Feature flags provide a way for .NET and ASP.NET Core applications to turn features on or off dynamically. Developers can use feature flags in simple use cases like conditional statements to more advanced scenarios like conditionally adding routes or MVC filters. Feature flags are built on top of the .NET Core configuration system. Any .NET Core configuration provider is capable of acting as the backbone for feature flags.
Here are some of the benefits of using this library:
- A common convention for feature management
- Low barrier-to-entry
- Built on
IConfiguration - Supports JSON file feature flag setup
- Built on
- Feature Flag lifetime management
- Configuration values can change in real-time, feature flags can be consistent across the entire request
- Simple to Complex Scenarios Covered
- Toggle on/off features through declarative configuration file
- Dynamically evaluate state of feature based on call to server
- API extensions for ASP.NET Core and MVC framework
- Routing
- Filters
- Action Attributes
API Reference: https://go.microsoft.com/fwlink/?linkid=2091700
- Feature Flags
- Consumption
- ASP.NET Core Integration
- Implement a Feature Filter
- Providing a Context For Feature Evaluation
- Built-in Feature Filters
- Targeting
- Variants
- Telemetry
- Caching
- Custom Feature Providers
Feature flags are composed of two parts, a name and a list of feature-filters that are used to turn the feature on.
Feature filters define a scenario for when a feature should be enabled. When a feature is evaluated for whether it is on or off, its list of feature filters is traversed until one of the filters decides the feature should be enabled. At this point the feature is considered enabled and traversal through the feature filters stops. If no feature filter indicates that the feature should be enabled, then it will be considered disabled.
As an example, a Microsoft Edge browser feature filter could be designed. This feature filter would activate any features it is attached to as long as an HTTP request is coming from Microsoft Edge.
The .NET Core configuration system is used to determine the state of feature flags. The foundation of this system is IConfiguration. Any provider for IConfiguration can be used as the feature state provider for the feature flag library. This enables scenarios ranging from appsettings.json to Azure App Configuration and more.
The feature management library supports appsettings.json as a feature flag source since it is a provider for .NET Core's IConfiguration system. Below we have an example of the format used to set up feature flags in a json file.
{
"Logging": {
"LogLevel": {
"Default": "Warning"
}
},
// Define feature flags in a json file
"FeatureManagement": {
"FeatureT": {
"EnabledFor": [
{
"Name": "AlwaysOn"
}
]
},
"FeatureU": {
"EnabledFor": []
},
"FeatureV": {
"EnabledFor": [
{
"Name": "TimeWindow",
"Parameters": {
"Start": "Wed, 01 May 2019 13:59:59 GMT",
"End": "Mon, 01 July 2019 00:00:00 GMT"
}
}
]
}
}
}The FeatureManagement section of the json document is used by convention to load feature flag settings. In the section above, we see that we have provided three different features. Features define their feature filters using the EnabledFor property. In the feature filters for FeatureT we see AlwaysOn. This feature filter is built-in and if specified will always enable the feature. The AlwaysOn feature filter does not require any configuration, so it only has the Name property. FeatureU has no filters in its EnabledFor property and thus will never be enabled. Any functionality that relies on this feature being enabled will not be accessible as long as the feature filters remain empty. However, as soon as a feature filter is added that enables the feature it can begin working. FeatureV specifies a feature filter named TimeWindow. This is an example of a configurable feature filter. We can see in the example that the filter has a Parameters property. This is used to configure the filter. In this case, the start and end times for the feature to be active are configured.
Advanced: The usage of colon ':' in feature flag names is forbidden.
The following snippet demonstrates an alternative way to define a feature that can be used for on/off features.
{
"Logging": {
"LogLevel": {
"Default": "Warning"
}
},
// Define feature flags in config file
"FeatureManagement": {
"FeatureT": true, // On feature
"FeatureX": false // Off feature
}
}The RequirementType property of a feature flag is used to determine if the filters should use Any or All logic when evaluating the state of a feature. If RequirementType is not specified, the default value is Any.
Anymeans only one filter needs to evaluate to true for the feature to be enabled.Allmeans every filter needs to evaluate to true for the feature to be enabled.
A RequirementType of All changes the traversal. First, if there are no filters, the feature will be disabled. Then, the feature-filters are traversed until one of the filters decides that the feature should be disabled. If no filter indicates that the feature should be disabled, then it will be considered enabled.
