Getting Started

First, make sure you have the driver properly installed.

Connecting to Cassandra

Before we can start executing any queries against a Cassandra cluster we need to setup an instance of Cluster. As the name suggests, you will typically have one instance of Cluster for each Cassandra cluster you want to interact with.

The simplest way to create a Cluster is like this:

from cassandra.cluster import Cluster

cluster = Cluster()

This will attempt to connection to a Cassandra instance on your local machine ( You can also specify a list of IP addresses for nodes in your cluster:

from cassandra.cluster import Cluster

cluster = Cluster(['', ''])

The set of IP addresses we pass to the Cluster is simply an initial set of contact points. After the driver connects to one of these nodes it will automatically discover the rest of the nodes in the cluster and connect to them, so you don’t need to list every node in your cluster.

If you need to use a non-standard port, use SSL, or customize the driver’s behavior in some other way, this is the place to do it:

from cassandra.cluster import Cluster
from cassandra.policies import DCAwareRoundRobinPolicy

cluster = Cluster(
    ['', '', ''],

You can find a more complete list of options in the Cluster documentation.

Instantiating a Cluster does not actually connect us to any nodes. To establish connections and begin executing queries we need a Session, which is created by calling Cluster.connect():

cluster = Cluster()
session = cluster.connect()

The connect() method takes an optional keyspace argument which sets the default keyspace for all queries made through that Session:

cluster = Cluster()
session = cluster.connect('mykeyspace')

You can always change a Session’s keyspace using set_keyspace() or by executing a USE query:

# or you can do this instead
session.execute('USE users')

Executing Queries

Now that we have a Session we can begin to execute queries. The simplest way to execute a query is to use execute():

rows = session.execute('SELECT name, age, email FROM users')
for user_row in rows:
    print, user_row.age,

This will transparently pick a Cassandra node to execute the query against and handle any retries that are necessary if the operation fails.

By default, each row in the result set will be a namedtuple. Each row will have a matching attribute for each column defined in the schema, such as name, age, and so on. You can also treat them as normal tuples by unpacking them or accessing fields by position. The following three examples are equivalent:

rows = session.execute('SELECT name, age, email FROM users')
for row in rows:
    print, row.age,
rows = session.execute('SELECT name, age, email FROM users')
for (name, age, email) in rows:
    print name, age, email
rows = session.execute('SELECT name, age, email FROM users')
for row in rows:
    print row[0], row[1], row[2]

If you prefer another result format, such as a dict per row, you can change the row_factory attribute.

For queries that will be run repeatedly, you should use Prepared statements.

Passing Parameters to CQL Queries

When executing non-prepared statements, the driver supports two forms of parameter place-holders: positional and named.

Positional parameters are used with a %s placeholder. For example, when you execute:

    INSERT INTO users (name, credits, user_id)
    VALUES (%s, %s, %s)
    ("John O'Reilly", 42, uuid.uuid1())

It is translated to the following CQL query:

INSERT INTO users (name, credits, user_id)
VALUES ('John O''Reilly', 42, 2644bada-852c-11e3-89fb-e0b9a54a6d93)

Note that you should use %s for all types of arguments, not just strings. For example, this would be wrong:

session.execute("INSERT INTO USERS (name, age) VALUES (%s, %d)", ("bob", 42))  # wrong

Instead, use %s for the age placeholder.

If you need to use a literal % character, use %%.

Note: you must always use a sequence for the second argument, even if you are only passing in a single variable:

session.execute("INSERT INTO foo (bar) VALUES (%s)", "blah")  # wrong
session.execute("INSERT INTO foo (bar) VALUES (%s)", ("blah"))  # wrong
session.execute("INSERT INTO foo (bar) VALUES (%s)", ("blah", ))  # right
session.execute("INSERT INTO foo (bar) VALUES (%s)", ["blah"])  # right

Note that the second line is incorrect because in Python, single-element tuples require a comma.

Named place-holders use the %(name)s form:

    INSERT INTO users (name, credits, user_id, username)
    VALUES (%(name)s, %(credits)s, %(user_id)s, %(name)s)
    {'name': "John O'Reilly", 'credits': 42, 'user_id': uuid.uuid1()}

Note that you can repeat placeholders with the same name, such as %(name)s in the above example.

Only data values should be supplied this way. Other items, such as keyspaces, table names, and column names should be set ahead of time (typically using normal string formatting).

