fix(api): add sync safety guards against degenerate SDS responses

Add SYNC_GUARDS thresholds so a partial/degenerate upstream response never
triggers a destructive operation:
- skip stale category deletion when the fetched tree is suspiciously small
  vs the existing SDS category count
- skip delist detection unless both leaf-category and seen-product counts
  are healthy

Verified: 77 tests pass; live SDS returns 226 categories (guard off),
incident-case ratios (2/226, 2/2) are correctly blocked.
This commit is contained in:
yeuimu
2026-08-20 16:37:44 +08:00
parent 79fabd85f7
commit aed9afef92
6 changed files with 342 additions and 45 deletions
+1 -1
View File
@@ -42,7 +42,7 @@ PORT=3001
后台无需额外环境变量;Vite 代理已把 `/api/*` 转给 `http://localhost:3001`
### 3. 启动
### 3. 部署
启动顺序:先启动后端,再启动另两个。
@@ -0,0 +1,36 @@
import { Test } from '@nestjs/testing';
import { of } from 'rxjs';
import { HttpService } from '@nestjs/axios';
import { ConfigService } from '@nestjs/config';
import { SdsClientService } from './sds-client.service';
describe('SdsClientService', () => {
let service: SdsClientService;
let http: { post: jest.Mock; get: jest.Mock };
beforeEach(async () => {
http = { post: jest.fn(), get: jest.fn() };
const moduleRef = await Test.createTestingModule({
providers: [
SdsClientService,
{ provide: HttpService, useValue: http },
{ provide: ConfigService, useValue: { get: jest.fn(() => undefined) } },
],
}).compile();
service = moduleRef.get(SdsClientService);
});
describe('fetchCategoryTree', () => {
it('throws when the upstream returns a degenerate small tree', async () => {
http.post.mockReturnValue(of({ data: [{ id: 1, name: 'Only' }] }));
await expect(service.fetchCategoryTree()).rejects.toThrow(/degenerate/i);
});
it('returns the tree when it is healthy', async () => {
const tree = Array.from({ length: 20 }, (_, i) => ({ id: i + 1, name: `C${i}` }));
http.post.mockReturnValue(of({ data: tree }));
const result = await service.fetchCategoryTree();
expect(result).toHaveLength(20);
});
});
});
+13
View File
@@ -9,6 +9,13 @@ const POD_HEADERS = {
Referer: 'https://inkpod.vip/',
} as const;
/**
* Minimum number of category nodes a healthy `category/tree/3` response contains.
* Below this the response is treated as degenerate and rejected so the caller
* never runs a destructive sync against a partial tree.
*/
export const MIN_SDS_CATEGORY_NODES = 10;
export interface SdsCategoryTreeNode {
id: number | string | null;
name?: string;
@@ -109,6 +116,12 @@ export class SdsClientService {
if (!Array.isArray(data)) {
throw new Error(`SDS category tree returned ${typeof data}, expected array`);
}
if (data.length < MIN_SDS_CATEGORY_NODES) {
throw new Error(
`SDS category tree is degenerate (${data.length} nodes < ${MIN_SDS_CATEGORY_NODES}) — ` +
`aborting to avoid destructive sync`,
);
}
return data as SdsCategoryTreeNode[];
}
+97 -1
View File
@@ -1,6 +1,10 @@
import { Test } from '@nestjs/testing';
import { ConfigModule } from '@nestjs/config';
import { SyncService } from './sync.service';
import {
SyncService,
shouldRunDelistDetection,
shouldSkipStaleDeletion,
} from './sync.service';
import { SdsClientService } from './sds-client.service';
import { PrismaService } from '../prisma/prisma.service';
@@ -168,6 +172,98 @@ describe('SyncService', () => {
});
});
describe('sync guard thresholds', () => {
describe('shouldSkipStaleDeletion', () => {
it('skips stale deletion when the fetched count is below the hard floor', () => {
expect(shouldSkipStaleDeletion(2, 226)).toBe(true);
expect(shouldSkipStaleDeletion(9, 226)).toBe(true);
});
it('skips stale deletion when fetched is far smaller than existing (ratio guard)', () => {
expect(shouldSkipStaleDeletion(100, 250)).toBe(true);
});
