Overview

This page describes how to instrument a Java application with the Datadog Feature Flags SDK. Datadog feature flags provide a unified way to remotely control feature availability in your app, experiment safely, and deliver new experiences with confidence.

The Java SDK integrates feature flags directly into the Datadog SDK and implements the OpenFeature standard for maximum flexibility and compatibility.

Compatibility requirements

The Datadog Feature Flags SDK for Java requires:

• Java 11 or higher
• Datadog Java SDK: Version 1.57.0 or later
• OpenFeature SDK: Version 1.18.2 or later
• Datadog Agent: Version 7.x or later with Remote Configuration enabled
• Datadog API key: Configured on the Agent (not the application) for Remote Configuration

For a full list of Datadog’s Java version and framework support, read Compatibility Requirements.

Getting started

Before you begin, make sure you’ve already installed and configured the Agent.

Installation

Feature flagging is integrated into the Datadog Java SDK. You need the SDK JAR and the OpenFeature SDK dependencies.

dependencies {
    // OpenFeature SDK for flag evaluation
    implementation 'dev.openfeature:sdk:1.18.2'

    // Datadog OpenFeature Provider
    implementation 'com.datadoghq:dd-openfeature:1.57.0'
}

dependencies {
    // OpenFeature SDK for flag evaluation
    implementation("dev.openfeature:sdk:1.18.2")

    // Datadog OpenFeature Provider
    implementation("com.datadoghq:dd-openfeature:1.57.0")
}

<dependencies>
    <!-- OpenFeature SDK for flag evaluation -->
    <dependency>
        <groupId>dev.openfeature</groupId>
        <artifactId>sdk</artifactId>
        <version>1.18.2</version>
    </dependency>

    <!-- Datadog OpenFeature Provider -->
    <dependency>
        <groupId>com.datadoghq</groupId>
        <artifactId>dd-openfeature</artifactId>
        <version>1.57.0</version>
    </dependency>
</dependencies>

Configuration

If your Datadog Agent already has Remote Configuration enabled for other features (like Dynamic Instrumentation or Application Security), you can skip the Agent configuration and go directly to Application configuration.

Agent configuration

Configure your Datadog Agent to enable Remote Configuration:

# Enable Remote Configuration
remote_configuration:
  enabled: true

# Set your API key
api_key: <YOUR_API_KEY>

Application configuration

If your application already runs with -javaagent:dd-java-agent.jar and has Remote Configuration enabled (DD_REMOTE_CONFIG_ENABLED=true), you only need to add the feature flagging variable (DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED=true). Skip the SDK download and JVM configuration steps.

Configure your Java application with the required environment variables or system properties:

# Required: Enable Remote Configuration in the SDK
export DD_REMOTE_CONFIG_ENABLED=true

# Required: Enable the feature flagging provider
# The EXPERIMENTAL_ prefix is historical; the provider is no longer experimental.
export DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED=true

# Required: Service name
export DD_SERVICE=<YOUR_SERVICE_NAME>

# Required: Environment (e.g., prod, staging, dev)
export DD_ENV=<YOUR_ENVIRONMENT>

# Optional: Version
export DD_VERSION=<YOUR_APP_VERSION>

# Start your application with the SDK
java -javaagent:path/to/dd-java-agent.jar -jar your-application.jar

java -javaagent:path/to/dd-java-agent.jar \
  -Ddd.remote.config.enabled=true \
  -Ddd.experimental.flagging.provider.enabled=true \
  -Ddd.service=<YOUR_SERVICE_NAME> \
  -Ddd.env=<YOUR_ENVIRONMENT> \
  -Ddd.version=<YOUR_APP_VERSION> \
  -jar your-application.jar

The Datadog feature flagging system starts automatically when the tracer is initialized with both Remote Configuration and the feature flagging provider enabled. No additional initialization code is required in your application.

Add the Java tracer to the JVM

For instructions on how to add the -javaagent argument to your application server or framework, see Add the Java SDK to the JVM.

