Basic module usage

The basic Psycopg usage is common to all the database adapters implementing the DB API 2.0 protocol. Here is an interactive session showing some of the basic commands:

>>> import psycopg2

# Connect to an existing database
>>> conn = psycopg2.connect("dbname=test user=postgres")

# Open a cursor to perform database operations
>>> cur = conn.cursor()

# Execute a command: this creates a new table
>>> cur.execute("CREATE TABLE test (id serial PRIMARY KEY, num integer, data varchar);")

# Pass data to fill a query placeholders and let Psycopg perform
# the correct conversion (no more SQL injections!)
>>> cur.execute("INSERT INTO test (num, data) VALUES (%s, %s)",
...      (100, "abc'def"))

# Query the database and obtain data as Python objects
>>> cur.execute("SELECT * FROM test;")
>>> cur.fetchone()
(1, 100, "abc'def")

# Make the changes to the database persistent
>>> conn.commit()

# Close communication with the database
>>> cur.close()
>>> conn.close()

The main entry points of Psycopg are:

Passing parameters to SQL queries

Psycopg casts Python variables to SQL literals by type. Many standard Python types are already adapted to the correct SQL representation.

Example: the Python function call:

>>> cur.execute(
...     """INSERT INTO some_table (an_int, a_date, a_string)
...         VALUES (%s, %s, %s);""",
...     (10,, 11, 18), "O'Reilly"))

is converted into the SQL command:

INSERT INTO some_table (an_int, a_date, a_string)
 VALUES (10, '2005-11-18', 'O''Reilly');

Named arguments are supported too using %(name)s placeholders. Using named arguments the values can be passed to the query in any order and many placeholders can use the same values:

>>> cur.execute(
...     """INSERT INTO some_table (an_int, a_date, another_date, a_string)
...         VALUES (%(int)s, %(date)s, %(date)s, %(str)s);""",
...     {'int': 10, 'str': "O'Reilly", 'date':, 11, 18)})

When parameters are used, in order to include a literal % in the query you can use the %% string. Using characters %, (, ) in the argument names is not supported.

While the mechanism resembles regular Python strings manipulation, there are a few subtle differences you should care about when passing parameters to a query:

  • The Python string operator % is not used: the execute() method accepts a tuple or dictionary of values as second parameter. Never use % or + to merge values into queries.

  • The variables placeholder must always be a %s, even if a different placeholder (such as a %d for integers or %f for floats) may look more appropriate:

    >>> cur.execute("INSERT INTO numbers VALUES (%d)", (42,)) # WRONG
    >>> cur.execute("INSERT INTO numbers VALUES (%s)", (42,)) # correct
  • For positional variables binding, the second argument must always be a sequence, even if it contains a single variable. And remember that Python requires a comma to create a single element tuple:

    >>> cur.execute("INSERT INTO foo VALUES (%s)", "bar")    # WRONG
    >>> cur.execute("INSERT INTO foo VALUES (%s)", ("bar"))  # WRONG
    >>> cur.execute("INSERT INTO foo VALUES (%s)", ("bar",)) # correct
    >>> cur.execute("INSERT INTO foo VALUES (%s)", ["bar"])  # correct
  • Only query values should be bound via this method: it shouldn’t be used to merge table or field names to the query. If you need to generate dynamically an SQL query (for instance choosing dynamically a table name) you can use the facilities provided by the psycopg2.sql module.

The problem with the query parameters

The SQL representation of many data types is often different from their Python string representation. The typical example is with single quotes in strings: in SQL single quotes are used as string literal delimiters, so the ones appearing inside the string itself must be escaped, whereas in Python single quotes can be left unescaped if the string is delimited by double quotes.

