Feature Flags

Feature flags are declared in Sentry's codebase. For self-hosted users, those flags are then configured via sentry.conf.py. For Sentry's SaaS deployment, Flagr is used to configure flags in production.

You can find a list of features available by looking at sentry/features/__init__.py. They're declared on the FeatureManager like so:

Copied
default_manager.add("organizations:onboarding", OrganizationFeature)  # NOQA

The feature can be enabled with the following in your sentry.conf.py, usually located at ~/.sentry/:

Copied
SENTRY_FEATURES["organizations:onboarding"] = True

You can modify the state of feature flags in tests using a context manager.

Generally you want your feature names to be unique to help in their removal. For example a feature flag like trends may prove difficult to find because trends may appear throughout the codebase. But a name like performance-trends-view is more likely to be unique and easier to remove later

Creating a new Feature Flag

Determine what scope the feature should have

Features can be scoped by organization, and projects. If you're not confident you want a project feature, create an organization level one. In this example we'll build a feature called test-feature scoped at the organization level.

Add your feature to server.py

conf/server.py contains many of the default settings in the application. Here you will add your feature, and decide what default value it should hold unless specified by the user.

The SENTRY_FEATURES dictionary contains all the features in the application with their corresponding scope. Your feature should start off disabled by default:

Copied
SENTRY_FEATURES = {
    'organizations:test-feature': False,
    'auth:register': True,
    # ...
    'projects:minidump': False,
}

Add your feature to the FeatureManager

The FeatureManager handles the application features. We add all the features to the FeatureManager, including the type of feature we want to add to the file /src/sentry/features/__init__.py.

In our case, it is

Copied
default_manager.add('organizations:test-feature', OrganizationFeature)

Add it to the Organization Model Serializer

The Organization model serializer (src/sentry/api/serializers/models/organization.py) builds a list called feature_list that is given to the front-end to use. By default the all features are checked and those that are present are added into the list. If your feature requires additional custom logic you will have to update the organization serializer

Using Model Flags (Less common)

Sometimes a flag on the model is used to indicate a feature flag as shown below. This is not recommended unless there is a specific reason to make changes to the model. For example, the require_2fa flag affects behavior on the backend to enforce two-factor authentication.

Copied
feature_list = []

if getattr(obj.flags, 'allow_joinleave'):
    feature_list.append('open-membership')
if not getattr(obj.flags, 'disable_shared_issues'):
    feature_list.append('shared-issues')
if getattr(obj.flags, 'require_2fa'):
    feature_list.append('require-2fa')

Checking your feature

In Python code

The FeatureManager's has method checks see if the feature exists. The has method takes in the feature's name, the object that corresponds to the scope of the feature (i.e. an organization for an organization level feature or a project for a project level feature), and the actor (aka user). In our case the feature will be added like:

Copied
if features.has('organizations:test-feature', obj, actor=user):
    feature_list.append('test-feature')

which only adds the feature to the feature_list if that feature is enabled for the organization and the type of user given. Note that when we give the feature to the frontend, we remove the scope prefix, and our 'organizations:test-feature' becomes 'test-feature'.

In JavaScript

There is a difference between using the flag in Sentry and in GetSentry. At this stage you're not quite ready to use your feature flag in GetSentry, but you are able to use it inside Sentry.

Declarative features with the Feature component

React uses a declarative programming paradigm. As such, we have a utility component that we use to hide components based on the feature flags available to a organization/project

Copied
import Feature from 'app/components/acl/feature';

const toRender = (
  <Feature features={['test-feature']}>
    <MyComponentToFlag />
  </Feature>
);

Imperative feature flag checks

There are some exceptions when React components are generated imperatively (e.g. headers/columns for Tables). In difficult times like these, the Organization / Project object has a array of the feature flags, which you can use in this way:

Copied
const {organization} = this.props;

// Method 2
organization.features.includes('test-feature'); // evals to True/False

Enabling features in development

In Sentry you can run sentry devserver to view your changes in development mode. If you would like to view a change behind a feature flag, you will need to open the file ~/.sentry/sentry.config.py on your local machine. This file contains your local settings for the sentry application, and can be viewed and edited. If you would like to toggle a flag on or off, add this to your configuration file:

Copied
SENTRY_FEATURES['organizations:test-feature'] = True

Where SENTRY_FEATURES will correspond to the SENTRY_FEATURES from step 2. Set it to True if you'd like the feature to be available and False if not.

Flagr in development

In general, you do not need to run flagr in development to test your feature flagging. If you do want to run flagr, you'll need to be running getsentry:

  1. Set the environment variable: export SENTRY_USE_FLAGR=true
  2. Start your devservices

Your local instance of flagr can be found at localhost:18000

Enabling your feature in production

Feature flags are declared in Sentry's codebase. For self-hosted users, those flags are then configured via sentry.conf.py. For Sentry's SaaS deployment, Flagr is used to configure flags in production.

First, add your feature to the flagr allowlist: in getsentry/features.py you will find the Flagr feature handler (FlagrFeatureHandler) then add your feature to the allowed_features property.

Copied
allowed_features = set(
    [
        ...
        "organizations:test-feature",
    ]
)

Building your feature in Flagr

Navigate to the Flagr UI and in the input box enter your feature name as the flag's description. Next, click the arrow on the right of the Create New Flag button and select Create Simple Boolean Flag. The flag's description is only used for easy finding of your flag on the Flagr homepage, but it is the Flag Key that the API uses to find your flag.

img/flagrNewFlag.png

You should then see your newly created flag in the list below, click your flag and enter your feature name as the flag key, optionally enter notes about your flag here, then click Save Flag. At this point, both your flag key and description should be your new feature. It is very important that your flag key matches exactly what is in sentry as that is what is used to look your feature up.