Make sure to include the feature flagging configuration flags:

• -Ddd.remote.config.enabled=true
• -Ddd.experimental.flagging.provider.enabled=true

Initialize the OpenFeature provider

Initialize the Datadog OpenFeature provider in your application startup code. The provider connects to the feature flagging system running in the Datadog SDK.

import dev.openfeature.sdk.OpenFeatureAPI;
import dev.openfeature.sdk.Client;
import datadog.trace.api.openfeature.Provider;
import dev.openfeature.sdk.exceptions.ProviderNotReadyError;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

public class App {
    private static final Logger logger = LoggerFactory.getLogger(App.class);
    private static Client client;

    public static void main(String[] args) throws Exception {
        // Initialize the Datadog provider
        logger.info("Initializing Datadog OpenFeature Provider...");
        OpenFeatureAPI api = OpenFeatureAPI.getInstance();

        try {
            // Set provider and wait for initial configuration (recommended)
            api.setProviderAndWait(new Provider());
            client = api.getClient("my-app");
            logger.info("OpenFeature provider initialized successfully");
        } catch (ProviderNotReadyError e) {
            // Handle gracefully - app will use default flag values
            logger.warn("Provider not ready (no tracer/config available), continuing with defaults", e);
            client = api.getClient("my-app");
            logger.info("App will use default flag values until provider is ready");
        } catch (Exception e) {
            logger.error("Failed to initialize OpenFeature provider", e);
            throw e;
        }

        // Your application code here
    }
}

Use setProviderAndWait() to block evaluation until the initial flag configuration is received from Remote Configuration. This helps ensure flags are ready before the application starts serving traffic. The default timeout is 30 seconds.

ProviderNotReadyError is an OpenFeature SDK exception thrown when the provider times out during initialization. Catching it allows the application to start with default flag values if Remote Configuration is unavailable. If not caught, the exception propagates and may prevent application startup. Handle this based on your availability requirements.

Asynchronous initialization

For non-blocking initialization, use setProvider() and listen for provider events:

import dev.openfeature.sdk.ProviderEvent;

OpenFeatureAPI api = OpenFeatureAPI.getInstance();
Client client = api.getClient();

// Listen for provider state changes
client.on(ProviderEvent.PROVIDER_READY, (event) -> {
    logger.info("Feature flags ready!");
});

client.on(ProviderEvent.PROVIDER_ERROR, (event) -> {
    logger.error("Provider error: {}", event.getMessage());
});

client.on(ProviderEvent.PROVIDER_STALE, (event) -> {
    logger.warn("Provider configuration is stale");
});

// Set provider asynchronously
api.setProvider(new Provider());

Set the evaluation context

The evaluation context defines the subject (user, device, session) for flag evaluation. It determines which flag variations are returned based on targeting rules.

import dev.openfeature.sdk.EvaluationContext;
import dev.openfeature.sdk.MutableContext;

// Create an evaluation context with a targeting key and attributes
EvaluationContext context = new MutableContext("user-123")
    .add("email", "user@example.com")
    .add("tier", "premium");

// Use the context for flag evaluations (see next section)

The targetingKey (for example, user-123) is the primary identifier used for consistent flag evaluations and percentage-based rollouts. It’s typically a user ID, session ID, or device ID.

Evaluate flags

Evaluate feature flags using the OpenFeature client. All flag types are supported: Boolean, string, integer, double, and object.

// Simple Boolean evaluation
boolean enabled = client.getBooleanValue("checkout.new", false, context);

if (enabled) {
    // New checkout flow
} else {
    // Old checkout flow
}

// Get detailed evaluation result
import dev.openfeature.sdk.FlagEvaluationDetails;

FlagEvaluationDetails<Boolean> details =
    client.getBooleanDetails("checkout.new", false, context);

logger.info("Value: {}", details.getValue());
logger.info("Variant: {}", details.getVariant());
logger.info("Reason: {}", details.getReason());

// Evaluate string flags (e.g., UI themes, API endpoints)
String theme = client.getStringValue("ui.theme", "light", context);

String apiEndpoint = client.getStringValue(
    "payment.api.endpoint",
    "https://api.example.com/v1",
    context
);

// Integer flags (e.g., limits, quotas)
int maxRetries = client.getIntegerValue("retries.max", 3, context);

// Double flags (e.g., thresholds, rates)
double discountRate = client.getDoubleValue("pricing.discount.rate", 0.0, context);

import dev.openfeature.sdk.Value;

// Evaluate object/JSON flags for complex configuration
Value config = client.getObjectValue("ui.config", new Value(), context);