Because of the difference, sometime subtle, between the data types representations, a naïve approach to query strings composition, such as using Python strings concatenation, is a recipe for terrible problems:

>>> SQL = "INSERT INTO authors (name) VALUES ('%s');" # NEVER DO THIS
>>> data = ("O'Reilly", )
>>> cur.execute(SQL % data) # THIS WILL FAIL MISERABLY
ProgrammingError: syntax error at or near "Reilly"
LINE 1: INSERT INTO authors (name) VALUES ('O'Reilly')

If the variables containing the data to send to the database come from an untrusted source (such as a form published on a web site) an attacker could easily craft a malformed string, either gaining access to unauthorized data or performing destructive operations on the database. This form of attack is called SQL injection and is known to be one of the most widespread forms of attack to database servers. Before continuing, please print this page as a memo and hang it onto your desk.

Psycopg can automatically convert Python objects to and from SQL literals: using this feature your code will be more robust and reliable. We must stress this point:


Never, never, NEVER use Python string concatenation (+) or string parameters interpolation (%) to pass variables to a SQL query string. Not even at gunpoint.

The correct way to pass variables in a SQL command is using the second argument of the execute() method:

>>> SQL = "INSERT INTO authors (name) VALUES (%s);" # Note: no quotes
>>> data = ("O'Reilly", )
>>> cur.execute(SQL, data) # Note: no % operator

Adaptation of Python values to SQL types

Many standard Python types are adapted into SQL and returned as Python objects when a query is executed.

The following table shows the default mapping between Python and PostgreSQL types:

Python PostgreSQL See also
None NULL Constants adaptation
bool bool
Numbers adaptation
Decimal numeric
Strings adaptation
Buffer protocol
bytea Binary adaptation
date date Date/Time objects adaptation
timedelta interval
list ARRAY Lists adaptation
Composite types
IN syntax
dict hstore Hstore data type
Psycopg’s Range range Range data types
Anything™ json JSON adaptation
UUID uuid UUID data type
ipaddress objects
Networking data types

The mapping is fairly customizable: see Adapting new Python types to SQL syntax and Type casting of SQL types into Python objects. You can also find a few other specialized adapters in the psycopg2.extras module.

Constants adaptation

Python None and boolean values True and False are converted into the proper SQL literals:

>>> cur.mogrify("SELECT %s, %s, %s;", (None, True, False))
'SELECT NULL, true, false;'

Numbers adaptation

Python numeric objects int, long, float, Decimal are converted into a PostgreSQL numerical representation:

>>> cur.mogrify("SELECT %s, %s, %s, %s;", (10, 10L, 10.0, Decimal("10.00")))
'SELECT 10, 10, 10.0, 10.00;'

Reading from the database, integer types are converted into int, floating point types are converted into float, numeric/decimal are converted into Decimal.


Sometimes you may prefer to receive numeric data as float instead, for performance reason or ease of manipulation: you can configure an adapter to cast PostgreSQL numeric to Python float. This of course may imply a loss of precision.

Strings adaptation

Python str and unicode are converted into the SQL string syntax. unicode objects (str in Python 3) are encoded in the connection encoding before sending to the backend: trying to send a character not supported by the encoding will result in an error. Data is usually received as str (i.e. it is decoded on Python 3, left encoded on Python 2). However it is possible to receive unicode on Python 2 too: see Unicode handling.

Unicode handling

Psycopg can exchange Unicode data with a PostgreSQL database. Python unicode objects are automatically encoded in the client encoding defined on the database connection (the PostgreSQL encoding, available in connection.encoding, is translated into a Python encoding using the encodings mapping):

>>> print u, type(u)
àèìòù€ <type 'unicode'>

>>> cur.execute("INSERT INTO test (num, data) VALUES (%s,%s);", (74, u))

When reading data from the database, in Python 2 the strings returned are usually 8 bit str objects encoded in the database client encoding:

>>> print conn.encoding

>>> cur.execute("SELECT data FROM test WHERE num = 74")
>>> x = cur.fetchone()[0]
>>> print x, type(x), repr(x)
àèìòù€ <type 'str'> '\xc3\xa0\xc3\xa8\xc3\xac\xc3\xb2\xc3\xb9\xe2\x82\xac'

