postgres-memory-server

Spin up a disposable real PostgreSQL or ParadeDB instance in tests with a tiny API inspired by mongodb-memory-server and redisjson-memory-server.

This package does not emulate Postgres. It starts an actual PostgreSQL process using native binaries — no Docker required.

Why this exists

For Postgres extension-heavy workloads, especially ParadeDB + pgvector, a real database is usually what you want in tests:

• no shared local database contamination
• deterministic integration tests
• realistic extension behavior
• straightforward CI/CD setup — no Docker daemon needed
• one API for plain Postgres and ParadeDB

Features

• Start a disposable Postgres instance with create() — no Docker
• PostgreSQL binaries bundled via embedded-postgres
• ParadeDB pg_search and pgvector extensions auto-downloaded from official releases
• Get a ready-to-use connection string with getUri()
• Run SQL strings, SQL files, or a migrations directory
• Snapshot and restore database state between tests
• Extension binaries cached in ~/.cache/postgres-memory-server/
• CLI for local scripts and debugging

Requirements

• Node.js 20+

That's it. No Docker, no system Postgres installation needed.

Install

npm install -D postgres-memory-server

The pg client library is included as a dependency.

Quick start

import { PostgresMemoryServer } from "postgres-memory-server";
import { Client } from "pg";

const db = await PostgresMemoryServer.create();

const client = new Client({ connectionString: db.getUri() });
await client.connect();
await client.query("select 1");
await client.end();

await db.stop();

ParadeDB quick start

import { PostgresMemoryServer } from "postgres-memory-server";

const db = await PostgresMemoryServer.createParadeDb();

await db.runSql(`
  CREATE TABLE documents (
    id bigserial primary key,
    title text not null,
    content text not null,
    embedding vector(3)
  );
`);

await db.stop();

By default, the ParadeDB preset automatically downloads and installs the extensions, then runs:

CREATE EXTENSION IF NOT EXISTS pg_search;
CREATE EXTENSION IF NOT EXISTS vector;

Extension binaries are downloaded once and cached in ~/.cache/postgres-memory-server/.

• pg_search is downloaded from ParadeDB GitHub releases
• pgvector is downloaded from Homebrew bottles (covers macOS + Linux)

How it works

Instead of Docker containers, this package uses:

1. embedded-postgres — bundles PostgreSQL binaries as npm packages (~10MB). The PostgreSQL version is determined by the installed embedded-postgres npm package version (e.g., embedded-postgres@18.x = PostgreSQL 18).

2. Native extension installation — for the ParadeDB preset, pg_search and pgvector extension binaries are downloaded from their official release channels, extracted, and installed into the embedded PostgreSQL directory.

3. Template database snapshots — snapshot() and restore() use PostgreSQL's native CREATE DATABASE ... TEMPLATE for fast, zero-copy test isolation.

PostgreSQL version

The PostgreSQL version is tied to the embedded-postgres npm package version. To use a specific PG version:

# PG 18 (default with latest embedded-postgres)
npm install -D embedded-postgres

# PG 17
npm install -D embedded-postgres@17.9.0-beta.16

# PG 16
npm install -D embedded-postgres@16.8.0-beta.16

Platform support

API

PostgresMemoryServer.create(options?)

Create a disposable Postgres instance.

const db = await PostgresMemoryServer.create({
  database: "testdb",
  username: "testuser",
  password: "testpassword",
});

PostgresMemoryServer.createPostgres(options?)

Convenience alias for the plain Postgres preset.

const db = await PostgresMemoryServer.createPostgres();

PostgresMemoryServer.createParadeDb(options?)

Starts a Postgres instance with ParadeDB extensions (pg_search + pgvector).

const db = await PostgresMemoryServer.createParadeDb();

Options

createJestGlobalSetup(options?)

Starts one disposable database process for a Jest run and injects its URI into DATABASE_URL by default.

createJestGlobalTeardown(options?)

Stops the process started by createJestGlobalSetup().