// Access structured data
if (config.isStructure()) {
    Value timeout = config.asStructure().getValue("timeout");
    Value endpoint = config.asStructure().getValue("endpoint");
}

Error handling

The OpenFeature SDK uses a default value pattern. If evaluation fails for any reason, the default value you provide is returned.

import dev.openfeature.sdk.ErrorCode;

// Check evaluation details for errors
FlagEvaluationDetails<Boolean> details =
    client.getBooleanDetails("checkout.new", false, context);

if (details.getErrorCode() != null) {
    switch (details.getErrorCode()) {
        case FLAG_NOT_FOUND:
            logger.warn("Flag does not exist: {}", "checkout.new");
            break;
        case PROVIDER_NOT_READY:
            logger.warn("Provider not initialized yet");
            break;
        case TARGETING_KEY_MISSING:
            logger.warn("Evaluation context missing targeting key");
            break;
        case TYPE_MISMATCH:
            logger.error("Flag value type doesn't match requested type");
            break;
        default:
            logger.error("Evaluation error for flag: {}", "checkout.new", details.getErrorCode());
    }
}

Common error codes

Advanced configuration

Custom initialization timeout

Configure how long the provider waits for initial configuration:

import datadog.trace.api.openfeature.Provider;
import java.util.concurrent.TimeUnit;

Provider.Options options = new Provider.Options()
    .initTimeout(10, TimeUnit.SECONDS);

api.setProviderAndWait(new Provider(options));

Configuration change events

Listen for configuration updates from Remote Configuration:

import dev.openfeature.sdk.ProviderEvent;

client.on(ProviderEvent.PROVIDER_CONFIGURATION_CHANGED, (event) -> {
    logger.info("Flag configuration updated: {}", event.getMessage());
    // Optionally re-evaluate flags or trigger cache refresh
});

PROVIDER_CONFIGURATION_CHANGED is an optional OpenFeature event. Check the Datadog provider documentation to verify this event is supported in your version.

Multiple clients

Use named clients to organize context and flags by domain or team:

// Named clients share the same provider instance but can have different contexts
Client checkoutClient = api.getClient("checkout");
Client analyticsClient = api.getClient("analytics");

// Each client can have its own evaluation context
EvaluationContext checkoutContext = new MutableContext("session-abc");
EvaluationContext analyticsContext = new MutableContext("user-123");

boolean newCheckout = checkoutClient.getBooleanValue(
    "checkout.ui.new", false, checkoutContext
);

boolean enhancedAnalytics = analyticsClient.getBooleanValue(
    "analytics.enhanced", false, analyticsContext
);

The Provider instance is shared globally. Client names are for organizational purposes only and don’t create separate provider instances. All clients use the same underlying Datadog provider and flag configurations.

Best practices

Initialize early

Initialize the OpenFeature provider as early as possible in your application lifecycle (for example, in main() or application startup). This helps ensure flags are ready before business logic executes.

Use meaningful default values

Always provide sensible default values that maintain safe behavior if flag evaluation fails:

// Good: Safe default that maintains current behavior
boolean useNewAlgorithm = client.getBooleanValue("algorithm.new", false, context);

// Good: Conservative default for limits
int rateLimit = client.getIntegerValue("rate.limit", 100, context);

Create context once

Create the evaluation context once per request/user/session and reuse it for all flag evaluations:

// In a web filter or request handler
EvaluationContext userContext = new MutableContext(userId)
    .add("email", user.getEmail())
    .add("tier", user.getTier());

// Reuse context for all flags in this request
boolean featureA = client.getBooleanValue("feature.a", false, userContext);
boolean featureB = client.getBooleanValue("feature.b", false, userContext);

Rebuilding the evaluation context for every flag evaluation adds unnecessary overhead. Create the context once at the start of the request lifecycle, then pass it to all subsequent flag evaluations.

Handle initialization failures (optional)

Consider handling initialization failures if your application can function with default flag values:

try {
    api.setProviderAndWait(new Provider());
} catch (ProviderNotReadyError e) {
    // Log error and continue with defaults
    logger.warn("Feature flags not ready, using defaults", e);
    // Application will use default values for all flags
}

If feature flags are critical for your application to function, let the exception propagate to prevent startup.

