Implementing feature toggles for a Spring Boot application - Part 3

Tue, Nov 29, 2016

In the third part of this series about implementing feature toggles for a Spring Boot application, we will take a look at exposing the state of feature flags as a Spring Boot management end point for monitoring and testing purposes.

Spring Boot Actuator exposes a number of end points to monitor and administer the application. The most commonly used of these in my experience are the info and health end points. These end points are used to communicate to load balancers that a particular instance is ready to accept traffic and to monitor the state of the application.

Features management end point

In our case, this end point will be used by administrators, QA engineers and sometimes business stake holders to see what features are available in a particular environment running the application. Of course, we could have re-used the end point we built in the [last blog post]() for this purpose, but creating an management end point allows us to:

  • Control the HTTP end point together with other end points by using the management.context-path property. This will allow us to provide certain nodes in our network access to only the management end points without having to expose application behavior to them.
  • Make this information available in a more human readable form for it’s consumers. Even though the information exposed in the /features end point is detailed, the format was designed for consumption by code.

Because this is for human consumption, the end point will display the feature state in the following format:

{
  "available": [
    "feature.hello",
    "feature.bar"
    ...
  ],
  "enabled": [
    "feature.hello"
    ...
  ]
}

Management end points in Spring Boot are created by implementing the EndPoint interface. In our case, the end point will depend on FeatureRepository to do the heavy lifting.

public class FeatureEndpoint implements Endpoint<HashMap<String, Set<String>>> {

  public static final String ID = "features";
  private final FeatureRepository featureRepository;

  public FeatureEndpoint(FeatureRepository featureRepository) {
    this.featureRepository = featureRepository;
  }

  @Override
  public String getId() {
    return ID;
  }

  @Override
  public boolean isEnabled() {
    return true;
  }

  @Override
  public boolean isSensitive() {
    return false;
  }

  @Override
  public HashMap<String, Set<String>> invoke() {
    HashMap<String, Set<String>> map = new HashMap<>();
    map.put("enabled", featureRepository.enabledKeys());
    map.put("available", featureRepository.featureKeys());
    return map;
  }
}

In order to retrieve the keys for features that are toggled on, we will create an enabledKeys method in the repository.

public class FeatureRepository {
  // ...
  public Set<String> enabledKeys() {
    return featureKeys().stream()
        .filter(f -> isOn(f))
        .collect(Collectors.toSet());
  }
}

With this end point in place and the management.context-path set to management in our properties, this end point will produce the following output.

> curl -s "http://localhost:8080/management/features" | jq .
{
  "available": [
    "feature.foo",
    "feature.hello"
  ],
  "enabled": [
    "feature.hello"
  ]
}

It is important to note that our end point has it’s sensitivity flag set to false by default. You should consider the security implications of that before choosing to leave it false.

In the next blog post in this series, we will explore how our feature toggle mechanism so far have introduced a dependency on the application being restarted and how to eliminate this dependency.