metabuilder/dbal/docs

Database Abstraction Layer (DBAL)

A language-agnostic database abstraction layer that provides a secure interface between client applications and database backends. The DBAL uses TypeScript for rapid development and testing, with a C++ production layer for enhanced security and performance.

Architecture Overview

┌─────────────────────────────────────────────────────────────────┐
│                     Client Application (Spark)                   │
│                         (TypeScript/React)                       │
└────────────────────────────────┬────────────────────────────────┘
                                 │
                                 ▼
┌─────────────────────────────────────────────────────────────────┐
│                         DBAL Client                              │
│              (TypeScript Dev / C++ Production)                   │
│  ┌────────────────────┬──────────────────┬────────────────────┐ │
│  │  Query Builder     │  Validation      │  Error Handling    │ │
│  └────────────────────┴──────────────────┴────────────────────┘ │
└────────────────────────────────┬────────────────────────────────┘
                                 │
                    ┌────────────┴────────────┐
                    │    IPC/RPC Bridge       │
                    │   (gRPC/WebSocket)      │
                    └────────────┬────────────┘
                                 │
┌─────────────────────────────────────────────────────────────────┐
│                      DBAL Daemon (C++)                           │
│              [Production Only - Sandboxed]                       │
│  ┌────────────────────┬──────────────────┬────────────────────┐ │
│  │  Auth/ACL          │  Query Executor  │  Connection Pool   │ │
│  └────────────────────┴──────────────────┴────────────────────┘ │
└────────────────────────────────┬────────────────────────────────┘
                                 │
                    ┌────────────┴────────────┐
                    │                         │
                    ▼                         ▼
           ┌────────────────┐       ┌────────────────┐
           │  Prisma Client │       │  SQLite Direct │
           │  (Server-side) │       │   (Embedded)   │
           └────────────────┘       └────────────────┘
                    │                         │
                    ▼                         ▼
           ┌────────────────┐       ┌────────────────┐
           │   PostgreSQL   │       │   SQLite DB    │
           │     MySQL      │       │                │
           │   SQL Server   │       │                │
           └────────────────┘       └────────────────┘

Supported Databases

The Prisma adapter behind DBAL already targets the databases you care about: PostgreSQL, MySQL, SQLite, and any other engine Prisma supports (SQL Server, CockroachDB, MongoDB, etc.). Switch between them by pointing DATABASE_URL at the desired backend and regenerating the Prisma client for your schema.

The TypeScript client exposes three Prisma-based adapters: PrismaAdapter, PostgresAdapter, and MySQLAdapter. Setting config.adapter to 'postgres' or 'mysql' constructs the dialect-specific adapter, which keeps the shared Prisma logic but tweaks the capabilities metadata (e.g., enabling full-text search where supported) and leaves the rest of the stack focused on validation, ACLs, and audit logging.

# PostgreSQL
export DATABASE_URL="postgresql://user:pass@db:5432/metabuilder"

# MySQL
export DATABASE_URL="mysql://user:pass@db:3306/metabuilder"

npx prisma generate

With config.adapter = 'prisma', DBAL sends every request through PrismaAdapter, and Prisma handles dialect differences, migrations, and connection pooling defined in prisma/schema.prisma (generated from DBAL) and prisma/migrations/. That keeps DBAL focused on validation, ACLs, and audit logging while it can still drive PostgreSQL, MySQL, or any other Prisma-supported store.

The C++ daemon still resides in Phase 3—the current implementation is backed by the in-memory store described in dbal/production/docs/PHASE3_DAEMON.md, so Postgres/MySQL adapters for the daemon are still future work.

Native Prisma bridge

The Phase 3 daemon can still leverage Prisma without bundling Node by calling NativePrismaAdapter. Each SQL plan is serialized as a JSON payload with the $n or ? placeholders plus parameters and sent to /api/native-prisma on the Next.js server. The API route validates DBAL_NATIVE_PRISMA_TOKEN, reconstructs a Prisma.sql template, executes the query through the shared Prisma client, and returns rows or affected counts so the daemon sees the same SqlRow/int values as a regular SQL adapter. Set the same DBAL_NATIVE_PRISMA_TOKEN (mirrored in frontends/nextjs/.env.example) when running the daemon so the bridge rejects unauthorized callers.

