690427:0812 Update Infras #01
CI / CD Pipeline / build (push) Successful in 5m51s
CI / CD Pipeline / deploy (push) Successful in 2m9s

This commit is contained in:
2026-04-27 08:12:28 +07:00
parent 9384581aee
commit a57fef4d44
68 changed files with 9750 additions and 468 deletions
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
+61 -11
View File
@@ -1,10 +1,12 @@
---
name: nestjs-best-practices
description: NestJS best practices and architecture patterns for building production-ready applications. This skill should be used when writing, reviewing, or refactoring NestJS code to ensure proper patterns for modules, dependency injection, security, and performance.
description: NestJS best practices and architecture patterns for building production-ready LCBP3-DMS backend code. Enforces ADR-009 (no TypeORM migrations), ADR-019 (hybrid UUID), ADR-016 (security), ADR-007 (error handling), ADR-008 (BullMQ), ADR-001/002 (workflow + numbering), ADR-018/020 (AI boundary), and ADR-021 (workflow context).
version: 1.8.9
scope: backend
user-invocable: false
license: MIT
metadata:
author: Kadajett
version: '1.1.0'
upstream: 'Kadajett/nestjs-best-practices v1.1.0 (forked + LCBP3-aligned)'
---
# NestJS Best Practices
@@ -110,6 +112,13 @@ Reference these guidelines when:
- `devops-use-logging` - Structured logging
- `devops-graceful-shutdown` - Zero-downtime deployments
### 11. LCBP3-Specific (CRITICAL — Project Overrides)
- `db-no-typeorm-migrations`**CRITICAL** ADR-009: edit SQL directly
- `lcbp3-workflow-engine`**CRITICAL** ADR-001/002/021: DSL state machine + double-lock numbering + workflow context
- `security-file-two-phase-upload`**CRITICAL** ADR-016: Upload → Temp → ClamAV → Commit
- `lcbp3-ai-boundary`**CRITICAL** ADR-018/020: Ollama on-prem only, human-in-the-loop
## NAP-DMS Project-Specific Rules (MUST FOLLOW)
These rules override general NestJS best practices for the NAP-DMS project:
@@ -120,21 +129,62 @@ These rules override general NestJS best practices for the NAP-DMS project:
- แก้ไข schema โดยตรงที่: `specs/03-Data-and-Storage/lcbp3-v1.8.0-schema-02-tables.sql`
- ใช้ n8n workflow สำหรับ data migration ถ้าจำเป็น
### ADR-019: Hybrid Identifier Strategy (CRITICAL)
### ADR-019: Hybrid Identifier Strategy (CRITICAL — March 2026 Pattern)
> **Updated pattern:** `UuidBaseEntity` exposes `publicId` **directly**. ห้ามใช้ `@Expose({ name: 'id' })` — API จะคืน `publicId` เป็น field name ตรงๆ.
```typescript
// ✅ CORRECT — ใช้ UuidBaseEntity
@Entity()
export class Project {
@PrimaryGeneratedColumn()
@Exclude() // ห้ามส่งออกทาง API
id: number; // INT AUTO_INCREMENT - internal only
export class Project extends UuidBaseEntity {
// publicId (string UUIDv7) + id (INT, @Exclude) สืบทอดจาก UuidBaseEntity
// API response → { publicId: "019505a1-7c3e-7000-8000-abc123..." }
@Column({ type: 'uuid' })
@Expose({ name: 'id' }) // ส่งออกเป็น 'id' ทาง API
publicId: string; // UUIDv7 - public API identifier
@Column()
projectCode: string;
@Column()
projectName: string;
}
```
```typescript
// ❌ WRONG — pattern เก่า ห้ามใช้
@Entity()
export class OldProject {
@PrimaryGeneratedColumn()
@Exclude()
id: number;
@Column({ type: 'uuid' })
@Expose({ name: 'id' }) // ❌ อย่า rename publicId เป็น 'id'
publicId: string;
}
```
**DTO Input (รับ UUID จาก Frontend):**
```typescript
export class CreateContractDto {
@IsUUID('7')
projectUuid: string; // รับ UUID string จาก client
}
// Controller resolves UUID → INT internally
@Post()
async create(@Body() dto: CreateContractDto) {
const projectId = await this.projectService.resolveInternalId(dto.projectUuid);
return this.contractService.create({ ...dto, projectId });
}
```
**ห้ามเด็ดขาด (CI Blocker):**
-`parseInt(projectPublicId)` — "019505…" → 19 (silently wrong)
-`Number(publicId)` / `+publicId` — NaN
-`@Expose({ name: 'id' })` บน `publicId` (pattern เก่า)
- ❌ Expose INT `id` ใน API response (ต้อง `@Exclude()` เสมอ)
### Two-Phase File Upload
```typescript
@@ -0,0 +1,24 @@
{
"version": "1.8.9",
"organization": "**NAP-DMS / LCBP3** — Laem Chabang Port Phase 3 Document Management System",
