feat(editor): merge multi-database query results into one view

This commit is contained in:
goodvecn-dev authored and GitHub committed 2026-09-23 10:36:40 +08:00
1 parent 9d07708617
commit 8b822e974b
31 files changed
+3085 -87

No files matched your search

@@ -0,0 +1,209 @@
import type { QueryResult } from "@/types/database";
/**
* Rows read from a single target of a multi-database run. The merged view and
* its Excel export need the query's body rather than just the editor's first
* page, so every source is fetched up to this cap; the user's global result-row
* limit still applies on top of it.
*/
export const MULTI_SOURCE_MAX_ROWS_PER_SOURCE = 10_000;
/**
* Upper bound on merged rows. The merged view is a quick inspection surface, so
* a fan-out over many connections must not be able to build an unbounded table;
* hitting the cap is reported instead of silently dropping rows.
*/
export const MULTI_SOURCE_MERGE_MAX_ROWS = 50_000;
export type MultiSourceCellValue = string | number | boolean | null;
export interface MultiSourceResultInput {
/** Stable identity of the contributing result (used as the row's origin key). */
key: string;
/** Human label written into the injected source column for every row. */
label: string;
result: QueryResult | undefined;
}
export interface MultiSourceMergeOptions {
/** Header of the injected first column; localized by the caller. */
sourceColumnLabel: string;
maxRows?: number;
}
export interface MultiSourceColumnSummary {
columnIndex: number;
column: string;
sum: number;
/** Number of contributing values; empty cells never count. */
count: number;
}
export interface MultiSourceMergeSource {
key: string;
label: string;
rowCount: number;
columnCount: number;
}
export interface MultiSourceMergeOutput {
columns: string[];
rows: MultiSourceCellValue[][];
/** One entry per all-numeric column, aligned by column ordinal. */
summaries: MultiSourceColumnSummary[];
sources: MultiSourceMergeSource[];
rowCount: number;
truncated: boolean;
/** Labels of inputs that carried no tabular payload and were left out. */
skippedLabels: string[];
}
interface MergePlan {
input: MultiSourceResultInput;
result: QueryResult;
/** Merged column ordinal for each source column ordinal. */
targetIndexes: Map<number, number>;
}
function usableInput(input: MultiSourceResultInput): boolean {
if (!input.result) return false;
if (input.result.execution_error === true) return false;
return (input.result.columns?.length ?? 0) > 0;
}
function mergedColumnName(column: string | undefined, sourceIndex: number): string {
const name = typeof column === "string" ? column.trim() : "";
return name === "" ? `column_${sourceIndex + 1}` : name;
}
function isEmptyCell(value: MultiSourceCellValue | undefined): boolean {
if (value === null || value === undefined) return true;
return typeof value === "string" && value.trim() === "";
}
/**
* A cell counts as numeric when it is a finite number or a non-empty string that
* parses to one (drivers return DECIMAL/BIGINT as strings). Booleans are not
* numeric, so a flag column never turns into a total.
*/
export function isNumericSummaryCellValue(value: MultiSourceCellValue | undefined): boolean {
if (typeof value === "number") return Number.isFinite(value);
if (typeof value === "string") {
const trimmed = value.trim();
return trimmed !== "" && Number.isFinite(Number(trimmed));
}
return false;
}
/**
* Simple result-level aggregates: sums every column whose non-empty cells are
* all numeric. A single non-numeric cell disqualifies the whole column, and a
* column without values produces no total.
*/
export function summarizeNumericColumns(columns: readonly string[], rows: readonly (readonly MultiSourceCellValue[])[], options: { skipColumnIndexes?: readonly number[] } = {}): MultiSourceColumnSummary[] {