"FeatureW": {
"RequirementType": "All",
"EnabledFor": [
{
"Name": "TimeWindow",
"Parameters": {
"Start": "Mon, 01 May 2023 13:59:59 GMT",
"End": "Sat, 01 July 2023 00:00:00 GMT"
}
},
{
"Name": "Percentage",
"Parameters": {
"Value": "50"
}
}
]
}
In the above example, FeatureW specifies a RequirementType of All, meaning all of it's filters must evaluate to true for the feature to be enabled. In this case, the feature will be enabled for 50% of users during the specified time window.
Status is an optional property of a feature flag that controls how a flag's enabled state is evaluated. By default, the status of a flag is Conditional, meaning that feature filters should be evaluated to determine if the flag is enabled. If the Status of a flag is set to Disabled then feature filters are not evaluated and the flag is always considered to be disabled.
"FeatureX": {
"Status": "Disabled",
"EnabledFor": [
{
"Name": "AlwaysOn"
}
]
}
In this example, even though the AlwaysOn filter would normally always make the feature enabled, the Status property is set to Disabled, so this feature will always be disabled.
The basic form of feature management is checking if a feature flag is enabled and then performing actions based on the result. This is done through the IFeatureManager's IsEnabledAsync method.
…
IFeatureManager featureManager;
…
if (await featureManager.IsEnabledAsync("FeatureX"))
{
// Do something
}Feature management relies on .NET Core dependency injection. We can register the feature management services using standard conventions.
using Microsoft.FeatureManagement;
public class Startup
{
public void ConfigureServices(IServiceCollection services)
{
services.AddFeatureManagement();
}
}By default, the feature manager retrieves feature flag configuration from the "FeatureManagement" section of the .NET Core configuration data. If the "FeatureManagement" section does not exist, the configuration will be considered empty.
You can also specify that feature flag configuration should be retrieved from a different configuration section by passing the section to AddFeatureManagement. The following example tells the feature manager to read from a different section called "MyFeatureFlags" instead:
services.AddFeatureManagement(configuration.GetSection("MyFeatureFlags"));When using the feature management library with MVC, the IFeatureManager can be obtained through dependency injection.
public class HomeController : Controller
{
private readonly IFeatureManager _featureManager;
public HomeController(IFeatureManager featureManager)
{
_featureManager = featureManager;
}
}The AddFeatureManagement method adds feature management services as singletons within the application, but there are scenarios where it may be necessary for feature management services to be added as scoped services instead. For example, users may want to use feature filters which consume scoped services for context information. In this case, the AddScopedFeatureManagement method should be used instead. This will ensure that feature management services, including feature filters, are added as scoped services.
services.AddScopedFeatureManagement();The feature management library provides functionality in ASP.NET Core and MVC to enable common feature flag scenarios in web applications. These capabilities are available by referencing the Microsoft.FeatureManagement.AspNetCore NuGet package.
MVC controller and actions can require that a given feature, or one of any list of features, be enabled in order to execute. This can be done by using a FeatureGateAttribute, which can be found in the Microsoft.FeatureManagement.Mvc namespace.
[FeatureGate("FeatureX")]
public class HomeController : Controller
{
…
}The HomeController above is gated by "FeatureX". "FeatureX" must be enabled before any action the HomeController contains can be executed.
[FeatureGate("FeatureX")]
public IActionResult Index()
{
return View();
}The Index MVC action above requires "FeatureX" to be enabled before it can be executed.
When an MVC controller or action is blocked because none of the features it specifies are enabled, a registered IDisabledFeaturesHandler will be invoked. By default, a minimalistic handler is registered which returns HTTP 404. This can be overridden using the IFeatureManagementBuilder when registering feature flags.
public interface IDisabledFeaturesHandler
{
Task HandleDisabledFeature(IEnumerable<string> features, ActionExecutingContext context);
}In MVC views <feature> tags can be used to conditionally render content based on whether a feature is enabled or not.
<feature name="FeatureX">
<p>This can only be seen if 'FeatureX' is enabled.</p>
</feature>You can also negate the tag helper evaluation to display content when a feature or set of features are disabled. By setting negate="true" in the example below, the content is only rendered if FeatureX is disabled.
<feature negate="true" name="FeatureX">
<p>This can only be seen if 'FeatureX' is disabled.</p>
</feature>The <feature> tag can reference multiple features by specifying a comma separated list of features in the name attribute.
<feature name="FeatureX,FeatureY">
<p>This can only be seen if 'FeatureX' and 'FeatureY' are enabled.</p>
</feature>By default, all listed features must be enabled for the feature tag to be rendered. This behavior can be overidden by adding the requirement attribute as seen in the example below.