"date": "2026-04-22",
"abstract": "Comprehensive NestJS best-practices guide compiled for the LCBP3-DMS backend. Contains 40+ rules across 11 categories (10 general + 1 project-specific), prioritized by impact. Forked from Kadajett/nestjs-best-practices (v1.1.0) and aligned to LCBP3 ADRs: ADR-001 (workflow engine), ADR-002 (document numbering), ADR-007 (error handling), ADR-008 (notifications/BullMQ), ADR-009 (no TypeORM migrations), ADR-016 (security), ADR-018/020 (AI boundary), ADR-019 (hybrid UUID identifier — March 2026 pattern), and ADR-021 (workflow context).\n\nThis document is the single, consolidated reference used by Cascade and other AI coding agents when writing, reviewing, or refactoring backend code in this repository. All LCBP3-specific overrides live in section 11.",
"references": [
"[AGENTS.md (root)](../../../AGENTS.md) — canonical AI agent rules",
"[CONTRIBUTING.md](../../../CONTRIBUTING.md) — spec authoring + PR process",
"[ADR-001 Unified Workflow Engine](../../../specs/06-Decision-Records/ADR-001-unified-workflow-engine.md)",
"[ADR-002 Document Numbering Strategy](../../../specs/06-Decision-Records/ADR-002-document-numbering-strategy.md)",
"[ADR-007 Error Handling Strategy](../../../specs/06-Decision-Records/ADR-007-error-handling-strategy.md)",
"[ADR-008 Email/Notification Strategy](../../../specs/06-Decision-Records/ADR-008-email-notification-strategy.md)",
"[ADR-009 Database Migration Strategy](../../../specs/06-Decision-Records/ADR-009-database-migration-strategy.md)",
"[ADR-016 Security & Authentication](../../../specs/06-Decision-Records/ADR-016-security-authentication.md)",
"[ADR-018 AI Boundary](../../../specs/06-Decision-Records/ADR-018-ai-boundary.md)",
"[ADR-019 Hybrid Identifier Strategy](../../../specs/06-Decision-Records/ADR-019-hybrid-identifier-strategy.md)",
"[ADR-020 AI Intelligence Integration](../../../specs/06-Decision-Records/ADR-020-ai-intelligence-integration.md)",
"[ADR-021 Workflow Context](../../../specs/06-Decision-Records/ADR-021-workflow-context.md)",
"[Backend Engineering Guidelines](../../../specs/05-Engineering-Guidelines/05-02-backend-guidelines.md)",
"[Schema — v1.8.0 Tables](../../../specs/03-Data-and-Storage/lcbp3-v1.8.0-schema-02-tables.sql)",
"[Data Dictionary](../../../specs/03-Data-and-Storage/03-01-data-dictionary.md)",
"Upstream: [Kadajett/nestjs-best-practices](https://github.com/Kadajett/nestjs-best-practices) v1.1.0"
]
}
@@ -5,20 +5,22 @@ impactDescription: Use INT PK internally + UUID for public API per project ADR-0
tags: database, uuid, identifier, adr-019, api-design, typeorm
---
## Hybrid Identifier Strategy (ADR-019)
## Hybrid Identifier Strategy (ADR-019) — March 2026 Pattern
**This project follows ADR-019: INT Primary Key (internal) + UUIDv7 (public API)**
Unlike standard practices that use UUID as the primary key, this project uses a **hybrid approach** optimized for MariaDB performance and API consistency.
> **Updated pattern (March 2026):** Entities extend `UuidBaseEntity`. The `publicId` column is exposed **directly** in API responses — ห้ามใช้ `@Expose({ name: 'id' })` เพื่อ rename.
### The Strategy
| Layer | Field | Type | Usage |
|-------|-------|------|-------|
| **Database PK** | `id` | `INT AUTO_INCREMENT` | Internal foreign keys only |
| **Public API** | `uuid` | `MariaDB UUID` (native) | External references, URLs |
| **DTO Input** | `xxxUuid` | `string` | Accept UUID in create/update |
| **DTO Output** | `id` | `string` | API returns UUID as `id` via `@Expose` |
| Layer | Field | Type | Usage |
| --------------- | ---------- | ----------------------------------- | ------------------------------------------------- |
| **Database PK** | `id` | `INT AUTO_INCREMENT` | Internal foreign keys only (marked `@Exclude()`) |
| **Public API** | `publicId` | `MariaDB UUID` (native, BINARY(16)) | External references, URLs — exposed as-is |
| **DTO Input** | `xxxUuid` | `string` (UUIDv7) | Accept UUID in create/update DTOs |
| **DTO Output** | `publicId` | `string` (UUIDv7) | API returns `publicId` field directly (no rename) |