Each flag defaults to disabled, and your feature will not appear until you enable it, this is done via the toggle at the top right of the flag page.

img/flagrDescription.png

In Flagr your feature needs a Variant and at least one Segment.

Variant

Variants represent the possible variations of a flag. Because flagr is currently only used to toggle features, the only supported variant is on, and this value is case-insensitive.

Segment

Segments represent the portions of our audience that your feature will be enabled for. Each segment can have multiple conditions all of which must be matched for a feature to be enabled.

Distribution

Represents the distribution of variants in a segment, because we'll only have one variant this value should always be 100% for each segment.

Creating a segment constraint

When creating a segment, without the distribution set, Flagr will respond as if the segment doesn't exist yet. This means that if you're creating or modifying a segment do not set the Distribution until you're ready for your feature to be enabled.

Here are the properties that are being sent to Flagr as well as the types of their values.

Each constraint requires a Property, Operator, and Value. Each segment can contain multiple constraints, all of which must be matched. A way to think about segments is that each segment is ORed together, and each constraint is ANDed together

Properties

Properties are a preset dictionary that get sent to flagr by sentry, these depend on the organization or project being passed to features.has

PropertyValue TypesNotes/Example config
organization_idint
organization_slugstringExample
organization_is-early-adopterboolUsers opt in from settings, example
features[string]All active features for org
subscription_is-freebool
subscription_is-trial-planbool
subscription_planstringExample
subscription_channelint
project_slugstringOnly passed for projects
project_idintOnly passed for projects
user_idint
user_domainstring
user_is-superuserbool
user_is-staffbool
team_slugs[string]Use with an organization condition, example

More properties can be added by updating FlagrFeatureHandler.get_context but be considerate of the performance of additional context. Properties cannot contain periods . as this is a special character used by flagr, in general use an underscore _ instead.

Operators

Operators are generally straightforward, for example, == means equal. There is the exception however of CONTAINS and IN which only operate on arrays, This means that if you want to check that for a substring you want to use the =~ operator.

Copied
property CONTAINS "foo"    =======> {"property": ["foo", "bar"]} -> True
                                    {"property": "foobar"}       -> False

property IN ["foo", "bar"] =======> {"property": ["foo", "bar"]} -> False
                                    {"property": "foo"}          -> True

property =~ "foo"          =======> {"property": ["foo", "bar"]} -> False
                                    {"property": "foobar"}       -> True

Values

  • Values that are strings must be wrapped in double-quotes ""
  • Values that are arrays must be wrapped in square brackets []
  • Values that are bools or ints should not be wrapped in quotes, for example, true or 1

Saving

New constraints must be saved by clicking Add Constraint, existing constraints are only updated after clicking the green Save button, the Save Segment Settings button does not include constraints.

Sample Flagr Configurations

Here are some example flagr configurations for methods to release a feature to different audiences

Releasing a feature to a subset of organizations

You can enable a feature for a specific list of organization slugs by setting the condition property to organization_slug, the operator to IN, and the value to an array of organizations, using square brackets [] for the array, and double quotes " for each organization.

img/flagrOrganizationSubset.png

Releasing a feature to a specific team

You can enable a feature for only a specific team within an organization by setting the condition property to team_slugs, the operator to CONTAINS, and the value to the slug of the team wrapped in double quotes ". Also include a condition in the same segment to match organization, with either organization_slug or organization_id. If this isn't included, any org that has the same team name will get the feature.

img/flagrTeamSubset.png

Releasing a feature to Early Adopters

You can enable a feature to organizations that have opted in to be early adopters. To do this set the constraint property to organization_is-early-adopter, the operator to ==, and the value to true.

img/flagrEarlyAdopterSubset.png

Releasing a feature to organizations with specific plans

Enabling a feature for customers on specific plans can be done by creating a segment with the property set to subscription_plan, the operator to IN, and the value to an array of plans

img/flagrPlanSubset.png

Releasing to organizations incrementally

When releasing a large or potentially disruptive feature you may want to enable it for a percentage of organizations incrementally. For example, on Monday it is available to 10% of users, and increasing the percentage of customers by 10% each day.

When creating your segment set the rollout rate to 0, then on each day that you increase it, enter a larger value, and click Save Segment Setting.

Note: You should not need to modify the distribution to incrementally release your feature.

After launch (Graduation)

After your feature has been mainlined and is available for all hosted customers:

  • If the feature cannot be disabled, or if no application specific overrides are required, remove the feature flag and all related checks from the Sentry code base. If necessary, also remove references to the feature from the onpremise and getsentry repositories.
  • If the feature flag controls a behavior specific to Sentry SaaS, leave it disabled in Sentry and add appropriate feature handlers in getsentry (see below). In this case, keep in mind that the feature is also disabled during development.
  • If the feature is generally available but is constrained to specific organizations or projects on Sentry SaaS, enable the feature by default in in conf/server.py and add appropriate feature handlers in getsentry. This ensures that the feature is enabled in development and self-hosted Sentry installations.
  • If the feature was enabled through flagr, remove it from FlagrFeatureHandler.allowed_features, and delete the feature from the Flagr UI as well this is done by navigating to the flag configuration page then clicking Delete Flag at the bottom of the page.

To override features in getsentry, follow these steps:

  1. Disable the feature in getsentry/conf/settings/defaults.py by updating SENTRY_FEATURES.
  2. Add a new feature handler class in getsentry/features.py that determines availability of the feature based on the organization or actor.
  3. Register the handler at the bottom of getsentry/features.py.
You can edit this page on GitHub.