<feature name="FeatureX,FeatureY" requirement="Any">
<p>This can only be seen if either 'FeatureX' or 'FeatureY' or both are enabled.</p>
</feature>The <feature> tag requires a tag helper to work. This can be done by adding the feature management tag helper to the ViewImports.cshtml file.
@addTagHelper *, Microsoft.FeatureManagement.AspNetCoreMVC action filters can be set up to conditionally execute based on the state of a feature. This is done by registering MVC filters in a feature aware manner.
The feature management pipeline supports async MVC Action filters, which implement IAsyncActionFilter.
services.AddMvc(o =>
{
o.Filters.AddForFeature<SomeMvcFilter>("FeatureX");
});The code above adds an MVC filter named SomeMvcFilter. This filter is only triggered within the MVC pipeline if the feature it specifies, "FeatureX", is enabled.
MVC Razor pages can require that a given feature, or one of any list of features, be enabled in order to execute. This can be done by using a FeatureGateAttribute, which can be found in the Microsoft.FeatureManagement.Mvc namespace.
[FeatureGate("FeatureX")]
public class IndexModel : PageModel
{
public void OnGet()
{
}
}The code above sets up a Razor page to require the "FeatureX" to be enabled. If the feature is not enabled, the page will generate an HTTP 404 (NotFound) result.
When used on Razor pages, the FeatureGateAttribute must be placed on the page handler type. It cannot be placed on individual handler methods.
The feature management library can be used to add application branches and middleware that execute conditionally based on feature state.
app.UseMiddlewareForFeature<ThirdPartyMiddleware>("FeatureX");With the above call, the application adds a middleware component that only appears in the request pipeline if the feature "FeatureX" is enabled. If the feature is enabled/disabled during runtime, the middleware pipeline can be changed dynamically.
This builds off the more generic capability to branch the entire application based on a feature.
app.UseForFeature(featureName, appBuilder =>
{
appBuilder.UseMiddleware<T>();
});Creating a feature filter provides a way to enable features based on criteria that you define. To implement a feature filter, the IFeatureFilter interface must be implemented. IFeatureFilter has a single method named EvaluateAsync. When a feature specifies that it can be enabled for a feature filter, the EvaluateAsync method is called. If EvaluateAsync returns true it means the feature should be enabled.
The following snippet demonstrates how to add a customized feature filter MyCriteriaFilter.
services.AddFeatureManagement()
.AddFeatureFilter<MyCriteriaFilter>();Feature filters are registered by calling AddFeatureFilter<T> on the IFeatureManagementBuilder returned from AddFeatureManagement. These feature filters have access to the services that exist within the service collection that was used to add feature flags. Dependency injection can be used to retrieve these services.
Note: When filters are referenced in feature flag settings (e.g. appsettings.json), the Filter part of the type name should be omitted. Please refer to the Filter Alias Attribute section below for more details.
Some feature filters require parameters to decide whether a feature should be turned on or not. For example, a browser feature filter may turn on a feature for a certain set of browsers. It may be desired that Edge and Chrome browsers enable a feature, while Firefox does not. To do this a feature filter can be designed to expect parameters. These parameters would be specified in the feature configuration, and in code would be accessible via the FeatureFilterEvaluationContext parameter of IFeatureFilter.EvaluateAsync.
public class FeatureFilterEvaluationContext
{
/// <summary>
/// The name of the feature being evaluated.
/// </summary>
public string FeatureName { get; set; }
/// <summary>
/// The settings provided for the feature filter to use when evaluating whether the feature should be enabled.
/// </summary>
public IConfiguration Parameters { get; set; }
}FeatureFilterEvaluationContext has a property named Parameters. These parameters represent a raw configuration that the feature filter can use to decide how to evaluate whether the feature should be enabled or not. To use the browser feature filter as an example once again, the filter could use Parameters to extract a set of allowed browsers that would have been specified for the feature and then check if the request is being sent from one of those browsers.
[FilterAlias("Browser")]
public class BrowserFilter : IFeatureFilter
{
…
public Task<bool> EvaluateAsync(FeatureFilterEvaluationContext context)
{
BrowserFilterSettings settings = context.Parameters.Get<BrowserFilterSettings>() ?? new BrowserFilterSettings();
//
// Here we would use the settings and see if the request was sent from any of BrowserFilterSettings.AllowedBrowsers
}
}When a feature filter is registered to be used for a feature flag, the alias used in configuration is the name of the feature filter type with the Filter suffix, if any, removed. For example, MyCriteriaFilter would be referred to as MyCriteria in configuration.