### Why Hybrid IDs?
@@ -27,31 +29,51 @@ Unlike standard practices that use UUID as the primary key, this project uses a
- **Compatibility**: UUID works well with distributed systems and external integrations
- **MariaDB Native**: Uses MariaDB's native UUID type (stored as BINARY(16), auto-converts to string)
### Entity Definition
### Entity Definition (Current Pattern)
```typescript
import { Entity, PrimaryGeneratedColumn, Column, Index } from 'typeorm';
import { Exclude, Expose } from 'class-transformer';
import { Entity, Column } from 'typeorm';
import { UuidBaseEntity } from '@/common/entities/uuid-base.entity';
@Entity('contracts')
export class Contract {
@PrimaryGeneratedColumn()
@Exclude() // Never expose in API response
id: number; // Internal INT PK - used for FK relationships
@Column({ type: 'uuid', unique: true })
@Expose({ name: 'id' }) // Exposed as 'id' in API
uuid: string; // Public UUIDv7 - what API consumers see
export class Contract extends UuidBaseEntity {
// publicId (string UUIDv7) + id (INT, @Exclude) สืบทอดจาก UuidBaseEntity
// API response → { publicId: "019505a1-7c3e-7000-8000-abc123...", contractCode: ..., ... }
@Column()
contractCode: string;
@Column()
contractName: string;
@Column({ name: 'project_id' })
projectId: number; // INT FK — internal, not exposed if marked @Exclude in UuidBaseEntity
}
```
### DTO Pattern (Accept UUID, Resolve to INT)
**`UuidBaseEntity` (shared base):**
```typescript
import { PrimaryGeneratedColumn, Column, CreateDateColumn, UpdateDateColumn } from 'typeorm';
import { Exclude } from 'class-transformer';
export abstract class UuidBaseEntity {
@PrimaryGeneratedColumn()
@Exclude() // ❗ CRITICAL: INT id must never leak to API
id: number;
@Column({ type: 'uuid', unique: true, generated: 'uuid' })
publicId: string; // UUIDv7, exposed as-is
@CreateDateColumn()
createdAt: Date;
@UpdateDateColumn()
updatedAt: Date;
}
```
### DTO Pattern (Accept UUID, Resolve to INT Internally)
```typescript
// dto/create-contract.dto.ts
@@ -59,8 +81,8 @@ import { IsUUID, IsNotEmpty } from 'class-validator';
export class CreateContractDto {
@IsNotEmpty()
@IsUUID('4')
projectUuid: string; // Accept UUID from client
@IsUUID('7') // UUIDv7 (MariaDB native)
projectUuid: string; // Accept UUID from client
@IsNotEmpty()
contractCode: string;
@@ -69,48 +91,38 @@ export class CreateContractDto {
contractName: string;
}
// dto/contract-response.dto.ts
import { Exclude, Expose } from 'class-transformer';
export class ContractResponseDto {
@Expose({ name: 'id' })
uuid: string; // Returned as 'id' field in JSON
contractCode: string;
contractName: string;
}
// ❌ NO Response DTO with @Expose rename needed.
// Entity class_transformer via TransformInterceptor will serialize publicId directly.
