	}
 	return c
 }
 

pkg/types/querybuildertypes/querybuildertypesv5/heatmap.go

@@ -0,0 +1,296 @@
+package querybuildertypesv5
+
+import (
+	"math"
+	"slices"
+
+	"github.com/SigNoz/signoz/pkg/errors"
+	"github.com/SigNoz/signoz/pkg/types/metrictypes"
+)
+
+const (
+	// HeatmapBucketColumn is the alias a heatmap statement gives the column holding
+	// a row's bucket upper bound. Every other aggregation returns a single numeric
+	// column the reader treats as the value; this name tells the two apart.
+	HeatmapBucketColumn = "__bucket"
+
+	DefaultNumBuckets = 60
+
+	// MaxLogScale is the resolution ClickHouse buckets every log heatmap at:
+	// 2^MaxLogScale bands per doubling. It is both the default and the finest
+	// available, since a coarser LogBucketsSpec.Scale folds down from it.
+	MaxLogScale = 4
+	// MinLogScale is one band per 16x, the coarsest axis worth rendering.
+	MinLogScale = -4
+
+	// A positive value approaching zero runs its band index off to -inf, so
+	// without a clamp one near-zero sample would stretch the axis by thousands
+	// of bands once AddHeatmapBucketsWithNoCounts spans it.
+	MinLogBandIndex = -512 // 2^-32, about 2.3e-10
+	MaxLogBandIndex = 1024 // 2^64, about 1.8e19
+)
+
+// MinLogUpperBound and MaxLogUpperBound are the ends the log axis is clamped
+// to. They do not vary with the requested scale.
+var (
+	MinLogUpperBound = math.Exp2(float64(MinLogBandIndex) / math.Exp2(MaxLogScale))
+	MaxLogUpperBound = math.Exp2(float64(MaxLogBandIndex) / math.Exp2(MaxLogScale))
+)
+
+// HeatmapBucketing is the bucket axis a heatmap statement builds in ClickHouse,
+// resolved from BucketOptions once the metric type is known. It stays nil for
+// histograms, whose upper bounds come from their own `le` labels.
+type HeatmapBucketing struct {
+	Kind BucketsKind
+	// LogScale is the resolution the caller asked for. ClickHouse always buckets
+	// at MaxLogScale, and postprocessing folds the axis down to this.
+	LogScale int
+	// MaxValue and NumBuckets are linear only.
+	MaxValue   float64
+	NumBuckets int
+}
+
+// ToHeatmapBucketing fills in what the caller left unset. An absent
+// BucketOptions resolves to the finest log axis, the one kind that needs nothing
+// from the caller.
+func (b *BucketOptions) ToHeatmapBucketing() HeatmapBucketing {
+	resolved := HeatmapBucketing{
+		Kind:       BucketsKindLog,
+		LogScale:   MaxLogScale,
+		NumBuckets: DefaultNumBuckets,
+	}
+	if b == nil {
+		return resolved
+	}
+
+	switch spec := b.Spec.(type) {
+	case LinearBucketsSpec:
+		resolved.Kind = BucketsKindLinear
+		resolved.MaxValue = spec.MaxValue
+		if spec.NumBuckets > 0 {
+			resolved.NumBuckets = spec.NumBuckets
+		}
+	case LogBucketsSpec:
+		if spec.Scale != nil {
+			resolved.LogScale = *spec.Scale
+		}
+	}
+
+	return resolved
+}
+
+// This cannot be called in validateHeatmap cuz type is resolved in querier.go.
+func (a *MetricAggregation) VerifyAndApplyBucketOptions(bucketOptions *BucketOptions) error {
+	switch a.Type {
+	case metrictypes.HistogramType:
+		if bucketOptions != nil {
+			return errors.NewInvalidInputf(errors.CodeInvalidInput,
+				"bucketOptions are not supported for histogram metrics: %q takes its bucket axis from its own `le` labels, so nothing in the spec would be applied", a.MetricName)
+		}
+		a.HeatmapBucketing = nil
+		return nil
+	// A summary carries no upper bounds of its own either, and its samples reach
+	// the final select the same way a gauge's do, so it buckets identically.
+	case metrictypes.GaugeType, metrictypes.SumType, metrictypes.SummaryType:
+		bucketing := bucketOptions.ToHeatmapBucketing()
+		a.HeatmapBucketing = &bucketing
+		return nil
+	case metrictypes.UnspecifiedType:
+		return errors.NewInvalidInputf(errors.CodeInvalidInput,
+			"heatmaps need a metric whose type is known: no type is recorded for %q, so its bucket axis cannot be chosen", a.MetricName)
+	case metrictypes.ExpHistogramType:
+		return errors.Newf(errors.TypeUnsupported, errors.CodeUnsupported,
+			"heatmaps are not supported for exponential histograms yet: %q keeps its bucket counts in a sketch column, which needs its own reader", a.MetricName)
+	default:
+		return errors.Newf(errors.TypeUnsupported, errors.CodeUnsupported,
+			"heatmaps are not supported for %s metrics", a.Type.StringValue())
+	}
+}
+
+func MergeBucketUpperBounds(tsData ...*TimeSeriesData) map[int][]float64 {
+	upperBoundsByAggregation := map[int][]float64{}
+
+	for _, data := range tsData {
+		if data == nil {
+			continue
+		}
+		for _, aggBucket := range data.Aggregations {
+			if len(aggBucket.Meta.Buckets) == 0 {
+				continue
+			}
+			upperBoundsByAggregation[aggBucket.Index] = append(upperBoundsByAggregation[aggBucket.Index], aggBucket.Meta.Buckets...)
+		}
+	}
+
+	for index, upperBounds := range upperBoundsByAggregation {
+		slices.Sort(upperBounds)
+		upperBoundsByAggregation[index] = slices.Compact(upperBounds)
+	}
+
+	return upperBoundsByAggregation
+}
+
+// DownscaleHeatmapResolution folds the MaxLogScale axis ClickHouse buckets at
+// down to toScale, merging every 2^(MaxLogScale-toScale) adjacent bands into
+// one. The coarser upper bounds are a subset of the finer ones, so the fold is
+// exact.
+func DownscaleHeatmapResolution(tsData *TimeSeriesData, toScale int) {
+	if tsData == nil || toScale >= MaxLogScale {
+		return
+	}
+	for _, aggBucket := range tsData.Aggregations {
+		downscaleHeatmapResolutionForAggregation(aggBucket, toScale)
+	}
+}
+
+func downscaleHeatmapResolutionForAggregation(aggBucket *AggregationBucket, toScale int) {
+	if aggBucket == nil || len(aggBucket.Meta.Buckets) == 0 {
+		return
+	}
+
+	factor := int(math.Exp2(float64(MaxLogScale - toScale)))
+
+	// Merging is by index in the exponential mapping, not by position in
+	// Meta.Buckets, which lists only the upper bounds some series reached.
+	coarseUpperBounds := make([]float64, 0, len(aggBucket.Meta.Buckets))
+	upperBoundToCoarseIndex := make(map[float64]int, len(aggBucket.Meta.Buckets))
+	mergedInto := make([]int, len(aggBucket.Meta.Buckets))
+	for index, upperBound := range aggBucket.Meta.Buckets {
+		coarsened := coarsenUpperBound(upperBound, toScale, factor)
+		coarseIndex, ok := upperBoundToCoarseIndex[coarsened]
+		if !ok {
+			coarseIndex = len(coarseUpperBounds)
+			coarseUpperBounds = append(coarseUpperBounds, coarsened)
+			upperBoundToCoarseIndex[coarsened] = coarseIndex
+		}
+		mergedInto[index] = coarseIndex
+	}
+
+	overflowIndex := len(coarseUpperBounds)
+	for _, series := range aggBucket.Series {
+		for _, point := range series.Values {
+			if len(point.Values) == 0 {
+				continue
+			}
+			coarseCounts := make([]float64, overflowIndex+1)
+			for index, count := range point.Values {
+				if index >= len(mergedInto) {
+					coarseCounts[overflowIndex] += count
+					continue
+				}
+				coarseCounts[mergedInto[index]] += count
+			}
+			point.Values = coarseCounts
+		}
+	}
+	aggBucket.Meta.Buckets = coarseUpperBounds
+}
+
+// coarsenUpperBound moves an upper bound from the MaxLogScale exponential axis
+// onto the toScale one. The zero band has no exponent to rescale and stays put.
+func coarsenUpperBound(upperBound float64, toScale, factor int) float64 {
+	if upperBound <= 0 || math.IsInf(upperBound, 0) || math.IsNaN(upperBound) {
+		return upperBound
+	}
+	index := int(math.Round(math.Log2(upperBound) * math.Exp2(MaxLogScale)))
+	merged := int(math.Ceil(float64(index) / float64(factor)))
+	return math.Exp2(float64(merged) / math.Exp2(float64(toScale)))
+}
+
+// AddHeatmapBucketsWithNoCounts spans the range from the lowest upper bound some
+// series reached to the highest. Meta.Buckets leaves the ones in between out
+// entirely, so without this a gap renders with its two sides touching.
+//
+// Only a value-derived axis can be spanned: its upper bounds come from an index
+// that is a pure function of the value, so the ones in between are known without
+// having seen them. Nothing says what sits between two `le` labels.
+func AddHeatmapBucketsWithNoCounts(tsData *TimeSeriesData, bucketing HeatmapBucketing) {
+	if tsData == nil {
+		return
+	}
+	for _, aggBucket := range tsData.Aggregations {
+		addHeatmapBucketsWithNoCountsForAggregation(aggBucket, bucketing)
+	}
+}
+
+func addHeatmapBucketsWithNoCountsForAggregation(aggBucket *AggregationBucket, bucketing HeatmapBucketing) {
+	if aggBucket == nil || len(aggBucket.Meta.Buckets) == 0 {
+		return
+	}
+
+	// The zero bucket holds everything at or below zero. It has no index on either
+	// axis and sits below every other upper bound, so it keeps index 0 and the
+	// fill runs over the rest.
+	offset := 0
+	if aggBucket.Meta.Buckets[0] <= 0 {
+		offset = 1
+	}
+	positive := aggBucket.Meta.Buckets[offset:]
+	if len(positive) == 0 {
+		return
+	}
+
+	// Only finite upper bounds have an index, and the fill sizes a slice from
+	// one. Nothing should put +Inf or NaN on the axis, but bail if it happens.
+	indexes := make([]int, len(positive))
+	for i, upperBound := range positive {
+		if math.IsInf(upperBound, 0) || math.IsNaN(upperBound) {
+			return
+		}
+		indexes[i] = bucketing.calculateIndexOfUpperBound(upperBound)
+	}
+	lowest, highest := slices.Min(indexes), slices.Max(indexes)
+
+	denseUpperBounds := append([]float64{}, aggBucket.Meta.Buckets[:offset]...)
+	for index := lowest; index <= highest; index++ {
+		denseUpperBounds = append(denseUpperBounds, bucketing.calculateUpperBoundAtIndex(index))
+	}
+	if len(denseUpperBounds) == len(aggBucket.Meta.Buckets) {
+		return
+	}
+
+	// Counts map through their index rather than by matching upper bounds, so a
+	// regenerated upper bound differing from ClickHouse's in its last bit still
+	// lands where it came from.
+	shiftedTo := make([]int, len(aggBucket.Meta.Buckets))
+	for i, index := range indexes {
+		shiftedTo[i+offset] = index - lowest + offset
+	}
+
+	overflowIndex := len(denseUpperBounds)
+	for _, series := range aggBucket.Series {
+		for _, point := range series.Values {
+			if len(point.Values) == 0 {
+				continue
+			}
+			denseCounts := make([]float64, overflowIndex+1)
+			for index, count := range point.Values {
+				if index >= len(shiftedTo) {
+					denseCounts[overflowIndex] += count
+					continue
+				}
+				denseCounts[shiftedTo[index]] += count
+			}
+			point.Values = denseCounts
+		}
+	}
+	aggBucket.Meta.Buckets = denseUpperBounds
+}
+
+// calculateIndexOfUpperBound and calculateUpperBoundAtIndex are inverses over
+// the axis being returned, so they read h.LogScale rather than the MaxLogScale
+// ClickHouse bucketed at: k * maxValue / numBuckets on a linear axis,
+// 2^(k / 2^scale) on a log one.
+func (h HeatmapBucketing) calculateIndexOfUpperBound(upperBound float64) int {
+	if h.Kind == BucketsKindLinear {
+		return int(math.Round(upperBound * float64(h.NumBuckets) / h.MaxValue))
+	}
+	return int(math.Round(math.Log2(upperBound) * math.Exp2(float64(h.LogScale))))
+}
+
+func (h HeatmapBucketing) calculateUpperBoundAtIndex(index int) float64 {
+	if h.Kind == BucketsKindLinear {
+		return float64(index) * h.MaxValue / float64(h.NumBuckets)
+	}
+	return math.Exp2(float64(index) / math.Exp2(float64(h.LogScale)))
+}