Type Conversions

For non-prepared statements, Python types are cast to CQL literals in the following way:

Python Type

CQL Literal Type






  • float
  • double
  • int
  • long
  • int
  • bigint
  • varint
  • smallint
  • tinyint
  • counter



  • str
  • unicode
  • ascii
  • varchar
  • text
  • buffer
  • bytearray








  • list
  • tuple
  • generator


  • set
  • frozenset


  • dict
  • OrderedDict



  • timeuuid
  • uuid

Asynchronous Queries

The driver supports asynchronous query execution through execute_async(). Instead of waiting for the query to complete and returning rows directly, this method almost immediately returns a ResponseFuture object. There are two ways of getting the final result from this object.

The first is by calling result() on it. If the query has not yet completed, this will block until it has and then return the result or raise an Exception if an error occurred. For example:

from cassandra import ReadTimeout

query = "SELECT * FROM users WHERE user_id=%s"
future = session.execute_async(query, [user_id])

# ... do some other work

    rows = future.result()
    user = rows[0]
    print, user.age
except ReadTimeout:
    log.exception("Query timed out:")

This works well for executing many queries concurrently:

# build a list of futures
futures = []
query = "SELECT * FROM users WHERE user_id=%s"
for user_id in ids_to_fetch:
    futures.append(session.execute_async(query, [user_id])

# wait for them to complete and use the results
for future in futures:
    rows = future.result()
    print rows[0].name

Alternatively, instead of calling result(), you can attach callback and errback functions through the add_callback(), add_errback(), and add_callbacks(), methods. If you have used Twisted Python before, this is designed to be a lightweight version of that:

def handle_success(rows):
    user = rows[0]
        process_user(, user.age,
    except Exception:
        log.error("Failed to process user %s",
        # don't re-raise errors in the callback

def handle_error(exception):
    log.error("Failed to fetch user info: %s", exception)

future = session.execute_async(query)
future.add_callbacks(handle_success, handle_error)
There are a few important things to remember when working with callbacks:
  • Exceptions that are raised inside the callback functions will be logged and then ignored.

  • Your callback will be run on the event loop thread, so any long-running operations will prevent other requests from being handled

Setting a Consistency Level

The consistency level used for a query determines how many of the replicas of the data you are interacting with need to respond for the query to be considered a success.

By default, ConsistencyLevel.LOCAL_ONE will be used for all queries. You can specify a different default for the session on Session.default_consistency_level. To specify a different consistency level per request, wrap queries in a SimpleStatement:

from cassandra import ConsistencyLevel
from cassandra.query import SimpleStatement

query = SimpleStatement(
    "INSERT INTO users (name, age) VALUES (%s, %s)",
session.execute(query, ('John', 42))

Prepared Statements

Prepared statements are queries that are parsed by Cassandra and then saved for later use. When the driver uses a prepared statement, it only needs to send the values of parameters to bind. This lowers network traffic and CPU utilization within Cassandra because Cassandra does not have to re-parse the query each time.

To prepare a query, use Session.prepare():

user_lookup_stmt = session.prepare("SELECT * FROM users WHERE user_id=?")

users = []
for user_id in user_ids_to_query:
    user = session.execute(user_lookup_stmt, [user_id])

prepare() returns a PreparedStatement instance which can be used in place of SimpleStatement instances or literal string queries. It is automatically prepared against all nodes, and the driver handles re-preparing against new nodes and restarted nodes when necessary.

Note that the placeholders for prepared statements are ? characters. This is different than for simple, non-prepared statements (although future versions of the driver may use the same placeholders for both).

Setting a Consistency Level with Prepared Statements

To specify a consistency level for prepared statements, you have two options.

The first is to set a default consistency level for every execution of the prepared statement:

from cassandra import ConsistencyLevel

cluster = Cluster()
session = cluster.connect("mykeyspace")
user_lookup_stmt = session.prepare("SELECT * FROM users WHERE user_id=?")
user_lookup_stmt.consistency_level = ConsistencyLevel.QUORUM

# these will both use QUORUM
user1 = session.execute(user_lookup_stmt, [user_id1])[0]
user2 = session.execute(user_lookup_stmt, [user_id2])[0]

The second option is to create a BoundStatement from the PreparedStatement and binding parameters and set a consistency level on that:

# override the QUORUM default
user3_lookup = user_lookup_stmt.bind([user_id3])
user3_lookup.consistency_level = ConsistencyLevel.ALL
user3 = session.execute(user3_lookup)