```
### Service/Controller Pattern
```typescript
@Controller('contracts')
@UseGuards(JwtAuthGuard, CaslAbilityGuard)
export class ContractsController {
constructor(
private contractsService: ContractsService,
private uuidResolver: UuidResolver, // Helper to convert UUID → INT
private uuidResolver: UuidResolver
) {}
@Post()
async create(@Body() dto: CreateContractDto) {
// Resolve UUID to INT PK for database operations
// Resolve UUID INT PK for FK relationship
const projectId = await this.uuidResolver.resolveProject(dto.projectUuid);
// Create with INT FK
const contract = await this.contractsService.create({
...dto,
projectId, // INT for database
projectId,
});
// Response automatically transforms via @Expose
// Response: TransformInterceptor + @Exclude on id → publicId exposed directly
return contract;
}
@Get(':id')
async findOne(@Param('id') uuid: string) {
// Controller receives UUID string
// Service handles UUID → INT resolution internally
return this.contractsService.findByUuid(uuid);
@Get(':publicId')
async findOne(@Param('publicId', ParseUuidPipe) publicId: string) {
return this.contractsService.findOneByPublicId(publicId);
}
}
```
@@ -124,21 +136,21 @@ export class UuidResolver {
@InjectRepository(Project)
private projectRepo: Repository<Project>,
@InjectRepository(Contract)
private contractRepo: Repository<Contract>,
private contractRepo: Repository<Contract>
) {}
async resolveProject(uuid: string): Promise<number> {
async resolveProject(publicId: string): Promise<number> {
const project = await this.projectRepo.findOne({
where: { uuid },
select: ['id'], // Only fetch INT PK
where: { publicId },
select: ['id'], // Only INT PK for FK
});
if (!project) throw new NotFoundException('Project not found');
return project.id;
}
async resolveContract(uuid: string): Promise<number> {
async resolveContract(publicId: string): Promise<number> {
const contract = await this.contractRepo.findOne({
where: { uuid },
where: { publicId },
select: ['id'],
});
if (!contract) throw new NotFoundException('Contract not found');
@@ -147,20 +159,20 @@ export class UuidResolver {
}
```
### TransformInterceptor (Required)
### TransformInterceptor (Required — register ONCE)
```typescript
// Must be configured globally to handle @Exclude/@Expose
// Register via APP_INTERCEPTOR in CommonModule — ห้ามซ้ำใน main.ts
@Injectable()
export class TransformInterceptor implements NestInterceptor {
intercept(context: ExecutionContext, next: CallHandler): Observable<any> {
return next.handle().pipe(
map((data) => instanceToPlain(data)), // Applies class-transformer decorators
map((data) => instanceToPlain(data)) // Applies @Exclude / @Expose
);
}
}
// app.module.ts
// common.module.ts
@Module({
providers: [
{
@@ -169,40 +181,42 @@ export class TransformInterceptor implements NestInterceptor {
},
],
})
export class AppModule {}
export class CommonModule {}
```
> **Warning:** ห้ามเรียก `app.useGlobalInterceptors(new TransformInterceptor())` ใน `main.ts` ซ้ำ — จะทำให้ response double-wrap `{ data: { data: ... } }`.
### Critical: NEVER ParseInt on UUID
```typescript
// ❌ WRONG - parseInt on UUID gives garbage value
const id = parseInt(projectUuid); // "0195a1b2-..." → 195 (wrong!)
const id = parseInt(projectPublicId); // "0195a1b2-..." → 195 (wrong!)
// ❌ WRONG - Number() on UUID
const id = Number(projectUuid); // NaN
const id = Number(projectPublicId); // NaN
// ❌ WRONG - Unary plus on UUID
const id = +projectUuid; // NaN
const id = +projectPublicId; // NaN
// ✅ CORRECT - Resolve via database lookup
const projectId = await uuidResolver.resolveProject(projectUuid);
const projectId = await uuidResolver.resolveProject(projectPublicId);
// ✅ CORRECT - Use TypeORM find with UUID column
const project = await projectRepo.findOne({ where: { uuid: projectUuid } });
const id = project.id; // Get INT PK from entity
// ✅ CORRECT - Use TypeORM find with publicId column
const project = await projectRepo.findOne({ where: { publicId: projectPublicId } });
const id = project.id; // Get INT PK from entity
```
### Query with UUID (No Resolution Needed)
### Query with publicId (No Resolution Needed)
```typescript
// Direct UUID lookup in TypeORM
const project = await this.projectRepo.findOne({
where: { uuid: projectUuid }, // Query by UUID column
where: { publicId: projectPublicId },
});
// Relations use INT FK internally
const contracts = await this.contractRepo.find({
where: { projectId: project.id }, // INT for FK query
where: { projectId: project.id }, // INT for FK query
});
```
@@ -0,0 +1,100 @@
---
title: No TypeORM Migrations (ADR-009)
impact: CRITICAL
impactDescription: Edit SQL schema files directly; n8n handles data migration. Do not generate TypeORM migration files.
tags: database, schema, migration, adr-009, sql, n8n
---
## No TypeORM Migrations (ADR-009)
**This project does NOT use TypeORM migration files.**
All schema changes must be made **directly** in the canonical SQL file:
- `specs/03-Data-and-Storage/lcbp3-v1.8.0-schema-02-tables.sql`
Delta scripts (for incremental rollout to existing environments) go under:
- `specs/03-Data-and-Storage/deltas/YYYY-MM-DD-descriptive-name.sql`
Data migration (e.g., backfilling a new column) is handled by **n8n workflows**, not TypeORM's `QueryRunner`.
---
## Why No Migrations?
1. **Single source of truth** — The full SQL schema is always readable as one file. No need to replay a migration chain to understand current state.
2. **Review friendly** — Schema diff = git diff on the SQL file. Reviewers see the complete picture.