"MyFeature": {
"EnabledFor": [
{
"Name": "MyCriteria"
}
]
}This can be overridden through the use of the FilterAliasAttribute. A feature filter can be decorated with this attribute to declare the name that should be used in configuration to reference this feature filter within a feature flag.
If a feature is configured to be enabled for a specific feature filter and that feature filter hasn't been registered, then an exception will be thrown when the feature is evaluated. The exception can be disabled by using the feature management options.
services.Configure<FeatureManagementOptions>(options =>
{
options.IgnoreMissingFeatureFilters = true;
});Feature filters can evaluate whether a feature should be enabled based on the properties of an HTTP Request. This is performed by inspecting the HTTP Context. A feature filter can get a reference to the HTTP Context by obtaining an IHttpContextAccessor through dependency injection.
public class BrowserFilter : IFeatureFilter
{
private readonly IHttpContextAccessor _httpContextAccessor;
public BrowserFilter(IHttpContextAccessor httpContextAccessor)
{
_httpContextAccessor = httpContextAccessor ?? throw new ArgumentNullException(nameof(httpContextAccessor));
}
}The IHttpContextAccessor must be added to the dependency injection container on startup for it to be available. It can be registered in the IServiceCollection using the following method.
public void ConfigureServices(IServiceCollection services)
{
…
services.TryAddSingleton<IHttpContextAccessor, HttpContextAccessor>();
…
}Advanced: IHttpContextAccessor/HttpContext should not be used in the Razor components of server-side Blazor apps. The recommended approach for passing http context in Blazor apps is to copy the data into a scoped service. For Blazor apps, AddScopedFeatureManagement should be used to register the feature management services.
Please refer to the Scoped Feature Management Services section for more details.
In console applications there is no ambient context such as HttpContext that feature filters can acquire and utilize to check if a feature should be on or off. In this case, applications need to provide an object representing a context into the feature management system for use by feature filters. This is done by using IFeatureManager.IsEnabledAsync<TContext>(string featureName, TContext appContext). The appContext object that is provided to the feature manager can be used by feature filters to evaluate the state of a feature.
MyAppContext context = new MyAppContext
{
AccountId = current.Id;
}
if (await featureManager.IsEnabledAsync(feature, context))
{
…
}Contextual feature filters implement the IContextualFeatureFilter<TContext> interface. These special feature filters can take advantage of the context that is passed in when IFeatureManager.IsEnabledAsync<TContext> is called. The TContext type parameter in IContextualFeatureFilter<TContext> describes what context type the filter is capable of handling. This allows the developer of a contextual feature filter to describe what is required of those who wish to utilize it. Since every type is a descendant of object, a filter that implements IContextualFeatureFilter<object> can be called for any provided context. To illustrate an example of a more specific contextual feature filter, consider a feature that is enabled if an account is in a configured list of enabled accounts.
public interface IAccountContext
{
string AccountId { get; set; }
}
[FilterAlias("AccountId")]
class AccountIdFilter : IContextualFeatureFilter<IAccountContext>
{
public Task<bool> EvaluateAsync(FeatureFilterEvaluationContext featureEvaluationContext, IAccountContext accountId)
{
//
// Evaluate if the feature should be on with the help of the provided IAccountContext
}
}We can see that the AccountIdFilter requires an object that implements IAccountContext to be provided to be able to evalute the state of a feature. When using this feature filter, the caller needs to make sure that the passed in object implements IAccountContext.
Note: Only a single feature filter interface can be implemented by a single type. Trying to add a feature filter that implements more than a single feature filter interface will result in an ArgumentException.
Filters of IFeatureFilter and IContextualFeatureFilter can share the same alias. Specifically, you can have one filter alias shared by 0 or 1 IFeatureFilter and 0 or N IContextualFeatureFilter<ContextType>, so long as there is at most one applicable filter for ContextType.
The following passage describes the process of selecting a filter when contextual and non-contextual filters of the same name are registered in an application.
Let's say you have a non-contextual filter called FilterA and two contextual filters FilterB and FilterC which accept TypeB and TypeC contexts respectively. All three filters share the same alias SharedFilterName.
You also have a feature flag MyFeature which uses the feature filter SharedFilterName in its configuration.