>>> conn.set_client_encoding('LATIN9')

>>> cur.execute("SELECT data FROM test WHERE num = 74")
>>> x = cur.fetchone()[0]
>>> print type(x), repr(x)
<type 'str'> '\xe0\xe8\xec\xf2\xf9\xa4'

In Python 3 instead the strings are automatically decoded in the connection encoding, as the str object can represent Unicode characters. In Python 2 you must register a typecaster in order to receive unicode objects:

>>> psycopg2.extensions.register_type(psycopg2.extensions.UNICODE, cur)

>>> cur.execute("SELECT data FROM test WHERE num = 74")
>>> x = cur.fetchone()[0]
>>> print x, type(x), repr(x)
àèìòù€ <type 'unicode'> u'\xe0\xe8\xec\xf2\xf9\u20ac'

In the above example, the UNICODE typecaster is registered only on the cursor. It is also possible to register typecasters on the connection or globally: see the function register_type() and Type casting of SQL types into Python objects for details.


In Python 2, if you want to uniformly receive all your database input in Unicode, you can register the related typecasters globally as soon as Psycopg is imported:

import psycopg2
import psycopg2.extensions

and forget about this story.

Binary adaptation

Python types representing binary objects are converted into PostgreSQL binary string syntax, suitable for bytea fields. Such types are buffer (only available in Python 2), memoryview (available from Python 2.7), bytearray (available from Python 2.6) and bytes (only from Python 3: the name is available from Python 2.6 but it’s only an alias for the type str). Any object implementing the Revised Buffer Protocol should be usable as binary type where the protocol is supported (i.e. from Python 2.6). Received data is returned as buffer (in Python 2) or memoryview (in Python 3).

Changed in version 2.4: only strings were supported before.

Changed in version 2.4.1: can parse the ‘hex’ format from 9.0 servers without relying on the version of the client library.


In Python 2, if you have binary data in a str object, you can pass them to a bytea field using the psycopg2.Binary wrapper:

mypic = open('picture.png', 'rb').read()
curs.execute("insert into blobs (file) values (%s)",


Since version 9.0 PostgreSQL uses by default a new “hex” format to emit bytea fields. Starting from Psycopg 2.4.1 the format is correctly supported. If you use a previous version you will need some extra care when receiving bytea from PostgreSQL: you must have at least libpq 9.0 installed on the client or alternatively you can set the bytea_output configuration parameter to escape, either in the server configuration file or in the client session (using a query such as SET bytea_output TO escape;) before receiving binary data.

Date/Time objects adaptation

Python builtin datetime, date, time, timedelta are converted into PostgreSQL’s timestamp[tz], date, time[tz], interval data types. Time zones are supported too. The Egenix mx.DateTime objects are adapted the same way:

>>> dt =
>>> dt
datetime.datetime(2010, 2, 8, 1, 40, 27, 425337)

>>> cur.mogrify("SELECT %s, %s, %s;", (dt,, dt.time()))
"SELECT '2010-02-08T01:40:27.425337', '2010-02-08', '01:40:27.425337';"

>>> cur.mogrify("SELECT %s;", (dt - datetime.datetime(2010,1,1),))
"SELECT '38 days 6027.425337 seconds';"

Time zones handling

The PostgreSQL type timestamp with time zone (a.k.a. timestamptz) is converted into Python datetime objects with a tzinfo attribute set to a FixedOffsetTimezone instance.

>>> cur.execute("SET TIME ZONE 'Europe/Rome';")  # UTC + 1 hour
>>> cur.execute("SELECT '2010-01-01 10:30:45'::timestamptz;")
>>> cur.fetchone()[0].tzinfo, name=None)

Note that only time zones with an integer number of minutes are supported: this is a limitation of the Python datetime module. A few historical time zones had seconds in the UTC offset: these time zones will have the offset rounded to the nearest minute, with an error of up to 30 seconds.

