Neon serverless driver

The Neon serverless driver is a low-latency Postgres driver for JavaScript and TypeScript that allows you to query data from serverless and edge environments over HTTP or WebSockets in place of TCP. The driver's low-latency capability is due to message pipelining and other optimizations.

When to query over HTTP vs WebSockets:

HTTP: Querying over an HTTP fetch request is faster for single, non-interactive transactions, also referred to as "one-shot queries". Issuing multiple queries via a single, non-interactive transaction is also supported. See Use the driver over HTTP.
WebSockets: If you require session or interactive transaction support or compatibility with node-postgres (the popular npm pg package), use WebSockets. See Use the driver over WebSockets.

Install the Neon serverless driver

You can install the driver with your preferred JavaScript package manager. For example:

npm install @neondatabase/serverless

The driver includes TypeScript types (the equivalent of @types/pg). No additional installation is required.

note

The Neon serverless driver is also available as a JavaScript Registry (JSR) package: https://jsr.io/@neon/serverless. The JavaScript Registry (JSR) is a package registry for JavaScript and TypeScript. JSR works with many runtimes (Node.js, Deno, browsers, and more) and is backward compatible with npm.

Configure your Neon database connection

You can obtain a connection string for your database from the Connection Details widget on the Neon Dashboard. Your Neon connection string will look something like this:

DATABASE_URL=postgresql://[user]:[password]@[neon_hostname]/[dbname]

The examples that follow assume that your database connection string is assigned to a DATABASE_URL variable in your application's environment file.

Use the driver over HTTP

The Neon serverless driver uses the neon function for queries over HTTP.

You can use raw SQL queries or tools such as Drizzle-ORM, kysely, Zapatos, and others for type safety.

Node.js

Drizzle-ORM

Vercel Edge Function

Vercel Serverless Function

import { neon } from '@neondatabase/serverless';

const sql = neon(process.env.DATABASE_URL);
const posts = await sql('SELECT * FROM posts WHERE id = $1', [postId]);
// `post` is now [{ id: 12, title: 'My post', ... }] (or undefined)

note

The maximum request size and response size for queries over HTTP is 10 MB.

neon function configuration options

The neon(...) function returns a query function that can be used both as a tagged-template function and as an ordinary function:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL);

// as a tagged-template function
const rowsA = await sql`SELECT * FROM posts WHERE id = ${postId}`;

// as an ordinary function (exactly equivalent)
const rowsB = await sql('SELECT * FROM posts WHERE id = $1', [postId]);

By default, the query function returned by neon(...) returns only the rows resulting from the provided SQL query, and it returns them as an array of objects where the keys are column names. For example:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL);
const rows = await sql`SELECT * FROM posts WHERE id = ${postId}`;
// -> [{ id: 12, title: "My post", ... }]

However, you can customize the return format of the query function using the configuration options fullResults and arrayMode. These options are available both on the neon(...) function and on the query function it returns (but only when the query function is called as an ordinary function, not as a tagged-template function).

arrayMode: boolean, false by default

The default arrayMode value is false. When it is true, rows are returned as an array of arrays instead of an array of objects:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL, { arrayMode: true });
const rows = await sql`SELECT * FROM posts WHERE id = ${postId}`;
// -> [[12, "My post", ...]]

Or, with the same effect:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL);
const rows = await sql('SELECT * FROM posts WHERE id = $1', [postId], { arrayMode: true });
// -> [[12, "My post", ...]]

fullResults: boolean

The default fullResults value is false. When it is true, additional metadata is returned alongside the result rows, which are then found in the rows property of the return value. The metadata matches what would be returned by node-postgres:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL, { fullResults: true });
const results = await sql`SELECT * FROM posts WHERE id = ${postId}`;
/* -> {
  rows: [{ id: 12, title: "My post", ... }],
  fields: [
    { name: "id", dataTypeID: 23, ... },
    { name: "title", dataTypeID: 25, ... },
    ...
  ],
  rowCount: 1,
  rowAsArray: false,
  command: "SELECT"
} 
*/

Or, with the same effect:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL);
const results = await sql('SELECT * FROM posts WHERE id = $1', [postId], { fullResults: true });
// -> { ... same as above ... }

fetchOptions: Record<string, any>

The fetchOptions option can also be passed to either neon(...) or the query function. This option takes an object that is merged with the options to the fetch call.

For example, to increase the priority of every database fetch request:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL, { fetchOptions: { priority: 'high' } });
const rows = await sql`SELECT * FROM posts WHERE id = ${postId}`;

Or to implement a fetch timeout:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL);
const abortController = new AbortController();
const timeout = setTimeout(() => abortController.abort('timed out'), 10000);
const rows = await sql('SELECT * FROM posts WHERE id = $1', [postId], {
  fetchOptions: { signal: abortController.signal },
}); // throws an error if no result received within 10s
clearTimeout(timeout);

For additional details, see Options and configuration.