If all of three filters are registered:
- When you call IsEnabledAsync("MyFeature"), the
FilterAwill be used to evaluate the feature flag. - When you call IsEnabledAsync("MyFeature", context), if context's type is
TypeB,FilterBwill be used. If context's type isTypeC,FilterCwill be used. - When you call IsEnabledAsync("MyFeature", context), if context's type is
TypeF,FilterAwill be used.
There a few feature filters that come with the Microsoft.FeatureManagement package: PercentageFilter, TimeWindowFilter, ContextualTargetingFilter and TargetingFilter. All filters, except for the TargetingFilter, are added automatically when feature management is registered. The TargetingFilter is added with the WithTargeting method that is detailed in the Targeting section below.
Each of the built-in feature filters have their own parameters. Here is the list of feature filters along with examples.
This filter provides the capability to enable a feature based on a set percentage.
"EnhancedPipeline": {
"EnabledFor": [
{
"Name": "Microsoft.Percentage",
"Parameters": {
"Value": 50
}
}
]
}This filter provides the capability to enable a feature based on a time window. If only End is specified, the feature will be considered on until that time. If only Start is specified, the feature will be considered on at all points after that time.
"EnhancedPipeline": {
"EnabledFor": [
{
"Name": "Microsoft.TimeWindow",
"Parameters": {
"Start": "Wed, 01 May 2019 13:59:59 GMT",
"End": "Mon, 01 July 2019 00:00:00 GMT"
}
}
]
}This filter provides the capability to enable a feature for a target audience. An in-depth explanation of targeting is explained in the targeting section below. The filter parameters include an audience object which describes users, groups, excluded users/groups, and a default percentage of the user base that should have access to the feature. Each group object that is listed in the target audience must also specify what percentage of the group's members should have access. If a user is specified in the exclusion section, either directly or if the user is in an excluded group, the feature will be disabled. Otherwise, if a user is specified in the users section directly, or if the user is in the included percentage of any of the group rollouts, or if the user falls into the default rollout percentage then that user will have the feature enabled.
"EnhancedPipeline": {
"EnabledFor": [
{
"Name": "Microsoft.Targeting",
"Parameters": {
"Audience": {
"Users": [
"Jeff",
"Alicia"
],
"Groups": [
{
"Name": "Ring0",
"RolloutPercentage": 100
},
{
"Name": "Ring1",
"RolloutPercentage": 50
}
],
"DefaultRolloutPercentage": 20,
"Exclusion": {
"Users": [
"Ross"
],
"Groups": [
"Ring2"
]
}
}
}
}
]
}All of the built-in feature filter alias' are in the 'Microsoft' feature filter namespace. This is to prevent conflicts with other feature filters that may share the same simple alias. The segments of a feature filter namespace are split by the '.' character. A feature filter can be referenced by its fully qualified alias such as 'Microsoft.Percentage' or by the last segment which in the case of 'Microsoft.Percentage' is 'Percentage'.
Targeting is a feature management strategy that enables developers to progressively roll out new features to their user base. The strategy is built on the concept of targeting a set of users known as the target audience. An audience is made up of specific users, groups, excluded users/groups, and a designated percentage of the entire user base. The groups that are included in the audience can be broken down further into percentages of their total members.
The following steps demonstrate an example of a progressive rollout for a new 'Beta' feature:
- Individual users Jeff and Alicia are granted access to the Beta
- Another user, Mark, asks to opt-in and is included.
- Twenty percent of a group known as "Ring1" users are included in the Beta.
- The number of "Ring1" users included in the beta is bumped up to 100 percent.
- Five percent of the user base is included in the beta.
- The rollout percentage is bumped up to 100 percent and the feature is completely rolled out.
This strategy for rolling out a feature is built-in to the library through the included Microsoft.Targeting feature filter.
An example web application that uses the targeting feature filter is available in the FeatureFlagDemo example project.
To begin using the TargetingFilter in an application it must be added to the application's service collection just as any other feature filter. Unlike other built in filters, the TargetingFilter relies on another service to be added to the application's service collection. That service is an ITargetingContextAccessor.
The implementation type used for the ITargetingContextAccessor service must be implemented by the application that is using the targeting filter. Here is an example setting up feature management in a web application to use the TargetingFilter with an implementation of ITargetingContextAccessor called HttpContextTargetingContextAccessor.
services.AddFeatureManagement()
.WithTargeting<HttpContextTargetingContextAccessor>();The targeting context accessor and TargetingFilter are registered by calling WithTargeting<T> on the IFeatureManagementBuilder.
