gitlab-org--gitlab-foss/doc/operations/feature_flags.md

17 KiB

stage group info
Release Progressive Delivery To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#designated-technical-writers

Feature Flags (CORE)

With Feature Flags, you can deploy your application's new features to production in smaller batches. You can toggle a feature on and off to subsets of users, helping you achieve Continuous Delivery. Feature flags help reduce risk, allowing you to do controlled testing, and separate feature delivery from customer launch.

For an example of feature flags in action, see GitLab for Deploys, Feature Flags, and Error Tracking.

NOTE: Note: The Feature Flags GitLab offer as a feature (described in this document) is not the same method used for the development of GitLab.

How it works

GitLab uses Unleash, a feature toggle service.

By enabling or disabling a flag in GitLab, your application can determine which features to enable or disable.

You can create feature flags in GitLab and use the API from your application to get the list of feature flags and their statuses. The application must be configured to communicate with GitLab, so it's up to developers to use a compatible client library and integrate the feature flags in your app.

Create a feature flag

To create and enable a feature flag:

  1. Navigate to your project's Operations > Feature Flags.
  2. Click the New feature flag button.
  3. Enter a name that starts with a letter and contains only lowercase letters, digits, underscores (_), or dashes (-), and does not end with a dash (-) or underscore (_).
  4. Enter a description (optional, 255 characters max).
  5. Enter details about how the flag should be applied:
    • In GitLab 13.0 and earlier, add Environment specs. For each environment, include the Status (default enabled) and Rollout strategy (defaults to All users).
    • In GitLab 13.1 and later, add Feature Flag Strategies. For each strategy, include the Type (defaults to All users) and Environments (defaults to all environments).
  6. Click Create feature flag.

You can change these settings by clicking the {pencil} (edit) button next to any feature flag in the list.

Maximum number of feature flags

Introduced in GitLab 13.5.

The maximum number of feature flags per project on self-managed GitLab instances is 200. On GitLab.com, the maximum number is determined by GitLab.com tier:

Tier Number of feature flags per project
Free 50
Bronze 100
Silver 150
Gold 200

Feature flag strategies

  • Introduced in GitLab 13.0.
  • It was deployed behind a feature flag, disabled by default.
  • It became enabled by default in GitLab 13.2.
  • It's recommended for production use.
  • It's enabled on GitLab.com.
  • For GitLab self-managed instances, a GitLab administrator can choose to disable it. (CORE ONLY)

You can apply a feature flag strategy across multiple environments, without defining the strategy multiple times.

GitLab Feature Flags use Unleash as the feature flag engine. In Unleash, there are strategies for granular feature flag controls. GitLab Feature Flags can have multiple strategies, and the supported strategies are:

Strategies can be added to feature flags when creating a feature flag, or by editing an existing feature flag after creation by navigating to Operations > Feature Flags and clicking {pencil} (edit).

All users

Enables the feature for all users. It uses the default Unleash activation strategy.

Percent Rollout

Introduced in GitLab 13.5.

Enables the feature for a percentage of page views, with configurable consistency of behavior. This consistency is also known as stickiness. It uses the flexibleRollout Unleash activation strategy.

You can configure the consistency to be based on:

  • User IDs: Each user ID has a consistent behavior, ignoring session IDs.
  • Session IDs: Each session ID has a consistent behavior, ignoring user IDs.
  • Random: Consistent behavior is not guaranteed. The feature is enabled for the selected percentage of page views randomly. User IDs and session IDs are ignored.
  • Available ID: Consistent behavior is attempted based on the status of the user:
    • If the user is logged in, make behavior consistent based on user ID.
    • If the user is anonymous, make the behavior consistent based on the session ID.
    • If there is no user ID or session ID, then the feature is enabled for the selected percentage of page view randomly.

For example, set a value of 15% based on Available ID to enable the feature for 15% of page views. For authenticated users this is based on their user ID. For anonymous users with a session ID it would be based on their session ID instead as they do not have a user ID. Then if no session ID is provided, it falls back to random.

The rollout percentage can be from 0% to 100%.

Selecting a consistency based on User IDs functions the same as the percent of Users rollout.

CAUTION: Caution: Selecting Random provides inconsistent application behavior for individual users.

Percent of Users

Enables the feature for a percentage of authenticated users. It uses the Unleash activation strategy gradualRolloutUserId.

For example, set a value of 15% to enable the feature for 15% of authenticated users.

The rollout percentage can be from 0% to 100%.

Stickiness (consistent application behavior for the same user) is guaranteed for logged-in users, but not anonymous users.

Note that percent rollout with a consistency based on User IDs has the same behavior. We recommend using percent rollout because it's more flexible than percent of users

CAUTION: Caution: If the percent of users strategy is selected, then the Unleash client must be given a user ID for the feature to be enabled. See the Ruby example below.

User IDs

Enables the feature for a list of target users. It is implemented using the Unleash userWithId activation strategy.

Enter user IDs as a comma-separated list of values (for example, user@example.com, user2@example.com, or username1,username2,username3, and so on). Note that user IDs are identifiers for your application users. They do not need to be GitLab users.

CAUTION: Caution: The Unleash client must be given a user ID for the feature to be enabled for target users. See the Ruby example below.

User List

Introduced in GitLab 13.1.

Enables the feature for lists of users created in the Feature Flags UI, or with the Feature Flag User List API. Similar to User IDs, it uses the Unleash userWithId activation strategy.

It's not possible to disable a feature for members of a user list, but you can achieve the same effect by enabling a feature for a user list that doesn't contain the excluded users.

For example:

  • Full-user-list = User1A, User1B, User2A, User2B, User3A, User3B, ...
  • Full-user-list-excluding-B-users = User1A, User2A, User3A, ...

