Skip to main content

3. Evaluation Context

experimental

Overview

The evaluation context provides ambient information for the purposes of flag evaluation. Contextual data may be used as the basis for targeting, including rule-based evaluation, overrides for specific subjects, or fractional flag evaluation.

The context might contain information about the end-user, the application, the host, or any other ambient data that might be useful in flag evaluation. For example, a flag system might define rules that return a specific value based on the user's email address, locale, or the time of day. The context provides this information. The context can be optionally provided at evaluation, and mutated in before hooks.

3.1 Fields

Note

Field casing is not specified, and should be chosen in accordance with language idioms.

see: types

Requirement 3.1.1

The evaluation context structure MUST define an optional targeting key field of type string, identifying the subject of the flag evaluation.

The targeting key uniquely identifies the subject (end-user, or client service) of a flag evaluation. Providers may require this field for fractional flag evaluation, rules, or overrides targeting specific users. Such providers may behave unpredictably if a targeting key is not specified at flag resolution.

Requirement 3.1.2

The evaluation context MUST support the inclusion of custom fields, having keys of type string, and values of type boolean | string | number | datetime | structure.

see: structure, datetime

Requirement 3.1.3

The evaluation context MUST support fetching the custom fields by key and also fetching all key value pairs.

Requirement 3.1.4

The evaluation context fields MUST have an unique key.

The key uniquely identifies a field in the evaluation context and it should be unique across all types to avoid any collision when marshalling the evaluation context by the provider.

3.2 Context levels and merging

Condition 3.2.1

The implementation uses the dynamic-context paradigm.

see: dynamic-context paradigm

Conditional Requirement 3.2.1.1

The API, Client and invocation MUST have a method for supplying evaluation context.

API (global) evaluation context can be used to supply static data to flag evaluation, such as an application identifier, compute region, or hostname. Client and invocation evaluation context are ideal for dynamic data, such as end-user attributes.

Condition 3.2.2

experimental

The implementation uses the static-context paradigm.

see: static-context paradigm

Conditional Requirement 3.2.2.1

The API MUST have a method for setting the global evaluation context.

API (global) evaluation context can be used to supply data to flag evaluation, such as (but not limited to) user name, email, or user organization membership changes.

Conditional Requirement 3.2.2.2

The Client and invocation MUST NOT have a method for supplying evaluation context.

In the static-context paradigm, context is global. The client and invocation cannot supply evaluation context.

Conditional Requirement 3.2.2.3

The API MUST have a method for setting evaluation context for a domain.

In the static-context paradigm, provider specific context can be set using the associated domain.
The global context is used if there is no matching provider specific context.

See setting a provider, domain for details.

Conditional Requirement 3.2.2.4

The API MUST have a mechanism to manage evaluation context for an associated domain.

In the static-context paradigm, it's possible to create and remove provider-specific context.
See setting a provider, domain for details.

Requirement 3.2.3

Evaluation context MUST be merged in the order: API (global; lowest precedence) -> transaction -> client -> invocation -> before hooks (highest precedence), with duplicate values being overwritten.

Any fields defined in the transaction evaluation context will overwrite duplicate fields defined in the global evaluation context, any fields defined in the client evaluation context will overwrite duplicate fields defined in the transaction evaluation context, and fields defined in the invocation evaluation context will overwrite duplicate fields defined globally or on the client. Any resulting evaluation context from a before hook will overwrite duplicate fields defined globally, on the client, or in the invocation.

This describes the precedence of all evaluation context variants. Depending on the paradigm, not all variants might be available in an SDK implementation.

Condition 3.2.4

experimental

The implementation uses the static-context paradigm.

see: static-context paradigm

Conditional Requirement 3.2.4.1

When the global evaluation context is set, the on context changed handler MUST run.

The SDK implementation must run the on context changed handler on all registered provider that use the global evaluation context whenever it is mutated.

Conditional Requirement 3.2.4.2

When the evaluation context for a specific provider is set, the on context changed handler MUST only run on the associated provider.

The SDK implementation must run the on context changed handler only on the provider that is scoped to the mutated evaluation context.

3.3 Context Propagation

experimental

Transaction context is a container for transaction-specific evaluation context (e.g. user id, user agent, IP).
Transaction context can be set where specific data is available (e.g. an auth service or request handler) and by using the transaction context propagator it will automatically be applied to all flag evaluations within a transaction (e.g. a request or thread).

The following shows a possible TypeScript implementation using AsyncLocalStorage (async_hooks):

export class AsyncLocalStorageTransactionContext implements TransactionContextPropagator {
private asyncLocalStorage = new AsyncLocalStorage<EvaluationContext>();

getTransactionContext(): EvaluationContext {
return this.asyncLocalStorage.getStore() ?? {};
}
setTransactionContext(context: EvaluationContext, callback: () => void): void {
this.asyncLocalStorage.run(context, callback);
}
}

/**
* This example is based on an express middleware.
*/
app.use((req: Request, res: Response, next: NextFunction) => {
const ip = res.headers.get("X-Forwarded-For")
OpenFeature.setTransactionContext({ targetingKey: req.user.id, ipAddress: ip }, () => {
// The transaction context is used in any flag evaluation throughout the whole call chain of next
next();
});
})

Condition 3.3.1

The implementation uses the dynamic-context paradigm.

see: dynamic-context paradigm

Conditional Requirement 3.3.1.1

The API SHOULD have a method for setting a transaction context propagator.

If there already is a transaction context propagator, it is replaced with the new one.

Condition 3.3.1.2

The SDK implements context propagation.

A language may not have any applicable way of implementing transaction context propagation so the language SDK might not implement context propagation.

Conditional Requirement 3.3.1.2.1

The API MUST have a method for setting the evaluation context of the transaction context propagator for the current transaction.

If a transaction context propagator is set, the SDK will call the method defined in 3.3.1.3 with this evaluation context and so this evaluation context will be available during the current transaction.
If no transaction context propagator is set, this evaluation context is not used for evaluations.
This method then can be used for example in a request handler to add request-specific information to the evaluation context.

Conditional Requirement 3.3.1.2.2

A transaction context propagator MUST have a method for setting the evaluation context of the current transaction.

A transaction context propagator is responsible for persisting context for the duration of a single transaction.
Typically, a transaction context propagator will propagate the context using a language-specific carrier such as ThreadLocal (Java), async hooks (Node.js), Context (Go) or another similar mechanism.

Conditional Requirement 3.3.1.2.3

A transaction context propagator MUST have a method for getting the evaluation context of the current transaction.

This will be used by the SDK implementation when merging the context for evaluating a feature flag.

Condition 3.3.2

The implementation uses the static-context paradigm.

see: static-context paradigm

Conditional Requirement 3.3.2.1

The API MUST NOT have a method for setting a transaction context propagator.

In the static-context paradigm, context is global, so there must not be different contexts between transactions.