>>> cur.execute("SET TIME ZONE 'Asia/Calcutta';")  # offset was +5:53:20
>>> cur.execute("SELECT '1930-01-01 10:30:45'::timestamptz;")
>>> cur.fetchone()[0].tzinfo, name=None)

Changed in version 2.2.2: timezones with seconds are supported (with rounding). Previously such timezones raised an error. In order to deal with them in previous versions use psycopg2.extras.register_tstz_w_secs().

Infinite dates handling

PostgreSQL can store the representation of an “infinite” date, timestamp, or interval. Infinite dates are not available to Python, so these objects are mapped to date.max, datetime.max, interval.max. Unfortunately the mapping cannot be bidirectional so these dates will be stored back into the database with their values, such as 9999-12-31.

It is possible to create an alternative adapter for dates and other objects to map date.max to infinity, for instance:

class InfDateAdapter:
    def __init__(self, wrapped):
        self.wrapped = wrapped
    def getquoted(self):
        if self.wrapped ==
            return b"'infinity'::date"
        elif self.wrapped ==
            return b"'-infinity'::date"
            return psycopg2.extensions.DateFromPy(self.wrapped).getquoted()

psycopg2.extensions.register_adapter(, InfDateAdapter)

Of course it will not be possible to write the value of date.max in the database anymore: infinity will be stored instead.

Lists adaptation

Python lists are converted into PostgreSQL ARRAYs:

>>> cur.mogrify("SELECT %s;", ([10, 20, 30], ))
'SELECT ARRAY[10,20,30];'


You can use a Python list as the argument of the IN operator using the PostgreSQL ANY operator.

ids = [10, 20, 30]
cur.execute("SELECT * FROM data WHERE id = ANY(%s);", (ids,))

Furthermore ANY can also work with empty lists, whereas IN () is a SQL syntax error.


Reading back from PostgreSQL, arrays are converted to lists of Python objects as expected, but only if the items are of a known type. Arrays of unknown types are returned as represented by the database (e.g. {a,b,c}). If you want to convert the items into Python objects you can easily create a typecaster for array of unknown types.

Tuples adaptation

Python tuples are converted into a syntax suitable for the SQL IN operator and to represent a composite type:

>>> cur.mogrify("SELECT %s IN %s;", (10, (10, 20, 30)))
'SELECT 10 IN (10, 20, 30);'


SQL doesn’t allow an empty list in the IN operator, so your code should guard against empty tuples. Alternatively you can use a Python list.

If you want PostgreSQL composite types to be converted into a Python tuple/namedtuple you can use the register_composite() function.

New in version 2.0.6: the tuple IN adaptation.

Changed in version 2.0.14: the tuple IN adapter is always active. In previous releases it was necessary to import the extensions module to have it registered.

Changed in version 2.3: namedtuple instances are adapted like regular tuples and can thus be used to represent composite types.

Transactions control

In Psycopg transactions are handled by the connection class. By default, the first time a command is sent to the database (using one of the cursors created by the connection), a new transaction is created. The following database commands will be executed in the context of the same transaction – not only the commands issued by the first cursor, but the ones issued by all the cursors created by the same connection. Should any command fail, the transaction will be aborted and no further command will be executed until a call to the rollback() method.

The connection is responsible for terminating its transaction, calling either the commit() or rollback() method. Committed changes are immediately made persistent into the database. Closing the connection using the close() method or destroying the connection object (using del or letting it fall out of scope) will result in an implicit rollback.

It is possible to set the connection in autocommit mode: this way all the commands executed will be immediately committed and no rollback is possible. A few commands (e.g. CREATE DATABASE, VACUUM...) require to be run outside any transaction: in order to be able to run these commands from Psycopg, the connection must be in autocommit mode: you can use the autocommit property (set_isolation_level() in older versions).