it('does NOT skip when fetched count is healthy', () => {
expect(shouldSkipStaleDeletion(226, 226)).toBe(false);
expect(shouldSkipStaleDeletion(200, 226)).toBe(false);
});
it('does NOT skip when there are no existing SDS categories', () => {
expect(shouldSkipStaleDeletion(0, 0)).toBe(false);
expect(shouldSkipStaleDeletion(2, 0)).toBe(false);
});
});
describe('shouldRunDelistDetection', () => {
it('skips delist detection when leaf categories are too few', () => {
expect(shouldRunDelistDetection(2, 500)).toBe(false);
expect(shouldRunDelistDetection(9, 500)).toBe(false);
});
it('skips delist detection when the seen product count is too small', () => {
expect(shouldRunDelistDetection(148, 2)).toBe(false);
expect(shouldRunDelistDetection(148, 49)).toBe(false);
});
it('runs delist detection only when both metrics are healthy', () => {
expect(shouldRunDelistDetection(148, 500)).toBe(true);
expect(shouldRunDelistDetection(10, 50)).toBe(true);
});
});
it('category sync keeps existing SDS categories when upstream returns a degenerate tree', async () => {
const stamp = Date.now();
const keep = await prisma.category.create({
data: { sdsCategoryId: `keep-${stamp}`, categoryName: `Keep ${stamp}` },
});
createdSdsCategoryIds.push(`keep-${stamp}`);
sds.fetchCategoryTree.mockResolvedValueOnce([
{ id: `g-${stamp}-1`, name: 'Tiny 1' },
{ id: `g-${stamp}-2`, name: 'Tiny 2' },
]);
createdSdsCategoryIds.push(`g-${stamp}-1`, `g-${stamp}-2`);
const result = await service.syncCategories();
expect(result.deletedStale).toBe(0);
const still = await prisma.category.findUnique({ where: { id: keep.id } });
expect(still).not.toBeNull();
});
it('product sync does NOT delist origin goods when it sees too few products', async () => {
const stamp = Date.now();
const leaf = await prisma.category.create({
data: { sdsCategoryId: `leafguard-${stamp}`, categoryName: `LeafGuard ${stamp}` },
});
createdSdsCategoryIds.push(`leafguard-${stamp}`);
const active = await prisma.originGood.create({
data: { sdsGoodId: `active-${stamp}`, delisted: false, goodName: 'Active' },
});
createdSdsGoodIds.push(`active-${stamp}`);
sds.fetchProductsPage.mockImplementation(async (categoryId) => {
if (categoryId === `leafguard-${stamp}`) {
return {
content: [
{ id: `guardp-${stamp}-1`, name: 'P1' },
{ id: `guardp-${stamp}-2`, name: 'P2' },
],
};
}
return { content: [] };
});
const result = await service.syncProducts();
expect(result.delisted).toBe(0);
const still = await prisma.originGood.findUnique({ where: { id: active.id } });
expect(still?.delisted).toBe(false);
createdSdsGoodIds.push(`guardp-${stamp}-1`, `guardp-${stamp}-2`);
});
});
describe('getStatus', () => {
it('returns recent logs ordered by startedAt desc', async () => {
const logs = await service.getStatus(5);
+109 -43
View File
@@ -8,6 +8,7 @@ export interface CategorySyncResult {
inserted: number;
updated: number;
total: number;
deletedStale: number;
}
export interface ProductSyncResult {
@@ -15,6 +16,44 @@ export interface ProductSyncResult {
updated: number;
total: number;
leafCategories: number;
delisted: number;
}
/**
* Safety guards so a degenerate/partial SDS response never triggers a
* destructive operation (stale category deletion / mass delist marking).
* Upstream normally returns ~226 categories and ~150 leaf categories with
* hundreds of products — the floors below only trigger on abnormal responses.
*/
export const SYNC_GUARDS = {
MIN_CATEGORY_COUNT: 10,
MIN_CATEGORY_RATIO: 0.5,
MIN_LEAF_CATEGORIES: 10,
MIN_SEEN_GOODS: 50,
} as const;
/**
* True when the fetched category count is suspiciously small compared to the
* categories already synced from SDS, i.e. the upstream response is likely
* partial/degenerate. In that case stale deletion must be skipped.
