Skip to content

Creating an in-process flagd provider

An in-process flagd provider is designed to be embedded into the application, and therefore no communication outside the process of the application for feature flag evaluation is needed. This can be desirable in some cases, particularly if latency is a concern.

The in-process flagd provider is responsible for creating an abstraction between the JsonLogic based evaluation of flag configurations following the flag configuration scheme used by flagd and the OpenFeature SDK (for the chosen technology).


The Flag definition containing the feature flags and JsonLogic based targeting rules shall be retrieved by the in-process flagd provider via a gRPC client connection to a sync server, such as flagd-proxy.

Sync source

An implementation of an in-process flagd-provider must accept the following environment variables which determine the sync source:

  • FLAGD_SOURCE_URI: The URI identifying the sync source. Depending on the sync provider type, this can be the URI of a gRPC service providing the sync API required by the in-process flagd provider, or the name of a Custom Resource containing the flag definition.
  • FLAGD_SOURCE_PROVIDER_TYPE: The type of the provider. E.g. grpc or kubernetes.
  • FLAGD_SOURCE_SELECTOR: Optional selector for the feature flag definition of interest. This is used as a selector for the flagd-proxie's sync API to identify a flag definition within a collection of feature flag definitions.

An implementation of an in-process flagd provider should provide a source for retrieving the flag definition, namely a gRPC source. Other sources may be desired eventually, so separation of concerns should be maintained between the abstractions evaluating flags and those retrieving confirmation.

gRPC sources

gRPC sync sources are identified by the provider field set to grpc. When such a sync source is specified, the in-process flagd provider should connect to the gRPC service located at the uri of the sync source, and use its sync API to retrieve the feature flag definition. If the selector field of the sync source is set, that selector should be passed through to the Sync and FetchAllFlags requests sent to the gRPC server.


Protobuf schemas define the contract between a client (flagd or the in-process provider implementation) and server (flagd-proxy). flagd-proxy's schemas are defined here.

Code generation for gRPC sync

Leverage the buf CLI or protoc to generate a flagd-proxy client in the chosen technology:

Add the open-feature schema repository as a submodule

git submodule add --force

Create a buf.gen.{chosen language}.yaml for the chosen language in schemas/protobuf (if it doesn't already exist) using one of the other files as a template (find a plugin for the chosen language here) and create a pull request with this file.

Generate the code (this step ought to be automated in the build process for the chosen technology so that the generated code is never committed)

cd schemas/protobuf
buf generate --template buf.gen.{chosen language}.yaml

As an alternative to buf, use the .proto file directly along with whatever protoc-related tools or plugins avaialble for your language.

Move the generated code (following convention for the chosen language) and add its location to .gitignore

Note that for the in-process provider only the sync package will be relevant, as it does not communicate with flagd, but only with compliant gRPC services such as flagd-proxy.

JsonLogic evaluation

An in-process flagd provider should provide the feature set offered by JsonLogic to evaluate flag resolution requests for a given context. If available, the JsonLogic library for the chosen technology should be used. Additionally, it should also provide the custom JsonLogic evaluators and $flagd properties in the evaluation context described below.

Custom JsonLogic evaluators

In addition to the built-in evaluators provided by JsonLogic, the following custom targeting rules should be implemented by the provider:

  • Fractional operation: This evaluator allows splitting of the returned variants of a feature flag into different buckets, where each bucket can be assigned a percentage, representing how many requests will resolve to the corresponding variant. The sum of all weights must be 100, and the distribution must be performed by using the value of a referenced from the evaluation context to hash that value and map it to a value between [0, 100]. It is important to note that evaluations MUST be sticky, meaning that flag resolution requests containing the same value for the referenced property in their context MUST always resolve to the same variant. For calculating the hash value of the referenced evaluation context property, the MurmurHash3 hash function should be used. This is to ensure that flag resolution requests yield the same result, regardless of which implementation of the in-process flagd provider is being used. For more specific implementation guidelines, please refer to this document.
  • Semantic version evaluation: This evaluator checks if the given property within the evaluation context matches a semantic versioning condition. It returns 'true', if the value of the given property meets the condition, 'false' if not. For more specific implementation guidelines, please refer to this document.
  • StartsWith/EndsWith evaluation: This evaluator selects a variant based on whether the specified property within the evaluation context starts/ends with a certain string. For more specific implementation guidelines, please refer to this document.

Targeting key

An in-process provider should map the targeting-key into a top level property of the context used in rules, with the key "targetingKey".

$flagd properties in the evaluation context

An in-process flagd provider should also add the following properties to the JsonLogic evaluation context so that users can use them in their targeting rules. Conflicting properties in the context will be overwritten by the values below.

Property Description
$flagd.flagKey the identifier for the flag being evaluated
$flagd.timestamp a unix timestamp (in seconds) of the time of evaluation

Provider construction

(using Go as an example)

Create a provider struct/class/type (whichever is relevant to the chosen language) with an exported (public) constructor allowing configuration (e.g. flagd host). Give the provider an un-exported (private) client field, set this field as the client generated by the previous step.

