No results for ""
EXPAND ALL
  • Home
  • API docs

GIVE DOCS FEEDBACK

Python SDK 7.x to 8.0 migration guide

Read time: 9 minutes
Last edited: Mar 11, 2024

Overview

This topic explains the changes in the Python SDK 8.0 release and how to adapt code that uses a 7.x version of the Python SDK to use version 8.0 or later.

Version 8.0 includes several breaking changes. Additionally, if you use the Relay Proxy, you must update your Relay Proxy to version 7.0 before you update your SDK to version 8.0. To learn more, read the Relay Proxy 7.0 release notes. To upgrade to the latest Relay Proxy version, visit Relay Proxy releases on GitHub.

Understanding contexts

Many LaunchDarkly customers create targeting rules for feature flags based on a variety of different information, including attributes pertaining to users, organizations, devices, and more. In previous versions of the LaunchDarkly SDK, you could define this information in a user object, using a combination of built-in and custom attributes. Now you can define this information in a more structured way, using contexts.

Each context has a required attribute called kind that you can use to categorize context instances for targeting and Experimentation. You can also add other attributes. Attributes can be strings, booleans, numbers, arrays, or JSON objects.

When you evaluate a feature flag within your application, the flag's targeting rules use information from one or more kinds of contexts. For example, you may know:

  • the username, first name, last name, and email address of a person, as part of a context with kind of "user"
  • the company, department, and location of an organization, as part of a context with kind of "organization"
  • the device, model, and operating system of an environment, as part of a context with kind of "device"

This new version of the LaunchDarkly SDK requires you to evaluate feature flags using an evaluation context, which is an object containing one or more contexts.

To learn more about contexts, read Contexts.

Migrating from users to contexts

The 8.0 version of this SDK lets you use contexts. When you migrate from version 7.x, replace every instance of a user dictionary with a Context object . If there are any instances you do not replace, the 8.0 version of the Python SDK will convert each user dictionary it receives to a Context.

LaunchDarkly assumes older versions of the SDK use user contexts

A context always has a kind attribute. When older versions of the SDK send events to LaunchDarkly, LaunchDarkly will convert the users in those events to contexts with a kind of user.

If a flag configuration specifies any context kinds other than user, older versions of the Python SDK will not evaluate the flag correctly. You must upgrade your SDK if you are going to use context kinds other than user in your flag configurations.

The primary differences between working with users and working with contexts include the following:

  • Changes to flag evaluation: The methods for evaluating flags now require contexts, rather than users.
  • Create contexts, not users: Where you previously created users, now you must create contexts.
  • Changes to attributes: There are now fewer built-in attributes. You can still add as many custom attributes as you like, although the format has changed slightly. A flag's targeting rules can now address fields within a JSON object.
  • Changes to private attributes: You can mark specific attributes of a context as private, either across all contexts of any kind, or within a particular context or context kind.
  • Changes to alias events: The alias method has been removed.

To learn more about upgrading to contexts, read Best practices for upgrading users to contexts.

Understanding changes to flag evaluation

The methods for evaluating flags and determining flag evaluation reasons have changed slightly. The 8.0 version of the SDK includes the following changes:

  • The variation and variation_detail methods now take a context, rather than a user, as a parameter. To learn more, read the following API docs:

  • The USER_NOT_SPECIFIED evaluation error code was previously defined as, the user object or user key was not provided. It has been redefined to mean that the context was not provided or was invalid.

Here's how to evaluate a flag using a context:

flag_value = client.variation("flag-key-123abc", context, false)

Understanding differences between users and contexts

Where you previously created users, now you can create contexts.

The new Context type represents a context. There is no special type for users in the Python SDK. Users are still represented by a dictionary object.

Here's how to construct a basic context, as compared with constructing a user:

user = {"key": "user-key-123abc"}

Here's how to construct a basic context, with a context kind of something other than "user":

# use Context.create when including only the key and kind
context1 = Context.create("org-key-123abc", "organization")

Here's how to construct a multi-context, which includes multiple context kinds:

multi_context = Context.create_multi(
Context.create("user-key-123abc"),
Context.create("device-key-123abc", "device")
)

Understanding changes to built-in and custom attributes

This section describes the changes to built-in and custom attributes in the 8.0 version of the SDK.

Working with built-in and custom attributes



In previous SDK versions, the user object included several built-in attributes for describing the user. It also included optional custom attributes, which you could add to a custom object within the user object and then populate.