pkg/types/querybuildertypes/querybuildertypesv5/req.go

@@ -397,6 +397,8 @@ type QueryRangeRequest struct {
 	PromQLProvider string `json:"-"`
 
 	FormatOptions *FormatOptions `json:"formatOptions,omitempty"`
+
+	BucketOptions *BucketOptions `json:"bucketOptions,omitempty"`
 }
 
 // PrepareJSONSchema adds description to the QueryRangeRequest schema.
@@ -734,3 +736,130 @@ func (r *QueryRangeRequest) GetQueriesSupportingZeroDefault() map[string]bool {
 
 	return canDefaultZero
 }
+
+type BucketOptions struct {
+	Kind BucketsKind `json:"kind"`
+	Spec any         `json:"spec"`
+}
+
+type BucketsKind struct {
+	valuer.String
+}
+
+var (
+	BucketsKindLinear = BucketsKind{valuer.NewString("linear")}
+	BucketsKindLog    = BucketsKind{valuer.NewString("log")}
+)
+
+// Enum implements jsonschema.Enum.
+func (BucketsKind) Enum() []any {
+	return []any{
+		BucketsKindLinear,
+		BucketsKindLog,
+	}
+}
+
+// LinearBucketsSpec divides (0, MaxValue] into NumBuckets equal bands.
+type LinearBucketsSpec struct {
+	// Everything above MaxValue is counted in the trailing overflow band. Evenly
+	// spaced upper bounds have no top to divide without it, so it is required.
+	MaxValue   float64 `json:"maxValue" required:"true"`
+	NumBuckets int     `json:"numBuckets,omitempty"`
+}
+
+// LogBucketsSpec spaces upper bounds at 2^Scale bands per doubling, the mapping
+// an exponential histogram uses.
+type LogBucketsSpec struct {
+	// ClickHouse always buckets at MaxLogScale and the surplus is folded away
+	// afterwards, so every Scale reads the same cache entry. MaxLogScale applies
+	// when unset.
+	Scale *int `json:"scale,omitempty"`
+}
+
+func (b *BucketOptions) UnmarshalJSON(data []byte) error {
+	var shadow struct {
+		Kind BucketsKind     `json:"kind"`
+		Spec json.RawMessage `json:"spec"`
+	}
+	if err := binding.JSON.BindBody(bytes.NewReader(data), &shadow, binding.WithDisallowUnknownFields(true)); err != nil {
+		return err
+	}
+
+	b.Kind = shadow.Kind
+
+	// An absent spec is a malformed pair rather than a request for defaults;
+	// `"spec": {}` asks for those.
+	if len(shadow.Spec) == 0 {
+		return errors.NewInvalidInputf(errors.CodeInvalidInput,
+			"bucketOptions spec is required, use an empty object for the kind's defaults")
+	}
+
+	switch shadow.Kind {
+	case BucketsKindLinear:
+		var spec LinearBucketsSpec
+		if err := binding.JSON.BindBody(bytes.NewReader(shadow.Spec), &spec, binding.WithDisallowUnknownFields(true), binding.WithUnknownFieldContext("linear buckets spec")); err != nil {
+			return err
+		}
+		b.Spec = spec
+
+	case BucketsKindLog:
+		var spec LogBucketsSpec
+		if err := binding.JSON.BindBody(bytes.NewReader(shadow.Spec), &spec, binding.WithDisallowUnknownFields(true), binding.WithUnknownFieldContext("log buckets spec")); err != nil {
+			return err
+		}
+		b.Spec = spec
+
+	default:
+		return errors.NewInvalidInputf(
+			errors.CodeInvalidInput,
+			"invalid bucketOptions kind: %s",
+			shadow.Kind.StringValue(),
+		).WithAdditional(
+			"Valid bucket kinds are: linear, log",
+		)
+	}
+
+	return nil
+}
+
+// bucketOptionsLinear and bucketOptionsLog are the OpenAPI schemas for the two
+// BucketOptions variants. They have to be named types: the reflector turns an
+// anonymous one into an inline subschema, leaving the discriminator mapping in
+// PrepareJSONSchema pointing at components that were never emitted. `kind` is
+// required:"true" on both so oapi-codegen renders the discriminator non-pointer.
+type bucketOptionsLinear struct {
+	Kind BucketsKind       `json:"kind" required:"true" description:"How the upper bounds are spaced."`
+	Spec LinearBucketsSpec `json:"spec" required:"true" description:"The evenly spaced bucket specification."`
+}
+
+type bucketOptionsLog struct {
+	Kind BucketsKind    `json:"kind" required:"true" description:"How the upper bounds are spaced."`
+	Spec LogBucketsSpec `json:"spec" required:"true" description:"The logarithmic bucket specification."`
+}
+
+var _ jsonschema.OneOfExposer = BucketOptions{}
+
+func (BucketOptions) JSONSchemaOneOf() []any {
+	return []any{
+		bucketOptionsLinear{},
+		bucketOptionsLog{},
+	}
+}
+
+var _ jsonschema.Preparer = BucketOptions{}
+
+// PrepareJSONSchema marks the options as a `kind`-discriminated union;
+// signoz.attachDiscriminators promotes it and strips the base properties.
+func (BucketOptions) PrepareJSONSchema(s *jsonschema.Schema) error {
+	if s.ExtraProperties == nil {
+		s.ExtraProperties = map[string]any{}
+	}
+	s.ExtraProperties["x-signoz-discriminator"] = map[string]any{
+		"propertyName": "kind",
+		"mapping": map[string]string{
+			BucketsKindLinear.StringValue(): "#/components/schemas/Querybuildertypesv5BucketOptionsLinear",
+			BucketsKindLog.StringValue():    "#/components/schemas/Querybuildertypesv5BucketOptionsLog",
+		},
+	}
+	return nil
+}