To use the TargetingFilter in a web application, an implementation of ITargetingContextAccessor is required. This is because when a targeting evaluation is being performed, information such as what user is currently being evaluated is needed. This information is known as the targeting context. Different web applications may extract this information from different places. Some common examples of where an application may pull the targeting context are the request's HTTP context or a database.
An example that extracts targeting context information from the application's HTTP context is included in the FeatureFlagDemo example project. This method relies on the use of IHttpContextAccessor which is discussed here.
The targeting filter relies on a targeting context to evaluate whether a feature should be turned on. This targeting context contains information such as what user is currently being evaluated, and what groups the user in. In console applications there is typically no ambient context available to flow this information into the targeting filter, thus it must be passed directly when FeatureManager.IsEnabledAsync is called. This is supported through the use of the ContextualTargetingFilter. Applications that need to float the targeting context into the feature manager should use this instead of the TargetingFilter.
Since ContextualTargetingFilter is an IContextualTargetingFilter<ITargetingContext>, an implementation of ITargetingContext must be passed in to IFeatureManager.IsEnabledAsync for it to be able to evaluate and turn a feature on.
IFeatureManager fm;
…
// userId and groups defined somewhere earlier in application
TargetingContext targetingContext = new TargetingContext
{
UserId = userId,
Groups = groups;
}
await fm.IsEnabledAsync(featureName, targetingContext);The ContextualTargetingFilter still uses the feature filter alias Microsoft.Targeting, so the configuration for this filter is consistent with what is mentioned in that section.
An example that uses the ContextualTargetingFilter in a console application is available in the TargetingConsoleApp example project.
Options are available to customize how targeting evaluation is performed across all features. These options can be configured when setting up feature management.
services.Configure<TargetingEvaluationOptions>(options =>
{
options.IgnoreCase = true;
});When defining an Audience, users and groups can be excluded from the audience. This is useful when a feature is being rolled out to a group of users, but a few users or groups need to be excluded from the rollout. Exclusion is defined by adding a list of users and groups to the Exclusion property of the audience.
"Audience": {
"Users": [
"Jeff",
"Alicia"
],
"Groups": [
{
"Name": "Ring0",
"RolloutPercentage": 100
}
],
"DefaultRolloutPercentage": 0
"Exclusion": {
"Users": [
"Mark"
]
}
}In the above example, the feature will be enabled for users named Jeff and Alicia. It will also be enabled for users in the group named Ring0. However, if the user is named Mark, the feature will be disabled, regardless of if they are in the group Ring0 or not. Exclusions take priority over the rest of the targeting filter.
When new features are added to an application, there may come a time when a feature has multiple different proposed design options. A common solution for deciding on a design is some form of A/B testing, which involves providing a different version of the feature to different segments of the user base and choosing a version based on user interaction. In this library, this functionality is enabled by representing different configurations of a feature with variants.
Variants enable a feature flag to become more than a simple on/off flag. A variant represents a value of a feature flag that can be a string, a number, a boolean, or even a configuration object. A feature flag that declares variants should define under what circumstances each variant should be used, which is covered in greater detail in the Allocating Variants section.
public class Variant
{
/// <summary>
/// The name of the variant.
/// </summary>
public string Name { get; set; }
/// <summary>
/// The configuration of the variant.
/// </summary>
public IConfigurationSection Configuration { get; set; }
}For each feature, a variant can be retrieved using the IVariantFeatureManager's GetVariantAsync method.
…
IVariantFeatureManager featureManager;
…
Variant variant = await featureManager.GetVariantAsync(MyFeatureFlags.FeatureU, CancellationToken.None);
IConfigurationSection variantConfiguration = variant.Configuration;
// Do something with the resulting variant and its configurationOnce a variant is retrieved, the configuration of a variant can be used directly as an IConfigurationSection from the variant's Configuration property. Another option is to bind the configuration to an object using .NET's configuration binding pattern.
IConfigurationSection variantConfiguration = variant.Configuration;
MyFeatureSettings settings = new MyFeatureSettings();
variantConfiguration.Bind(settings);The variant returned is dependent on the user currently being evaluated, and that information is obtained from an instance of TargetingContext. This context can either be passed in when calling GetVariantAsync or it can be automatically retrieved from an implementation of ITargetingContextAccessor if one is registered.
