Data Access Abstraction

Build APIs with swappable data layers using the repository pattern. Develop locally with SQLite, deploy to Azure with Cosmos DB or PostgreSQL — same route code, different backend. Works in any language.

When to Use This Skill

• Building an API that needs to run locally (SQLite) and in Azure (Cosmos DB or PostgreSQL)
• Adding a new data provider to an existing app without changing routes or business logic
• Migrating from one database to another incrementally
• Any project following the AIMarket journey pattern (local dev → cloud deploy)

Pattern Overview

The pattern is the same regardless of language:

Routes/Controllers → Repository Interfaces → Factory → Implementations
                                                │
                                                ├── SQLite (local dev)
                                                ├── Cosmos DB (Azure deploy)
                                                ├── PostgreSQL (Azure deploy)
                                                └── In-memory (testing)

Three rules:

1. Define repository interfaces (or abstract classes / protocols) per entity
2. Routes depend only on the interfaces — never import a database client directly
3. A factory reads a config value (DATA_PROVIDER) and returns the right implementation

Adding a new database means writing a new implementation file. Zero changes to routes.

Environment Variable

All languages use the same convention:

DATA_PROVIDER=sqlite     # Local development (default)
DATA_PROVIDER=cosmos     # Azure Cosmos DB
DATA_PROVIDER=postgres   # Azure PostgreSQL

Node.js / TypeScript

Repository Interfaces

// data/interfaces.ts

export interface IProductRepository {
  getAll(params: {
    page: number; pageSize: number;
    category?: string; minPrice?: number; maxPrice?: number;
  }): Promise<{ data: Product[]; totalCount: number }>;

  getById(id: string): Promise<Product | null>;
  create(input: CreateProductInput): Promise<Product>;
  update(id: string, fields: Partial<Product>): Promise<Product | null>;
}

export interface IOrderRepository {
  create(input: CreateOrderInput): Promise<Order>;
  getById(id: string): Promise<Order | null>;
  getByUserId(userId: string, page: number, pageSize: number): Promise<{ data: Order[]; totalCount: number }>;
}

export interface IUserRepository {
  create(input: CreateUserInput): Promise<User>;
  getById(id: string): Promise<User | null>;
  getByEmail(email: string): Promise<User | null>;
}

export interface DataStore {
  products: IProductRepository;
  orders: IOrderRepository;
  users: IUserRepository;
  close?(): void;
}

Factory

// data/store.ts

export async function createStore(): Promise<DataStore> {
  const provider = process.env.DATA_PROVIDER || 'sqlite';
  switch (provider) {
    case 'sqlite':
      return (await import('./sqlite.js')).createSqliteStore();
    case 'cosmos':
      return (await import('./cosmos.js')).createCosmosStore();
    case 'postgres':
      return (await import('./postgres.js')).createPostgresStore();
    default:
      throw new Error(`Unknown DATA_PROVIDER: ${provider}`);
  }
}

SQLite Implementation

Use better-sqlite3 (synchronous API, fast). Wrap sync calls in Promise.resolve() or make routes await — sync values resolve immediately.

// data/sqlite.ts

import Database from 'better-sqlite3';

export function createSqliteStore(dbPath = 'app.db'): DataStore {
  const db = new Database(dbPath);
  db.pragma('journal_mode = WAL');
  db.pragma('foreign_keys = ON');
  // Create tables, seed if empty...

  return {
    products: {
      async getAll({ page, pageSize, category, minPrice, maxPrice }) {
        let where = 'WHERE status = ?';
        const params: any[] = ['active'];
        if (category) { where += ' AND category = ?'; params.push(category); }
        if (minPrice != null) { where += ' AND price >= ?'; params.push(minPrice); }
        if (maxPrice != null) { where += ' AND price <= ?'; params.push(maxPrice); }

        const { c: totalCount } = db.prepare(`SELECT COUNT(*) as c FROM products ${where}`).get(...params);
        const rows = db.prepare(`SELECT * FROM products ${where} LIMIT ? OFFSET ?`)
          .all(...params, pageSize, (page - 1) * pageSize);
        return { data: rows.map(parseRow), totalCount };
      },
      // ...
    },
    close: () => db.close(),
  };
}

SQLite-specific: Store arrays as JSON strings (JSON.stringify(tags)), parse on read. Use a junction table for order items.