pkg/types/querybuildertypes/querybuildertypesv5/request_type.go

@@ -19,11 +19,11 @@ func (r *RequestType) UnmarshalJSON(data []byte) error {
 	}
 	v := RequestType{valuer.NewString(s)}
 	switch v {
-	case RequestTypeScalar, RequestTypeTimeSeries, RequestTypeRaw, RequestTypeRawStream, RequestTypeTrace, RequestTypeDistribution:
+	case RequestTypeScalar, RequestTypeTimeSeries, RequestTypeRaw, RequestTypeRawStream, RequestTypeTrace, RequestTypeDistribution, RequestTypeHeatmap:
 		*r = v
 		return nil
 	default:
-		return errors.NewInvalidInputf(errors.CodeInvalidInput, "unknown request type %q; allowed values: %s", s, "`scalar`, `time_series`, `raw`, `raw_stream`, `trace`, `distribution`")
+		return errors.NewInvalidInputf(errors.CodeInvalidInput, "unknown request type %q; allowed values: %s", s, "`scalar`, `time_series`, `raw`, `raw_stream`, `trace`, `distribution`, `heatmap`")
 	}
 }
 
@@ -41,6 +41,9 @@ var (
 	RequestTypeTrace = RequestType{valuer.NewString("trace")}
 	// []Bucket (struct{Lower,Upper,Count float64}), example: histogram.
 	RequestTypeDistribution = RequestType{valuer.NewString("distribution")}
+	// TimeSeriesData carrying one count per histogram bucket at each timestamp,
+	// with the shared bucket upper bounds on the aggregation's meta.
+	RequestTypeHeatmap = RequestType{valuer.NewString("heatmap")}
 )
 
 // IsAggregation returns true for request types that produce aggregated results
@@ -49,7 +52,7 @@ var (
 // For non-aggregation types (raw, raw_stream, trace), those fields are ignored
 // and don't need to be validated.
 func (r RequestType) IsAggregation() bool {
-	return r == RequestTypeTimeSeries || r == RequestTypeScalar || r == RequestTypeDistribution
+	return r == RequestTypeTimeSeries || r == RequestTypeScalar || r == RequestTypeDistribution || r == RequestTypeHeatmap
 }
 
 // Enum implements jsonschema.Enum; returns the acceptable values for RequestType.
@@ -60,6 +63,7 @@ func (RequestType) Enum() []any {
 		RequestTypeRaw,
 		RequestTypeRawStream,
 		RequestTypeTrace,
+		RequestTypeHeatmap,
 		// RequestTypeDistribution,
 	}
 }

pkg/types/querybuildertypes/querybuildertypesv5/resp.go

@@ -138,12 +138,10 @@ type TimeSeriesData struct {
 }
 
 type AggregationBucket struct {
-	Index int    `json:"index"` // or string Alias
-	Alias string `json:"alias"`
-	Meta  struct {
-		Unit string `json:"unit,omitempty"`
-	} `json:"meta,omitempty"`
-	Series []*TimeSeries `json:"series"` // no extra nesting
+	Index  int             `json:"index"` // or string Alias
+	Alias  string          `json:"alias"`
+	Meta   AggregationMeta `json:"meta,omitempty"`
+	Series []*TimeSeries   `json:"series"` // no extra nesting
 
 	PredictedSeries  []*TimeSeries `json:"predictedSeries,omitempty"`
 	UpperBoundSeries []*TimeSeries `json:"upperBoundSeries,omitempty"`
@@ -151,6 +149,54 @@ type AggregationBucket struct {
 	AnomalyScores    []*TimeSeries `json:"anomalyScores,omitempty"`
 }
 
+// ReindexValuesToNewUpperBounds moves each count to the index its upper bound
+// holds in onto, a superset of Meta.Buckets. No count changes, only its position
+// in Values.
+func (a *AggregationBucket) ReindexValuesToNewUpperBounds(onto []float64) {
+	if a == nil {
+		return
+	}
+
+	from := a.Meta.Buckets
+	if len(onto) == 0 || slices.Equal(from, onto) {
+		return
+	}
+
+	upperBoundToIndex := make(map[float64]int, len(onto))
+	for index, upperBound := range onto {
+		upperBoundToIndex[upperBound] = index
+	}
+
+	for _, series := range a.Series {
+		for _, point := range series.Values {
+			if len(point.Values) == 0 {
+				continue
+			}
+			reindexed := make([]float64, len(onto)+1)
+			for index, count := range point.Values {
+				if index >= len(from) {
+					reindexed[len(onto)] = count
+					break
+				}
+				if newIndex, ok := upperBoundToIndex[from[index]]; ok {
+					reindexed[newIndex] = count
+				}
+			}
+			point.Values = reindexed
+		}
+	}
+
+	a.Meta.Buckets = onto
+}
+
+type AggregationMeta struct {
+	Unit string `json:"unit,omitempty"`
+	// Buckets holds ascending upper bounds shared by every series in the
+	// AggregationBucket, set only for heatmap results. Each point's Values holds
+	// len(Buckets)+1 counts: one per bound, then the open-above overflow.
+	Buckets []float64 `json:"buckets,omitempty"`
+}
+
 type TimeSeries struct {
 	Labels []*Label           `json:"labels,omitempty"`
 	Values []*TimeSeriesValue `json:"values"`
@@ -254,13 +300,9 @@ type TimeSeriesValue struct {
 	// on the client side, these partial values are rendered differently.
 	Partial bool `json:"partial,omitempty"`
 
-	// for the heatmap type chart
+	// Values holds one count per histogram bucket for heatmap results, in the
+	// order of the aggregation's Meta.Buckets. Value is unused in that case.
 	Values []float64 `json:"values,omitempty"`
-	Bucket *Bucket   `json:"bucket,omitempty"`
-}
-
-type Bucket struct {
-	Step float64 `json:"step"`
 }
 
 type ColumnType struct {

pkg/types/querybuildertypes/querybuildertypesv5/series_limit.go

@@ -127,7 +127,7 @@ func calculateSeriesValue(series *TimeSeries) float64 {
 
 	// For single-point series, return that value directly
 	if len(series.Values) == 1 {
-		value := series.Values[0].Value
+		value := calculatePointValue(series.Values[0])
 		if math.IsNaN(value) || math.IsInf(value, 0) {
 			return 0.0
 		}
@@ -139,10 +139,11 @@ func calculateSeriesValue(series *TimeSeries) float64 {
 	var count float64
 
 	for _, point := range series.Values {
-		if math.IsNaN(point.Value) || math.IsInf(point.Value, 0) {
+		value := calculatePointValue(point)
+		if math.IsNaN(value) || math.IsInf(value, 0) {
 			continue
 		}
-		sum += point.Value
+		sum += value
 		count++
 	}
 
@@ -154,6 +155,25 @@ func calculateSeriesValue(series *TimeSeries) float64 {
 	return sum / count
 }
 
