psycopg_pool – Connection pool implementations#

A connection pool is an object used to create and maintain a limited amount of PostgreSQL connections, reducing the time requested by the program to obtain a working connection and allowing an arbitrary large number of concurrent threads or tasks to use a controlled amount of resources on the server. See Connection pools for more details and usage pattern.

This package exposes a few connection pool classes:


The psycopg_pool package is distributed separately from the main psycopg package: use pip install "psycopg[pool]", or pip install psycopg_pool, to make it available. See Installing the connection pool.

The version numbers indicated in this page refer to the psycopg_pool package, not to psycopg.

The ConnectionPool class#

class psycopg_pool.ConnectionPool(conninfo='', *, open=True, connection_class=<class 'psycopg.Connection'>, configure=None, reset=None, **kwargs)#

This class implements a connection pool serving Connection instances (or subclasses). The constructor has alot of arguments, but only conninfo and min_size are the fundamental ones, all the other arguments have meaningful defaults and can probably be tweaked later, if required.

  • conninfo (str) – The connection string. See connect() for details.

  • min_size (int, default: 4) – The minimum number of connection the pool will hold. The pool will actively try to create new connections if some are lost (closed, broken) and will try to never go below min_size.

  • max_size (int, default: None) – The maximum number of connections the pool will hold. If None, or equal to min_size, the pool will not grow or shrink. If larger than min_size, the pool can grow if more than min_size connections are requested at the same time and will shrink back after the extra connections have been unused for more than max_idle seconds.

  • kwargs (dict) – Extra arguments to pass to connect(). Note that this is one dict argument of the pool constructor, which is expanded as connect() keyword parameters.

  • connection_class (type, default: Connection) – The class of the connections to serve. It should be a Connection subclass.

  • open (bool, default: True) – If True, open the pool, creating the required connections, on init. If False, open the pool when open() is called or when the pool context is entered. See the open() method documentation for more details.

  • configure (Callable[[Connection], None]) – A callback to configure a connection after creation. Useful, for instance, to configure its adapters. If the connection is used to run internal queries (to inspect the database) make sure to close an eventual transaction before leaving the function.

  • reset (Callable[[Connection], None]) – A callback to reset a function after it has been returned to the pool. The connection is guaranteed to be passed to the reset() function in “idle” state (no transaction). When leaving the reset() function the connection must be left in idle state, otherwise it is discarded.

  • name (str) – An optional name to give to the pool, useful, for instance, to identify it in the logs if more than one pool is used. if not specified pick a sequential name such as pool-1, pool-2, etc.

  • timeout (float, default: 30 seconds) – The default maximum time in seconds that a client can wait to receive a connection from the pool (using connection() or getconn()). Note that these methods allow to override the timeout default.

  • max_waiting (int, default: 0) – Maximum number of requests that can be queued to the pool, after which new requests will fail, raising TooManyRequests. 0 means no queue limit.

  • max_lifetime (float, default: 1 hour) – The maximum lifetime of a connection in the pool, in seconds. Connections used for longer get closed and replaced by a new one. The amount is reduced by a random 10% to avoid mass eviction.

  • max_idle (float, default: 10 minutes) – Maximum time, in seconds, that a connection can stay unused in the pool before being closed, and the pool shrunk. This only happens to connections more than min_size, if max_size allowed the pool to grow.

  • reconnect_timeout (float, default: 5 minutes) – Maximum time, in seconds, the pool will try to create a connection. If a connection attempt fails, the pool will try to reconnect a few times, using an exponential backoff and some random factor to avoid mass attempts. If repeated attempts fail, after reconnect_timeout second the connection attempt is aborted and the reconnect_failed() callback invoked.

  • reconnect_failed (Callable[[ConnectionPool], None]) – Callback invoked if an attempt to create a new connection fails for more than reconnect_timeout seconds. The user may decide, for instance, to terminate the program (executing sys.exit()). By default don’t do anything: restart a new connection attempt (if the number of connection fell below min_size).

  • num_workers (int, default: 3) – Number of background worker threads used to maintain the pool state. Background workers are used for example to create new connections and to clean up connections when they are returned to the pool.

Changed in version 3.1: added open parameter to init method.


In a future version, the default value for the open parameter might be changed to False. If you rely on this behaviour (e.g. if you don’t use the pool as a context manager) you might want to specify this parameter explicitly.


Context manager to obtain a connection from the pool.

Return the connection immediately if available, otherwise wait up to timeout or self.timeout seconds and throw PoolTimeout if a connection is not available in time.

Upon context exit, return the connection to the pool. Apply the normal connection context behaviour (commit/rollback the transaction in case of success/error). If the connection is no more in working state replace it with a new one.

Return type:


with my_pool.connection() as conn:

# the connection is now back in the pool
open(wait=False, timeout=30.0)#

Open the pool by starting connecting and and accepting clients.

If wait is False, return immediately and let the background worker fill the pool if min_size > 0. Otherwise wait up to timeout seconds for the requested number of connections to be ready (see wait() for details).

It is safe to call open() again on a pool already open (because the method was already called, or because the pool context was entered, or because the pool was initialized with open = True) but you cannot currently re-open a closed pool.

New in version 3.1.


Close the pool and make it unavailable to new clients.

All the waiting and future clients will fail to acquire a connection with a PoolClosed exception. Currently used connections will not be closed until returned to the pool.