Cosmos DB Implementation

// data/cosmos.ts

import { CosmosClient } from '@azure/cosmos';

export function createCosmosStore(): DataStore {
  const client = new CosmosClient({ endpoint: process.env.COSMOS_ENDPOINT!, key: process.env.COSMOS_KEY! });
  const db = client.database(process.env.COSMOS_DATABASE || 'aimarket');

  return {
    products: {
      async getAll({ page, pageSize, category, minPrice, maxPrice }) {
        let query = 'SELECT * FROM c WHERE c.status = @status';
        const parameters = [{ name: '@status', value: 'active' }];
        if (category) { query += ' AND c.category = @cat'; parameters.push({ name: '@cat', value: category }); }
        // ... build query
        const { resources } = await db.container('products').items.query({ query, parameters }).fetchAll();
        return { data: resources.slice((page-1)*pageSize, page*pageSize), totalCount: resources.length };
      },
      // ...
    },
  };
}

Cosmos-specific: Arrays and objects are native JSON — no serialization. Embed order items in the order document.

PostgreSQL Implementation

Use pg (node-postgres). PostgreSQL supports native arrays and JSONB, so it sits between SQLite and Cosmos in terms of data handling.

// data/postgres.ts

import pg from 'pg';

export function createPostgresStore(): DataStore {
  const pool = new pg.Pool({
    connectionString: process.env.POSTGRES_CONNECTION_STRING,
    ssl: process.env.POSTGRES_SSL === 'true' ? { rejectUnauthorized: false } : undefined,
  });

  return {
    products: {
      async getAll({ page, pageSize, category, minPrice, maxPrice }) {
        let where = 'WHERE status = $1';
        const params: any[] = ['active'];
        let idx = 2;

        if (category) { where += ` AND category = $${idx++}`; params.push(category); }
        if (minPrice != null) { where += ` AND price >= $${idx++}`; params.push(minPrice); }
        if (maxPrice != null) { where += ` AND price <= $${idx++}`; params.push(maxPrice); }

        const countRes = await pool.query(`SELECT COUNT(*) FROM products ${where}`, params);
        const totalCount = parseInt(countRes.rows[0].count);
        const dataRes = await pool.query(
          `SELECT * FROM products ${where} LIMIT $${idx++} OFFSET $${idx}`,
          [...params, pageSize, (page - 1) * pageSize]
        );
        return { data: dataRes.rows, totalCount };
      },
      // ...
    },
    close: () => pool.end(),
  };
}

PostgreSQL-specific: Use TEXT[] for arrays (native), JSONB for nested objects, and $1 parameterized queries. Azure PostgreSQL requires ssl: { rejectUnauthorized: false }.

Routes

// routes/products.ts — depends only on DataStore interface

export function productRoutes(store: DataStore): Router {
  const router = Router();
  router.get('/', async (req, res) => {
    const result = await store.products.getAll({ page: 1, pageSize: 20 });
    res.json(result);
  });
  return router;
}

Python / FastAPI

Repository Interfaces

Use Protocol (structural typing) or ABC (nominal typing). Protocol is more Pythonic.

