Execution Profiles

Execution profiles aim at making it easier to execute requests in different ways within a single connected Session. Execution profiles are being introduced to deal with the exploding number of configuration options, especially as the database platform evolves more complex workloads.

The legacy configuration remains intact, but legacy and Execution Profile APIs cannot be used simultaneously on the same client Cluster. Legacy configuration will be removed in the next major release (4.0).

An execution profile and its parameters should be unique across Cluster instances. For example, an execution profile and its LoadBalancingPolicy should not be applied to more than one Cluster instance.

This document explains how Execution Profiles relate to existing settings, and shows how to use the new profiles for request execution.

Mapping Legacy Parameters to Profiles

Execution profiles can inherit from cluster.ExecutionProfile, and currently provide the following options, previously input from the noted attributes:

When using the new API, these parameters can be defined by instances of cluster.ExecutionProfile.

Using Execution Profiles

Default 

from cassandra.cluster import Cluster
cluster = Cluster()
session = cluster.connect()
local_query = 'SELECT rpc_address FROM system.local'
for _ in cluster.metadata.all_hosts():
    print session.execute(local_query)[0]

Row(rpc_address='127.0.0.2')
Row(rpc_address='127.0.0.1')

The default execution profile is built from Cluster parameters and default Session attributes. This profile matches existing default parameters.

Initializing cluster with profiles 

from cassandra.cluster import ExecutionProfile
from cassandra.policies import WhiteListRoundRobinPolicy

node1_profile = ExecutionProfile(load_balancing_policy=WhiteListRoundRobinPolicy(['127.0.0.1']))
node2_profile = ExecutionProfile(load_balancing_policy=WhiteListRoundRobinPolicy(['127.0.0.2']))

profiles = {'node1': node1_profile, 'node2': node2_profile}
session = Cluster(execution_profiles=profiles).connect()
for _ in cluster.metadata.all_hosts():
    print session.execute(local_query, execution_profile='node1')[0]

Row(rpc_address='127.0.0.1')
Row(rpc_address='127.0.0.1')

for _ in cluster.metadata.all_hosts():
    print session.execute(local_query, execution_profile='node2')[0]

Row(rpc_address='127.0.0.2')
Row(rpc_address='127.0.0.2')

for _ in cluster.metadata.all_hosts():
    print session.execute(local_query)[0]

Row(rpc_address='127.0.0.2')
Row(rpc_address='127.0.0.1')

Note that, even when custom profiles are injected, the default TokenAwarePolicy(DCAwareRoundRobinPolicy()) is still present. To override the default, specify a policy with the EXEC_PROFILE_DEFAULT key.

from cassandra.cluster import EXEC_PROFILE_DEFAULT
profile = ExecutionProfile(request_timeout=30)
cluster = Cluster(execution_profiles={EXEC_PROFILE_DEFAULT: profile})

Adding named profiles

New profiles can be added constructing from scratch, or deriving from default:

locked_execution = ExecutionProfile(load_balancing_policy=WhiteListRoundRobinPolicy(['127.0.0.1']))
node1_profile = 'node1_whitelist'
cluster.add_execution_profile(node1_profile, locked_execution)

for _ in cluster.metadata.all_hosts():
    print session.execute(local_query, execution_profile=node1_profile)[0]

Row(rpc_address='127.0.0.1')
Row(rpc_address='127.0.0.1')

See Cluster.add_execution_profile() for details and optional parameters.

Passing a profile instance without mapping

We also have the ability to pass profile instances to be used for execution, but not added to the mapping:

from cassandra.query import tuple_factory

tmp = session.execution_profile_clone_update('node1', request_timeout=100, row_factory=tuple_factory)

print session.execute(local_query, execution_profile=tmp)[0]
print session.execute(local_query, execution_profile='node1')[0]

('127.0.0.1',)
Row(rpc_address='127.0.0.1')

The new profile is a shallow copy, so the tmp profile shares a load balancing policy with one managed by the cluster. If reference objects are to be updated in the clone, one would typically set those attributes to a new instance.