Each variant has two properties: a name and a configuration. The name is used to refer to a specific variant, and the configuration is the value of that variant. The configuration can be set using either the ConfigurationReference or ConfigurationValue properties. ConfigurationReference is a string path that references a section of the current configuration that contains the feature flag declaration. ConfigurationValue is an inline configuration that can be a string, number, boolean, or configuration object. If both are specified, ConfigurationValue is used. If neither are specified, the returned variant's Configuration property will be null.
A list of all possible variants is defined for each feature under the Variants property.
{
"FeatureManagement":
{
"MyFlag":
{
"Variants": [
{
"Name": "Big",
"ConfigurationReference": "ShoppingCart:Big"
},
{
"Name": "Small",
"ConfigurationValue": {
"Size": 300
}
}
],
"EnabledFor": [
{
"Name": "AlwaysOn"
}
]
}
}
}
The process of allocating a feature's variants is determined by the Allocation property of the feature.
"Allocation": {
"DefaultWhenEnabled": "Small",
"DefaultWhenDisabled": "Small",
"User": [
{
"Variant": "Big",
"Users": [
"Marsha"
]
}
],
"Group": [
{
"Variant": "Big",
"Groups": [
"Ring1"
]
}
],
"Percentile": [
{
"Variant": "Big",
"From": 0,
"To": 10
}
],
"Seed": "13973240"
},
"Variants": [
{
"Name": "Big",
"ConfigurationReference": "ShoppingCart:Big"
},
{
"Name": "Small",
"ConfigurationValue": "300px"
}
]
The Allocation setting of a feature flag has the following properties:
| Property | Description |
|---|---|
DefaultWhenDisabled |
Specifies which variant should be used when a variant is requested while the feature is considered disabled. |
DefaultWhenEnabled |
Specifies which variant should be used when a variant is requested while the feature is considered enabled and no other variant was assigned to the user. |
User |
Specifies a variant and a list of users to whom that variant should be assigned. |
Group |
Specifies a variant and a list of groups the current user has to be in for that variant to be assigned. |
Percentile |
Specifies a variant and a percentage range the user's calculated percentage has to fit into for that variant to be assigned. |
Seed |
The value which percentage calculations for Percentile are based on. The percentage calculation for a specific user will be the same across all features if the same Seed value is used. If no Seed is specified, then a default seed is created based on the feature name. |
In the above example, if the feature is not enabled, the feature manager will assign the variant marked as DefaultWhenDisabled to the current user, which is Small in this case.
If the feature is enabled, the feature manager will check the User, Group, and Percentile allocations in that order to assign a variant. For this particular example, if the user being evaluated is named Marsha, in the group named Ring1, or the user happens to fall between the 0 and 10th percentile, then the specified variant is assigned to the user. In this case, all of these would return the Big variant. If none of these allocations match, the user is assigned the DefaultWhenEnabled variant, which is Small.
Allocation logic is similar to the Microsoft.Targeting feature filter, but there are some parameters that are present in targeting that aren't in allocation, and vice versa. The outcomes of targeting and allocation are not related.
You can use variants to override the enabled state of a feature flag. This gives variants an opportunity to extend the evaluation of a feature flag. If a caller is checking whether a flag that has variants is enabled, the feature manager will check if the variant assigned to the current user is set up to override the result. This is done using the optional variant property StatusOverride. By default, this property is set to None, which means the variant doesn't affect whether the flag is considered enabled or disabled. Setting StatusOverride to Enabled allows the variant, when chosen, to override a flag to be enabled. Setting StatusOverride to Disabled provides the opposite functionality, therefore disabling the flag when the variant is chosen. A feature with a Status of Disabled cannot be overridden.
If you are using a feature flag with binary variants, the StatusOverride property can be very helpful. It allows you to continue using APIs like IsEnabledAsync and FeatureGateAttribute in your application, all while benefiting from the new features that come with variants, such as percentile allocation and seed.
"Allocation": {
"Percentile": [{
"Variant": "On",
"From": 10,
"To": 20
}],
"DefaultWhenEnabled": "Off",
"Seed": "Enhanced-Feature-Group"
},
"Variants": [
{
"Name": "On"
},
{
"Name": "Off",
"StatusOverride": "Disabled"
}
],
"EnabledFor": [
{
"Name": "AlwaysOn"
}
]
In the above example, the feature is enabled by the AlwaysOn filter. If the current user is in the calculated percentile range of 10 to 20, then the On variant is returned. Otherwise, the Off variant is returned and because StatusOverride is equal to Disabled, the feature will now be considered disabled.