*/
export function shouldSkipStaleDeletion(fetched: number, existingSds: number): boolean {
if (existingSds <= 0) return false;
return (
fetched < SYNC_GUARDS.MIN_CATEGORY_COUNT ||
fetched < SYNC_GUARDS.MIN_CATEGORY_RATIO * existingSds
);
}
/**
* True only when both the leaf-category count and the number of seen products
* are healthy enough to trust the "not seen upstream => delisted" conclusion.
*/
export function shouldRunDelistDetection(leafCategories: number, seenGoods: number): boolean {
return (
leafCategories >= SYNC_GUARDS.MIN_LEAF_CATEGORIES &&
seenGoods >= SYNC_GUARDS.MIN_SEEN_GOODS
);
}
@Injectable()
@@ -82,8 +121,23 @@ export class SyncService {
this.logger.log(`Fetched ${flat.length} SDS categories`);
const seenSdsIds = new Set(flat.map((n) => n.sdsId));
// Guard: if the upstream tree is suspiciously small vs what we already
// have from SDS, skip stale deletion entirely — a partial response must
// never wipe the category library.
const existingSdsCount = await this.prisma.category.count({
where: { sdsCategoryId: { not: null } },
});
const skipStaleDeletion = shouldSkipStaleDeletion(flat.length, existingSdsCount);
if (skipStaleDeletion) {
this.logger.warn(
`Skipping stale category deletion: fetched=${flat.length} existingSds=${existingSdsCount} ` +
`(below guard thresholds)`,
);
}
// Single transaction: upsert + wire parents + delete stale.
// SDS tree is the source of truth — anything not in the response gets deleted.
// SDS tree is the source of truth — anything not in the response gets deleted
// (unless the response looks degenerate, see guard above).
const { inserted, updated, deletedStale } = await this.prisma.$transaction(async (tx) => {
let ins = 0;
let upd = 0;
@@ -133,50 +187,55 @@ export class SyncService {
// 3. Delete stale categories (in DB but not in SDS response)
// Detach parent links first, then delete leaf-first to respect FK constraints.
const staleCats = await tx.category.findMany({
where: { sdsCategoryId: { notIn: [...seenSdsIds] } },
select: { id: true },
});
const staleIds = staleCats.map((c) => c.id);
// Protect categories that have configured goods — onDelete: Restrict
const goodsInStale = await tx.good.groupBy({
by: ['categoryId'],
where: { categoryId: { in: staleIds } },
});
const protectedIds = new Set(goodsInStale.map((g) => g.categoryId));
const deletableIds = staleIds.filter((id) => !protectedIds.has(id));
// Detach all deletable categories from their parents
if (deletableIds.length > 0) {
await tx.category.updateMany({
where: { id: { in: deletableIds } },
data: { parentCategoryId: null },
// Skipped entirely when the response looks degenerate (see guard above).
let deletedStale = 0;
if (!skipStaleDeletion) {
const staleCats = await tx.category.findMany({
where: { sdsCategoryId: { notIn: [...seenSdsIds] } },
select: { id: true },
});
// Also detach any non-deletable children pointing to deletable parents
await tx.category.updateMany({
where: { parentCategoryId: { in: deletableIds } },
data: { parentCategoryId: null },
const staleIds = staleCats.map((c) => c.id);
// Protect categories that have configured goods — onDelete: Restrict
const goodsInStale = await tx.good.groupBy({
by: ['categoryId'],
where: { categoryId: { in: staleIds } },
});
// Delete leaf-first (repeatedly remove nodes with no children)
let remaining = [...deletableIds];
while (remaining.length > 0) {
const withChildren = await tx.category.findMany({
where: { parentCategoryId: { in: remaining } },
select: { parentCategoryId: true },
distinct: ['parentCategoryId'],
const protectedIds = new Set(goodsInStale.map((g) => g.categoryId));
const deletableIds = staleIds.filter((id) => !protectedIds.has(id));
deletedStale = deletableIds.length;
// Detach all deletable categories from their parents
if (deletableIds.length > 0) {
await tx.category.updateMany({
where: { id: { in: deletableIds } },
data: { parentCategoryId: null },
});
const hasChildSet = new Set(
withChildren.filter((c) => c.parentCategoryId).map((c) => c.parentCategoryId!.toString()),
);