3. **Ops alignment** — DBAs and operators work in SQL, not TypeScript.
4. **n8n for data** — Business-meaningful data transforms live in n8n where they can be versioned, retried, and orchestrated with monitoring.
---
## ✅ Workflow for a Schema Change
1. **Update Data Dictionary** first:
- `specs/03-Data-and-Storage/03-01-data-dictionary.md` — add field meaning + business rules.
2. **Update the canonical schema**:
- Edit `lcbp3-v1.8.0-schema-02-tables.sql` — add/alter column, constraint, index.
3. **Add a delta script** (if deploying to existing env):
- `specs/03-Data-and-Storage/deltas/2026-04-22-add-rfa-revision-column.sql`
```sql
-- Delta: Add revision column to rfa table
ALTER TABLE rfa
ADD COLUMN revision INT NOT NULL DEFAULT 1 AFTER status;
CREATE INDEX idx_rfa_revision ON rfa(revision);
```
4. **Update the Entity** (`backend/src/.../entities/rfa.entity.ts`):
```typescript
@Column({ type: 'int', default: 1 })
revision: number;
```
5. **If data backfill needed** → create n8n workflow, not TypeScript migration.
---
## ❌ Forbidden
```bash
# ❌ DO NOT generate migrations
pnpm typeorm migration:generate ./src/migrations/AddRevision
# ❌ DO NOT run migrations
pnpm typeorm migration:run
```
```typescript
// ❌ DO NOT write migration classes
export class AddRevision1730000000000 implements MigrationInterface {
async up(queryRunner: QueryRunner): Promise<void> { /* ... */ }
async down(queryRunner: QueryRunner): Promise<void> { /* ... */ }
}
```
---
## ✅ TypeORM Config (runtime only)
```typescript
// ormconfig.ts
export default {
type: 'mariadb',
// ...
synchronize: false, // ❗ NEVER true (would auto-sync entity ↔ schema)
migrationsRun: false, // ❗ NEVER true
// ❌ Do NOT specify `migrations:` entries
};
```
`synchronize: false` is mandatory because the canonical SQL file is authoritative — TypeORM should never mutate the schema.
---
## Reference
- [ADR-009 Database Migration Strategy](../../../../specs/06-Decision-Records/ADR-009-database-migration-strategy.md)
- [Data Dictionary](../../../../specs/03-Data-and-Storage/03-01-data-dictionary.md)
- [Schema Tables](../../../../specs/03-Data-and-Storage/lcbp3-v1.8.0-schema-02-tables.sql)
@@ -0,0 +1,157 @@
---
title: AI Integration Boundary (ADR-018 / ADR-020)
impact: CRITICAL
impactDescription: AI runs on Admin Desktop only; AI → DMS API → DB (never direct); human-in-the-loop validation mandatory; full audit trail.
tags: ai, ollama, boundary, adr-018, adr-020, privacy, audit
---
## AI Integration Boundary
LCBP3 uses **on-premises AI only** (Ollama on Admin Desktop) with strict isolation from data layers.
---
## The Boundary
```
┌────────────────────────────────────────────────────────────┐
│ User Browser (Next.js) │
└─────────────────────────┬──────────────────────────────────┘
│ (authenticated HTTPS)
┌─────────────────────────▼──────────────────────────────────┐
│ DMS API (NestJS) ◀── enforces CASL, validation, audit │
│ ├─ AiGateway (proxies to Ollama) │
│ └─ DB + Storage (Elasticsearch, MariaDB, File System) │
└─────────────────────────┬──────────────────────────────────┘
│ (HTTP → Admin Desktop, internal)
┌─────────────────────────▼──────────────────────────────────┐
│ Admin Desktop (Desk-5439) │
│ ├─ Ollama (Gemma 4) │
│ ├─ PaddleOCR (Thai + English) │
│ └─ n8n orchestration │
└────────────────────────────────────────────────────────────┘
```
**❗ Admin Desktop has NO network access to MariaDB, no SMB to storage, no shared secrets.** It receives base64-encoded file bytes over HTTPS and returns extracted text + suggestions.