Design Principles

1. Language Agnostic: API contracts defined in YAML/Proto, not tied to any language
2. Security First: C++ daemon sandboxes all database access with ACL enforcement
3. Development Speed: TypeScript implementation for rapid iteration
4. Zero Trust: User code never touches database credentials or raw connections
5. Capability-based: Adapters declare what they support (transactions, joins, TTL, etc.)
6. Testable: Shared test vectors ensure both implementations behave identically

Repository Structure

dbal/
├── api/                    # Language-agnostic contracts (source of truth)
│   ├── schema/            # Entity and operation definitions
│   ├── idl/               # Optional: Proto/FlatBuffers schemas
│   └── versioning/        # Compatibility rules
├── common/                # Shared test vectors and fixtures
├── ts/                    # TypeScript implementation (development)
├── cpp/                   # C++ implementation (production)
├── backends/              # Backend-specific assets
├── tools/                 # Code generation and build tools
└── scripts/               # Cross-platform build scripts

Quick Start

Development Mode (TypeScript)

cd dbal/development
npm install
npm run build
npm test

Production Mode (C++ Daemon)

cd dbal/production
mkdir build && cd build
cmake ..
make
./dbal_daemon --config=../config/prod.yaml

GitHub Spark Integration

For GitHub Spark deployments, the DBAL daemon runs as a sidecar service:

# In your Spark deployment config
services:
  dbal:
    image: your-org/dbal-daemon:latest
    ports:
      - "50051:50051"  # gRPC endpoint
    environment:
      - DBAL_MODE=production
      - DBAL_SANDBOX=strict

Monitoring & Daemon UI

frontends/dbal is a dedicated Next.js mini-app that showcases the C++ daemon's architecture, deployment readiness, and the ServerStatusPanel. The main frontends/nextjs app re-exports the @dbal-ui component at /dbal-daemon, and the panel polls /api/status (the shared feed lives in frontends/dbal/src/status.ts). Keep this page covered with frontends/nextjs/e2e/dbal-daemon/daemon.spec.ts and playwright.dbal-daemon.config.ts, or run npm run test:e2e:dbal-daemon after touching the UI.

Security Model

Sandboxing Strategy

1. Process Isolation: Daemon runs in separate process with restricted permissions
2. Capability-based Security: Each request checked against user ACL
3. Query Validation: All queries parsed and validated before execution
4. Credential Protection: DB credentials never exposed to client code
5. Audit Logging: All operations logged for security review

ACL System

user: "user_123"
role: "editor"
permissions:
  - entity: "posts"
    operations: [create, read, update]
    filters:
      author_id: "$user.id"  # Row-level security
  - entity: "comments"
    operations: [create, read]

API Contract Example

HTTP Utilities

For outbound integrations the daemon can use the new requests-inspired helper runtime::RequestsClient. It wraps the cpr HTTP helpers, exposes get/post helpers, parses JSON responses, and throws clean timeouts so code paths stay predictable.

Native Prisma calls route through NativePrismaAdapter, which currently POSTs to the /api/native-prisma Next.js API and returns the raw JSON rows or affected count using that helper. When the daemon calls runQuery/runNonQuery, the response is mapped back into SqlRow results so the rest of the stack stays unaware of the HTTP transport.

using namespace dbal::runtime;

RequestsClient http("https://api.prisma.example");
auto response = http.post("/rpc/execute", jsonPayload.dump(), {{"Authorization", "Bearer ..."}});
if (response.statusCode == 200) {
  const auto result = response.json["result"];
  // handle Prisma response
}

Entity Definition (YAML)

# api/schema/entities/post.yaml
entity: Post
version: "1.0"
fields:
  id:
    type: uuid
    primary: true
    generated: true
  title:
    type: string
    required: true
    max_length: 200
  content:
    type: text
    required: true
  author_id:
    type: uuid
    required: true
    foreign_key:
      entity: User
      field: id
  created_at:
    type: datetime
    generated: true
  updated_at:
    type: datetime
    auto_update: true

Operations (YAML)