In version 8.0, the only built-in attributes are are kind, key, name, and anonymous. Kind, key, and name are strings, and anonymous is a boolean.

You can define additional attributes for a context by passing in a name and value for each. Additional attributes can be any JSON type, including boolean, number, string, array, or object. In version 8.0, you do not need to add custom attributes within a custom object.

Here's how to construct a context with additional attributes, as compared with constructing a similar user:

user = {
"key": "user-key-123abc",
"name": "Sandy",
"email": "sandy@example.com",
"custom": {
"groups": ["admin"]
}
}

Referencing properties of an attribute object



In previous versions of the SDK, if you set the value of a user's custom attribute to a JSON object, you could not reference that object in evaluations. In version 8.0, if a context attribute's value is a JSON object, you can reference properties of that object as the attribute in the targeting rules for a flag or segment.

Here's how to add object attributes to a context. A JSON object is represented as a dictionary in Python.

address_data = {"street": "Main St", "city": "Springfield"}
context = Context.builder("user-key-123abc")
.set("address", address_data)
.build()

In your flag or segment targeting, use / as the delimiter to refer to specific object fields. For example, you can use /address/city in your targeting. To learn more, read Target with flags.

Removing the secondary attribute



In previous versions of the SDK, you could set the value of a user's secondary attribute, as an optional secondary key for a user. The SDK would incorporate this attribute into the variation bucket assignment hash.

In version 8.0, the secondary attribute has been removed. If you were previously using this attribute as part of distinguishing percentage rollouts, that will no longer work for your users.

Understanding changes to private attributes

As in previous versions of the SDK, you can mark specific attributes of a context as private. This restricts the context data your application sends to LaunchDarkly, while still using that data for flag targeting.

In version 8.0, there are two scopes for which you can mark attributes as private:

  • Across all contexts of any context kind. You might use this if you want to ensure that the SDK never stores an "email" attribute in LaunchDarkly, no matter whether it occurs in a user context, an organization context, or something else.
  • Within a particular context or context kind. You might use this if you want an "email" attribute to be private in a user context, but not in an organization context.

In the first example, all attributes are marked private for all contexts. Only the context key and kind are sent to LaunchDarkly. In the second example, the "email" and "address" attributes are private for all contexts:

config = Config(all_attributes_private=True)

To learn more, read Config.all_attributes_private) Config.private_attributes).

Here's how to mark an attribute as private for a particular context:

user = {
"key": "user-key-123abc",
"name": "Sandy",
"email": "sandy@example.com",
"privateAttributeNames": ["email"]
}

For attributes that are objects, you can mark specific fields private, using the / delimiter followed by the attribute name, then the / delimiter followed by the JSON property within the value.

For example, for the attribute "address": { "street": "Main St", "city": "Springfield" }, you could set just the /address/street as private.

Understanding changes to alias events

In previous versions of the SDK, multiple user objects could represent one person. For example, this could happen the first time a person logged in to your application. The person might be represented by an anonymous user before they logged in, and a different user after they logged in. You could associate these two LaunchDarkly users by sending an alias event in the SDK.

With the introduction of contexts, the person in this scenario is represented by two different context kinds. For example, before they log in, they might be represented by a device context. After they log in, they might be represented by a multi-context, for example, by one context kind based on their device and simultaneously by another context kind based on their user information.

The 8.0 version of the SDK removes the ability to send an alias event. If you currently alias users, you will need to remove this code when you migrate to version 8.0.

If you want to continue associating two contexts with each other, you can use two different context kinds, and then identify a multi-context that includes both individual contexts when you want the association to occur. Unlike the aliasing method, the association doesn't persist between calls. You must send the contexts you want to associate in each variation or identify call and each track call.

Here's how:

context1 = Context.create("user-key-123abc")
context2 = Context.create("device-key123abc", "device")
multi_context = Context.create_multi(context1, context2)
client.identify(multi_context)

To learn more, read Context.

Understanding changes to configuration options

In the 8.0 version of the SDK, several configuration options have changed:

  • The inline_users_in_events option has been removed.
  • The user_keys_capacity and user_keys_flush_interval options have been renamed context_keys_capacity and context_keys_flush_)interval, respectively.

To learn more, read Config.

Understanding what was deprecated

All types and methods that were marked as deprecated in the last 7.x release have been removed from the 8.0 release. If you were using these with a recent version previously, you should already have received deprecation warnings at compile time, with suggestions about their recommended replacements.

For a full list of deprecated modules, read the Deprecated modules section of the 7.x SDK API documentation.