---
## Required Patterns
### 1. AiGateway Module (backend)
```typescript
@Module({
controllers: [AiController],
providers: [AiService, AiGateway, AiAuditLogger],
exports: [AiService],
})
export class AiModule {}
@Injectable()
export class AiService {
async extractMetadata(fileId: number, user: User): Promise<ExtractedMetadata> {
// 1. Authorize (CASL: user can read this file)
await this.ability.ensureCan(user, 'read', File, fileId);
// 2. Load file (DMS API, inside the boundary)
const fileBytes = await this.storageService.read(fileId);
// 3. Call Admin Desktop AI over HTTP
const raw = await this.aiGateway.extract(fileBytes);
// 4. Validate AI output schema (Zod)
const parsed = ExtractedMetadataSchema.parse(raw);
// 5. Audit log (who, what, when, model, confidence)
await this.auditLogger.log({
userId: user.id,
action: 'ai.extract_metadata',
fileId,
model: raw.model,
confidence: parsed.confidence,
});
// 6. Return — frontend MUST render for human confirmation
return parsed;
}
}
```
### 2. Human-in-the-Loop
AI output is **never persisted directly**. Users must confirm via `DocumentReviewForm`:
```tsx
<DocumentReviewForm
document={doc}
aiSuggestions={suggestions}
onConfirm={(reviewed) => saveMetadata(reviewed)} // user edits applied
/>
```
The `user_confirmed_at` timestamp and diff (AI suggestion → final value) are stored in the audit log.
### 3. Rate Limiting
```typescript
@Post('ai/extract')
@UseGuards(JwtAuthGuard, CaslAbilityGuard, ThrottlerGuard)
@Throttle({ default: { limit: 10, ttl: 60_000 } }) // 10 req/min/user
async extract(@Body() dto: ExtractDto) { /* ... */ }
```
---
## ❌ Forbidden
```typescript
// ❌ AI container connecting to DB
// docker-compose.yml inside ai-service:
// environment:
// DATABASE_URL: mysql://... ← NEVER
// ❌ AI SDK calling cloud API
import OpenAI from 'openai'; // ❌ No cloud AI SDKs in production code
const client = new OpenAI({ apiKey: ... });
// ❌ Persisting AI output without human confirm
async extractAndSave(fileId: number) {
const metadata = await this.ai.extract(fileId);
await this.repo.save({ fileId, ...metadata }); // ❌ skips human review
}
// ❌ Skipping audit log
const result = await this.aiGateway.extract(bytes); // no logging
return result;
```
---
## Audit Log Schema
```sql
CREATE TABLE ai_audit_log (
id INT AUTO_INCREMENT PRIMARY KEY,
public_id UUID UNIQUE NOT NULL,
user_id INT NOT NULL,
action VARCHAR(64) NOT NULL, -- 'ai.extract_metadata', 'ai.classify', etc.
file_id INT,
model VARCHAR(64), -- 'gemma-4:7b', 'paddleocr-v3'
confidence DECIMAL(4,3),
input_hash CHAR(64), -- SHA-256 of input for replay detection
output_summary JSON,
human_confirmed_at DATETIME,
created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
INDEX idx_user_created (user_id, created_at),
INDEX idx_file (file_id)
);
```
---
## Reference
- [ADR-018 AI Boundary](../../../../specs/06-Decision-Records/ADR-018-ai-boundary.md)
- [ADR-020 AI Intelligence Integration](../../../../specs/06-Decision-Records/ADR-020-ai-intelligence-integration.md)
- [ADR-017 Ollama Data Migration](../../../../specs/06-Decision-Records/ADR-017-ollama-data-migration.md)
@@ -0,0 +1,181 @@
---
title: Workflow Engine + Document Numbering + Workflow Context (ADR-001 / 002 / 021)
impact: CRITICAL
impactDescription: DSL-based state machine; double-lock numbering; integrated workflow context exposed to clients.
tags: workflow, numbering, redlock, version-column, adr-001, adr-002, adr-021
---
## Workflow Engine + Numbering + Context
LCBP3 uses a **unified workflow engine** (DSL-based state machine) across RFA, Transmittal, Correspondence, Circulation, and Shop Drawing. Every state transition goes through the same engine — no per-type routing tables.
