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 APM tracer 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 APM Tracer: 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: Required 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 APM tracer. You need the tracer 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 experimental feature flag (DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED=true). Skip the tracer download and JVM configuration steps.

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

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

# Required: Enable experimental feature flagging support
export DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED=true

# Required: Your Datadog API key
export DD_API_KEY=<YOUR_API_KEY>

# 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 tracer
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.api.key=<YOUR_API_KEY> \
  -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 experimental 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 Tracer 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 tracer.

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 ensures 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 ensures 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()
);

Troubleshooting

Start here: verify prerequisites

Before investigating specific errors, confirm these prerequisites are in place:

1. The Datadog Agent is healthy and reachable: See APM Connection Errors to verify Agent connectivity.
2. The experimental flagging provider is enabled on the tracer: Set DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED=true.
3. Required tracer environment variables are set: DD_API_KEY, DD_ENV, and DD_SITE.
4. Your DD_ENV value appears in the Feature Flag environments list: Confirm your environment is visible in the Feature Flag Environments settings.

After confirming all prerequisites, continue with the following sections if feature flags still aren’t working.

Debug flag evaluations

If flags evaluate but return unexpected values, use getBooleanDetails() instead of getBooleanValue(). The Details variant of each evaluation method returns a FlagEvaluationDetails object that exposes the provider’s internal state, including the reason, variant, and any error code.

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 the logged output to understand why the provider returned a particular result.

Monitor provider state changes

Add event listeners early in your application startup to observe provider life cycle transitions:

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

Provider not ready

Problem: PROVIDER_NOT_READY errors when evaluating flags

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:

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

Solutions:

1. Enable debug logging to see the feature flagging system startup sequence. These messages are emitted at DEBUG level—set DD_TRACE_DEBUG=true to see them:

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

2. Wait for Remote Configuration sync (can take 30-60 seconds after publishing flags)
3. Verify flags are published in Datadog UI to the correct service and environment
4. If none of these apply, verify the Datadog Agent is healthy and reachable. See APM Connection Errors.

EVP proxy not available error

Problem: Logs show Cannot create backend API client since agentless mode is disabled, and agent does not support EVP proxy.

Verify the Datadog Agent is healthy and reachable. See APM Connection Errors.

No exposures in Datadog

Problem: Experiment exposures aren’t appearing in Datadog

Solution: Verify the flag is associated with an experiment in the Datadog UI. Exposures are only recorded for flags that are part of an experiment—standard feature flags without an experiment association do not generate exposure events.

Further reading

Additional helpful documentation, links, and articles:

Server-Side Feature FlagsDOCUMENTATION Java APM and Distributed TracingDOCUMENTATION