More advanced topics - Psycopg 2.7.3 documentation

Connection and cursor factories

Psycopg exposes two new-style classes that can be sub-classed and expanded to adapt them to the needs of the programmer: psycopg2.extensions.cursor and psycopg2.extensions.connection . The connection class is usually sub-classed only to provide an easy way to create customized cursors but other uses are possible. cursor is much more interesting, because it is the class where query building, execution and result type-casting into Python variables happens.

The extras module contains several examples of connection and cursor subclasses .

An example of cursor subclass performing logging is:

import psycopg2
import psycopg2.extensions
import logging

class LoggingCursor(psycopg2.extensions.cursor):
    def execute(self, sql, args=None):
        logger = logging.getLogger('sql_debug')
        logger.info(self.mogrify(sql, args))

        try:
            psycopg2.extensions.cursor.execute(self, sql, args)
        except Exception, exc:
            logger.error("%s: %s" % (exc.__class__.__name__, exc))
            raise

conn = psycopg2.connect(DSN)
cur = conn.cursor(cursor_factory=LoggingCursor)
cur.execute("INSERT INTO mytable VALUES (%s, %s, %s);",
             (10, 20, 30))

Adapting new Python types to SQL syntax

Any Python class or type can be adapted to an SQL string. Adaptation mechanism is similar to the Object Adaptation proposed in the PEP 246 and is exposed by the psycopg2.extensions.adapt() function.

The execute() method adapts its arguments to the ISQLQuote protocol. Objects that conform to this protocol expose a getquoted() method returning the SQL representation of the object as a string (the method must return bytes in Python 3). Optionally the conform object may expose a prepare() method.

There are two basic ways to have a Python object adapted to SQL:

• the object itself is conform, or knows how to make itself conform. Such object must expose a __conform__() method that will be called with the protocol object as argument. The object can check that the protocol is ISQLQuote , in which case it can return self (if the object also implements getquoted() ) or a suitable wrapper object. This option is viable if you are the author of the object and if the object is specifically designed for the database (i.e. having Psycopg as a dependency and polluting its interface with the required methods doesn’t bother you). For a simple example you can take a look at the source code for the psycopg2.extras.Inet object.
• If implementing the ISQLQuote interface directly in the object is not an option (maybe because the object to adapt comes from a third party library), you can use an adaptation function , taking the object to be adapted as argument and returning a conforming object. The adapter must be registered via the register_adapter() function. A simple example wrapper is psycopg2.extras.UUID_adapter used by the register_uuid() function.

A convenient object to write adapters is the AsIs wrapper, whose getquoted() result is simply the str() ing conversion of the wrapped object.

Example: mapping of a Point class into the point PostgreSQL geometric type:

>>> from psycopg2.extensions import adapt, register_adapter, AsIs

>>> class Point(object):
...    def __init__(self, x, y):
...        self.x = x
...        self.y = y

>>> def adapt_point(point):
...     x = adapt(point.x).getquoted()
...     y = adapt(point.y).getquoted()
...     return AsIs("'(%s, %s)'" % (x, y))

>>> register_adapter(Point, adapt_point)

>>> cur.execute("INSERT INTO atable (apoint) VALUES (%s)",
...             (Point(1.23, 4.56),))

The above function call results in the SQL command:

INSERT INTO atable (apoint) VALUES ('(1.23, 4.56)');

Type casting of SQL types into Python objects

PostgreSQL objects read from the database can be adapted to Python objects through an user-defined adapting function. An adapter function takes two arguments: the object string representation as returned by PostgreSQL and the cursor currently being read, and should return a new Python object. For example, the following function parses the PostgreSQL point representation into the previously defined Point class:

>>> def cast_point(value, cur):
...    if value is None:
...        return None
...
...    # Convert from (f1, f2) syntax using a regular expression.
...    m = re.match(r"\(([^)]+),([^)]+)\)", value)
...    if m:
...        return Point(float(m.group(1)), float(m.group(2)))
...    else:
...        raise InterfaceError("bad point representation: %r" % value)

In order to create a mapping from a PostgreSQL type (either standard or user-defined), its OID must be known. It can be retrieved either by the second column of the cursor.description :

>>> cur.execute("SELECT NULL::point")
>>> point_oid = cur.description[0][1]
>>> point_oid
600

or by querying the system catalog for the type name and namespace (the namespace for system objects is pg_catalog ):

