Fix: Group By on ratio charts

.changeset/ratio-group-by.md

@@ -0,0 +1,21 @@
+---
+"@hyperdx/common-utils": patch
+---
+
+fix: support Group By on ratio charts
+
+A ratio chart (`seriesReturnType: 'ratio'`) with a Group By previously collapsed
+to a single line. Two issues in the multi-series merge: (1) rows were keyed by
+time bucket only, so groups at the same bucket overwrote each other, and (2) the
+ratio computation dropped every non-value column, discarding the group
+dimension. The merge now keys by (time bucket + group dimensions) and the ratio
+result carries the group columns through, so a grouped ratio renders one series
+per group.
+
+Grouped ratios use share-of-total semantics: each group's denominator is the
+total of the denominator column across all groups in the same time bucket, so
+the grouped lines are each group's contribution to the overall ratio and sum to
+the ungrouped value (e.g. each tenant's share of the overall error rate), rather
+than each group's in-group rate. Ungrouped ratios are unchanged (one row per
+bucket → the bucket total is that row's denominator). A group absent from the
+filtered numerator (e.g. a tenant with zero errors) contributes 0%, not N/A.

packages/app/src/components/DBEditTimeChartForm/ChartEditorControls.tsx

@@ -1,10 +1,11 @@
-import { useState } from 'react';
+import { useMemo, useState } from 'react';
 import {
   Control,
   FieldArrayWithId,
   FieldErrors,
   UseFormClearErrors,
   UseFormSetValue,
+  useWatch,
 } from 'react-hook-form';
 import { TableConnection } from '@hyperdx/common-utils/dist/core/metadata';
 import {
@@ -32,6 +33,7 @@ import SourceSchemaPreview, {
 import { SourceSelectControlled } from '@/components/SourceSelect';
 import { SQLInlineEditorControlled } from '@/components/SQLEditor/SQLInlineEditor';
 import { IS_LOCAL_MODE } from '@/config';
+import { getMetricTableName } from '@/utils';
 import { DEFAULT_TILE_ALERT } from '@/utils/alerts';
 
 import { OnClickFormButton } from './OnClickForm/OnClickFormButton';
@@ -105,6 +107,28 @@ export function ChartEditorControls({
   const [isSourceSchemaPreviewOpen, setIsSourceSchemaPreviewOpen] =
     useState(false);
 
+  // Metric sources have no single `from.tableName` (they fan out to per-type
+  // metric tables), so the default tableConnection can't drive attribute
+  // autocomplete for the chart-level Group By. Point it at the selected
+  // metric's table + name so the editor can look up `Attributes` keys while
+  // typing, matching the per-series inputs.
+  const series = useWatch({ control, name: 'series' });
+  const groupByTableConnection = useMemo<TableConnection>(() => {
+    if (tableSource?.kind === SourceKind.Metric && Array.isArray(series)) {
+      const metric = series.find(s => s?.metricType && s?.metricName);
+      const metricTable = getMetricTableName(tableSource, metric?.metricType);
+      if (metricTable) {
+        return {
+          databaseName: tableSource.from.databaseName,
+          tableName: metricTable,
+          connectionId: tableSource.connection,
+          metricName: metric?.metricName,
+        };
+      }
+    }
+    return tableConnection;
+  }, [tableSource, series, tableConnection]);
+
   return (
     <>
       <Flex mb="md" align="center" justify="space-between">
@@ -211,7 +235,7 @@ export function ChartEditorControls({
                 </div>
                 <div>
                   <SQLInlineEditorControlled
-                    tableConnection={tableConnection}
+                    tableConnection={groupByTableConnection}
                     control={control}
                     name={`groupBy`}
                     placeholder="SQL Columns"