+// calculatePointValue returns what a point contributes to its series' rank.
+// Heatmap points carry one count per bucket in Values and leave Value at zero,
+// so they rank on the total across buckets.
+func calculatePointValue(point *TimeSeriesValue) float64 {
+	if len(point.Values) == 0 {
+		return point.Value
+	}
+
+	var total float64
+	for _, value := range point.Values {
+		if math.IsNaN(value) || math.IsInf(value, 0) {
+			continue
+		}
+		total += value
+	}
+
+	return total
+}
+
 // convertValueToString converts various types to string for comparison.
 func convertValueToString(value any) string {
 	switch v := value.(type) {

pkg/types/querybuildertypes/querybuildertypesv5/series_limit_test.go

@@ -1,10 +1,12 @@
 package querybuildertypesv5
 
 import (
+	"math"
 	"testing"
 
 	"github.com/SigNoz/signoz/pkg/types/telemetrytypes"
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func TestApplySeriesLimit(t *testing.T) {
@@ -232,3 +234,81 @@ func TestApplySeriesLimit(t *testing.T) {
 		assert.Equal(t, 40.0, result[2].Values[0].Value)
 	})
 }
+
+func TestApplySeriesLimitRanksHeatmapSeriesByBucketTotals(t *testing.T) {
+	// A reshaped heatmap point leaves Value at zero and holds one count per
+	// bucket in Values, so ranking has to sum the buckets to see any difference.
+	series := []*TimeSeries{
+		{
+			Labels: []*Label{{
+				Key:   telemetrytypes.TelemetryFieldKey{Name: "service.name"},
+				Value: "quiet",
+			}},
+			Values: []*TimeSeriesValue{
+				{Timestamp: 1000, Values: []float64{1, 2, 0}},
+				{Timestamp: 1060, Values: []float64{0, 1, 0}},
+			},
+		},
+		{
+			Labels: []*Label{{
+				Key:   telemetrytypes.TelemetryFieldKey{Name: "service.name"},
+				Value: "busy",
+			}},
+			Values: []*TimeSeriesValue{
+				{Timestamp: 1000, Values: []float64{40, 60, 5}},
+				{Timestamp: 1060, Values: []float64{30, 70, 5}},
+			},
+		},
+		{
+			Labels: []*Label{{
+				Key:   telemetrytypes.TelemetryFieldKey{Name: "service.name"},
+				Value: "middling",
+			}},
+			Values: []*TimeSeriesValue{
+				{Timestamp: 1000, Values: []float64{5, 5, 0}},
+				{Timestamp: 1060, Values: []float64{4, 6, 0}},
+			},
+		},
+	}
+
+	result := ApplySeriesLimit(series, nil, 2)
+
+	require.Len(t, result, 2)
+	assert.Equal(t, "busy", result[0].Labels[0].Value)
+	assert.Equal(t, "middling", result[1].Labels[0].Value)
+}
+
+func TestCalculatePointValue(t *testing.T) {
+	testCases := []struct {
+		description   string
+		point         *TimeSeriesValue
+		expectedValue float64
+	}{
+		{
+			description:   "a plain time series point ranks on its single value",
+			point:         &TimeSeriesValue{Timestamp: 1000, Value: 7},
+			expectedValue: 7,
+		},
+		{
+			description:   "a heatmap point ranks on the total across its buckets",
+			point:         &TimeSeriesValue{Timestamp: 1000, Values: []float64{1, 12, 14, 3}},
+			expectedValue: 30,
+		},
+		{
+			description:   "non-finite bucket counts are skipped",
+			point:         &TimeSeriesValue{Timestamp: 1000, Values: []float64{2, math.NaN(), math.Inf(1), 3}},
+			expectedValue: 5,
+		},
+		{
+			description:   "an empty bucket list falls back to the single value",
+			point:         &TimeSeriesValue{Timestamp: 1000, Value: 4, Values: []float64{}},
+			expectedValue: 4,
+		},
+	}
+
+	for _, testCase := range testCases {
+		t.Run(testCase.description, func(t *testing.T) {
+			assert.Equal(t, testCase.expectedValue, calculatePointValue(testCase.point))
+		})
+	}
+}

pkg/types/querybuildertypes/querybuildertypesv5/validation.go

@@ -2,6 +2,7 @@ package querybuildertypesv5
 
 import (
 	"fmt"
+	"math"
 	"slices"
 	"strings"
 
@@ -65,6 +66,8 @@ func wrapValidationError(cause error, contextIdentifier string, errorFormat stri
 const (
 	// Maximum limit for query results.
 	MaxQueryLimit = 10000
+
+	MaxNumBuckets = 512
 )
 
 // ValidationOption is a functional option for configuring validation behaviour.
@@ -581,7 +584,7 @@ func (r *QueryRangeRequest) Validate(opts ...ValidationOption) error {
 
 	// Validate request type
 	switch r.RequestType {
-	case RequestTypeRaw, RequestTypeRawStream, RequestTypeTrace, RequestTypeTimeSeries, RequestTypeScalar:
+	case RequestTypeRaw, RequestTypeRawStream, RequestTypeTrace, RequestTypeTimeSeries, RequestTypeScalar, RequestTypeHeatmap:
 		opts = append(opts, GetValidationOptions(r.RequestType)...)
 	default:
 		return errors.NewInvalidInputf(
@@ -589,10 +592,14 @@ func (r *QueryRangeRequest) Validate(opts ...ValidationOption) error {
 			"invalid request type: %s",
 			r.RequestType,
 		).WithAdditional(
-			"Valid request types are: raw, timeseries, scalar",
+			"Valid request types are: raw, timeseries, scalar, heatmap",
 		)
 	}
 
+	if err := r.validateHeatmap(); err != nil {
+		return err
+	}
+
 	// raw/trace request types don't support metric queries;
 	// metrics are always aggregated and there is no raw form.
 	if r.RequestType == RequestTypeRaw || r.RequestType == RequestTypeRawStream || r.RequestType == RequestTypeTrace {
@@ -630,11 +637,15 @@ func (r *QueryRangeRequest) ValidateRequestScope() ([]ValidationOption, error) {
 
 	var opts []ValidationOption
 	switch r.RequestType {
-	case RequestTypeRaw, RequestTypeRawStream, RequestTypeTrace, RequestTypeTimeSeries, RequestTypeScalar:
+	case RequestTypeRaw, RequestTypeRawStream, RequestTypeTrace, RequestTypeTimeSeries, RequestTypeScalar, RequestTypeHeatmap:
 		opts = GetValidationOptions(r.RequestType)
 	default:
 		return nil, errors.NewInvalidInputf(errors.CodeInvalidInput, "invalid request type: %s", r.RequestType).
-			WithAdditional("Valid request types are: raw, timeseries, scalar")
+			WithAdditional("Valid request types are: raw, timeseries, scalar, heatmap")
+	}
+
+	if err := r.validateHeatmap(); err != nil {
+		return nil, err
 	}
 
 	if r.RequestType == RequestTypeRaw || r.RequestType == RequestTypeRawStream || r.RequestType == RequestTypeTrace {
@@ -838,9 +849,135 @@ func validateQueryEnvelope(envelope QueryEnvelope, opts ...ValidationOption) err
 	}
 }
 