Create a user list

Introduced in GitLab 13.3.

To create a user list:

  1. In your project, navigate to Operations > Feature Flags.
  2. Click on New list.
  3. Enter a name for the list.
  4. Click Create.

You can view a list's User IDs by clicking the {pencil} (edit) button next to it. When viewing a list, you can rename it by clicking the Edit button.

Add users to a user list

Introduced in GitLab 13.3.

To add users to a user list:

  1. In your project, navigate to Operations > Feature Flags.
  2. Click on the {pencil} (edit) button next to the list you want to add users to.
  3. Click on Add Users.
  4. Enter the user IDs as a comma-separated list of values. For example, user@example.com, user2@example.com, or username1,username2,username3, and so on.
  5. Click on Add.

Remove users from a user list

Introduced in GitLab 13.3.

To remove users from a user list:

  1. In your project, navigate to Operations > Feature Flags.
  2. Click on the {pencil} (edit) button next to the list you want to change.
  3. Click on the {remove} (remove) button next to the ID you want to remove.

Enable or disable feature flag strategies

This feature is under development, but is ready for production use. It's deployed behind a feature flag that is enabled by default. GitLab administrators with access to the GitLab Rails console can disable it for your instance.

To disable it:

Feature.disable(:feature_flags_new_version)

To enable it:

Feature.enable(:feature_flags_new_version)

Rollout strategy (legacy)

In GitLab 13.0 and earlier, the Rollout strategy setting affects which users will experience the feature as enabled. Choose the percentage of users that the feature will be enabled for. The rollout strategy will have no effect if the environment spec is disabled.

It can be set to:

  • All users
  • Percent of users
    • Optionally, you can click the Include additional user IDs checkbox and add a list of specific users IDs to enable the feature for.
  • User IDs

Disable a feature flag for a specific environment

In GitLab 13.0 and earlier, to disable a feature flag for a specific environment:

  1. Navigate to your project's Operations > Feature Flags.

  2. For the feature flag you want to disable, click the Pencil icon.

  3. To disable the flag:

    • In GitLab 13.0 and earlier: Slide the Status toggle for the environment. Or, to delete the environment spec, on the right, click the Remove (X) icon.
    • In GitLab 13.1 and later: For each strategy it applies to, under Environments, delete the environment.
  4. Click Save changes.

Disable a feature flag for all environments

To disable a feature flag for all environments:

  1. Navigate to your project's Operations > Feature Flags.
  2. For the feature flag you want to disable, slide the Status toggle to Disabled.

The feature flag is displayed on the Disabled tab.

Integrate feature flags with your application

To use feature flags with your application, get access credentials from GitLab. Then prepare your application with a client library.

Get access credentials

To get the access credentials that your application needs to communicate with GitLab:

  1. Navigate to your project's Operations > Feature Flags.
  2. Click the Configure button to view the following:
    • API URL: URL where the client (application) connects to get a list of feature flags.

    • Instance ID: Unique token that authorizes the retrieval of the feature flags.

    • Application name: The name of the environment the application runs in (not the name of the application itself).

      For example, if the application runs for a production server, the Application name could be production or similar. This value is used for the environment spec evaluation.

Note that the meaning of these fields might change over time. For example, we're not sure if Instance ID will be single token or multiple tokens, assigned to the Environment. Also, Application name could describe the application version instead of the running environment.

Choose a client library

GitLab implements a single backend that is compatible with Unleash clients.

With the Unleash client, developers can define, in the application code, the default values for flags. Each feature flag evaluation can express the desired outcome if the flag isn't present in the provided configuration file.

Unleash currently offers many SDKs for various languages and frameworks.

Feature flags API information

For API content, see:

Golang application example

Here's an example of how to integrate feature flags in a Golang application:

package main

import (
    "io"
    "log"
    "net/http"

    "github.com/Unleash/unleash-client-go"
)

type metricsInterface struct {
}

func init() {
    unleash.Initialize(
        unleash.WithUrl("https://gitlab.com/api/v4/feature_flags/unleash/42"),
        unleash.WithInstanceId("29QmjsW6KngPR5JNPMWx"),
        unleash.WithAppName("production"), // Set to the running environment of your application
        unleash.WithListener(&metricsInterface{}),
    )
}

func helloServer(w http.ResponseWriter, req *http.Request) {
    if unleash.IsEnabled("my_feature_name") {
        io.WriteString(w, "Feature enabled\n")
    } else {
        io.WriteString(w, "hello, world!\n")
    }
}

func main() {
    http.HandleFunc("/", helloServer)
    log.Fatal(http.ListenAndServe(":8080", nil))
}

Ruby application example

Here's an example of how to integrate feature flags in a Ruby application.

The Unleash client is given a user ID for use with a Percent rollout (logged in users) rollout strategy or a list of Target Users.

#!/usr/bin/env ruby

require 'unleash'
require 'unleash/context'

unleash = Unleash::Client.new({
  url: 'http://gitlab.com/api/v4/feature_flags/unleash/42',
  app_name: 'production', # Set to the running environment of your application
  instance_id: '29QmjsW6KngPR5JNPMWx'
})

unleash_context = Unleash::Context.new
# Replace "123" with the ID of an authenticated user.
# Note that the context's user ID must be a string:
# https://unleash.github.io/docs/unleash_context
unleash_context.user_id = "123"

if unleash.is_enabled?("my_feature_name", unleash_context)
  puts "Feature enabled"
else
  puts "hello, world!"
end

You can link related issues to a feature flag. In the Linked issues section, click the + button and input the issue reference number or the full URL of the issue.

This feature is similar to the related issues feature.