---
## ADR-001: Unified Workflow Engine
### State Transition Pattern
```typescript
@Injectable()
export class WorkflowEngine {
async transition(
instanceId: string,
action: WorkflowAction,
actor: User,
context?: WorkflowContext,
): Promise<WorkflowInstance> {
// 1. Load current state from DB (never trust client-provided state)
const instance = await this.repo.findOneByPublicId(instanceId);
if (!instance) throw new NotFoundException();
// 2. Validate transition against DSL
const dsl = await this.dslService.load(instance.workflowTypeId);
const nextState = dsl.resolve(instance.currentState, action);
if (!nextState) {
throw new BusinessException(
`Action ${action} not allowed from state ${instance.currentState}`,
'ไม่สามารถดำเนินการนี้ได้ในสถานะปัจจุบัน',
'กรุณาตรวจสอบขั้นตอนการอนุมัติ',
'WF_INVALID_TRANSITION',
);
}
// 3. Apply transition atomically (optimistic lock via @VersionColumn)
instance.currentState = nextState;
await this.repo.save(instance); // throws OptimisticLockVersionMismatchError on race
// 4. Emit event for listeners (notifications via BullMQ — ADR-008)
this.eventBus.publish(new WorkflowTransitionedEvent(instance, action, actor));
return instance;
}
}
```
### ❌ Anti-Patterns
- ❌ Hard-coded `switch (state)` in controllers/services
- ❌ Trusting `currentState` from request body
- ❌ Creating separate routing tables per document type
---
## ADR-002: Document Numbering (Double-Lock)
Concurrent requests for a new document number **must** use both:
1. **Redis Redlock** — distributed lock across app instances
2. **TypeORM `@VersionColumn`** — optimistic lock on counter row
### Counter Entity
```typescript
@Entity('document_number_counters')
@Unique(['projectId', 'documentTypeId'])
export class DocumentNumberCounter extends UuidBaseEntity {
@Column({ name: 'project_id' })
projectId: number;
@Column({ name: 'document_type_id' })
documentTypeId: number;
@Column({ name: 'last_number', default: 0 })
lastNumber: number;
@VersionColumn()
version: number; // ❗ Optimistic lock — do not rename, do not remove
}
```
### Service Pattern
```typescript
@Injectable()
export class DocumentNumberingService {
constructor(
@InjectRepository(DocumentNumberCounter)
private counterRepo: Repository<DocumentNumberCounter>,
private redlock: RedlockService,
private readonly logger: Logger,
) {}
async generateNext(ctx: NumberingContext): Promise<string> {
const lockKey = `doc_num:${ctx.projectId}:${ctx.documentTypeId}`;
// Distributed lock — 3s TTL, up to 5 retries
const lock = await this.redlock.acquire([lockKey], 3000);
try {
// Optimistic lock via @VersionColumn
const counter = await this.counterRepo.findOne({
where: { projectId: ctx.projectId, documentTypeId: ctx.documentTypeId },
});
if (!counter) {
throw new NotFoundException('Counter not initialized for this project/type');
}
counter.lastNumber += 1;
await this.counterRepo.save(counter); // may throw OptimisticLockVersionMismatchError
return this.formatNumber(ctx, counter.lastNumber);
} catch (err) {
if (err instanceof OptimisticLockVersionMismatchError) {
this.logger.warn(`Numbering race detected for ${lockKey}, retrying`);
// Let caller retry via BullMQ retry policy
}
throw err;
} finally {
await lock.release();
}
}
private formatNumber(ctx: NumberingContext, seq: number): string {
// e.g. "LCBP3-RFA-0042"
return `${ctx.projectCode}-${ctx.typeCode}-${String(seq).padStart(4, '0')}`;
}
}
```
### ❌ Anti-Patterns
- ❌ App-side counter only (`let counter = 0; counter++`)
- ❌ Using `findOne` + `update` without `@VersionColumn`
- ❌ Using only Redis lock without DB optimistic lock (race if Redis fails)
---
## ADR-021: Integrated Workflow Context
Every workflow-aware API response **must** expose:
```typescript
export class WorkflowEnvelope<T> {
data: T;
workflow: {
instancePublicId: string;
currentState: string; // e.g. 'pending_review'
availableActions: string[]; // e.g. ['approve', 'reject', 'request-revision']
canEdit: boolean; // computed from CASL + current state
lastTransitionAt: string; // ISO 8601
};
stepAttachments?: Array<{ // files produced by the current/previous step
publicId: string;
fileName: string;
stepCode: string;
downloadUrl: string;
}>;
}
```
Frontend uses `workflow.availableActions` to render buttons — no client-side state machine logic.
---
## Reference
- [ADR-001 Unified Workflow Engine](../../../../specs/06-Decision-Records/ADR-001-unified-workflow-engine.md)
- [ADR-002 Document Numbering Strategy](../../../../specs/06-Decision-Records/ADR-002-document-numbering-strategy.md)
- [ADR-021 Workflow Context](../../../../specs/06-Decision-Records/ADR-021-workflow-context.md)
@@ -0,0 +1,137 @@
---
title: Two-Phase File Upload + ClamAV (ADR-016)
impact: CRITICAL
impactDescription: Upload → Temp → ClamAV scan → Commit → Permanent. Whitelist + 50MB cap. StorageService only.