# data/interfaces.py

from typing import Protocol
from models import Product, Order, User

class ProductRepository(Protocol):
    def get_all(self, *, page: int = 1, page_size: int = 20,
                category: str | None = None, min_price: float | None = None,
                max_price: float | None = None) -> dict:
        """Returns { 'data': list[Product], 'total_count': int }"""
        ...

    def get_by_id(self, id: str) -> Product | None: ...
    def create(self, data: dict) -> Product: ...
    def update(self, id: str, fields: dict) -> Product | None: ...

class OrderRepository(Protocol):
    def create(self, data: dict) -> Order: ...
    def get_by_id(self, id: str) -> Order | None: ...
    def get_by_user_id(self, user_id: str, page: int, page_size: int) -> dict: ...

class UserRepository(Protocol):
    def create(self, data: dict) -> User: ...
    def get_by_id(self, id: str) -> User | None: ...
    def get_by_email(self, email: str) -> User | None: ...

class DataStore(Protocol):
    products: ProductRepository
    orders: OrderRepository
    users: UserRepository

Factory

# data/store.py

import os

def create_store() -> DataStore:
    provider = os.getenv("DATA_PROVIDER", "sqlite")
    if provider == "sqlite":
        from data.sqlite import create_sqlite_store
        return create_sqlite_store()
    elif provider == "cosmos":
        from data.cosmos import create_cosmos_store
        return create_cosmos_store()
    elif provider == "postgres":
        from data.postgres import create_postgres_store
        return create_postgres_store()
    else:
        raise ValueError(f"Unknown DATA_PROVIDER: {provider}")

SQLite Implementation

# data/sqlite.py

import sqlite3, json, uuid
from models import Product

class SqliteProductRepository:
    def __init__(self, db: sqlite3.Connection):
        self.db = db

    def get_all(self, *, page=1, page_size=20, category=None, min_price=None, max_price=None):
        where, params = "WHERE status = ?", ["active"]
        if category:
            where += " AND category = ?"; params.append(category)
        if min_price is not None:
            where += " AND price >= ?"; params.append(min_price)
        if max_price is not None:
            where += " AND price <= ?"; params.append(max_price)

        total = self.db.execute(f"SELECT COUNT(*) FROM products {where}", params).fetchone()[0]
        rows = self.db.execute(
            f"SELECT * FROM products {where} LIMIT ? OFFSET ?",
            params + [page_size, (page - 1) * page_size]
        ).fetchall()

        return {"data": [self._parse(r) for r in rows], "total_count": total}

    def _parse(self, row) -> Product:
        d = dict(row)
        d["tags"] = json.loads(d["tags"])  # JSON string → list
        return Product(**d)

def create_sqlite_store(db_path="app.db"):
    db = sqlite3.connect(db_path)
    db.row_factory = sqlite3.Row
    db.execute("PRAGMA journal_mode=WAL")
    db.execute("PRAGMA foreign_keys=ON")
    # Create tables, seed if empty...
    return SqliteDataStore(
        products=SqliteProductRepository(db),
        orders=SqliteOrderRepository(db),
        users=SqliteUserRepository(db),
    )

Cosmos DB Implementation

# data/cosmos.py

from azure.cosmos import CosmosClient
import os

class CosmosProductRepository:
    def __init__(self, container):
        self.container = container

    def get_all(self, *, page=1, page_size=20, category=None, min_price=None, max_price=None):
        query = "SELECT * FROM c WHERE c.status = @status"
        params = [{"name": "@status", "value": "active"}]
        if category:
            query += " AND c.category = @cat"
            params.append({"name": "@cat", "value": category})
        # ... build query
        items = list(self.container.query_items(query=query, parameters=params, enable_cross_partition_query=True))
        start = (page - 1) * page_size
        return {"data": items[start:start+page_size], "total_count": len(items)}

def create_cosmos_store():
    client = CosmosClient(os.environ["COSMOS_ENDPOINT"], os.environ["COSMOS_KEY"])
    db = client.get_database_client(os.getenv("COSMOS_DATABASE", "aimarket"))
    return CosmosDataStore(
        products=CosmosProductRepository(db.get_container_client("products")),
        orders=CosmosOrderRepository(db.get_container_client("orders")),
        users=CosmosUserRepository(db.get_container_client("users")),
    )

FastAPI Routes

# routes/products.py — depends only on DataStore protocol

from fastapi import APIRouter, Depends
from data.store import create_store

router = APIRouter(prefix="/api/products")