const leaves = remaining.filter((id) => !hasChildSet.has(id.toString()));
if (leaves.length === 0) break; // safety: circular dependency
await tx.category.deleteMany({ where: { id: { in: leaves } } });
remaining = remaining.filter((id) => !leaves.some((l) => l === id));
// Also detach any non-deletable children pointing to deletable parents
await tx.category.updateMany({
where: { parentCategoryId: { in: deletableIds } },
data: { parentCategoryId: null },
});
// Delete leaf-first (repeatedly remove nodes with no children)
let remaining = [...deletableIds];
while (remaining.length > 0) {
const withChildren = await tx.category.findMany({
where: { parentCategoryId: { in: remaining } },
select: { parentCategoryId: true },
distinct: ['parentCategoryId'],
});
const hasChildSet = new Set(
withChildren.filter((c) => c.parentCategoryId).map((c) => c.parentCategoryId!.toString()),
);
const leaves = remaining.filter((id) => !hasChildSet.has(id.toString()));
if (leaves.length === 0) break; // safety: circular dependency
await tx.category.deleteMany({ where: { id: { in: leaves } } });
remaining = remaining.filter((id) => !leaves.some((l) => l === id));
}
}
}
return { inserted: ins, updated: upd, deletedStale: deletableIds.length };
return { inserted: ins, updated: upd, deletedStale };
});
await this.prisma.syncLog.update({
@@ -187,7 +246,7 @@ export class SyncService {
message: `inserted=${inserted} updated=${updated} total=${flat.length} staleDeleted=${deletedStale}`,
},
});
return { inserted, updated, total: flat.length };
return { inserted, updated, total: flat.length, deletedStale };
} catch (err) {
const message = err instanceof Error ? err.message : String(err);
await this.prisma.syncLog.update({
@@ -256,9 +315,13 @@ export class SyncService {
// Detect delisted products: mark origin goods not seen in upstream as delisted,
// and re-activate any previously delisted goods that reappeared.
// Guard: only trust this conclusion when the sync covered a healthy number of
// leaf categories and saw a healthy number of products — otherwise a partial
// sync must never mass-delist the product library.
let delistedCount = 0;
let reactivatedCount = 0;
if (seenSdsGoodIds.size > 0) {
const runDelist = shouldRunDelistDetection(leafRows.length, seenSdsGoodIds.size);
if (runDelist) {
const delistedResult = await this.prisma.originGood.updateMany({
where: { sdsGoodId: { notIn: [...seenSdsGoodIds] }, delisted: false },
data: { delisted: true },
@@ -270,7 +333,10 @@ export class SyncService {
delistedCount = delistedResult.count;
reactivatedCount = reactivatedResult.count;
} else {
this.logger.warn('No products seen from SDS — skipping delist detection');
this.logger.warn(
`Skipping delist detection: leafCategories=${leafRows.length} seenGoods=${seenSdsGoodIds.size} ` +
`(below guard thresholds)`,
);
}
await this.prisma.syncLog.update({
@@ -281,7 +347,7 @@ export class SyncService {
message: `inserted=${inserted} updated=${updated} total=${total} delisted=${delistedCount} reactivated=${reactivatedCount} leafCategories=${leafRows.length}`,
},
});
return { inserted, updated, total, leafCategories: leafRows.length };
return { inserted, updated, total, leafCategories: leafRows.length, delisted: delistedCount };
} catch (err) {
const message = err instanceof Error ? err.message : String(err);
await this.prisma.syncLog.update({
+86
View File
@@ -0,0 +1,86 @@
# SDS Sync Guard Fix 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:** Prevent the SDS (mapi.sdspod.com) sync from deleting the category/product library when the upstream API returns a degenerate/partial response, and recover the data lost in the 2026-08-20 incident.
**Background / Root cause:**
- On 2026-08-20 06:31 a manual sync ran while `category/tree/3` returned only 2 nodes (normal: 226).
- `syncCategories` treats SDS as the single source of truth: `staleDeleted=225` deleted all previously synced SDS categories.
- With SDS categories gone, `syncProducts` had `leafCategories=0/2``delisted=523` marked 523/536 origin goods as delisted.