packages/common-utils/src/__tests__/clickhouse.test.ts

@@ -440,6 +440,69 @@ describe('computeResultSetRatio', () => {
       /Unable to compute ratio/,
     );
   });
+
+  it('computes a grouped ratio as each group share of the per-bucket total (lines sum to the overall)', () => {
+    // Joined meta seeds the two value columns first, then the group + timestamp
+    // columns (mirrors queryChartConfig's merge output).
+    const mockResultSet: ResponseJSON<any> = {
+      meta: [
+        { name: 'errors', type: 'UInt64' },
+        { name: 'total', type: 'UInt64' },
+        { name: 'tenant', type: 'String' },
+        { name: 'timestamp', type: 'DateTime' },
+      ],
+      data: [
+        { errors: '20', total: '100', tenant: 'acme', timestamp: 't0' },
+        { errors: '30', total: '100', tenant: 'globex', timestamp: 't0' },
+        { errors: '10', total: '100', tenant: 'acme', timestamp: 't1' },
+      ],
+      rows: 3,
+      statistics: { elapsed: 0.1, rows_read: 3, bytes_read: 100 },
+    };
+
+    const result = computeResultSetRatio(mockResultSet);
+
+    // ratio column + carried-through group + timestamp
+    expect(result.meta.map(m => m.name)).toEqual([
+      'errors/total',
+      'tenant',
+      'timestamp',
+    ]);
+    // Denominator is the per-bucket total across all groups (t0: 200, t1: 100),
+    // so each group is its contribution to the overall rate.
+    expect(result.data).toEqual([
+      { 'errors/total': 0.1, tenant: 'acme', timestamp: 't0' }, // 20/200
+      { 'errors/total': 0.15, tenant: 'globex', timestamp: 't0' }, // 30/200
+      { 'errors/total': 0.1, tenant: 'acme', timestamp: 't1' }, // 10/100
+    ]);
+    // t0 contributions sum to the overall error rate for that bucket (50/200).
+    const t0 = result.data.filter(r => r.timestamp === 't0');
+    expect(t0.reduce((s, r) => s + r['errors/total'], 0)).toBeCloseTo(0.25);
+  });
+
+  it('renders a zero-error group as 0% (missing numerator), not N/A', () => {
+    const mockResultSet: ResponseJSON<any> = {
+      meta: [
+        { name: 'errors', type: 'UInt64' },
+        { name: 'total', type: 'UInt64' },
+        { name: 'tenant', type: 'String' },
+        { name: 'timestamp', type: 'DateTime' },
+      ],
+      data: [
+        // has errors -> contributes 20/200
+        { errors: '20', total: '100', tenant: 'acme', timestamp: 't0' },
+        // denominator only (no rows in the error-filtered numerator query) -> 0%
+        { total: '100', tenant: 'globex', timestamp: 't0' },
+      ],
+      rows: 2,
+      statistics: { elapsed: 0.1, rows_read: 2, bytes_read: 100 },
+    };
+
+    const result = computeResultSetRatio(mockResultSet);
+
+    expect(result.data[0]['errors/total']).toBe(0.1); // 20 / (100+100)
+    expect(result.data[1]['errors/total']).toBe(0); // 0 / 200
+  });
 });
 
 describe('processClickhouseSettings - optimization settings', () => {

packages/common-utils/src/clickhouse/index.ts

@@ -374,47 +374,64 @@ export const computeRatio = (
 };
 