Instance methods

• getUri() — connection string (postgres://...)
• getHost() — always localhost
• getPort() — randomly assigned free port
• getDatabase()
• getUsername()
• getPassword()
• getConnectionOptions() — { host, port, database, user, password }
• getImage() — descriptive label (e.g., "postgres:18")
• query(text, params?) — execute a query and return rows
• withClient(callback) — direct access to a pg.Client
• runSql(sql) — execute one or more SQL statements
• runSqlFile(filePath) — execute SQL from a file
• runMigrationsDir(dirPath) — run .sql files in lexicographic order
• snapshot() — create a restore point
• restore() — restore to last snapshot
• stop() — shut down the PostgreSQL process

Snapshots

const db = await PostgresMemoryServer.create();

await db.runSql([
  "create table users (id serial primary key, email text not null)",
  "insert into users (email) values ('first@example.com')",
]);

await db.snapshot();

await db.runSql("insert into users (email) values ('second@example.com')");

await db.restore();

const result = await db.query<{ count: string }>(
  "select count(*)::text as count from users",
);
console.log(result.rows[0]?.count); // 1

Snapshots use PostgreSQL template databases under the hood. Use a non-system database name (the default testdb works).

Running SQL files or migrations

await db.runSqlFile("./sql/001_init.sql");
await db.runMigrationsDir("./sql/migrations");

Migration files are run in lexicographic order.

CLI

npx postgres-memory-server --preset paradedb

Example output:

POSTGRES_MEMORY_SERVER_URI=postgres://testuser:testpassword@localhost:54329/testdb
POSTGRES_MEMORY_SERVER_HOST=localhost
POSTGRES_MEMORY_SERVER_PORT=54329
POSTGRES_MEMORY_SERVER_DATABASE=testdb
POSTGRES_MEMORY_SERVER_USERNAME=testuser
POSTGRES_MEMORY_SERVER_PASSWORD=testpassword

Press Ctrl+C to stop the server.

CLI flags

--preset postgres|paradedb
--version <tag>
--database <name>
--username <name>
--password <password>
--extension <name>      # repeatable
--init-file <path>      # repeatable
--json

Jest global setup

// jest.global-setup.ts
import { createJestGlobalSetup } from "postgres-memory-server";

export default createJestGlobalSetup({
  preset: "paradedb",
});

// jest.global-teardown.ts
import { createJestGlobalTeardown } from "postgres-memory-server";

export default createJestGlobalTeardown();

// jest.config.ts
import type { Config } from "jest";

const config: Config = {
  globalSetup: "./jest.global-setup.ts",
  globalTeardown: "./jest.global-teardown.ts",
  testEnvironment: "node",
};

export default config;

After setup runs, your tests can connect through process.env.DATABASE_URL.

Vitest example

import { afterAll, beforeAll, beforeEach, describe, expect, it } from "vitest";
import { PostgresMemoryServer } from "postgres-memory-server";

describe("db", () => {
  let db: PostgresMemoryServer;

  beforeAll(async () => {
    db = await PostgresMemoryServer.create();
    await db.runSql(`
      create table notes (
        id serial primary key,
        body text not null
      );
    `);
    await db.snapshot();
  });

  beforeEach(async () => {
    await db.restore();
  });

  afterAll(async () => {
    await db.stop();
  });

  it("starts from a clean snapshot", async () => {
    await db.runSql("insert into notes (body) values ('hello')");
    const result = await db.query<{ count: string }>(
      "select count(*)::text as count from notes",
    );
    expect(result.rows[0]?.count).toBe("1");
  });
});

Migrating from v0.1.0 (Docker-based)

v0.2.0 replaces Docker containers with native binaries. Key changes:

Breaking changes:

• The image option is deprecated and ignored. Remove it from your code.
• The version option for the postgres preset is ignored — install the desired embedded-postgres@<version> package instead.
• The version option for the paradedb preset now refers to the ParadeDB extension version (e.g., "0.22.5"), not the Docker image tag.

No changes needed if you only used create(), createPostgres(), or createParadeDb() with just database, username, password, or extensions options.

Caveats

• The PostgreSQL version is determined by the installed embedded-postgres npm package, not by a runtime option.
• ParadeDB pg_search requires macOS arm64 or Linux. Intel Macs and Windows are not supported for ParadeDB.
• Snapshot and restore terminate other connections to the database during the operation.
• The package is ESM-only.

Test scripts

npm run test:postgres
npm run test:paradedb

License

MIT