Cassandra 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 if the cluster is configured in legacy mode (not using execution profiles). Otherwise this can be done by setting the ExecutionProfile.consistency_level for the execution profile with key EXEC_PROFILE_DEFAULT. To specify a different consistency level per request, wrap queries in a SimpleStatement:Copy

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():Copy

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:Copy

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:Copy

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


Spcifies how many replicas must respond for an operation to be considered a success. By default, ONE is used for all operations.



= 0

Only requires that one replica receives the write or the coordinator stores a hint to replay later. Valid only for writes.


= 1

Only one replica needs to respond to consider the operation a success


= 2

Two replicas must respond to consider the operation a success


= 3

Three replicas must respond to consider the operation a success


= 4

ceil(RF/2) replicas must respond to consider the operation a success


= 5

All replicas must respond to consider the operation a success


= 6

Requires a quorum of replicas in the local datacenter


= 7

Requires a quorum of replicas in each datacenter


= 8

For conditional inserts/updates that utilize Cassandra’s lightweight transactions, this requires consensus among all replicas for the modified data.


= 9

Like SERIAL, but only requires consensus among replicas in the local datacenter.


= 10

Sends a request only to replicas in the local datacenter and waits for one response.