@router.get("/")
def list_products(page: int = 1, page_size: int = 20, category: str | None = None,
                  store: DataStore = Depends(create_store)):
    result = store.products.get_all(page=page, page_size=page_size, category=category)
    return {**result, "page": page, "page_size": page_size}

.NET / C# (Minimal APIs or Controllers)

Repository Interfaces

// Data/Interfaces.cs

public interface IProductRepository
{
    Task<(List<Product> Data, int TotalCount)> GetAllAsync(
        int page = 1, int pageSize = 20,
        string? category = null, decimal? minPrice = null, decimal? maxPrice = null);
    Task<Product?> GetByIdAsync(string id);
    Task<Product> CreateAsync(CreateProductInput input);
    Task<Product?> UpdateAsync(string id, UpdateProductInput input);
}

public interface IOrderRepository
{
    Task<Order> CreateAsync(CreateOrderInput input);
    Task<Order?> GetByIdAsync(string id);
    Task<(List<Order> Data, int TotalCount)> GetByUserIdAsync(string userId, int page, int pageSize);
}

public interface IUserRepository
{
    Task<User> CreateAsync(CreateUserInput input);
    Task<User?> GetByIdAsync(string id);
    Task<User?> GetByEmailAsync(string email);
}

Dependency Injection (Factory)

// Program.cs

var provider = builder.Configuration["DATA_PROVIDER"] ?? "sqlite";

if (provider == "sqlite")
{
    builder.Services.AddSingleton<IProductRepository, SqliteProductRepository>();
    builder.Services.AddSingleton<IOrderRepository, SqliteOrderRepository>();
    builder.Services.AddSingleton<IUserRepository, SqliteUserRepository>();
}
else if (provider == "cosmos")
{
    builder.Services.AddSingleton<IProductRepository, CosmosProductRepository>();
    builder.Services.AddSingleton<IOrderRepository, CosmosOrderRepository>();
    builder.Services.AddSingleton<IUserRepository, CosmosUserRepository>();
}
else if (provider == "postgres")
{
    builder.Services.AddSingleton<IProductRepository, PostgresProductRepository>();
    builder.Services.AddSingleton<IOrderRepository, PostgresOrderRepository>();
    builder.Services.AddSingleton<IUserRepository, PostgresUserRepository>();
}

.NET uses built-in DI instead of a manual factory. The pattern is the same: register implementations based on config, inject interfaces into routes.

SQLite Implementation

// Data/SqliteProductRepository.cs

using Microsoft.Data.Sqlite;
using System.Text.Json;

public class SqliteProductRepository : IProductRepository
{
    private readonly SqliteConnection _db;

    public SqliteProductRepository(IConfiguration config)
    {
        _db = new SqliteConnection(config.GetConnectionString("Sqlite") ?? "Data Source=app.db");
        _db.Open();
        // Create tables, seed if empty...
    }

    public async Task<(List<Product> Data, int TotalCount)> GetAllAsync(
        int page = 1, int pageSize = 20,
        string? category = null, decimal? minPrice = null, decimal? maxPrice = null)
    {
        var where = "WHERE status = 'active'";
        if (category != null) where += $" AND category = @category";
        if (minPrice != null) where += $" AND price >= @minPrice";
        if (maxPrice != null) where += $" AND price <= @maxPrice";

        // Execute count + select queries...
        // Parse tags from JSON string: JsonSerializer.Deserialize<string[]>(row["tags"])
    }
}

Cosmos DB Implementation

// Data/CosmosProductRepository.cs

using Microsoft.Azure.Cosmos;

public class CosmosProductRepository : IProductRepository
{
    private readonly Container _container;

    public CosmosProductRepository(CosmosClient client, IConfiguration config)
    {
        var db = client.GetDatabase(config["COSMOS_DATABASE"] ?? "aimarket");
        _container = db.GetContainer("products");
    }

    public async Task<(List<Product> Data, int TotalCount)> GetAllAsync(
        int page = 1, int pageSize = 20,
        string? category = null, decimal? minPrice = null, decimal? maxPrice = null)
    {
        var query = new QueryDefinition("SELECT * FROM c WHERE c.status = @status")
            .WithParameter("@status", "active");
        // Add filters, execute, paginate...
    }
}