# api/schema/operations/post.ops.yaml
operations:
  create:
    input: [title, content, author_id]
    output: Post
    acl_required: ["post:create"]
    
  read:
    input: [id]
    output: Post
    acl_required: ["post:read"]
    
  update:
    input: [id, title?, content?]
    output: Post
    acl_required: ["post:update"]
    row_level_check: "author_id = $user.id"
    
  delete:
    input: [id]
    output: boolean
    acl_required: ["post:delete"]
    row_level_check: "author_id = $user.id OR $user.role = 'admin'"
    
  list:
    input: [filter?, sort?, page?, limit?]
    output: Post[]
    acl_required: ["post:read"]

Client Usage

TypeScript Client

import { DBALClient } from '@metabuilder/dbal'

const client = new DBALClient({
  mode: 'development', // or 'production'
  endpoint: 'localhost:50051',
  auth: {
    user: currentUser,
    session: currentSession
  }
})

// CRUD operations
const post = await client.posts.create({
  title: 'Hello World',
  content: 'This is my first post',
  author_id: user.id
})

const posts = await client.posts.list({
  filter: { author_id: user.id },
  sort: { created_at: 'desc' },
  limit: 10
})

const updated = await client.posts.update(post.id, {
  title: 'Updated Title'
})

await client.posts.delete(post.id)

Development Workflow

1. Define Schema: Edit YAML files in api/schema/
2. Generate Code: python tools/codegen/gen_types.py
3. Implement Adapter: Add backend support in ts/src/adapters/
4. Write Tests: Create conformance tests in common/fixtures/
5. Run Tests: npm run test:conformance
6. Build C++ Daemon: cd cpp && cmake --build build
7. Deploy: Use Docker/Kubernetes to deploy daemon

Testing

Conformance Testing

The DBAL includes comprehensive conformance tests that ensure both TypeScript and C++ implementations behave identically:

# Run all conformance tests
python tools/conformance/run_all.py

# Run TS tests only
cd ts && npm run test:conformance

# Run C++ tests only
cd cpp && ./build/tests/conformance_tests

Test Vectors

Shared test vectors in common/fixtures/ ensure consistency:

# common/contracts/conformance_cases.yaml
- name: "Create and read post"
  operations:
    - action: create
      entity: Post
      input:
        title: "Test Post"
        content: "Test content"
        author_id: "user_123"
      expected:
        status: success
        output:
          id: "<uuid>"
          title: "Test Post"
    - action: read
      entity: Post
      input:
        id: "$prev.id"
      expected:
        status: success
        output:
          title: "Test Post"

Migration from Current System

Phase 1: Development Mode (Complete)

• Use TypeScript DBAL client in development
• Direct Prisma access (no daemon)
• Validates API contract compliance

Phase 2: Hybrid Mode (Current Implementation)

• Complete TypeScript DBAL client with Prisma adapter
• WebSocket bridge for remote daemon communication (prepared for C++)
• ACL enforcement and audit logging in TypeScript
• Runs entirely in GitHub Spark environment
• Prepares architecture for C++ daemon migration

Phase 3: Full Production (Future)

• All environments use C++ daemon
• TypeScript client communicates via WebSocket/gRPC
• Maximum security and performance
• Requires infrastructure beyond GitHub Spark

Capabilities System

Different backends support different features:

# api/schema/capabilities.yaml
adapters:
  prisma:
    transactions: true
    joins: true
    full_text_search: false
    ttl: false
    json_queries: true
    
  sqlite:
    transactions: true
    joins: true
    full_text_search: true
    ttl: false
    json_queries: true
    
  mongodb:
    transactions: true
    joins: false
    full_text_search: true
    ttl: true
    json_queries: true

Client code can check capabilities:

if (await client.capabilities.hasJoins()) {
  // Use join query
} else {
  // Fall back to multiple queries
}

Error Handling

Standardized errors across all implementations:

# api/schema/errors.yaml
errors:
  NOT_FOUND:
    code: 404
    message: "Entity not found"
    
  CONFLICT:
    code: 409
    message: "Entity already exists"
    
  UNAUTHORIZED:
    code: 401
    message: "Authentication required"
    
  FORBIDDEN:
    code: 403
    message: "Insufficient permissions"
    
  VALIDATION_ERROR:
    code: 422
    message: "Validation failed"
    fields:
      - field: string
        error: string

Contributing

See CONTRIBUTING.md for development guidelines.

License

MIT License - see LICENSE