9.9 KiB
Goods Data Cleaning Implementation Plan
For agentic workers: REQUIRED SUB-SKILL: Use executing-plans to implement this plan task-by-task. Steps use checkbox (
- [ ]) syntax for tracking.
Goal: 用「备份 → 导出 → 脚本转换 → 导入 → 验证」的流水线,把右侧 SDS 原产品库(origin_goods)按规则批量配置为左侧官网商品(goods),替代后台手动逐个点击。直接作用于生产库。
Background:
- 后台 GoodsView 已实现手动配置能力:右侧树选原产品 → 选国家/品类/标签 → 生成
goods记录。 - 手动逐个点击效率低,需要脚本化批量处理。
- 同步规则(品类映射 / 国家分配 / 标签分配 / 优先级策略)尚未确定,将在后续数据清洗计划中明确后填入
data-cleaning/sync-rules.config.mjs。
Architecture: 三步式流水线,复用现有 export/import 脚本,只新写转换脚本与独立规则文件。规则文件占位先行——改规则不改代码。转换脚本为纯函数式(JSON in → JSON out),不直接连接数据库;导入前输出变更摘要报告作为人工检查点;导入前必须完成全量备份保证可回滚。
Tech Stack: Node.js ESM 脚本、@prisma/client(仅 export/import 使用)、PostgreSQL。
目录约定:
- 清洗相关脚本/规则/产物统一放根目录
data-cleaning/(见 data-cleaning/README.md) - 复用现有 apps/api/scripts/export-data.mjs 与 apps/api/scripts/import-data.mjs
流水线总览
[生产库]
│ ① pg_dump 全量备份
▼
② node scripts/export-data.mjs data-cleaning/runs/<date>/export.json
▼
③ node data-cleaning/sync-transform.mjs data-cleaning/runs/<date>/export.json
→ 输出 transformed.json + change-report.md(人工检查点)
▼
④ node scripts/import-data.mjs data-cleaning/runs/<date>/transformed.json
▼
⑤ 验证:行数对比 / 抽查商品配置 / 官网公开 API 抽查
Task 1: 创建规则文件占位 sync-rules.config.mjs
Files:
-
Create:
data-cleaning/sync-rules.config.mjs -
Test: 无(纯数据文件,由 Task 2 的测试覆盖加载逻辑)
-
Step 1: 创建规则文件骨架
/**
* 商品数据清洗 — 同步规则配置
*
* 规则尚未确定。确定后只修改本文件,不改 sync-transform.mjs。
* 字段语义在规则确定时补充说明。
*/
export const rules = {
/** 品类映射:SDS 品类 → 本地品类(待定) */
categoryMapping: {
// '<sds_category_id 或名称>': '<本地 category_id 或名称>',
},
/** 国家分配:每条原产品生成哪些国家的 good(待定) */
countryAssignment: {
mode: 'none', // none | all | fixed | perCategory
fixedCountryIds: [],
perCategory: {},
},
/** 标签分配:新 good 挂哪些 tag(待定) */
tagAssignment: {
mode: 'none', // none | fixed | perCategory
fixedTagIds: [],
perCategory: {},
},
/** 优先级策略(待定) */
priority: {
defaultPriority: 0,
},
};
- Step 2: Commit
git add data-cleaning/sync-rules.config.mjs
git commit -m "feat(data-cleaning): add sync rules config placeholder"
Task 2: 创建转换脚本 sync-transform.mjs(TDD)
Files:
-
Create:
data-cleaning/sync-transform.mjs -
Test:
data-cleaning/sync-transform.test.mjs -
Step 1: 写失败测试
// data-cleaning/sync-transform.test.mjs
import { describe, it, expect } from 'vitest';
import { transform } from './sync-transform.mjs';
import { rules } from './sync-rules.config.mjs';
const baseDump = () => ({
exportedAt: new Date().toISOString(),
tables: {
users: [], countries: [], categories: [], tag_groups: [], tags: [],
positions: [],
origin_goods: [
{ id: '1', sds_good_id: 'SDS-A', good_name: 'A', delisted: 'false', is_custom: 'false' },
{ id: '2', sds_good_id: 'SDS-B', good_name: 'B', delisted: 'true', is_custom: 'false' },
],
origin_good_variants: [], origin_good_details: [],
goods: [], good_tags: [], sync_logs: [],
},
});
describe('transform', () => {
it('rules 为空时原样返回且报告零变更', () => {
const dump = baseDump();
const { result, report } = transform(dump, rules);
expect(result.tables.goods).toHaveLength(0);
expect(report.created).toBe(0);
expect(result.tables.origin_goods).toHaveLength(2);
});
it('跳过已下架原产品', () => {
const dump = baseDump();
const testRules = { ...rules, countryAssignment: { mode: 'all' } };
const { report } = transform(dump, testRules);
// 只有未下架的 id=1 会生成 good
expect(report.created).toBe(1);
expect(report.skippedDelisted).toBe(1);
});
it('幂等:已存在的 (originGoodId,countryId) 不重复创建', () => {
const dump = baseDump();
dump.tables.goods = [
{ id: '100', origin_good_id: '1', country_id: '9', good_name: 'A', good_priority: '0' },
];
dump.tables.countries = [{ id: '9', country_name: 'US' }];
const testRules = {
...rules,
countryAssignment: { mode: 'fixed', fixedCountryIds: ['9'] },
};
const { report } = transform(dump, testRules);
expect(report.created).toBe(0);