When a feature flag change is deployed, it is often important to analyze its effect on an application. For example, here are a few questions that may arise:
- Are my flags enabled/disabled as expected?
- Are targeted users getting access to a certain feature as expected?
- Which variant is a particular user seeing?
These types of questions can be answered through the emission and analysis of feature flag evaluation events. This library supports emitting these events through telemetry publishers. One or many telemetry publishers can be registered to publish events whenever feature flags are evaluated.
By default, feature flags will not have telemetry emitted. To publish telemetry for a given feature flag, the flag MUST declare that it is enabled for telemetry emission.
For flags defined in appsettings.json, that is done by using the Telemetry property on feature flags.
{
"FeatureManagement":
{
"MyFlag":
{
"Telemetry": {
"Enabled": true
},
"EnabledFor": [
{
"Name": "AlwaysOn"
}
]
}
}
}
The appsettings snippet above defines a feature flag named MyFlag that is enabled for telemetry. This is indicated by the Telemetry object which sets Enabled to true. The value of the Enabled property must be true to publish telemetry for the flag.
The Telemetry section of a feature flag has the following properties:
| Property | Description |
|---|---|
Enabled |
Specifies whether telemetry should be published for the feature flag. |
Metadata |
A collection of key-value pairs, modeled as a dictionary, that can be used to attach custom metadata about the feature flag to evaluation events. |
Custom handling of feature flag telemetry is made possible by implementing an ITelemetryPublisher and registering it in the feature manager. Whenever a feature flag that has telemetry enabled is evaluated the registered telemetry publisher will get a chance to publish the corresponding evaluation event.
public interface ITelemetryPublisher
{
ValueTask PublishEvent(EvaluationEvent evaluationEvent, CancellationToken cancellationToken);
}The EvaluationEvent type can be found here for reference.
Registering telemetry publishers is done when calling AddFeatureManagement(). Here is an example setting up feature management to emit telemetry with an implementation of ITelemetryPublisher called MyTelemetryPublisher.
builder.services
.AddFeatureManagement()
.AddTelemetryPublisher<MyTelemetryPublisher>();The Microsoft.FeatureManagement.Telemetry.ApplicationInsights package provides a built-in telemetry publisher implementation that sends feature flag evaluation data to Application Insights. To take advantage of this, add a reference to the package and register the Application Insights telemetry publisher as shown below.
builder.services
.AddFeatureManagement()
.AddTelemetryPublisher<ApplicationInsightsTelemetryPublisher>();Note: The base Microsoft.FeatureManagement package does not include this telemetry publisher.
An example of its usage can be found in the EvaluationDataToApplicationInsights example.
This telemetry publisher depends on Application Insights already being setup and registered as an application service. For example, that is done here in the example application.
Feature state is provided by the IConfiguration system. Any caching and dynamic updating is expected to be handled by configuration providers. The feature manager asks IConfiguration for the latest value of a feature's state whenever a feature is checked to be enabled.
There are scenarios which require the state of a feature to remain consistent during the lifetime of a request. The values returned from the standard IFeatureManager may change if the IConfiguration source which it is pulling from is updated during the request. This can be prevented by using IFeatureManagerSnapshot. IFeatureManagerSnapshot can be retrieved in the same manner as IFeatureManager. IFeatureManagerSnapshot implements the interface of IFeatureManager, but it caches the first evaluated state of a feature during a request and will return the same state of a feature during its lifetime.
Implementing a custom feature provider enables developers to pull feature flags from sources such as a database or a feature management service. The included feature provider that is used by default pulls feature flags from .NET Core's configuration system. This allows for features to be defined in an appsettings.json file or in configuration providers like Azure App Configuration. This behavior can be substituted to provide complete control of where feature definitions are read from.
To customize the loading of feature definitions, one must implement the IFeatureDefinitionProvider interface.
public interface IFeatureDefinitionProvider
{
Task<FeatureDefinition> GetFeatureDefinitionAsync(string featureName);
IAsyncEnumerable<FeatureDefinition> GetAllFeatureDefinitionsAsync();
}To use an implementation of IFeatureDefinitionProvider it must be added into the service collection before adding feature management. The following example adds an implementation of IFeatureDefinitionProvider named InMemoryFeatureDefinitionProvider.
services.AddSingleton<IFeatureDefinitionProvider, InMemoryFeatureDefinitionProvider>()
.AddFeatureManagement()This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.
When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.