Issue multiple queries with the transaction() function

The transaction(queriesOrFn, options) function is exposed as a property on the query function. It allows multiple queries to be executed within a single, non-interactive transaction.

The first argument to transaction(), queriesOrFn, is either an array of queries or a non-async function that receives a query function as its argument and returns an array of queries.

The array-of-queries case looks like this:

import { neon } from '@neondatabase/serverless';
const sql = neon(process.env.DATABASE_URL);
const showLatestN = 10;

const [posts, tags] = await sql.transaction(
  [sql`SELECT * FROM posts ORDER BY posted_at DESC LIMIT ${showLatestN}`, sql`SELECT * FROM tags`],
  {
    isolationLevel: 'RepeatableRead',
    readOnly: true,
  }
);

Or as an example of the function case:

const [authors, tags] = await neon(process.env.DATABASE_URL).transaction((txn) => [
  txn`SELECT * FROM authors`,
  txn`SELECT * FROM tags`,
]);

The optional second argument to transaction(), options, has the same keys as the options to the ordinary query function -- arrayMode, fullResults and fetchOptions — plus three additional keys that concern the transaction configuration. These transaction-related keys are: isolationMode, readOnly and deferrable.

Note that options cannot be supplied for individual queries within a transaction. Query and transaction options must instead be passed as the second argument of the transaction() function. For example, this arrayMode setting is ineffective (and TypeScript won't compile it): await sql.transaction([sql('SELECT now()', [], { arrayMode: true })]). Instead, use await sql.transaction([sql('SELECT now()')], { arrayMode: true }).

isolationMode

This option selects a Postgres transaction isolation mode. If present, it must be one of ReadUncommitted, ReadCommitted, RepeatableRead, or Serializable.
readOnly

If true, this option ensures that a READ ONLY transaction is used to execute the queries passed. This is a boolean option. The default value is false.
deferrable

If true (and if readOnly is also true, and isolationMode is Serializable), this option ensures that a DEFERRABLE transaction is used to execute the queries passed. This is a boolean option. The default value is false.

For additional details, see transaction(...) function.

Advanced configuration options

For advanced configuration options, see neonConfig configuration, in the Neon serverless driver GitHub readme.

Use the driver over WebSockets

The Neon serverless driver supports the Pool and Client constructors for querying over WebSockets.

The Pool and Client constructors, provide session and transaction support, as well as node-postgres compatibility. You can find the API guide for the Pool and Client constructors in the node-postgres documentation.

Consider using the driver with Pool or Client in the following scenarios:

You already use node-postgres in your code base and would like to migrate to using @neondatabase/serverless.
You are writing a new code base and want to use a package that expects a node-postgres-compatible driver.
Your backend service uses sessions / interactive transactions with multiple queries per connection.

You can use the Neon serverless driver in the same way you would use node-postgres with Pool and Client. Where you usually import pg, import @neondatabase/serverless instead.

Node.js

Prisma

Drizzle-ORM

Vercel Edge Function

Vercel Serverless Function

import { Pool } from '@neondatabase/serverless';

const pool = new Pool({ connectionString: process.env.DATABASE_URL });
const posts = await pool.query('SELECT * FROM posts WHERE id =$1', [postId]);
pool.end();

Pool and Client usage notes

In Node.js and some other environments, there's no built-in WebSocket support. In these cases, supply a WebSocket constructor function.
```
import { Pool, neonConfig } from '@neondatabase/serverless';
import ws from 'ws';
neonConfig.webSocketConstructor = ws;
```
In serverless environments such as Vercel Edge Functions or Cloudflare Workers, WebSocket connections can't outlive a single request. That means Pool or Client objects must be connected, used and closed within a single request handler. Don't create them outside a request handler; don't create them in one handler and try to reuse them in another; and to avoid exhausting available connections, don't forget to close them.

For examples that demonstrate these points, see Pool and Client.

Advanced configuration options

For advanced configuration options, see neonConfig configuration, in the Neon serverless driver GitHub readme.

Example applications

Explore the example applications that use the Neon serverless driver.

UNESCO World Heritage sites app

Neon provides an example application to help you get started with the Neon serverless driver. The application generates a JSON listing of the 10 nearest UNESCO World Heritage sites using IP geolocation (data copyright © 1992 – 2022 UNESCO/World Heritage Centre).

UNESCO World Heritage sites app

There are different implementations of the application to choose from.

Ping Thing

The Ping Thing application pings a Neon Serverless Postgres database using a Vercel Edge Function and shows the journey your request makes. You can read more about this application in the accompanying blog post: How to use Postgres at the Edge

Ping Thing
Ping a Neon Serverless Postgres database using a Vercel Edge Function to see the journey your request makes