expect(report.duplicates).toBe(1);
});
});
注:若仓库未配置 vitest 运行
.mjs,可用node --test(node:test)替代,断言等价改写。
- Step 2: 运行测试确认 RED
Run: cd apps/api && npx vitest run ../../data-cleaning/sync-transform.test.mjs(或 node --test data-cleaning/)
Expected: FAIL(模块不存在)
- Step 3: 最小实现
// data-cleaning/sync-transform.mjs
/**
* 数据转换脚本(纯函数式):读取导出 JSON + 规则文件,
* 输出待导入 JSON(transformed.json)与变更摘要(change-report.md)。
*
* Usage:
* node data-cleaning/sync-transform.mjs <export.json>
*/
import { readFileSync, writeFileSync, mkdirSync } from 'node:fs';
import { dirname } from 'node:path';
import { rules } from './sync-rules.config.mjs';
/** 纯函数:dump + rules → { result, report } */
export function transform(dump, rulesConfig) {
const tables = JSON.parse(JSON.stringify(dump.tables)); // deep clone
const countries = tables.countries;
const existingKeys = new Set(
tables.goods.map((g) => `${g.origin_good_id}:${g.country_id}`),
);
const report = { created: 0, duplicates: 0, skippedDelisted: 0, details: [] };
const activeCountries = countries.filter((c) => c.id);
for (const og of tables.origin_goods) {
if (og.delisted === 'true' || og.delisted === true) {
if (og.is_custom !== 'true' && og.is_custom !== true) report.skippedDelisted++;
continue;
}
let targets = [];
if (rulesConfig.countryAssignment.mode === 'all') {
targets = activeCountries.map((c) => c.id);
} else if (rulesConfig.countryAssignment.mode === 'fixed') {
targets = rulesConfig.countryAssignment.fixedCountryIds;
}
for (const countryId of targets) {
const key = `${og.id}:${countryId}`;
if (existingKeys.has(key)) { report.duplicates++; continue; }
const newId = String(
tables.goods.reduce((m, g) => Math.max(m, Number(g.id || 0)), 0) + 1,
);
tables.goods.push({
id: newId,
origin_good_id: og.id,
country_id: countryId,
category_id: rulesConfig.categoryMapping.default ?? '',
good_name: og.good_name,
good_priority: String(rulesConfig.priority.defaultPriority ?? 0),
});
existingKeys.add(key);
report.created++;
report.details.push(`+ good[${newId}] origin=${og.sds_good_id} country=${countryId}`);
}
}
return { result: { exportedAt: dump.exportedAt, tables }, report };
}
function main() {
const input = process.argv[2];
if (!input) {
console.error('Usage: node data-cleaning/sync-transform.mjs <export.json>');
process.exit(1);
}
const dump = JSON.parse(readFileSync(input, 'utf8'));
const { result, report } = transform(dump, rules);
mkdirSync(`${dirname(input)}/out`, { recursive: true });
writeFileSync(`${dirname(input)}/out/transformed.json`, JSON.stringify(result));
writeFileSync(
`${dirname(input)}/out/change-report.md`,
['# 变更摘要', `- 新增 goods: ${report.created}`, `- 重复跳过: ${report.duplicates}`,
`- 下架跳过: ${report.skippedDelisted}`, '', ...report.details.map((d) => `- ${d}`)].join('\n'),
);
console.log(JSON.stringify(report, null, 2));
}
// 测试环境下不自动执行 main
if (process.env.NODE_ENV !== 'test' && import.meta.url === `file://${process.argv[1]}`) {
main();
}
- Step 4: 运行测试确认 GREEN
Run: cd apps/api && npx vitest run ../../data-cleaning/sync-transform.test.mjs
Expected: PASS(3 个用例全过)
- Step 5: Commit
git add data-cleaning/sync-transform.mjs data-cleaning/sync-transform.test.mjs
git commit -m "feat(data-cleaning): add pure transform script with tests"
Task 3: 生产执行手册写入 README 并联调演练
Files:
-
Modify:
data-cleaning/README.md -
Step 1: 补充执行命令段(备份 / 导出 / 转换 / 人工检查 / 导入 / 验证 六个步骤的确切命令与预期输出)
-
Step 2: 本地或预发演练一次全流程(规则为空应零变更),确认 change-report 为空、行数一致
-
Step 3: Commit
git add data-cleaning/README.md
git commit -m "docs(data-cleaning): add production runbook"
待定事项(规则确定后回填)
| 项 | 状态 | 回填位置 |
|---|---|---|
| 品类映射规则 | 未定 | data-cleaning/sync-rules.config.mjs#categoryMapping |
| 国家分配策略 | 未定 | ...#countryAssignment |
| 标签分配策略 | 未定 | ...#tagAssignment |
| 优先级策略 | 未定 | ...#priority |
| 规则细节文档 | 未定 | 后续数据清洗计划文件夹(本目录)内新建规则说明 md |