Minimal API Routes

// Routes depend only on interfaces — injected via DI

app.MapGet("/api/products", async (IProductRepository repo,
    int page = 1, int pageSize = 20, string? category = null) =>
{
    var (data, total) = await repo.GetAllAsync(page, pageSize, category);
    return Results.Ok(new { data, page, pageSize, totalCount = total });
});

Java / Spring Boot

Repository Interfaces

// data/ProductRepository.java

public interface ProductRepository {
    PaginatedResult<Product> getAll(int page, int pageSize,
        String category, Double minPrice, Double maxPrice);
    Optional<Product> getById(String id);
    Product create(CreateProductInput input);
    Optional<Product> update(String id, UpdateProductInput input);
}

// data/OrderRepository.java

public interface OrderRepository {
    Order create(CreateOrderInput input);
    Optional<Order> getById(String id);
    PaginatedResult<Order> getByUserId(String userId, int page, int pageSize);
}

// data/UserRepository.java

public interface UserRepository {
    User create(CreateUserInput input);
    Optional<User> getById(String id);
    Optional<User> getByEmail(String email);
}

Factory via Spring Profiles

Spring profiles replace the manual factory. Set DATA_PROVIDER as the active profile:

# application.yml
spring:
  profiles:
    active: ${DATA_PROVIDER:sqlite}

// data/sqlite/SqliteProductRepository.java

@Repository
@Profile("sqlite")
public class SqliteProductRepository implements ProductRepository {
    private final JdbcTemplate jdbc;

    public SqliteProductRepository(DataSource dataSource) {
        this.jdbc = new JdbcTemplate(dataSource);
    }

    @Override
    public PaginatedResult<Product> getAll(int page, int pageSize,
            String category, Double minPrice, Double maxPrice) {
        var where = new StringBuilder("WHERE status = 'active'");
        var params = new ArrayList<>();

        if (category != null) { where.append(" AND category = ?"); params.add(category); }
        if (minPrice != null) { where.append(" AND price >= ?"); params.add(minPrice); }
        if (maxPrice != null) { where.append(" AND price <= ?"); params.add(maxPrice); }

        int total = jdbc.queryForObject(
            "SELECT COUNT(*) FROM products " + where, Integer.class, params.toArray());
        var rows = jdbc.query(
            "SELECT * FROM products " + where + " LIMIT ? OFFSET ?",
            this::mapRow,
            Stream.concat(params.stream(), Stream.of(pageSize, (page - 1) * pageSize)).toArray());

        return new PaginatedResult<>(rows, total);
    }

    private Product mapRow(ResultSet rs, int rowNum) throws SQLException {
        return new Product(
            rs.getString("id"),
            rs.getString("name"),
            // ...
            objectMapper.readValue(rs.getString("tags"), new TypeReference<List<String>>() {}),  // JSON → List
            // ...
        );
    }
}

// data/cosmos/CosmosProductRepository.java

@Repository
@Profile("cosmos")
public class CosmosProductRepository implements ProductRepository {
    private final CosmosContainer container;

    public CosmosProductRepository(CosmosClient client,
            @Value("${cosmos.database:aimarket}") String dbName) {
        this.container = client.getDatabase(dbName).getContainer("products");
    }

