Dev guideAPI Reference
Dev guideAPI ReferenceUser GuideGitHubNuGetDev CommunitySubmit a ticketLog In
GitHubNuGetDev CommunitySubmit a ticket

Event batching for the Python SDK

How the Optimizely Feature Experimentation Python SDK uses the event processor to batch impressions and conversion events into a single payload before sending it to Optimizely.

The Optimizely Feature Experimentation Python SDK lets you batch events and includes options to set a maximum batch size and flush interval timeout. The benefit of event batching means less network traffic for the same number of decision and conversion events tracked.



By default, event batching is disabled in Python SDK 3.3.0.

Event batching works with both out-of-the-box and custom event dispatchers.

Make sure that you are not sending personally identifiable information (PII) to Optimizely Feature Experimentation. The event batching process does not remove PII from events.

Configure event batching

Event batching can be enabled through the usage of BatchEventProcessor. Optimizely provides two main options to configure event batching: batch_size and flush_interval. You can pass in both these options when creating instance of BatchEventProcessor and pass the created instance during Optimizely client creation. When using BatchEventProcessor, events are held in a queue until either:

  • The number of events reaches the defined batch_size.
  • The oldest event has been in the queue for longer than the defined flush_interval, which is specified in seconds. The queue is then flushed and all queued events are sent to Optimizely Feature Experimentation in a single network request.
  • A new datafile revision is received. This occurs only when live datafile updates are enabled.
from optimizely import optimizely
from optimizely import event_dispatcher as optimizely_event_dispatcher
from optimizely.event import event_processor

# Set event dispatcher that the 
# You can reference your own implementation of event dispatcher here
event_dispatcher = optimizely_event_dispatcher.EventDispatcher

# Create instance of BatchEventProcessor.
# In this example here we set batch size to 15 events
# and flush interval to 50 seconds. 
# Setting start_on_init starts the consumer
# thread to start receiving events.
# See table below for explanation of these
# and other configuration options.  
batch_processor = event_processor.BatchEventProcessor(

# Create Optimizely client and pass in instance 
# of BatchEventProcessor to enable batching.
optimizely_client = optimizely.Optimizely(
    sdk_key='<Your SDK Key here>', 
    datafile='<Your datafile here>',

The table below defines these and other options that you can use to configure the BatchEventProcessor.

NameDefault valueDescription
event_dispatcher No default value. Required parameter.An event handler to manage network calls.

Provides a dispatch_event method that takes in URL and parameters and dispatches request.
NoOpLoggerA logger implementation to log issues.
30The maximum duration in seconds that an event can exist in the queue before being flushed.
10The maximum number of events to hold in the queue. Once this number is reached, all queued events are flushed and sent to Optimizely Feature Experimentation.
FalseBoolean, which, if set to true starts the thread consuming and queuing events on initializing the BatchEventProcessor.

By default, the value is False, so the consumer thread is not ready to receive any events. One can always start the consumer thread by calling start on the instance of BatchEventProcessor
5The number representing time interval in seconds before joining the consumer thread.
six.moves.queue.QueueComponent that allows you to accumulate events until they are dispatched.
NotificationCenterInstance of optimizely.notification_center.NotificationCenter.

By default an instance of notification_center.NotificationCenter is created.

For more information, see Initialize the Python SDK.



The maximum payload size is 3.5 MB. Optimizely rejects requests with a 400 response code, Bad Request Error, if the batch payload exceeds this limit.

The size limitation is because of the Optimizely Events API, which Feature Experimentation uses to send data to Optimizely.

The most common cause of a large payload size is a high batch size. If your payloads exceed the size limit, try configuring a smaller batch size.

Side effects

The table lists other Optimizely Feature Experimentation functionality that may be triggered by using this class.

LogEventWhenever the event processor produces a batch of events, a LogEvent object will be created using the EventFactory.

It contains batch of conversion and decision events.

This object will be dispatched using the provided event dispatcher and also it will be sent to the notification subscribers.
Notification ListenersFlush invokes the LOG_EVENT notification listener if this listener is subscribed to.

Registering LogEvent listener

To register a LogEvent notification listener:

def on_log_event():

optimizely_client.notification_center.add_notification_listener(enums.NotificationTypes.LOG_EVENT, on_log_event)


LogEvent object gets created using EventFactory. It represents the batch of decision and conversion events we send to the Optimizely Feature Experimentation backend.

Required (non null)
StringThe HTTP verb to use when dispatching the log event. It can be GET or POST.
Required (non null)
StringURL to dispatch log event to.
Required (non null)
DictEvent Batch. It contains all the information regarding every event which is batched. including list of visitors which contains UserEvent.
headersDictRequest Headers

Stop BatchEventProcessor on application exit

If you enable event batching, you must call the 'stop' method, batch_processor.stop(), before exiting. This ensures that queued events are flushed as soon as possible to avoid data loss.



Because the BatchEventProcessor maintains a buffer of queued events, you must call stop() on the BatchEventProcessor instance before shutting down your application.

stop()Stops and flushes the event queue.

Note: Optimizely recommends that you connect this method to a kill signal for the running process.