Use consistent targeting keys

Use consistent, stable identifiers as targeting keys:

• Good: User IDs, session IDs, device IDs
• Avoid: Timestamps, random values, frequently changing IDs

Monitor flag evaluation

Use the detailed evaluation results for logging and debugging:

FlagEvaluationDetails<Boolean> details =
    client.getBooleanDetails("feature.critical", false, context);

logger.info("Flag: {} | Value: {} | Variant: {} | Reason: {}",
    "feature.critical",
    details.getValue(),
    details.getVariant(),
    details.getReason()
);

Testing

You can test against a dedicated Datadog test environment with the real DatadogProvider, or swap it for OpenFeature’s InMemoryProvider to control flag values directly in test code. This section shows the in-memory approach, which keeps tests hermetic and offline. InMemoryProvider ships in dev.openfeature:sdk (already a test-scope dependency), so no additional library is required. Add dev.openfeature:sdk to your test configuration if it is not already present.

import dev.openfeature.sdk.Client;
import dev.openfeature.sdk.OpenFeatureAPI;
import dev.openfeature.sdk.providers.memory.Flag;
import dev.openfeature.sdk.providers.memory.InMemoryProvider;
import java.util.Map;
import org.junit.jupiter.api.*;
import static org.junit.jupiter.api.Assertions.*;

class CheckoutFlagTest {
    private Client client;

    @BeforeEach
    void setUp() {
        Map<String, Flag<?>> flags = Map.of(
            "new-checkout-flow", Flag.<Boolean>builder()
                .variant("on", true)
                .variant("off", false)
                .defaultVariant("on")
                .build(),
            "ui-theme", Flag.<String>builder()
                .variant("dark", "dark")
                .variant("light", "light")
                .defaultVariant("light")
                .build()
        );

        OpenFeatureAPI api = OpenFeatureAPI.getInstance();
        api.setProviderAndWait(new InMemoryProvider(flags));
        client = api.getClient();
    }

    @AfterEach
    void tearDown() {
        OpenFeatureAPI.getInstance().shutdown();
    }

    @Test
    void newCheckoutEnabledByDefault() {
        assertTrue(client.getBooleanValue("new-checkout-flow", false));
    }

    @Test
    void missingFlagReturnsDefault() {
        assertFalse(client.getBooleanValue("does-not-exist", false));
    }
}

OpenFeatureAPI.getInstance() is a singleton. Always call shutdown() in @AfterEach (or equivalent); otherwise, provider state leaks between test classes and causes flaky suites.

In Spring Boot tests, register the InMemoryProvider through a @TestConfiguration bean or in a @BeforeAll hook on an @SpringBootTest class — the OpenFeature API singleton persists for the lifetime of the Spring context, so initialization only needs to run once.

Troubleshooting

This guide follows the flag data path from the Flagging Platform (where flags are configured) through Remote Configuration and the Datadog Agent to the Java SDK, and back to the Flagging Platform (where evaluations and exposures appear). Work through each step in sequence to isolate where a problem originates.

1. Flagging platform: Verify flag configuration

Before checking infrastructure, confirm the flag itself is set up correctly:

1. The flag is enabled for the target environment, not disabled. Flags are disabled by default in each environment.
2. The flag targets the correct environment (DD_ENV). Flags do not target specific services—they apply to all services within the enabled environment.
3. Your DD_ENV value appears in Feature Flag Environments. If it is absent, the environment has not received any flag traffic yet.

2. Remote Configuration: Verify the path

Remote Configuration delivers flag configurations from the Datadog backend to the Agent.

1. RC is enabled on the Agent: Set remote_configuration.enabled: true in datadog.yaml or DD_REMOTE_CONFIG_ENABLED=true. See Remote Configuration.
2. DD_API_KEY is valid on the Agent and belongs to the target organization.
3. DD_SITE is set correctly on the Agent (site in datadog.yaml or DD_SITE env var). See Agent Site Issues.
4. Fleet Automation: Open Fleet Automation, select the Agent your application connects to, and confirm Remote Configuration is active.
5. Agent CLI: Run datadog-agent status and review the Remote Configuration section of the output. See Agent Commands.