+func (r *QueryRangeRequest) validateHeatmap() error {
+	if r.RequestType != RequestTypeHeatmap {
+		if r.BucketOptions != nil {
+			return errors.NewInvalidInputf(
+				errors.CodeInvalidInput,
+				"bucketOptions are only supported for heatmap requests, got %s",
+				r.RequestType,
+			)
+		}
+		return nil
+	}
+
+	if r.FormatOptions != nil && r.FormatOptions.FillGaps {
+		return errors.NewInvalidInputf(errors.CodeInvalidInput,
+			"fillGaps is not supported for heatmap requests: an absent column means collection stopped, which a zero-filled column would hide")
+	}
+
+	if err := r.BucketOptions.validateBucketOptions(); err != nil {
+		return err
+	}
+
+	enabled := 0
+	for _, envelope := range r.CompositeQuery.Queries {
+		switch spec := envelope.Spec.(type) {
+		case QueryBuilderQuery[MetricAggregation]:
+			if err := validateHeatmapQuery(spec.Functions, spec.Having); err != nil {
+				return err
+			}
+			if spec.Disabled {
+				continue
+			}
+			enabled++
+		case QueryBuilderFormula:
+			if err := validateHeatmapQuery(spec.Functions, spec.Having); err != nil {
+				return err
+			}
+			if spec.Disabled {
+				continue
+			}
+			enabled++
+		case ClickHouseQuery:
+			if spec.Disabled {
+				continue
+			}
+			enabled++
+		case PromQuery:
+			if r.BucketOptions != nil {
+				return errors.NewInvalidInputf(errors.CodeInvalidInput,
+					"bucketOptions are not supported for promql heatmap requests: the bucket axis comes from the `le` labels the query returns, so nothing in the spec would be applied")
+			}
+			if spec.Disabled {
+				continue
+			}
+			enabled++
+		// An AI query decodes to the traces spec, so it lands here too. Admitting
+		// either signal means capping Aggregations at one: each carries its own
+		// Meta.Buckets, and a heatmap renders against a single bucket axis.
+		case QueryBuilderQuery[LogAggregation], QueryBuilderQuery[TraceAggregation]:
+			return errors.New(errors.TypeUnsupported, errors.CodeUnsupported,
+				"heatmaps are not supported for the logs and traces signals yet")
+		default:
+			return errors.NewInvalidInputf(errors.CodeInvalidInput,
+				"heatmap requests support one metrics builder query, one formula over them, one clickhouse query, or one promql query, got %q", envelope.Type.StringValue())
+		}
+	}
+
+	if enabled != 1 {
+		return errors.NewInvalidInputf(errors.CodeInvalidInput,
+			"heatmap requests need exactly one enabled query, got %d", enabled)
+	}
+
+	return nil
+}
+
+func (b *BucketOptions) validateBucketOptions() error {
+	if b == nil {
+		return nil
+	}
+
+	switch spec := b.Spec.(type) {
+	case LinearBucketsSpec:
+		if math.IsNaN(spec.MaxValue) || math.IsInf(spec.MaxValue, 0) || spec.MaxValue <= 0 {
+			return errors.NewInvalidInputf(errors.CodeInvalidInput,
+				"linear buckets need a finite maxValue greater than 0, got %v", spec.MaxValue)
+		}
+		if spec.NumBuckets < 0 || spec.NumBuckets > MaxNumBuckets {
+			return errors.NewInvalidInputf(errors.CodeInvalidInput,
+				"numBuckets must be between 1 and %d, got %d", MaxNumBuckets, spec.NumBuckets)
+		}
+
+	case LogBucketsSpec:
+		if spec.Scale != nil && (*spec.Scale < MinLogScale || *spec.Scale > MaxLogScale) {
+			return errors.NewInvalidInputf(errors.CodeInvalidInput,
+				"scale must be between %d and %d, got %d", MinLogScale, MaxLogScale, *spec.Scale)
+		}
+
+	default:
+		return errors.NewInvalidInputf(
+			errors.CodeInvalidInput,
+			"invalid bucketOptions kind: %s",
+			b.Kind.StringValue(),
+		).WithAdditional(
+			"Valid bucket kinds are: linear, log",
+		)
+	}
+
+	return nil
+}
+
+// validateHeatmapQuery refuses the per-query settings that cannot mean anything
+// on a heatmap. It runs on disabled queries too: a disabled query is a formula
+// input, so whatever it does still reaches the cells.
+func validateHeatmapQuery(functions []Function, having *Having) error {
+	if len(functions) > 0 {
+		return errors.NewInvalidInputf(errors.CodeInvalidInput,
+			"functions are not supported for heatmap requests: a heatmap point is a count per bucket, not a single value")
+	}
+
+	if having != nil && having.Expression != "" {
+		return errors.NewInvalidInputf(errors.CodeInvalidInput,
+			"having is not supported for heatmap requests: it filters individual cells, which breaks the cumulative differencing")
+	}
+
+	return nil
+}
+
 func GetValidationOptions(requestType RequestType) []ValidationOption {
 	switch requestType {
-	case RequestTypeTimeSeries:
+	case RequestTypeTimeSeries, RequestTypeHeatmap:
 		return []ValidationOption{WithSkipSelectFieldValidation(), WithTimestampGroupByValidation()}
 	case RequestTypeScalar:
 		return []ValidationOption{WithSkipSelectFieldValidation(), WithReduceToValidation()}

tests/fixtures/querier.py

@@ -33,6 +33,7 @@ class RequestType:
     TIME_SERIES = "time_series"
     SCALAR = "scalar"
     TABLE = "table"
+    HEATMAP = "heatmap"
 
 
 @dataclass
@@ -173,6 +174,7 @@ def make_query_request(
     request_type: str = RequestType.TIME_SERIES,
     format_options: dict | None = None,
     variables: dict | None = None,
+    bucket_options: dict | None = None,
     no_cache: bool = True,
     timeout: int = QUERY_TIMEOUT,
     headers: dict | None = None,
@@ -191,6 +193,8 @@ def make_query_request(
     }
     if variables:
         payload["variables"] = variables
+    if bucket_options is not None:
+        payload["bucketOptions"] = bucket_options
 
     return requests.post(
         signoz.self.host_configs["8080"].get("/api/v5/query_range"),
@@ -359,6 +363,20 @@ def build_formula_query(
     return {"type": "builder_formula", "spec": spec}
 
 
+def build_log_bucket_options(scale: int | None = None) -> dict:
+    spec: dict[str, Any] = {}
+    if scale is not None:
+        spec["scale"] = scale
+    return {"kind": "log", "spec": spec}
+
+
+def build_linear_bucket_options(max_value: float, num_buckets: int | None = None) -> dict:
+    spec: dict[str, Any] = {"maxValue": max_value}
+    if num_buckets is not None:
+        spec["numBuckets"] = num_buckets
+    return {"kind": "linear", "spec": spec}
+
+
 def build_function(name: str, *args: Any) -> dict:
     func: dict[str, Any] = {"name": name}
     if args:
@@ -393,6 +411,24 @@ def get_all_series(response_json: dict, query_name: str) -> list[dict]:
     return aggregations[0].get("series", [])
 
 
+def get_heatmap_buckets(response_json: dict, query_name: str) -> list[float]:
+    """The ascending bucket upper bounds a heatmap result's counts are positional against.
+    Each point holds one more count than there are bounds: the trailing one is the open-above overflow."""
+    results = response_json.get("data", {}).get("data", {}).get("results", [])
+    result = find_named_result(results, query_name)
+    if not result:
+        return []
+    aggregations = result.get("aggregations", [])
+    if not aggregations:
+        return []
+    return aggregations[0].get("meta", {}).get("buckets", [])
+
+
+def get_heatmap_columns(response_json: dict, query_name: str) -> list[dict]:
+    """A heatmap result's points for its single series, oldest first."""
+    return sorted(get_series_values(response_json, query_name), key=lambda point: point["timestamp"])
+
+
 def get_scalar_value(response_json: dict, query_name: str) -> float | None:
     values = get_series_values(response_json, query_name)
     if values:

tests/integration/testdata/heatmap_histogram_3m.jsonl

@@ -0,0 +1,24 @@
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "1"}, "timestamp": "2025-01-10T10:00:00+00:00", "value": 10, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "2"}, "timestamp": "2025-01-10T10:00:00+00:00", "value": 20, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "4"}, "timestamp": "2025-01-10T10:00:00+00:00", "value": 30, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "+Inf"}, "timestamp": "2025-01-10T10:00:00+00:00", "value": 40, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "1"}, "timestamp": "2025-01-10T10:00:00+00:00", "value": 50, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "2"}, "timestamp": "2025-01-10T10:00:00+00:00", "value": 60, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "4"}, "timestamp": "2025-01-10T10:00:00+00:00", "value": 70, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "+Inf"}, "timestamp": "2025-01-10T10:00:00+00:00", "value": 80, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "1"}, "timestamp": "2025-01-10T10:01:00+00:00", "value": 11, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "2"}, "timestamp": "2025-01-10T10:01:00+00:00", "value": 23, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "4"}, "timestamp": "2025-01-10T10:01:00+00:00", "value": 33, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "+Inf"}, "timestamp": "2025-01-10T10:01:00+00:00", "value": 44, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "1"}, "timestamp": "2025-01-10T10:01:00+00:00", "value": 50, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "2"}, "timestamp": "2025-01-10T10:01:00+00:00", "value": 61, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "4"}, "timestamp": "2025-01-10T10:01:00+00:00", "value": 72, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "+Inf"}, "timestamp": "2025-01-10T10:01:00+00:00", "value": 82, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "1"}, "timestamp": "2025-01-10T10:02:00+00:00", "value": 13, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "2"}, "timestamp": "2025-01-10T10:02:00+00:00", "value": 25, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "4"}, "timestamp": "2025-01-10T10:02:00+00:00", "value": 39, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "api", "le": "+Inf"}, "timestamp": "2025-01-10T10:02:00+00:00", "value": 50, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "1"}, "timestamp": "2025-01-10T10:02:00+00:00", "value": 51, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "2"}, "timestamp": "2025-01-10T10:02:00+00:00", "value": 62, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "4"}, "timestamp": "2025-01-10T10:02:00+00:00", "value": 73, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}
+{"metric_name": "heatmap_request_duration_bucket", "labels": {"__temporality__": "Cumulative", "service": "web", "le": "+Inf"}, "timestamp": "2025-01-10T10:02:00+00:00", "value": 85, "temporality": "Cumulative", "type_": "Histogram", "is_monotonic": true, "flags": 0, "description": "", "unit": "", "env": "default", "resource_attrs": {}, "scope_attrs": {}}

