About Connection pooling

Neon uses PgBouncer to support connection pooling, enabling up to 10,000 concurrent connections. PgBouncer is a lightweight connection pooler for Postgres.

How to use connection pooling

To use connection pooling with Neon, use a pooled connection string instead of a regular connection string. A pooled connection string adds the -pooler option to your endpoint ID, as shown below:

postgresql://alex:AbC123dEf@ep-cool-darkness-123456-pooler.us-east-2.aws.neon.tech/dbname?sslmode=require

The Connect to your database modal, which you can access by clicking the Connect button on your Project Dashboard, provides Connection pooling toggle that adds the -pooler option to a connection string for you. You can copy a pooled connection string from the Dashboard or manually add the -pooler option to the endpoint ID in an existing connection string.

Connection limits without connection pooling

Each Postgres connection creates a new process in the operating system, which consumes resources. Postgres limits the number of open connections for this reason. The Postgres connection limit is defined by the Postgres max_connections parameter. In Neon, max_connections is set according to your compute size or autoscaling configuration — you can find the formula here: Parameter settings that differ by compute size.

You can check the max_connections limit for your compute by running the following query from the Neon SQL Editor or a client connected to Neon:

SHOW max_connections;

SELECT usename FROM pg_stat_activity WHERE datname = '<database_name>';

Even with the largest compute size, the max_connections limit may not be sufficient for some applications, such as those that use serverless functions. To increase the number of connections that Neon supports, you can use connection pooling. All Neon plans, including the Neon Free Plan, support connection pooling.

Connection pooling

Some applications open numerous connections, with most eventually becoming inactive. This behavior can often be attributed to database driver limitations, running many instances of an application, or applications with serverless functions. With regular Postgres, new connections are rejected when reaching the max_connections limit. To overcome this limitation, Neon supports connection pooling using PgBouncer, which allows Neon to support up to 10,000 concurrent connections.

Connection pooling, however, is not a magic bullet: As the name implies, connections share a pool of connections to Postgres — a pool of connections sitting in front of a limited number of direct connections to Postgres.

To ensure that direct access to Postgres is still possible for administrative tasks or similar, the pooler is configured to only open up a certain number of direct Postgres connections for each user to each database. This number of direct Postgres connections is determined by the PgBouncer default_pool_size setting, which is in turn determined by your compute's max_connections setting. For example, if default_pool_size is 100, there can be only 100 active connections from role alex to any particular database through the pooler. All other connections by alex to that database will have to wait for one of those 100 active connections to complete their transactions before the next connection's work is started.

At the same time, role dana will also be able to connect to the same database database through the pooler and have up to 100 concurrent active transactions across the same number of connections.

Similarly, even if role alex has 100 concurrently active transactions through the pooler to the same database database, that role can still start up to 100 concurrent transactions to a different database when connected through the pooler.

The max_connections setting still applies for direct Postgres connections.

PgBouncer

PgBouncer is an open-source connection pooler for Postgres. When an application needs to connect to a database, PgBouncer provides a connection from the pool. Connections in the pool are routed to a smaller number of actual Postgres connections. When a connection is no longer required, it is returned to the pool and is available to be used again. Maintaining a pool of available connections improves performance by reducing the number of connections that need to be created and torn down to service incoming requests. Connection pooling also helps avoid rejected connections. When all connections in the pool are being used, PgBouncer queues a new request until a connection from the pool becomes available.

Neon PgBouncer configuration settings

Neon's PgBouncer configuration is shown below. The settings are not user-configurable, but if you are a paid plan user and require a different setting, please contact Neon Support. For example, Neon sometimes raises the default_pool_size setting for users who support a large number of concurrent connections and repeatedly hit PgBouncer's pool size limit.

[pgbouncer]
pool_mode=transaction
max_client_conn=10000
default_pool_size=0.9 * max_connections
max_prepared_statements=0
query_wait_timeout=120

where max_connections is a Postgres setting.

The following list describes each setting. For a full explanation of each parameter, please refer to the official PgBouncer documentation.

• pool_mode=transaction: The pooling mode PgBouncer uses, set to transaction pooling.
• max_client_conn=10000: Maximum number of client connections allowed.
• default_pool_size: Default number of server connections to allow per user/database pair. The formula is 0.9 * max_connections. For max_connections details, see Parameter settings.
• max_prepared_statements=0: Maximum number of prepared statements a connection is allowed to have at the same time. 0 means prepared statements are disabled.
• query_wait_timeout=120: Maximum time queries are allowed to spend waiting for execution. Neon uses the default setting of 120 seconds.

Connection pooling in transaction mode

As mentioned above, Neon uses PgBouncer in transaction mode (pool_mode=transaction), which limits some functionality in Postgres. Functionality NOT supported in transaction mode includes:

• SET/RESET
• LISTEN
• WITH HOLD CURSOR
• PREPARE / DEALLOCATE
• PRESERVE / DELETE ROWS temp tables
• LOAD statement
• Session-level advisory locks

These session-level features are not supported transaction mode because:

1. In this mode, database connections are allocated from the pool on a per-transaction basis
2. Session states are not persisted across transactions

For the official list of limitations, refer to the "SQL feature map for pooling modes" section in the pgbouncer.org Features documentation.

Connection pooling with schema migration tools

We recommend using a direct (non-pooled) connection string when performing migrations using Object Relational Mappers (ORMs) and similar schema migration tools. With the exception of recent versions of Prisma ORM, which support using a pooled connection string with Neon, using a pooled connection string for migrations is likely not supported or prone to errors. Before attempting to perform migrations over a pooled connection string, please refer to your tool's documentation to determine if pooled connections are supported.

Optimize queries with PgBouncer and prepared statements

Protocol-level prepared statements are supported with Neon and PgBouncer as of the PgBouncer 1.22.0 release. Using prepared statements can help boost query performance while providing an added layer of protection against potential SQL injection attacks.

Understanding prepared statements

A prepared statement in Postgres allows for the optimization of an SQL query by defining its structure once and executing it multiple times with varied parameters. Here's an SQL-level example to illustrate. Note that direct SQL-level PREPARE and EXECUTE are not supported with PgBouncer (see below), so you can't use this query from the SQL Editor. It is meant to give you a clear idea of how a prepared statement works. Refer to the protocol-level samples below to see how this SQL-level example translates to different protocol-level examples.

PREPARE fetch_plan (TEXT) AS
SELECT * FROM users WHERE username = $1;

EXECUTE fetch_plan('alice');

fetch_plan here is the prepared statement's name, and $1 acts as a parameter placeholder.

The benefits of using prepared statements include:

• Performance: Parsing the SQL and creating the execution plan happens just once, speeding up subsequent executions. This performance benefit would be most noticeable on databases with heavy and repeated traffic.
• Security: By sending data values separately from the query, prepared statements reduce the risk of SQL injection attacks.

You can learn more about prepared statements in the PostgreSQL documentation. See PREPARE.

Use prepared statements with PgBouncer

Since pgBouncer supports protocol-level prepared statements only, you must rely on PostgreSQL client libraries instead (direct SQL-level PREPARE and EXECUTE are not supported). Fortunately, most PostgreSQL client libraries support prepared statements. Here are a couple of examples showing how to use prepared statements with Javascript and Python client libraries:

const query = {
  // give the query a unique name
  name: 'fetch-plan',
  text: 'SELECT * FROM users WHERE username = $1',
  values: ['alice'],
};
client.query(query);