3. Agent: Verify Agent health and connectivity

1. Agent is running and reachable: See APM Connection Errors for steps to verify Agent connectivity from the tracer.

2. Agent version: Feature flagging requires Agent 7.x or later with EVP Proxy support.

3. EVP proxy is available: Query the Agent’s info endpoint and confirm the response includes evp_proxy/v4/ and v0.7/config in the endpoints array:

   curl http://localhost:8126/info
   

   If the Agent logs show the following error, the Agent either started after the tracer connected or does not support EVP Proxy. In this case, check the Agent version and restart order.

   Cannot create backend API client since Agentless mode is disabled, and agent does not support EVP proxy
   

4. SDK: Verify Java SDK state

Enable debug logging

All feature flagging startup messages are emitted at DEBUG level. Set DD_TRACE_DEBUG=true and look for the startup sequence:

[dd.trace] Feature Flagging system starting
[dd.trace] Feature Flagging system started

Then confirm the EVP Proxy and config endpoints were discovered:

discovered ... evpProxyEndpoint=evp_proxy/v4/ configEndpoint=v0.7/config

If these messages are absent, verify DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED=true is set and the tracer started correctly.

Monitor provider state changes

Add event listeners early in application startup to observe provider life cycle transitions. Event listeners detect connectivity changes after initialization:

import dev.openfeature.sdk.ProviderEvent;

client.on(ProviderEvent.PROVIDER_READY, (event) -> {
    logger.info("Feature flag provider is ready");
});

client.on(ProviderEvent.PROVIDER_ERROR, (event) -> {
    logger.error("Feature flag provider error: {}", event.getMessage());
});

client.on(ProviderEvent.PROVIDER_STALE, (event) -> {
    logger.warn("Feature flag provider configuration is stale");
});

client.on(ProviderEvent.PROVIDER_CONFIGURATION_CHANGED, (event) -> {
    logger.info("Feature flag configuration updated");
});

A PROVIDER_ERROR or PROVIDER_STALE event after a period of normal operation indicates a loss of connectivity to the Agent or a Remote Configuration disruption.

Provider not ready

PROVIDER_NOT_READY is returned when flag evaluation is attempted before the provider has received its first configuration from Remote Configuration. This state persists until the tracer receives its initial flag configuration payload from the Agent.

Common causes:

• Async initialization: setProvider() was used instead of setProviderAndWait(). Evaluations that happen before the first Remote Configuration payload arrives return PROVIDER_NOT_READY.
• Initialization timeout: setProviderAndWait() timed out (default 30 seconds) and threw ProviderNotReadyError, which was caught. The application continues evaluating flags while waiting for the first configuration.

Remote Configuration sync can take 30-60 seconds after publishing flags. If PROVIDER_NOT_READY persists beyond that, re-check steps 2 and 3.

Debug flag evaluations

If flags return unexpected values, use getBooleanDetails() instead of getBooleanValue(). The Details variant returns a FlagEvaluationDetails object exposing the provider’s internal state:

FlagEvaluationDetails<Boolean> details =
    client.getBooleanDetails("your.flag.key", false, context);

logger.info("Flag evaluation details: value={}, variant={}, reason={}, errorCode={}",
    details.getValue(),
    details.getVariant(),
    details.getReason(),
    details.getErrorCode());

Review reason and errorCode to understand why the provider returned a given result.

Type mismatch errors

TYPE_MISMATCH is returned when the evaluation method does not match the flag’s configured type. Use the correct method for each flag type: getBooleanValue(), getStringValue(), getIntegerValue(), getDoubleValue().

5. Flagging platform: Verify data appears in Datadog

Flag evaluation metrics

Flag evaluation counts appear in Datadog when DD_METRICS_OTEL_ENABLED=true is set on the tracer. Each evaluation emits a feature_flag.evaluations counter metric tagged with the flag key, result variant, and evaluation reason. If this metric does not appear, verify the setting is enabled and the tracer version supports it.

Experiment exposures

Exposures appear in Datadog only for flags associated with an experiment. Standard feature flags without an experiment association do not generate exposure events. If exposures are missing:

1. Verify the flag is associated with an experiment in the Datadog UI.
2. Verify the Agent’s DD_API_KEY is correct and the Agent is receiving events.

Further reading

Additional helpful documentation, links, and articles:

Server-Side Feature FlagsDOCUMENTATION Java APM and Distributed TracingDOCUMENTATION