tests/integration/tests/queriermetrics/15_heatmap.py

@@ -0,0 +1,883 @@
+from collections.abc import Callable
+from datetime import UTC, datetime, timedelta
+from http import HTTPStatus
+from uuid import uuid4
+
+import pytest
+
+from fixtures import types
+from fixtures.auth import USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD
+from fixtures.fs import get_testdata_file_path
+from fixtures.metrics import Metrics
+from fixtures.querier import (
+    RequestType,
+    assert_identical_query_response,
+    build_builder_query,
+    build_formula_query,
+    build_function,
+    build_linear_bucket_options,
+    build_log_bucket_options,
+    get_all_series,
+    get_error_message,
+    get_heatmap_buckets,
+    get_heatmap_columns,
+    index_series_by_label,
+    make_query_request,
+)
+
+HISTOGRAM_FILE = get_testdata_file_path("histogram_data_1h.jsonl")
+HISTOGRAM_COUNTERS_FILE = get_testdata_file_path("heatmap_histogram_3m.jsonl")
+MINUTE_MS = 60_000
+
+# the request-level rules below are checked before the metric is resolved, so
+# they need no data behind the query
+MISSING_METRIC = "test_heatmap_metric_that_is_never_written"
+
+
+def test_gauge_heatmap(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    minutes = 3
+    start_ms = int((now - timedelta(minutes=minutes + 1)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_gauge"
+
+    value_by_host = {f"host-{host:02d}": (200, 400, 800)[host // 8] + host for host in range(24)}
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric_name,
+                labels={"host": host},
+                timestamp=now - timedelta(minutes=minutes - minute),
+                value=value,
+                type_="Gauge",
+                is_monotonic=False,
+            )
+            for host, value in value_by_host.items()
+            for minute in range(minutes)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "max", "max", group_by=["host"])],
+        request_type=RequestType.HEATMAP,
+        bucket_options=build_log_bucket_options(0),
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    data = response.json()
+    # a group by gives one series per host, and all of them are counted against
+    # this one axis
+    assert get_heatmap_buckets(data, "A") == pytest.approx([256.0, 512.0, 1024.0])
+
+    columns_by_host = {
+        host: sorted(series["values"], key=lambda column: column["timestamp"])
+        for host, series in index_series_by_label(get_all_series(data, "A"), "host").items()
+    }
+    assert len(columns_by_host) == len(value_by_host)
+
+    # a column holds one count per bucket plus a trailing one for the overflow
+    for host, columns in columns_by_host.items():
+        occupied = int(host.removeprefix("host-")) // 8
+        assert [column["values"] for column in columns] == [[1 if slot == occupied else 0 for slot in range(4)]] * minutes
+
+    # summed across the hosts, a column is the spread of the 24 of them
+    for minute in range(minutes):
+        assert [sum(columns[minute]["values"][slot] for columns in columns_by_host.values()) for slot in range(4)] == [8, 8, 8, 0]
+
+
+def test_sum_heatmap(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    minutes = 3
+    start_ms = int((now - timedelta(minutes=minutes + 1)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_sum"
+
+    value_by_endpoint = {f"/endpoint-{endpoint:02d}": (100, 800)[endpoint // 8] + endpoint for endpoint in range(16)}
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric_name,
+                labels={"endpoint": endpoint},
+                timestamp=now - timedelta(minutes=minutes - minute),
+                value=value,
+                temporality="Cumulative",
+                type_="Sum",
+            )
+            for endpoint, value in value_by_endpoint.items()
+            for minute in range(minutes)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "max", "max", temporality="cumulative", group_by=["endpoint"])],
+        request_type=RequestType.HEATMAP,
+        bucket_options=build_log_bucket_options(0),
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    data = response.json()
+    assert get_heatmap_buckets(data, "A") == pytest.approx([128.0, 256.0, 512.0, 1024.0])
+
+    columns_by_endpoint = {
+        endpoint: sorted(series["values"], key=lambda column: column["timestamp"])
+        for endpoint, series in index_series_by_label(get_all_series(data, "A"), "endpoint").items()
+    }
+    assert len(columns_by_endpoint) == len(value_by_endpoint)
+
+    for endpoint, columns in columns_by_endpoint.items():
+        occupied = 0 if int(endpoint.removeprefix("/endpoint-")) < 8 else 3
+        assert [column["values"] for column in columns] == [[1 if slot == occupied else 0 for slot in range(5)]] * minutes
+
+    for minute in range(minutes):
+        assert [sum(columns[minute]["values"][slot] for columns in columns_by_endpoint.values()) for slot in range(5)] == [8, 0, 0, 8, 0]
+
+
+def test_histogram_heatmap(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    start_ms = int((now - timedelta(minutes=65)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_histogram"
+
+    insert_metrics(
+        Metrics.load_from_file(
+            HISTOGRAM_FILE,
+            base_time=now - timedelta(minutes=60),
+            metric_name_override=metric_name,
+        )
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "increase", "p50", group_by=["le"], filter_expression='endpoint = "/health"')],
+        request_type=RequestType.HEATMAP,
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    data = response.json()
+    # a histogram's axis is its recorded `le` bounds exactly, since nothing is
+    # known about what sits between two of them; `le=+Inf` has no finite bound
+    # and is counted in the trailing overflow slot
+    assert get_heatmap_buckets(data, "A") == [1000, 1500, 2000, 4000, 5000, 6000, 8000]
+
+    columns = get_heatmap_columns(data, "A")
+    assert columns
+    for column in columns:
+        assert len(column["values"]) == 8
+        assert all(count >= 0 for count in column["values"])
+    assert any(sum(column["values"]) > 0 for column in columns)
+
+
+def test_linear_buckets(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    start_ms = int((now - timedelta(minutes=30)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_linear"
+
+    # 100 wide buckets: 100 lands on the first, 250 on the third, and 1500 is
+    # past maxValue so it counts in the overflow
+    values = [100, 250, 1500]
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric_name,
+                labels={"service": "api"},
+                timestamp=now - timedelta(minutes=len(values) - minute),
+                value=value,
+                type_="Gauge",
+                is_monotonic=False,
+            )
+            for minute, value in enumerate(values)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "max", "max")],
+        request_type=RequestType.HEATMAP,
+        bucket_options=build_linear_bucket_options(1000, 10),
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    data = response.json()
+    # 200 is on the axis although nothing reached it, so the gap between 100 and
+    # 300 renders as a gap
+    assert get_heatmap_buckets(data, "A") == pytest.approx([100.0, 200.0, 300.0])
+
+    columns = get_heatmap_columns(data, "A")
+    assert [column["values"] for column in columns] == [
+        [1, 0, 0, 0],
+        [0, 0, 1, 0],
+        [0, 0, 0, 1],
+    ]
+
+def test_zero_bucket(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    start_ms = int((now - timedelta(minutes=30)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_zero_bucket"
+
+    values = [0, 256, 1024]
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric_name,
+                labels={"service": "api"},
+                timestamp=now - timedelta(minutes=len(values) - minute),
+                value=value,
+                type_="Gauge",
+                is_monotonic=False,
+            )
+            for minute, value in enumerate(values)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "max", "max")],
+        request_type=RequestType.HEATMAP,
+        bucket_options=build_log_bucket_options(0),
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    data = response.json()
+    # a log axis cannot place a value at or below zero, so those share a bucket
+    # of their own beneath the rest, and 512 is spanned above it
+    assert get_heatmap_buckets(data, "A") == pytest.approx([0.0, 256.0, 512.0, 1024.0])
+
+    columns = get_heatmap_columns(data, "A")
+    assert [column["values"] for column in columns] == [
+        [1, 0, 0, 0, 0],
+        [0, 1, 0, 0, 0],
+        [0, 0, 0, 1, 0],
+    ]
+
+
+def test_single_bucket(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    start_ms = int((now - timedelta(minutes=30)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_single_bucket"
+
+    values = [256, 256, 256]
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric_name,
+                labels={"service": "api"},
+                timestamp=now - timedelta(minutes=len(values) - minute),
+                value=value,
+                type_="Gauge",
+                is_monotonic=False,
+            )
+            for minute, value in enumerate(values)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "max", "max")],