By default even a simple SELECT will start a transaction: in long-running programs, if no further action is taken, the session will remain “idle in transaction”, an undesirable condition for several reasons (locks are held by the session, tables bloat...). For long lived scripts, either make sure to terminate a transaction as soon as possible or use an autocommit connection.

A few other transaction properties can be set session-wide by the connection: for instance it is possible to have read-only transactions or change the isolation level. See the set_session() method for all the details.

with statement

Starting from version 2.5, psycopg2’s connections and cursors are context managers and can be used with the with statement:

with psycopg2.connect(DSN) as conn:
    with conn.cursor() as curs:

When a connection exits the with block, if no exception has been raised by the block, the transaction is committed. In case of exception the transaction is rolled back.

When a cursor exits the with block it is closed, releasing any resource eventually associated with it. The state of the transaction is not affected.

Note that, unlike file objects or other resources, exiting the connection’s with block doesn’t close the connection but only the transaction associated with it: a connection can be used in more than a with statement and each with block is effectively wrapped in a separate transaction:

conn = psycopg2.connect(DSN)

with conn:
    with conn.cursor() as curs:

with conn:
    with conn.cursor() as curs:


Server side cursors

When a database query is executed, the Psycopg cursor usually fetches all the records returned by the backend, transferring them to the client process. If the query returned an huge amount of data, a proportionally large amount of memory will be allocated by the client.

If the dataset is too large to be practically handled on the client side, it is possible to create a server side cursor. Using this kind of cursor it is possible to transfer to the client only a controlled amount of data, so that a large dataset can be examined without keeping it entirely in memory.

Server side cursor are created in PostgreSQL using the DECLARE command and subsequently handled using MOVE, FETCH and CLOSE commands.

Psycopg wraps the database server side cursor in named cursors. A named cursor is created using the cursor() method specifying the name parameter. Such cursor will behave mostly like a regular cursor, allowing the user to move in the dataset using the scroll() method and to read the data using fetchone() and fetchmany() methods. Normally you can only scroll forward in a cursor: if you need to scroll backwards you should declare your cursor scrollable.

Named cursors are also iterable like regular cursors. Note however that before Psycopg 2.4 iteration was performed fetching one record at time from the backend, resulting in a large overhead. The attribute itersize now controls how many records are fetched at time during the iteration: the default value of 2000 allows to fetch about 100KB per roundtrip assuming records of 10-20 columns of mixed number and strings; you may decrease this value if you are dealing with huge records.

Named cursors are usually created WITHOUT HOLD, meaning they live only as long as the current transaction. Trying to fetch from a named cursor after a commit() or to create a named cursor when the connection transaction isolation level is set to AUTOCOMMIT will result in an exception. It is possible to create a WITH HOLD cursor by specifying a True value for the withhold parameter to cursor() or by setting the withhold attribute to True before calling execute() on the cursor. It is extremely important to always close() such cursors, otherwise they will continue to hold server-side resources until the connection will be eventually closed. Also note that while WITH HOLD cursors lifetime extends well after commit(), calling rollback() will automatically close the cursor.


It is also possible to use a named cursor to consume a cursor created in some other way than using the DECLARE executed by execute(). For example, you may have a PL/pgSQL function returning a cursor:

CREATE FUNCTION reffunc(refcursor) RETURNS refcursor AS $$
    OPEN $1 FOR SELECT col FROM test;
    RETURN $1;
$$ LANGUAGE plpgsql;

You can read the cursor content by calling the function with a regular, non-named, Psycopg cursor:

cur1 = conn.cursor()
cur1.callproc('reffunc', ['curname'])

and then use a named cursor in the same transaction to “steal the cursor”:

cur2 = conn.cursor('curname')
for record in cur2:     # or cur2.fetchone, fetchmany...
    # do something with record

Thread and process safety

The Psycopg module and the connection objects are thread-safe: many threads can access the same database either using separate sessions and creating a connection per thread or using the same connection and creating separate cursors. In DB API 2.0 parlance, Psycopg is level 2 thread safe.