Create methods for the provider to satisfy the chosen language SDK's provider interface. These methods ought to wrap the built client's methods.

type Provider struct {
    evaluator IJsonEvaluator

type ProviderOption func(*Provider)

func NewProvider(options ...ProviderOption) *Provider {
    provider := &Provider{}
    for _, opt := range opts {

    // create a store that is responsible for retrieving the flag configurations
    // from the sources that are given to the provider via the options
    s := store.NewFlags()
    s.FlagSources = append(s.FlagSources, os.Getenv("FLAGD_SOURCE_URI"))
    s.SourceMetadata[provider.URI] = store.SourceDetails{
        Source:   os.Getenv("FLAGD_SOURCE_URI"),
        Selector: os.Getenv("FLAGD_SOURCE_SELECTOR")),

    // derive evaluator
    provider.evaluator := setupJSONEvaluator(logger, s)

    return provider

func WithHost(host string) ProviderOption {
    return func(p *Provider) {
        p.flagdHost = host

func (p *Provider) BooleanEvaluation(
    ctx context.Context, flagKey string, defaultValue bool, evalCtx of.FlattenedContext,
) of.BoolResolutionDetail {

    res, err := p.evaluator.ResolveBoolean(ctx, flagKey, context)

    if err != nil {
        return of.BoolResolutionDetail{
            Value: defaultValue,
            ProviderResolutionDetail: of.ProviderResolutionDetail{
                ResolutionError: of.NewGeneralResolutionError(err.Error()),
                Reason:          of.Reason(res.Reason),
                Variant:         res.Variant,

    return of.BoolResolutionDetail{
        Value: defaultValue,
        ProviderResolutionDetail: of.ProviderResolutionDetail{
            Reason:          of.Reason(res.Reason),
            Variant:         res.Variant,

// ...

Provider lifecycle, initialization and shutdown

With the release of the v0.6.0 spec, OpenFeature now outlines a lifecycle for in-process flagd provider initialization and shutdown.

In-process flagd providers should do the following to make use of OpenFeature v0.6.0 features:

  • start in a NOT_READY state
  • fetch the flag definition specified in the sync provider sources and set state to READY or ERROR in the initialization function
  • note that the SDK will automatically emit PROVIDER_READY/PROVIDER_ERROR according to the termination of the initialization function
  • throw an exception or terminate abnormally if a connection cannot be established during initialization
  • For gRPC based sources (i.e. flagd-proxy), attempt to restore the streaming connection to flagd-proxy (if the connection cannot be established or is broken):
  • If flag definition have been retrieved previously, go into STALE state to indicate that flag resolution responses are based on potentially outdated Flag definition.
  • reconnection should be attempted with an exponential back-off, with a max-delay of maxSyncRetryInterval (see configuration)
  • reconnection should be attempted up to maxSyncRetryDelay times (see configuration)
  • PROVIDER_READY and PROVIDER_CONFIGURATION_CHANGED should be emitted, in that order, after successful reconnection
  • For Kubernetes sync sources, retry to retrieve the FlagConfiguration resource, using an exponential back-off strategy, with a max-delay of maxSyncRetryInterval (see configuration)
  • emit PROVIDER_CONFIGURATION_CHANGED event and update the ruleset when a configuration_change message is received on the streaming connection
  • close the streaming connection in the shutdown function
    [*] --> NOT_READY
    NOT_READY --> READY: initialize(), stream connected, flag configurations retrieved
    NOT_READY --> ERROR: initialize(), unable to connect (retry)
    READY --> STALE: previously retrieved flag configurations can not be retrieved anymore (emit stale*)
    STALE --> READY: connection to flag source reestablished, and latest flag configurations retrieved (emit ready*, changed*)
    STALE --> ERROR: connection reattempt failed after maxSyncRetries reached (emit error*)
    READY --> READY: configuration_change (emit changed*)
    ERROR --> READY: reconnect successful (emit ready*, changed*)
    ERROR --> ERROR: maxSyncRetries reached
    ERROR --> [*]: shutdown(), stream disconnected



Expose means to configure the provider aligned with the following priority system (highest to lowest).

flowchart LR
    constructor-parameters -->|highest priority| environment-variables -->|lowest priority| defaults

Explicit declaration

This takes the form of parameters to the provider's constructor, it has the highest priority.

Environment variables

Read environment variables with sensible defaults (before applying the values explicitly declared to the constructor).

Option name Environment variable name Type Options Default
host FLAGD_PROXY_HOST string localhost
port FLAGD_PROXY_PORT number 8013
tls FLAGD_PROXY_TLS boolean false
sourceProviderType FLAGD_SOURCE_PROVIDER_TYPE string grpc
sourceSelector FLAGD_SOURCE_SELECTOR string
maxSyncRetries FLAGD_MAX_SYNC_RETRIES int 0 (0 means unlimited)
maxSyncRetryInterval FLAGD_MAX_SYNC_RETRY_INTERVAL int 60s

Error handling

Handle flag evaluation errors by using the error constructors exported by the SDK (e.g. openfeature.NewProviderNotReadyResolutionError(ConnectionError)), thereby allowing the SDK to parse and handle the error appropriately.

Post creation

The following steps will extend the reach of the newly created provider to other developers of the chosen technology.

Open an issue to document the provider

Create an issue here for adding the provider to