>>> cur.execute("""
...    SELECT pg_type.oid
...      FROM pg_type JOIN pg_namespace
...             ON typnamespace = pg_namespace.oid
...     WHERE typname = %(typename)s
...       AND nspname = %(namespace)s""",
...    {'typename': 'point', 'namespace': 'pg_catalog'})
>>> point_oid = cur.fetchone()[0]
>>> point_oid
600

After you know the object OID, you can create and register the new type:

>>> POINT = psycopg2.extensions.new_type((point_oid,), "POINT", cast_point)
>>> psycopg2.extensions.register_type(POINT)

The new_type() function binds the object OIDs (more than one can be specified) to the adapter function. register_type() completes the spell. Conversion is automatically performed when a column whose type is a registered OID is read:

>>> cur.execute("SELECT '(10.2,20.3)'::point")
>>> point = cur.fetchone()[0]
>>> print type(point), point.x, point.y
 10.2 20.3

A typecaster created by new_type() can be also used with new_array_type() to create a typecaster converting a PostgreSQL array into a Python list.

Asynchronous notifications

Psycopg allows asynchronous interaction with other database sessions using the facilities offered by PostgreSQL commands LISTEN and NOTIFY . Please refer to the PostgreSQL documentation for examples about how to use this form of communication.

Notifications are instances of the Notify object made available upon reception in the connection.notifies list. Notifications can be sent from Python code simply executing a NOTIFY command in an execute() call.

Because of the way sessions interact with notifications (see NOTIFY documentation), you should keep the connection in autocommit mode if you wish to receive or send notifications in a timely manner.

Notifications are received after every query execution. If the user is interested in receiving notifications but not in performing any query, the poll() method can be used to check for new messages without wasting resources.

A simple application could poll the connection from time to time to check if something new has arrived. A better strategy is to use some I/O completion function such as select() to sleep until awakened by the kernel when there is some data to read on the connection, thereby using no CPU unless there is something to read:

import select
import psycopg2
import psycopg2.extensions

conn = psycopg2.connect(DSN)
conn.set_isolation_level(psycopg2.extensions.ISOLATION_LEVEL_AUTOCOMMIT)

curs = conn.cursor()
curs.execute("LISTEN test;")

print "Waiting for notifications on channel 'test'"
while 1:
    if select.select([conn],[],[],5) == ([],[],[]):
        print "Timeout"
    else:
        conn.poll()
        while conn.notifies:
            notify = conn.notifies.pop(0)
            print "Got NOTIFY:", notify.pid, notify.channel, notify.payload

Running the script and executing a command such as NOTIFY test, 'hello' in a separate psql shell, the output may look similar to:

Waiting for notifications on channel 'test'
Timeout
Timeout
Got NOTIFY: 6535 test hello
Timeout
...

Note that the payload is only available from PostgreSQL 9.0: notifications received from a previous version server will have the payload attribute set to the empty string.

Asynchronous support

Psycopg can issue asynchronous queries to a PostgreSQL database. An asynchronous communication style is established passing the parameter async =1 to the connect() function: the returned connection will work in asynchronous mode .

In asynchronous mode, a Psycopg connection will rely on the caller to poll the socket file descriptor, checking if it is ready to accept data or if a query result has been transferred and is ready to be read on the client. The caller can use the method fileno() to get the connection file descriptor and poll() to make communication proceed according to the current connection state.

The following is an example loop using methods fileno() and poll() together with the Python select() function in order to carry on asynchronous operations with Psycopg:

def wait(conn):
    while 1:
        state = conn.poll()
        if state == psycopg2.extensions.POLL_OK:
            break
        elif state == psycopg2.extensions.POLL_WRITE:
            select.select([], [conn.fileno()], [])
        elif state == psycopg2.extensions.POLL_READ:
            select.select([conn.fileno()], [], [])
        else:
            raise psycopg2.OperationalError("poll() returned %s" % state)

The above loop of course would block an entire application: in a real asynchronous framework, select() would be called on many file descriptors waiting for any of them to be ready. Nonetheless the function can be used to connect to a PostgreSQL server only using nonblocking commands and the connection obtained can be used to perform further nonblocking queries. After poll() has returned POLL_OK , and thus wait() has returned, the connection can be safely used:

>>> aconn = psycopg2.connect(database='test', async=1)
>>> wait(aconn)
>>> acurs = aconn.cursor()

Note that there are a few other requirements to be met in order to have a completely non-blocking connection attempt: see the libpq documentation for PQconnectStart() .