const skipped = new Set(options.skipColumnIndexes ?? []);
const sums: number[] = Array.from({ length: columns.length }, () => 0);
const counts: number[] = Array.from({ length: columns.length }, () => 0);
const stillNumeric: boolean[] = Array.from({ length: columns.length }, () => true);
for (const row of rows) {
for (let columnIndex = 0; columnIndex < columns.length; columnIndex += 1) {
if (skipped.has(columnIndex) || !stillNumeric[columnIndex]) continue;
const value = row[columnIndex];
if (isEmptyCell(value)) continue;
if (!isNumericSummaryCellValue(value)) {
stillNumeric[columnIndex] = false;
continue;
}
sums[columnIndex] += typeof value === "number" ? value : Number(value);
counts[columnIndex] += 1;
}
}
const summaries: MultiSourceColumnSummary[] = [];
for (let columnIndex = 0; columnIndex < columns.length; columnIndex += 1) {
if (skipped.has(columnIndex) || !stillNumeric[columnIndex] || counts[columnIndex] === 0) continue;
summaries.push({ columnIndex, column: columns[columnIndex] ?? "", sum: sums[columnIndex]!, count: counts[columnIndex]! });
}
return summaries;
}
/**
* Merges the results of one query executed against several connections /
* databases into a single table.
*
* Columns are aligned by name (union, first-seen order) and every row keeps a
* leading provenance column naming the source it came from. A column a source
* does not have is padded with null. Duplicate column names inside one result
* keep their own value streams by matching the Nth occurrence of a name.
*/
export function mergeMultiSourceResults(inputs: readonly MultiSourceResultInput[], options: MultiSourceMergeOptions): MultiSourceMergeOutput {
const maxRows = Math.max(1, Math.trunc(options.maxRows ?? MULTI_SOURCE_MERGE_MAX_ROWS));
const columns: string[] = [options.sourceColumnLabel];
// A merged column is identified by (name, occurrence-in-source) so duplicate
// names in one result still keep both of their value streams.
const indexByLayoutKey = new Map<string, number>();
const plans: MergePlan[] = [];
const skippedLabels: string[] = [];
for (const input of inputs) {
if (!usableInput(input)) {
skippedLabels.push(input.label);
continue;
}
const result = input.result!;
const occurrences = new Map<string, number>();
const targetIndexes = new Map<number, number>();
const hiddenColumnIndexes = new Set(result.hidden_column_indexes ?? []);
result.columns.forEach((column, sourceIndex) => {
if (hiddenColumnIndexes.has(sourceIndex)) return;
const name = mergedColumnName(column, sourceIndex);
const occurrence = occurrences.get(name) ?? 0;
occurrences.set(name, occurrence + 1);
const layoutKey = `${name}\u0000${occurrence}`;
let targetIndex = indexByLayoutKey.get(layoutKey);
if (targetIndex === undefined) {
targetIndex = columns.length;
indexByLayoutKey.set(layoutKey, targetIndex);
columns.push(name);
}
targetIndexes.set(sourceIndex, targetIndex);
});
plans.push({ input, result, targetIndexes });
}
const rows: MultiSourceCellValue[][] = [];
let truncated = false;
for (const plan of plans) {
for (const sourceRow of plan.result.rows) {
if (rows.length >= maxRows) {
truncated = true;
break;
}
const row: MultiSourceCellValue[] = Array.from({ length: columns.length }, () => null);
row[0] = plan.input.label;
plan.targetIndexes.forEach((targetIndex, sourceIndex) => {
row[targetIndex] = sourceRow[sourceIndex] ?? null;
});
rows.push(row);
}
if (truncated) break;
}
return {
columns,
rows,
// The injected provenance column is text by construction, but skip it
// explicitly so a numeric-looking label can never become a total.
summaries: summarizeNumericColumns(columns, rows, { skipColumnIndexes: [0] }),
sources: plans.map((plan) => ({
key: plan.input.key,
label: plan.input.label,
rowCount: plan.result.rows.length,
columnCount: plan.targetIndexes.size,
})),
rowCount: rows.length,
truncated,
skippedLabels,
};
}