psycopg2/doc/src/advanced.rst

More advanced topics
====================

.. sectionauthor:: Daniele Varrazzo <daniele.varrazzo@gmail.com>

.. testsetup:: *

    import re
    import select

    cur.execute("CREATE TABLE atable (apoint point)")
    conn.commit()

    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)

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

.. index::
    double: Subclassing; Cursor
    double: Subclassing; Connection

.. _subclassing-connection:
.. _subclassing-cursor:

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.

.. index::
    single: Example; Cursor subclass

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))



.. index::
    single: Objects; Creating new adapters
    single: Adaptation; Creating new adapters
    single: Data types; Creating new adapters

.. _adapting-new-types:

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 `~cursor.execute()` method adapts its arguments to the
`~psycopg2.extensions.ISQLQuote` protocol.  Objects that conform to this
protocol expose a `!getquoted()` method returning the SQL representation
of the object as a string.

The easiest way to adapt an object to an SQL string is to register an adapter
function via the `~psycopg2.extensions.register_adapter()` function.  The
adapter function must take the value to be adapted as argument and return a
conform object.  A convenient object is the `~psycopg2.extensions.AsIs`
wrapper, whose `!getquoted()` result is simply the `!str()`\ ing
conversion of the wrapped object.

.. index::
    single: Example; Types adaptation

Example: mapping of a `!Point` class into the |point|_ PostgreSQL
geometric type:

.. doctest::

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

    >>> register_adapter(Point, adapt_point)

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


.. |point| replace:: :sql:`point`
.. _point: http://www.postgresql.org/docs/8.4/static/datatype-geometric.html#AEN6084

The above function call results in the SQL command::

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



.. index:: Type casting

.. _type-casting-from-sql-to-python:

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 :sql:`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 :sql:`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 `~psycopg2.extensions.new_type()` function binds the object OIDs
(more than one can be specified) to the adapter function.
`~psycopg2.extensions.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
    <class 'Point'> 10.2 20.3



.. index::
    pair: Asynchronous; Notifications
    pair: LISTEN; SQL command
    pair: NOTIFY; SQL command

.. _async-notify:

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 of how to use this form of
communications.

Notifications received are made available in the `connection.notifies`
list. Notifications can be sent from Python code simply using a :sql:`NOTIFY`
command in an `~cursor.execute()` call.

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

.. |LISTEN| replace:: :sql:`LISTEN`
.. _LISTEN: http://www.postgresql.org/docs/8.4/static/sql-listen.html
.. |NOTIFY| replace:: :sql:`NOTIFY`
.. _NOTIFY: http://www.postgresql.org/docs/8.4/static/sql-notify.html

Notification are received after every query execution. If the user is interested
in receiveing notification but not in performing any query, the
`~connection.poll()` method can be used to check for notification 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 awaken from 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 'NOTIFY test'"
    while 1:
        if select.select([conn],[],[],5) == ([],[],[]):
            print "Timeout"
        else:
            conn.poll()
            while conn.notifies:
                print "Got NOTIFY:", conn.notifies.pop()

Running the script and executing the command :sql:`NOTIFY test` in a separate
:program:`psql` shell, the output may look similar to::

    Waiting for 'NOTIFY test'
    Timeout
    Timeout
    Got NOTIFY: (6535, 'test')
    Timeout
    ...



.. index::
    double: Asynchronous; Connection

.. _async-support:

Asynchronous support
--------------------

.. versionadded:: 2.2.0

Psycopg can issue asynchronous queries to a PostgreSQL database. An asynchronous
communication style is established passing the parameter *async*\=1 to the
`~psycopg2.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 `~connection.fileno()` to get the connection file
descriptor and `~connection.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)

.. |select()| replace:: `!select()`
.. _select(): http://docs.python.org/library/select.html#select.select

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 `~psycopg2.extensions.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()

Notice 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|_.

.. |PQconnectStart| replace:: `!PQconnectStart()`
.. _PQconnectStart: http://www.postgresql.org/docs/8.4/static/libpq-connect.html#AEN33199

The same loop should be also used to perform nonblocking queries: after
sending a query via `~cursor.execute()` or `~cursor.callproc()`, call
`!poll()` on the connection available from `cursor.connection` until it
returns `!POLL_OK`, at which pont 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 :ref:`autocommit <autocommit>` mode and it is not
possible to change it using `~connection.set_isolation_level()`. So a
transaction is not implicitly started at the first query and is not possible
to use methods `~connection.commit()` and `~connection.rollback()`: you can
manually control transactions using `~cursor.execute()` to send database
commands such as :sql:`BEGIN`, :sql:`COMMIT` and :sql:`ROLLBACK`.

With asynchronous connections it is also not possible to use
`~connection.set_client_encoding()`, `~cursor.executemany()`, :ref:`large
objects <large-objects>`, :ref:`named cursors <server-side-cursors>`.

:ref:`COPY commands <copy>` are not supported either in asynchronous mode, but
this will be probably implemented in a future release.




.. index::
    single: Greenlet, Coroutine, eventlet, gevent, Wait callback

.. _green-support:

Support to coroutine libraries
------------------------------

.. versionadded:: 2.2.0

Psycopg can be used together with coroutine_\-based libraries, and participate
to cooperative multithread.

Coroutine-based libraries (such as Eventlet_ or gevent_) can usually patch the
Python standard library in order to enable a coroutine switch in presence of
blocking I/O: the process is usually referred as making the system *green*, in
reference to greenlet_, the basic Python micro-thread library.

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 available data.

This way of working is less flexible of complete asynchronous I/O, but has the
advantage of maintaining a complete |DBAPI| semantics: from the point of view
of the end user, all Psycopg functions and objects will work transparently
in the coroutine environment (the calling coroutine will be blocked while
other coroutines can be scheduled to run), allowing non modified code and
third party libraries (such as SQLAlchemy_) to be used in coroutine-based
programs.

Notice that, while I/O correctly yields control to other coroutines, each
connection has a lock allowing a single cursor at time to communicate with the
backend: such lock is not *green*, so blocking against it would block the
entire program waiting for data, not the single coroutine. Therefore,
programmers are advised to either avoid to share connections between coroutines
or to use a library-friendly lock to synchronize shares connections, e.g. for
pooling.

Coroutine libraries authors should provide a callback implementation (and
probably 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)

.. _coroutine: http://en.wikipedia.org/wiki/Coroutine
.. _greenlet: http://pypi.python.org/pypi/greenlet
.. _Eventlet: http://eventlet.net/
.. _gevent: http://www.gevent.org/
.. _SQLAlchemy: http://www.sqlalchemy.org/
.. __: http://www.postgresql.org/docs/8.4/static/libpq-async.html



.. testcode::
    :hide:

    aconn.close()
    conn.rollback()
    cur.execute("DROP TABLE atable")
    conn.commit()
    cur.close()
    conn.close()