Wait timeout seconds for threads to terminate their job, if positive. If the timeout expires the pool is closed anyway, although it may raise some warnings on exit.


The pool can be also used as a context manager, in which case it will be opened (if necessary) on entering the block and closed on exiting it:

with ConnectionPool(...) as pool:
    # code using the pool

Wait for the pool to be full (with min_size connections) after creation.

Close the pool, and raise PoolTimeout, if not ready within timeout sec.

Calling this method is not mandatory: you can try and use the pool immediately after its creation. The first client will be served as soon as a connection is ready. You can use this method if you prefer your program to terminate in case the environment is not configured properly, rather than trying to stay up the hardest it can.

name: str#

The name of the pool set on creation, or automatically generated if not set.

min_size: int#
max_size: int#

The current minimum and maximum size of the pool. Use resize() to change them at runtime.

resize(min_size, max_size=None)#

Change the size of the pool during runtime.


Verify the state of the connections currently in the pool.

Test each connection: if it works return it to the pool, otherwise dispose of it and create a new one.


Return current stats about the pool usage.

Return type:

Dict[str, int]


Return current stats about the pool usage.

After the call, all the counters are reset to zero.

Return type:

Dict[str, int]

See Pool stats for the metrics returned.

Functionalities you may not need


Obtain a connection from the pool.

You should preferably use connection(). Use this function only if it is not possible to use the connection as context manager.

After using this function you must call a corresponding putconn(): failing to do so will deplete the pool. A depleted pool is a sad pool: you don’t want a depleted pool.

Return type:



Return a connection to the loving hands of its pool.

Use this function only paired with a getconn(). You don’t need to use it if you use the much more comfortable connection() context manager.

Pool exceptions#

class psycopg_pool.PoolTimeout(*args, info=None, encoding='utf-8', pgconn=None)#

The pool couldn’t provide a connection in acceptable time.

Subclass of OperationalError

class psycopg_pool.PoolClosed(*args, info=None, encoding='utf-8', pgconn=None)#

Attempt to get a connection from a closed pool.

Subclass of OperationalError

class psycopg_pool.TooManyRequests(*args, info=None, encoding='utf-8', pgconn=None)#

Too many requests in the queue waiting for a connection from the pool.

Subclass of OperationalError

The AsyncConnectionPool class#

AsyncConnectionPool has a very similar interface to the ConnectionPool class but its blocking methods are implemented as async coroutines. It returns instances of AsyncConnection, or of its subclass if specified so in the connection_class parameter.

Only the functions with different signature from ConnectionPool are listed here.

class psycopg_pool.AsyncConnectionPool(conninfo='', *, open=True, connection_class=<class 'psycopg.AsyncConnection'>, configure=None, reset=None, **kwargs)#
  • connection_class (type, default: AsyncConnection) – The class of the connections to serve. It should be an AsyncConnection subclass.

  • configure (async Callable[[AsyncConnection], None]) – A callback to configure a connection after creation.

  • reset (async Callable[[AsyncConnection], None]) – A callback to reset a function after it has been returned to the pool.

Return type:


async with my_pool.connection() as conn:
    await conn.execute(...)

# the connection is now back in the pool
async open(wait=False, timeout=30.0)#
async close(timeout=5.0)#


The pool can be also used as an async context manager, in which case it will be opened (if necessary) on entering the block and closed on exiting it:

async with AsyncConnectionPool(...) as pool:
    # code using the pool

All the other constructor parameters are the same of ConnectionPool.

async wait(timeout=30.0)#
async resize(min_size, max_size=None)#
async check()#
async getconn(timeout=None)#
Return type:


async putconn(conn)#

Null connection pools#

New in version 3.1.

The NullConnectionPool is a ConnectionPool subclass which doesn’t create connections preemptively and doesn’t keep unused connections in its state. See Null connection pools for further details.

The interface of the object is entirely compatible with its parent class. Its behaviour is similar, with the following differences:

class psycopg_pool.NullConnectionPool(conninfo='', min_size=0, *args, **kwargs)#

All the other constructor parameters are the same as in ConnectionPool.

  • min_size (int, default: 0) – Always 0, cannot be changed.

  • max_size (int, default: None) – If None or 0, create a new connection at every request, without a maximum. If greater than 0, don’t create more than max_size connections and queue the waiting clients.

  • reset (Callable[[Connection], None]) – It is only called when there are waiting clients in the queue, before giving them a connection already open. If no client is waiting, the connection is closed and discarded without a fuss.

  • max_idle – Ignored, as null pools don’t leave idle connections sitting around.


Create a connection for test.

Calling this function will verify that the connectivity with the database works as expected. However the connection will not be stored in the pool.

Close the pool, and raise PoolTimeout, if not ready within timeout sec.

resize(min_size, max_size=None)#

Change the size of the pool during runtime.

Only max_size can be changed; min_size must remain 0.


No-op, as the pool doesn’t have connections in its state.

The AsyncNullConnectionPool is, similarly, an AsyncConnectionPool subclass with the same behaviour of the NullConnectionPool.

class psycopg_pool.AsyncNullConnectionPool(conninfo='', min_size=0, *args, **kwargs)#

The interface is the same of its parent class AsyncConnectionPool. The behaviour is different in the same way described for NullConnectionPool.