+        request_type=RequestType.HEATMAP,
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    # a single bucket leaves no gap to span
+    data = response.json()
+    assert get_heatmap_buckets(data, "A") == [256.0]
+    assert [column["values"] for column in get_heatmap_columns(data, "A")] == [[1, 0], [1, 0], [1, 0]]
+
+
+# the values seeded below are 256 and 512, so each axis here runs from the bucket
+# holding 256 to the one holding 512 at that option's resolution: 16 log buckets
+# to the 2x, one bucket per 16x, or a linear bucket every maxValue/numBuckets
+@pytest.mark.parametrize(
+    "bucket_options, expected_buckets",
+    [
+        (None, [256 * 2 ** (step / 16) for step in range(17)]),
+        ({"kind": "log", "spec": {}}, [256 * 2 ** (step / 16) for step in range(17)]),
+        (build_log_bucket_options(4), [256 * 2 ** (step / 16) for step in range(17)]),
+        (build_log_bucket_options(-4), [2**16]),
+        (build_linear_bucket_options(1024), [step * 1024 / 60 for step in range(15, 31)]),
+        (build_linear_bucket_options(1024, 512), [step * 1024 / 512 for step in range(128, 257)]),
+    ],
+    ids=["absent", "log_defaults", "the_finest_scale", "the_coarsest_scale", "linear_without_num_buckets", "the_most_buckets"],
+)
+def test_bucket_option_limits(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+    bucket_options: dict | None,
+    expected_buckets: list[float],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    start_ms = int((now - timedelta(minutes=30)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_bucket_option_limits"
+
+    values = [256, 512]
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric_name,
+                labels={"service": "api"},
+                timestamp=now - timedelta(minutes=len(values) - minute),
+                value=value,
+                type_="Gauge",
+                is_monotonic=False,
+            )
+            for minute, value in enumerate(values)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "max", "max")],
+        request_type=RequestType.HEATMAP,
+        bucket_options=bucket_options,
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    data = response.json()
+    assert get_heatmap_buckets(data, "A") == pytest.approx(expected_buckets)
+
+    columns = get_heatmap_columns(data, "A")
+    assert len(columns) == len(values)
+    for column in columns:
+        assert len(column["values"]) == len(expected_buckets) + 1
+        assert sum(column["values"]) == 1
+
+
+def test_formula_heatmap(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    start_ms = int((now - timedelta(minutes=30)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_formula"
+
+    values = [100, 260, 295, 512, 1500]
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric_name,
+                labels={"service": "api"},
+                timestamp=now - timedelta(minutes=len(values) - minute),
+                value=value,
+                type_="Gauge",
+                is_monotonic=False,
+            )
+            for minute, value in enumerate(values)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    from_metric = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "max", "max")],
+        request_type=RequestType.HEATMAP,
+    )
+    assert from_metric.status_code == HTTPStatus.OK, from_metric.text
+
+    # a formula over the same query has to land its counts on the same axis
+    from_formula = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [
+            build_builder_query("A", metric_name, "max", "max", disabled=True),
+            build_formula_query("F1", "A"),
+        ],
+        request_type=RequestType.HEATMAP,
+    )
+    assert from_formula.status_code == HTTPStatus.OK, from_formula.text
+
+    assert get_heatmap_buckets(from_formula.json(), "F1") == pytest.approx(get_heatmap_buckets(from_metric.json(), "A"))
+    assert [column["values"] for column in get_heatmap_columns(from_formula.json(), "F1")] == [column["values"] for column in get_heatmap_columns(from_metric.json(), "A")]
+
+
+def test_promql_heatmap(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    end_ms = (int((datetime.now(tz=UTC) - timedelta(minutes=5)).timestamp() * 1000) // MINUTE_MS) * MINUTE_MS
+    start_ms = end_ms - MINUTE_MS
+
+    # the file's three columns are one minute apart, and the first sits a minute
+    # before the query window so the earliest step has something to increase over.
+    # Every counter in it stays above its own rise across a window, below which
+    # increase clips its back-extrapolation at the counter's zero point.
+    insert_metrics(Metrics.load_from_file(HISTOGRAM_COUNTERS_FILE, base_time=datetime.fromtimestamp((start_ms - MINUTE_MS) / 1000, tz=UTC)))
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [{"type": "promql", "spec": {"name": "A", "query": "sum by (le) (increase(heatmap_request_duration_bucket[2m]))", "step": 60}}],
+        request_type=RequestType.HEATMAP,
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    data = response.json()
+    assert get_heatmap_buckets(data, "A") == [1, 2, 4]
+
+    series = get_all_series(data, "A")
+    assert len(series) == 1
+    # `le` is what the bucket axis is read off, so it is never a group label
+    assert series[0].get("labels") in (None, [])
+
+    # what the query returns per `le` is still cumulative across `le`, so each
+    # count here is its own minus the one below it, and `le=+Inf` has no finite
+    # bound to sit on and lands in the trailing slot. increase over a 2m window of
+    # minutely samples extrapolates one minute's rise to two, hence the scaling.
+    assert [column["values"] for column in get_heatmap_columns(data, "A")] == [[2, 6, 2, 2], [6, 0, 8, 4]]
+
+
+def test_promql_heatmap_without_le(
+    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 = f"promql_heatmap_gauge_{uuid4().hex[:8]}"
+    end_ms = (int((datetime.now(tz=UTC) - timedelta(minutes=5)).timestamp() * 1000) // MINUTE_MS) * MINUTE_MS
+    start_ms = end_ms - MINUTE_MS
+
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric,
+                labels={"service": "api"},
+                timestamp=datetime.fromtimestamp(ts_ms / 1000, tz=UTC),
+                value=42.0,
+                type_="Gauge",
+                is_monotonic=False,
+            )
+            for ts_ms in range(start_ms, end_ms + 1, MINUTE_MS)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [{"type": "promql", "spec": {"name": "A", "query": metric, "step": 60}}],
+        request_type=RequestType.HEATMAP,
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    # the bucket axis of a promql heatmap comes from `le`, and a series without
+    # it has no bucket to sit in
+    assert get_heatmap_buckets(response.json(), "A") == []
+    assert get_heatmap_columns(response.json(), "A") == []
+
+
+def test_clickhouse_heatmap(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    start = now - timedelta(minutes=2)
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+
+    # a clickhouse statement names its bucket upper bound column `bucket`
+    response = make_query_request(
+        signoz,
+        token,
+        int(start.timestamp() * 1000),
+        int(now.timestamp() * 1000),
+        [
+            {
+                "type": "clickhouse_sql",
+                "spec": {
+                    "name": "A",
+                    "query": (f"SELECT toDateTime({int(start.timestamp())}) AS ts, toFloat64(10) AS bucket, toFloat64(3) AS `__result_0` UNION ALL SELECT toDateTime({int(start.timestamp())}) AS ts, toFloat64(20) AS bucket, toFloat64(7) AS `__result_0`"),
+                    "disabled": False,
+                },
+            }
+        ],
+        request_type=RequestType.HEATMAP,
+    )
+    assert response.status_code == HTTPStatus.OK, response.text
+
+    data = response.json()
+    assert get_heatmap_buckets(data, "A") == [10, 20]
+    # a clickhouse heatmap takes no bucketOptions, so the counts land exactly
+    # where the statement put them, plus the overflow slot
+    assert [column["values"] for column in get_heatmap_columns(data, "A")] == [[3, 7, 0]]
+
+
+def test_cached_heatmap_matches_uncached(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    metric_name = "test_heatmap_cache"
+
+    # the first half sits 16x below the second, so the cached range and the fresh
+    # one reach disjoint parts of the axis and neither may lose its counts
+    insert_metrics(
+        [
+            Metrics(
+                metric_name=metric_name,
+                labels={"service": "api"},
+                timestamp=now - timedelta(minutes=60 - minute),
+                value=256 if minute < 15 else 4096,
+                type_="Gauge",
+                is_monotonic=False,
+            )
+            for minute in range(30)
+        ]
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    query = [build_builder_query("A", metric_name, "max", "max")]
+    wide_start_ms = int((now - timedelta(minutes=60)).timestamp() * 1000)
+    wide_end_ms = int((now - timedelta(minutes=30)).timestamp() * 1000)
+
+    warmup = make_query_request(
+        signoz,
+        token,
+        wide_start_ms,
+        int((now - timedelta(minutes=45)).timestamp() * 1000),
+        query,
+        request_type=RequestType.HEATMAP,
+        no_cache=False,
+    )
+    assert warmup.status_code == HTTPStatus.OK, warmup.text