The same loop should be also used to perform nonblocking queries: after sending a query via execute() or callproc() , call poll() on the connection available from cursor.connection until it returns POLL_OK , at which point the query has been completely sent to the server and, if it produced data, the results have been transferred to the client and available using the regular cursor methods:

>>> acurs.execute("SELECT pg_sleep(5); SELECT 42;")
>>> wait(acurs.connection)
>>> acurs.fetchone()[0]
42

When an asynchronous query is being executed, connection.isexecuting() returns True . Two cursors can’t execute concurrent queries on the same asynchronous connection.

There are several limitations in using asynchronous connections: the connection is always in autocommit mode and it is not possible to change it. So a transaction is not implicitly started at the first query and is not possible to use methods commit() and rollback() : you can manually control transactions using execute() to send database commands such as BEGIN , COMMIT and ROLLBACK . Similarly set_session() can’t be used but it is still possible to invoke the SET command with the proper default_transaction_... parameter.

With asynchronous connections it is also not possible to use set_client_encoding() , executemany() , large objects , named cursors .

COPY commands are not supported either in asynchronous mode, but this will be probably implemented in a future release.

Support for coroutine libraries

Psycopg can be used together with coroutine -based libraries and participate in cooperative multithreading.

Coroutine-based libraries (such as Eventlet or gevent ) can usually patch the Python standard library in order to enable a coroutine switch in the presence of blocking I/O: the process is usually referred as making the system green , in reference to the green threads .

Because Psycopg is a C extension module, it is not possible for coroutine libraries to patch it: Psycopg instead enables cooperative multithreading by allowing the registration of a wait callback using the psycopg2.extensions.set_wait_callback() function. When a wait callback is registered, Psycopg will use libpq non-blocking calls instead of the regular blocking ones, and will delegate to the callback the responsibility to wait for the socket to become readable or writable.

Working this way, the caller does not have the complete freedom to schedule the socket check whenever they want as with an asynchronous connection , but has the advantage of maintaining a complete DB API 2.0 semantics: from the point of view of the end user, all Psycopg functions and objects will work transparently in the coroutine environment (blocking the calling green thread and giving other green threads the possibility to be scheduled), allowing non modified code and third party libraries (such as SQLAlchemy ) to be used in coroutine-based programs.

Coroutine libraries authors should provide a callback implementation (and possibly a method to register it) to make Psycopg as green as they want. An example callback (using select() to block) is provided as psycopg2.extras.wait_select() : it boils down to something similar to:

def wait_select(conn):
    while 1:
        state = conn.poll()
        if state == extensions.POLL_OK:
            break
        elif state == extensions.POLL_READ:
            select.select([conn.fileno()], [], [])
        elif state == extensions.POLL_WRITE:
            select.select([], [conn.fileno()], [])
        else:
            raise OperationalError("bad state from poll: %s" % state)

Providing callback functions for the single coroutine libraries is out of psycopg2 scope, as the callback can be tied to the libraries’ implementation details. You can check the psycogreen project for further informations and resources about the topic.

Replication protocol support

Modern PostgreSQL servers (version 9.0 and above) support replication. The replication protocol is built on top of the client-server protocol and can be operated using libpq , as such it can be also operated by psycopg2 . The replication protocol can be operated on both synchronous and asynchronous connections.

Server version 9.4 adds a new feature called Logical Replication .

from __future__ import print_function
import sys
import psycopg2
import psycopg2.extras

conn = psycopg2.connect('dbname=psycopg2_test user=postgres',
   connection_factory=psycopg2.extras.LogicalReplicationConnection)
cur = conn.cursor()
try:
    # test_decoding produces textual output
    cur.start_replication(slot_name='pytest', decode=True)
except psycopg2.ProgrammingError:
    cur.create_replication_slot('pytest', output_plugin='test_decoding')
    cur.start_replication(slot_name='pytest', decode=True)

class DemoConsumer(object):
    def __call__(self, msg):
        print(msg.payload)
        msg.cursor.send_feedback(flush_lsn=msg.data_start)

democonsumer = DemoConsumer()

print("Starting streaming, press Control-C to end...", file=sys.stderr)
try:
   cur.consume_stream(democonsumer)
except KeyboardInterrupt:
   cur.close()
   conn.close()
   print("The slot 'pytest' still exists. Drop it with "
      "SELECT pg_drop_replication_slot('pytest'); if no longer needed.",
      file=sys.stderr)
   print("WARNING: Transaction logs will accumulate in pg_xlog "
      "until the slot is dropped.", file=sys.stderr)