    @Override
    public PaginatedResult<Product> getAll(int page, int pageSize,
            String category, Double minPrice, Double maxPrice) {
        var query = new StringBuilder("SELECT * FROM c WHERE c.status = @status");
        var params = new ArrayList<SqlParameter>();
        params.add(new SqlParameter("@status", "active"));

        if (category != null) {
            query.append(" AND c.category = @cat");
            params.add(new SqlParameter("@cat", category));
        }
        // ... build query

        var options = new CosmosQueryRequestOptions();
        var items = container.queryItems(
            new SqlQuerySpec(query.toString(), params), options, Product.class);

        var all = items.stream().collect(Collectors.toList());
        int start = (page - 1) * pageSize;
        return new PaginatedResult<>(
            all.subList(start, Math.min(start + pageSize, all.size())), all.size());
    }
}

SQLite DataSource Config

// config/SqliteConfig.java

@Configuration
@Profile("sqlite")
public class SqliteConfig {
    @Bean
    public DataSource dataSource() {
        var ds = new SQLiteDataSource();
        ds.setUrl("jdbc:sqlite:app.db");
        return ds;
    }

    @Bean
    public CommandLineRunner initDb(DataSource dataSource) {
        return args -> {
            var jdbc = new JdbcTemplate(dataSource);
            jdbc.execute("PRAGMA journal_mode=WAL");
            jdbc.execute("PRAGMA foreign_keys=ON");
            // Create tables, seed if empty...
        };
    }
}

Controller Routes

// controller/ProductController.java — depends only on ProductRepository interface

@RestController
@RequestMapping("/api/products")
public class ProductController {
    private final ProductRepository repo;

    public ProductController(ProductRepository repo) {  // Spring injects the active profile's impl
        this.repo = repo;
    }

    @GetMapping
    public ResponseEntity<?> list(
            @RequestParam(defaultValue = "1") int page,
            @RequestParam(defaultValue = "20") int pageSize,
            @RequestParam(required = false) String category) {
        var result = repo.getAll(page, pageSize, category, null, null);
        return ResponseEntity.ok(Map.of(
            "data", result.data(), "page", page, "pageSize", pageSize,
            "totalCount", result.totalCount(),
            "totalPages", (int) Math.ceil((double) result.totalCount() / pageSize)
        ));
    }
}

Key: Spring's @Profile + DI eliminates the manual factory. Set DATA_PROVIDER=postgres or DATA_PROVIDER=cosmos as the active profile and Spring wires the matching implementations automatically. Add @Profile("postgres") to a PostgresProductRepository that uses JdbcTemplate with a PostgreSQL DataSource — same JdbcTemplate API as SQLite, different driver and SQL dialect.

SQLite Patterns (All Languages)

SQLite doesn't support arrays or nested objects. These patterns apply regardless of language:

Pragmas to always set:

PRAGMA journal_mode = WAL;    -- Better concurrent read performance
PRAGMA foreign_keys = ON;     -- Enforce referential integrity

Cosmos DB Patterns (All Languages)

Partition key strategy: Choose a field that distributes data evenly. Common choices: /id (default), /category (products), /userId (orders).

PostgreSQL Patterns (All Languages)

PostgreSQL supports native arrays and JSONB, so it handles complex types better than SQLite but uses SQL instead of Cosmos DB's document model.

Azure PostgreSQL requirements:

# Connection string must include sslmode
POSTGRES_CONNECTION_STRING=postgresql://user:pass@host.postgres.database.azure.com:5432/aimarket?sslmode=require

# For Azure Flexible Server, also set
POSTGRES_SSL=true

Libraries by language:

Anti-Patterns

Checklist

• Repository interfaces defined per entity (typed, no database-specific imports)
• Factory/DI reads DATA_PROVIDER and returns the correct implementation
• Routes depend only on interfaces, never on implementation files
• SQLite implementation handles JSON serialization for arrays and objects
• PostgreSQL implementation uses native TEXT[] for arrays and JSONB for objects
• Cosmos DB implementation uses native JSON types (no serialization)
• DATA_PROVIDER=sqlite works locally
• DATA_PROVIDER=postgres works with Azure PostgreSQL (SSL enabled)
• DATA_PROVIDER=cosmos works with Azure Cosmos DB
• .env.example documents all providers and their required variables
• An in-memory implementation exists for unit tests