+
+    from_cache = make_query_request(signoz, token, wide_start_ms, wide_end_ms, query, request_type=RequestType.HEATMAP, no_cache=False)
+    assert from_cache.status_code == HTTPStatus.OK, from_cache.text
+
+    uncached = make_query_request(signoz, token, wide_start_ms, wide_end_ms, query, request_type=RequestType.HEATMAP, no_cache=True)
+    assert uncached.status_code == HTTPStatus.OK, uncached.text
+
+    assert_identical_query_response(from_cache, uncached)
+
+    # 256 and 4096 are 16x apart, which the axis covers at 16 buckets per 2x, and
+    # every column holds the one value its minute recorded
+    assert len(get_heatmap_buckets(uncached.json(), "A")) == 65
+    assert [sum(column["values"]) for column in get_heatmap_columns(uncached.json(), "A")] == [1] * 30
+
+
+def test_histogram_rejects_bucket_options(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    insert_metrics: Callable[[list[Metrics]], None],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    start_ms = int((now - timedelta(minutes=65)).timestamp() * 1000)
+    end_ms = int(now.timestamp() * 1000)
+    metric_name = "test_heatmap_histogram_with_bucket_options"
+
+    insert_metrics(
+        Metrics.load_from_file(
+            HISTOGRAM_FILE,
+            base_time=now - timedelta(minutes=60),
+            metric_name_override=metric_name,
+        )
+    )
+
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+    response = make_query_request(
+        signoz,
+        token,
+        start_ms,
+        end_ms,
+        [build_builder_query("A", metric_name, "increase", "p50")],
+        request_type=RequestType.HEATMAP,
+        bucket_options=build_log_bucket_options(2),
+    )
+    assert response.status_code == HTTPStatus.BAD_REQUEST, response.text
+    assert "bucketOptions are not supported for histogram metrics" in get_error_message(response.json())
+
+
+@pytest.mark.parametrize(
+    "request_type",
+    [RequestType.TIME_SERIES, RequestType.SCALAR, RequestType.RAW],
+    ids=["time_series", "scalar", "raw"],
+)
+def test_bucket_options_outside_a_heatmap(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    request_type: str,
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+
+    response = make_query_request(
+        signoz,
+        token,
+        int((now - timedelta(minutes=30)).timestamp() * 1000),
+        int(now.timestamp() * 1000),
+        [build_builder_query("A", MISSING_METRIC, "max", "max")],
+        request_type=request_type,
+        bucket_options=build_log_bucket_options(2),
+    )
+    assert response.status_code == HTTPStatus.BAD_REQUEST, response.text
+    assert "bucketOptions are only supported for heatmap requests" in get_error_message(response.json())
+
+
+def test_fill_gaps_is_rejected(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+
+    response = make_query_request(
+        signoz,
+        token,
+        int((now - timedelta(minutes=30)).timestamp() * 1000),
+        int(now.timestamp() * 1000),
+        [build_builder_query("A", MISSING_METRIC, "max", "max")],
+        request_type=RequestType.HEATMAP,
+        format_options={"formatTableResultForUI": False, "fillGaps": True},
+    )
+    assert response.status_code == HTTPStatus.BAD_REQUEST, response.text
+    assert "fillGaps is not supported for heatmap requests" in get_error_message(response.json())
+
+
+@pytest.mark.parametrize(
+    "queries",
+    [
+        [
+            build_builder_query("A", MISSING_METRIC, "max", "max"),
+            build_builder_query("B", MISSING_METRIC, "min", "min"),
+        ],
+        [build_builder_query("A", MISSING_METRIC, "max", "max", disabled=True)],
+        [
+            build_builder_query("A", MISSING_METRIC, "max", "max"),
+            build_builder_query("B", MISSING_METRIC, "min", "min", disabled=True),
+            build_formula_query("F1", "B"),
+        ],
+        [],
+    ],
+    ids=["two_enabled_queries", "only_a_disabled_query", "a_formula_beside_an_enabled_query", "no_queries"],
+)
+def test_wrong_number_of_enabled_queries(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    queries: list[dict],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+
+    response = make_query_request(
+        signoz,
+        token,
+        int((now - timedelta(minutes=30)).timestamp() * 1000),
+        int(now.timestamp() * 1000),
+        queries,
+        request_type=RequestType.HEATMAP,
+    )
+    assert response.status_code == HTTPStatus.BAD_REQUEST, response.text
+
+
+@pytest.mark.parametrize(
+    "queries, expected_message",
+    [
+        (
+            [build_builder_query("A", MISSING_METRIC, "max", "max", functions=[build_function("absolute")])],
+            "functions are not supported for heatmap requests",
+        ),
+        (
+            [
+                build_builder_query("A", MISSING_METRIC, "max", "max", disabled=True, functions=[build_function("absolute")]),
+                build_formula_query("F1", "A"),
+            ],
+            "functions are not supported for heatmap requests",
+        ),
+        (
+            [
+                build_builder_query("A", MISSING_METRIC, "max", "max", disabled=True),
+                build_formula_query("F1", "A", functions=[build_function("absolute")]),
+            ],
+            "functions are not supported for heatmap requests",
+        ),
+        (
+            [
+                {
+                    "type": "builder_query",
+                    "spec": {
+                        "name": "A",
+                        "signal": "metrics",
+                        "aggregations": [{"metricName": MISSING_METRIC, "timeAggregation": "max", "spaceAggregation": "max"}],
+                        "stepInterval": 60,
+                        "having": {"expression": "value > 1"},
+                    },
+                }
+            ],
+            "having is not supported for heatmap requests",
+        ),
+    ],
+    ids=["functions_on_the_query", "functions_on_a_disabled_formula_input", "functions_on_the_formula", "having_on_the_query"],
+)
+def test_functions_and_having_are_rejected(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    queries: list[dict],
+    expected_message: str,
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+
+    response = make_query_request(
+        signoz,
+        token,
+        int((now - timedelta(minutes=30)).timestamp() * 1000),
+        int(now.timestamp() * 1000),
+        queries,
+        request_type=RequestType.HEATMAP,
+    )
+    assert response.status_code == HTTPStatus.BAD_REQUEST, response.text
+    assert expected_message in get_error_message(response.json())
+
+
+def test_promql_rejects_bucket_options(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+
+    response = make_query_request(
+        signoz,
+        token,
+        int((now - timedelta(minutes=30)).timestamp() * 1000),
+        int(now.timestamp() * 1000),
+        [{"type": "promql", "spec": {"name": "A", "query": MISSING_METRIC}}],
+        request_type=RequestType.HEATMAP,
+        bucket_options=build_log_bucket_options(2),
+    )
+    assert response.status_code == HTTPStatus.BAD_REQUEST, response.text
+    assert "bucketOptions are not supported for promql heatmap requests" in get_error_message(response.json())
+
+
+@pytest.mark.parametrize(
+    "bucket_options, expected_message",
+    [
+        ({"kind": "quadratic", "spec": {}}, "invalid bucketOptions kind"),
+        ({"kind": "log"}, "bucketOptions spec is required"),
+        ({"kind": "linear"}, "bucketOptions spec is required"),
+        ({"kind": "linear", "spec": {"maxValue": 1000, "scale": 2}}, 'unknown field "scale" in linear buckets spec'),
+        ({"kind": "log", "spec": {"scale": 5}}, "scale must be between -4 and 4"),
+        ({"kind": "log", "spec": {"scale": -5}}, "scale must be between -4 and 4"),
+        ({"kind": "linear", "spec": {"maxValue": 0}}, "linear buckets need a finite maxValue greater than 0"),
+        ({"kind": "linear", "spec": {"maxValue": -10}}, "linear buckets need a finite maxValue greater than 0"),
+        ({"kind": "linear", "spec": {"maxValue": 1000, "numBuckets": 513}}, "numBuckets must be between 1 and 512"),
+    ],
+    ids=[
+        "unknown_kind",
+        "log_without_a_spec",
+        "linear_without_a_spec",
+        "scale_under_the_linear_kind",
+        "scale_above_the_maximum",
+        "scale_below_the_minimum",
+        "zero_max_value",
+        "negative_max_value",
+        "too_many_buckets",
+    ],
+)
+def test_malformed_bucket_options(
+    signoz: types.SigNoz,
+    create_user_admin: None,  # pylint: disable=unused-argument
+    get_token: Callable[[str, str], str],
+    bucket_options: dict,
+    expected_message: str,
+) -> None:
+    now = datetime.now(tz=UTC).replace(second=0, microsecond=0)
+    token = get_token(USER_ADMIN_EMAIL, USER_ADMIN_PASSWORD)
+
+    response = make_query_request(
+        signoz,
+        token,
+        int((now - timedelta(minutes=30)).timestamp() * 1000),
+        int(now.timestamp() * 1000),
+        [build_builder_query("A", MISSING_METRIC, "max", "max")],
+        request_type=RequestType.HEATMAP,
+        bucket_options=bucket_options,
+    )
+    assert response.status_code == HTTPStatus.BAD_REQUEST, response.text
+    assert expected_message in get_error_message(response.json())