 export const computeResultSetRatio = (resultSet: ResponseJSON<any>) => {
-  const _meta = resultSet.meta;
+  const _meta = resultSet.meta ?? [];
   const _data = resultSet.data;
-  const timestampColumn = inferTimestampColumn(_meta ?? []);
-  const _restColumns = _meta?.filter(m => m.name !== timestampColumn?.name);
-  const firstColumn = _restColumns?.[0];
-  const secondColumn = _restColumns?.[1];
-  if (!firstColumn || !secondColumn) {
+  // The numerator/denominator are the two value columns. The joined meta seeds
+  // them first (see queryChartConfig), so the first two numeric columns are the
+  // ratio operands; numeric group-by dimensions (if any) come after.
+  const numericColumns = inferNumericColumn(_meta);
+  const numerator = numericColumns?.[0];
+  const denominator = numericColumns?.[1];
+  if (!numerator || !denominator) {
     throw new Error(
       `Unable to compute ratio - meta information: ${JSON.stringify(_meta)}.`,
     );
   }
-  const ratioColumnName = `${firstColumn.name}/${secondColumn.name}`;
-  const result = {
+  const ratioColumnName = `${numerator.name}/${denominator.name}`;
+  // Carry through every non-operand column — the timestamp and any group-by
+  // dimensions — so a grouped ratio renders one series per group instead of
+  // collapsing into a single line.
+  const passthroughColumns = _meta.filter(
+    m => m.name !== numerator.name && m.name !== denominator.name,
+  );
+  const timestampColumn = inferTimestampColumn(_meta);
+
+  // Share-of-total semantics: each group's denominator is the total of the
+  // denominator column across ALL groups in the same time bucket. So a grouped
+  // ratio shows each group's contribution to the overall ratio (the lines sum
+  // to the ungrouped value) rather than each group's own in-group rate. With no
+  // grouping there's one row per bucket, so the bucket total equals that row's
+  // denominator and the result is unchanged.
+  const bucketKey = (row: Record<string, any>) =>
+    timestampColumn ? String(row[timestampColumn.name]) : '__all__';
+  const totalDenominatorByBucket = new Map<string, number>();
+  for (const row of _data) {
+    const denom = castToNumber(row[denominator.name]);
+    if (!Number.isNaN(denom)) {
+      const key = bucketKey(row);
+      totalDenominatorByBucket.set(
+        key,
+        (totalDenominatorByBucket.get(key) ?? 0) + denom,
+      );
+    }
+  }
+
+  return {
     ...resultSet,
-    data: _data.map(row => ({
-      [ratioColumnName]: computeRatio(
-        row[firstColumn.name],
-        row[secondColumn.name],
-      ),
-      ...(timestampColumn
-        ? {
-            [timestampColumn.name]: row[timestampColumn.name],
-          }
-        : {}),
-    })),
-    meta: [
-      {
-        name: ratioColumnName,
-        type: 'Float64',
-      },
-      ...(timestampColumn
-        ? [
-            {
-              name: timestampColumn.name,
-              type: timestampColumn.type,
-            },
-          ]
-        : []),
-    ],
+    data: _data.map(row => {
+      // A group absent from the (filtered) numerator query contributes zero, not
+      // "no data" — so a zero-error group reads 0%, not N/A.
+      const numeratorValue = row[numerator.name] ?? 0;
+      const bucketTotal = totalDenominatorByBucket.get(bucketKey(row));
+      return {
+        [ratioColumnName]: computeRatio(numeratorValue, bucketTotal ?? NaN),
+        ...Object.fromEntries(
+          passthroughColumns.map(c => [c.name, row[c.name]]),
+        ),
+      };
+    }),
+    meta: [{ name: ratioColumnName, type: 'Float64' }, ...passthroughColumns],
   };
-  return result;
 };
 
 export interface QueryInputs<Format extends DataFormat> {
@@ -703,20 +720,29 @@ export abstract class BaseClickhouseClient {
                 ),
               )
             : { ...row };
-          const ts =
-            timestampColumn != null
+          // When the series are grouped, two rows at the same time bucket but
+          // different group values must stay distinct — key by (bucket + group
+          // dims) via the hash of the row minus its value column. Without a
+          // group dimension this collapses to the timestamp (or a fixed key),
+          // preserving the original behavior.
+          const hasGroupCols = Object.keys(_rowWithoutValue).some(
+            key => key !== timestampColumn?.name,
+          );
+          const mergeKey = hasGroupCols
+            ? objectHash(_rowWithoutValue)
+            : timestampColumn != null
               ? row[timestampColumn.name]
               : isTimeSeries
                 ? objectHash(_rowWithoutValue)
                 : '__FIXED_TIMESTAMP__';
-          if (tsBucketMap.has(ts)) {
-            const existingRow = tsBucketMap.get(ts);
-            tsBucketMap.set(ts, {
+          if (tsBucketMap.has(mergeKey)) {
+            const existingRow = tsBucketMap.get(mergeKey);
+            tsBucketMap.set(mergeKey, {
               ...existingRow,
               ...row,
             });
           } else {
-            tsBucketMap.set(ts, row);
+            tsBucketMap.set(mergeKey, row);
           }
         }
       }