The difference between the above two approaches is that, using different connections, the commands will be executed in different sessions and will be served by different server processes. On the other hand, using many cursors on the same connection, all the commands will be executed in the same session (and in the same transaction if the connection is not in autocommit mode), but they will be serialized.

The above observations are only valid for regular threads: they don’t apply to forked processes nor to green threads. libpq connections shouldn’t be used by a forked processes, so when using a module such as multiprocessing or a forking web deploy method such as FastCGI make sure to create the connections after the fork.

Connections shouldn’t be shared either by different green threads: see Support for coroutine libraries for further details.


Psycopg cursor objects provide an interface to the efficient PostgreSQL COPY command to move data from files to tables and back.

Currently no adaptation is provided between Python and PostgreSQL types on COPY: the file can be any Python file-like object but its format must be in the format accepted by PostgreSQL COPY command (data fromat, escaped characters, etc).

The methods exposed are:

Reads data from a file-like object appending them to a database table (COPY table FROM file syntax). The source file must provide both read() and readline() method.
Writes the content of a table to a file-like object (COPY table TO file syntax). The target file must have a write() method.
Allows to handle more specific cases and to use all the COPY features available in PostgreSQL.

Please refer to the documentation of the single methods for details and examples.

Access to PostgreSQL large objects

PostgreSQL offers support for large objects, which provide stream-style access to user data that is stored in a special large-object structure. They are useful with data values too large to be manipulated conveniently as a whole.

Psycopg allows access to the large object using the lobject class. Objects are generated using the connection.lobject() factory method. Data can be retrieved either as bytes or as Unicode strings.

Psycopg large object support efficient import/export with file system files using the lo_import() and lo_export() libpq functions.

Changed in version 2.6: added support for large objects greated than 2GB. Note that the support is enabled only if all the following conditions are verified:

  • the Python build is 64 bits;
  • the extension was built against at least libpq 9.3;
  • the server version is at least PostgreSQL 9.3 (server_version must be >= 90300).

If Psycopg was built with 64 bits large objects support (i.e. the first two contidions above are verified), the psycopg2.__version__ constant will contain the lo64 flag. If any of the contition is not met several lobject methods will fail if the arguments exceed 2GB.

Two-Phase Commit protocol support

New in version 2.3.

Psycopg exposes the two-phase commit features available since PostgreSQL 8.1 implementing the two-phase commit extensions proposed by the DB API 2.0.

The DB API 2.0 model of two-phase commit is inspired by the XA specification, according to which transaction IDs are formed from three components:

  • a format ID (non-negative 32 bit integer)
  • a global transaction ID (string not longer than 64 bytes)
  • a branch qualifier (string not longer than 64 bytes)

For a particular global transaction, the first two components will be the same for all the resources. Every resource will be assigned a different branch qualifier.

According to the DB API 2.0 specification, a transaction ID is created using the connection.xid() method. Once you have a transaction id, a distributed transaction can be started with connection.tpc_begin(), prepared using tpc_prepare() and completed using tpc_commit() or tpc_rollback(). Transaction IDs can also be retrieved from the database using tpc_recover() and completed using the above tpc_commit() and tpc_rollback().

PostgreSQL doesn’t follow the XA standard though, and the ID for a PostgreSQL prepared transaction can be any string up to 200 characters long. Psycopg’s Xid objects can represent both XA-style transactions IDs (such as the ones created by the xid() method) and PostgreSQL transaction IDs identified by an unparsed string.

The format in which the Xids are converted into strings passed to the database is the same employed by the PostgreSQL JDBC driver: this should allow interoperation between tools written in Python and in Java. For example a recovery tool written in Python would be able to recognize the components of transactions produced by a Java program.

For further details see the documentation for the above methods.