Syncs
Syncs are a core part of flagd; they are the abstraction that enables different sources for feature flag definitions. flagd can connect to one or more sync sources.
Available syncs
Filepath sync
The file path sync provider reads and watch the source file for updates(ex: changes and deletions). It's important to note that most file operations result in multiple file system events. For production use-cases, a symbolic link is recommended for the watched file, which enables atomic modification. See the relevant troubleshooting entry.
In this example, etc/featureflags.json is a valid feature flag definition file accessible by the flagd process.
See sync source configuration for details.
HTTP sync
The HTTP sync provider fetch flags from a remote source and periodically poll the source for flag definition updates.
In this example, https://my-flag-source.json is a remote endpoint responding valid feature flag definition when
invoked with HTTP GET request.
The polling interval, port, TLS settings, and authentication information can be configured.
See sync source configuration for details.
gRPC sync
The gRPC sync provider streams flag definitions from a gRPC sync provider implementation. This stream connection is defined by the sync service protobuf definition.
In this example, grpc-sync-source is a grpc target implementing sync.proto definition.
See sync source configuration for details.
Kubernetes sync
The Kubernetes sync provider allows flagd to connect to a Kubernetes cluster and evaluate flags against a specified FeatureFlag resource as defined within the open-feature-operator spec. This configuration is best used in conjunction with the OpenFeature Operator.
To use an existing FeatureFlag custom resource, start flagd with the following command:
In this example, default/my_example expected to be a valid FeatureFlag resource, where default is the
namespace and my_example being the resource name.
See sync source configuration for details.
GCS sync
The GCS sync provider fetches flags from a GCS blob and periodically poll the GCS for the flag definition updates. It uses application default credentials if they are configured to authorize the calls to GCS.
In this example, gs://my-bucket/my-flags.json is expected to be a valid GCS URI accessible by the flagd
(either by being public or together with application default credentials).
The polling interval can be configured.
See sync source configuration for details.
Merging
Flagd can be configured to read from multiple sources at once, when this is the case flagd will merge all flag definition into a single merged state.
For example:
In this example, source-A and source-B provide a single flag definition, the foo flag and the bar flag respectively.
The merge logic for this definition is simple, both flag definition are added to the store.
In most scenarios, these flag sources will be supplying n number of definition, using a unique flag key for each definition.
However, as multiple sources are being used, there is the opportunity for keys to be duplicated, intentionally or not, between flag sources.
In these situations flagd uses a merge priority order to ensure that its behavior is consistent.
Merge order is dictated by the order that sync-providers and uris are defined, with the latest defined source taking precedence over those defined before it, as an example:
When flagd is started with the command defined above, source-B takes priority over source-A, whilst source-C takes priority over both source-B and source-A.
Using the above example, if a flag key is duplicated across all 3 sources, then the definition from source-C would be the only one stored in the merged state.
State Resync Events
Given the above example, the source-A and source-B 'versions' of flag definition the foo have been discarded, so if a delete event in source-C results in the removal of the fooflag, there will no longer be any reference of foo in flagd's store.
As a result of this flagd will return FLAG_NOT_FOUND errors, and the OpenFeature SDK will always return the default value.
To prevent flagd falling out of sync with its flag sources during delete events, resync events are used.
When a delete event results in a flag definition being removed from the merged state, the full set of definition is requested from all flag sources, and the merged state is rebuilt.
As a result, the value of the foo flag from source-B will be stored in the merged state, preventing flagd from returning FLAG_NOT_FOUND errors.
In the example above, a delete event results in a resync event being fired, as source-C has deleted its 'version' of the foo, this results in a new merge state being formed from the remaining definition.
Resync events may lead to further resync events if the returned flag definition result in further delete events, however the state will eventually be resolved correctly.