- Result: all 536 origin goods `delisted=true`; all 240 website `Good` rows reference delisted origin goods → `/public/goods` returns `total: 0` → product-center page appears empty.
**Architecture:** Add safety guards to `SyncService` and `SdsClientService` so a partial/failed upstream response never triggers destructive operations (`staleDeleted` / delist marking). Recovery is a data fix (re-enable delisted origin goods) followed by normal re-sync.
**Tech Stack:** NestJS, TypeScript, Prisma, Jest.
---
### Task 1: Data recovery — re-enable all origin goods
**Files:**
- Run ad-hoc prisma script (read+write to DB via apps/api @prisma/client)
- [ ] **Step 1: Reset `delisted=false` for all `originGood` rows**
All 536 rows were mis-flagged by the 08-20 partial sync. Run
`prisma.originGood.updateMany({ where: {}, data: { delisted: false } })`.
- [ ] **Step 2: Trigger category sync then product sync**
Verify `/sync/status` logs show SUCCESS with healthy numbers
(categories ≈ 225+, products ≈ 519+).
- [ ] **Step 3: Verify `/public/goods` non-empty and product-center page renders products.**
### Task 2: Hardening — category sync stale-deletion guard
**Files:**
- Modify: `apps/api/src/sync/sync.service.ts`
- Modify: `apps/api/src/sync/sync.service.spec.ts`
- [ ] **Step 1: Write failing test**
Given existing SDS-linked categories in DB, when `fetchCategoryTree` returns a
degenerate small tree (< MIN_CATEGORY_COUNT), `syncCategories` must NOT delete
existing categories (no stale deletion) and must log a warning.
- [ ] **Step 2: Run test, verify RED.**
- [ ] **Step 3: Implement guard**
Before the stale-deletion block, compute `existingSdsCount = categories with sdsCategoryId`.
Skip stale deletion when `flat.length < MIN_CATEGORY_COUNT (10)` OR
`flat.length < MIN_CATEGORY_RATIO (0.5) * existingSdsCount`. Log warning with counts.
- [ ] **Step 4: Verify GREEN + no regressions.**
### Task 3: Hardening — product sync delist guard
**Files:**
- Modify: `apps/api/src/sync/sync.service.ts`
- Modify: `apps/api/src/sync/sync.service.spec.ts`
- [ ] **Step 1: Write failing test**
When there are very few leaf categories (< MIN_LEAF_CATEGORIES) or very few seen
products (< MIN_SEEN_GOODS), `syncProducts` must skip the delist/reactivate marking
and log a warning instead of mass-delisting.
- [ ] **Step 2: Run test, verify RED.**
- [ ] **Step 3: Implement guard**
Only run delist detection when `leafRows.length >= MIN_LEAF_CATEGORIES (10)` AND
`seenSdsGoodIds.size >= MIN_SEEN_GOODS (50)`. Otherwise skip and log warning.
- [ ] **Step 4: Verify GREEN + no regressions.**
### Task 4: Hardening — SDS client response sanity validation
**Files:**
- Modify: `apps/api/src/sync/sds-client.service.ts`
- Modify: `apps/api/src/sync/sds-client.service.spec.ts` (create if absent)
- [ ] **Step 1: Write failing test**
`fetchCategoryTree` throws when the returned array is degenerate (e.g. fewer than
`MIN_SDS_CATEGORY_NODES`), so callers never operate on a bad tree.
- [ ] **Step 2: Run test, verify RED.**
- [ ] **Step 3: Implement**
If the fetched category array length < MIN_SDS_CATEGORY_NODES (10), throw an Error
with the received count so sync logs FAILED (non-destructive) instead of deleting data.
- [ ] **Step 4: Verify GREEN + no regressions.**
### Task 5: Full verification
- [ ] **Step 1:** Run `apps/api` full jest suite — all green.
- [ ] **Step 2:** Typecheck `apps/api` (`tsc -p tsconfig.json --noEmit` or nest build).
- [ ] **Step 3:** Trigger a real sync via the admin UI / HTTP; confirm sync logs show
healthy category/product counts and `/public/goods` returns data.
- [ ] **Step 4:** Confirm product-center page at `http://192.168.124.137:3000/product-center` renders products.