tags: file-upload, clamav, security, adr-016, storage
---
## Two-Phase File Upload (ADR-016)
**Never write uploaded files directly to permanent storage.** All uploads must go through:
```
Client → Upload endpoint → Temp storage → ClamAV scan → Commit endpoint → Permanent storage
```
---
## Constraints (non-negotiable)
| Rule | Value |
| --- | --- |
| Allowed MIME types | `application/pdf`, `image/vnd.dwg`, `application/vnd.openxmlformats-officedocument.wordprocessingml.document`, `application/vnd.openxmlformats-officedocument.spreadsheetml.sheet`, `application/zip` |
| Allowed extensions | `.pdf`, `.dwg`, `.docx`, `.xlsx`, `.zip` |
| Max size | 50 MB |
| Temp TTL | 24 h (purged by cron) |
| Virus scan | ClamAV (blocking) |
| Mover | `StorageService` only — never `fs.rename` directly from controller |
---
## Phase 1: Upload to Temp
```typescript
@Post('upload')
@UseGuards(JwtAuthGuard, ThrottlerGuard)
@UseInterceptors(FileInterceptor('file', {
limits: { fileSize: 50 * 1024 * 1024 }, // 50 MB
}))
async uploadTemp(
@UploadedFile() file: Express.Multer.File,
@CurrentUser() user: User,
): Promise<{ tempId: string; expiresAt: string }> {
// 1. Validate MIME + extension (defense in depth)
this.fileValidator.assertAllowed(file);
// 2. Scan with ClamAV
const scanResult = await this.clamavService.scan(file.buffer);
if (!scanResult.clean) {
throw new BusinessException(
`ClamAV rejected: ${scanResult.signature}`,
'ไฟล์ไม่ปลอดภัย ระบบตรวจพบความเสี่ยง',
'กรุณาตรวจสอบไฟล์และลองใหม่อีกครั้ง',
'FILE_INFECTED',
);
}
// 3. Save to temp (encrypted at rest)
const tempId = await this.storageService.saveToTemp(file, user.id);
return {
tempId,
expiresAt: addHours(new Date(), 24).toISOString(),
};
}
```
---
## Phase 2: Commit in Transaction
The business operation (e.g., creating a Correspondence) promotes temp files to permanent **in the same DB transaction**.
```typescript
async createCorrespondence(dto: CreateCorrespondenceDto, user: User) {
return this.dataSource.transaction(async (manager) => {
// 1. Create domain entity
const entity = await manager.save(Correspondence, {
...dto,
createdById: user.id,
});
// 2. Commit temp files → permanent (ACID together with entity)
await this.storageService.commitFiles(
dto.tempFileIds,
{ entityId: entity.id, entityType: 'correspondence' },
manager,
);
return entity;
});
}
```
If the transaction rolls back, temp files remain and expire in 24h — no orphaned permanent files.
---
## StorageService Contract
```typescript
export interface StorageService {
saveToTemp(file: Express.Multer.File, ownerId: number): Promise<string>;
commitFiles(
tempIds: string[],
target: { entityId: number; entityType: string },
manager: EntityManager,
): Promise<FileRecord[]>;
purgeExpiredTemp(): Promise<number>; // called by cron
getPermanentPath(fileId: number): Promise<string>;
}
```
---
## ❌ Forbidden
```typescript
// ❌ Direct write to permanent
fs.writeFileSync(`/var/storage/${file.originalname}`, file.buffer);
// ❌ Skip ClamAV
await this.storageService.savePermanent(file);
// ❌ Non-whitelist MIME
@UseInterceptors(FileInterceptor('file')) // no size or type limit
// ❌ Commit outside transaction
const entity = await this.repo.save(...);
await this.storageService.commitFiles(tempIds, ...); // race: entity exists, files may fail
```
---
## Reference
- [ADR-016 Security & Authentication](../../../../specs/06-Decision-Records/ADR-016-security-authentication.md)
- [Edge Cases](../../../../specs/01-Requirements/01-06-edge-cases-and-rules.md) — file upload scenarios
@@ -32,6 +32,7 @@ const CATEGORIES = [
{ prefix: 'api-', name: 'API Design', impact: 'MEDIUM', section: 8 },
{ prefix: 'micro-', name: 'Microservices', impact: 'MEDIUM', section: 9 },
{ prefix: 'devops-', name: 'DevOps & Deployment', impact: 'LOW-MEDIUM', section: 10 },
{ prefix: 'lcbp3-', name: 'LCBP3 Project-Specific', impact: 'CRITICAL', section: 11 },
];
interface RuleFrontmatter {
@@ -50,8 +51,10 @@ interface Rule {
}
function parseFrontmatter(content: string): { frontmatter: RuleFrontmatter | null; body: string } {
// Normalize CRLF → LF so the regex works on Windows-authored files
const normalized = content.replace(/\r\n/g, '\n');
const frontmatterRegex = /^---\n([\s\S]*?)\n---\n([\s\S]*)$/;
const match = content.match(frontmatterRegex);
const match = normalized.match(frontmatterRegex);
if (!match) {
return { frontmatter: null, body: content };