Caching

The Java SDK supports caching to reduce API calls and improve performance.

Caffeine Cache

Caffeine provides high-performance in-memory caching, ideal for single-instance applications.

Installation

<dependency>
    <groupId>io.toggly</groupId>
    <artifactId>toggly-cache-caffeine</artifactId>
    <version>1.7.0</version>
</dependency>

Configuration

SnapshotProvider httpProvider = new HttpSnapshotProvider(config);

SnapshotProvider cachedProvider = new CaffeineCachingSnapshotProvider(
    httpProvider,
    CaffeineCacheConfig.builder()
        .expireAfterWrite(Duration.ofMinutes(5))
        .refreshAfterWrite(Duration.ofMinutes(1))
        .maximumSize(100)
        .recordStats()
        .build()
);

TogglyClient client = new TogglyClient(config, cachedProvider);

The examples assume an existing TogglyConfig config. Keep the client/provider for the application's lifetime. Because the provider was supplied to the client, close client first and cachedProvider second at shutdown; the outer cache closes the HTTP delegate. HTTP polling/live updates do not evict this outer snapshot cache, so its TTL/refresh policy can delay the definitions visible to evaluation. Call the outer provider's refresh() to fetch through it explicitly.

Configuration Options

Option	Default	Description
expireAfterWrite	-	Time after which entries expire
refreshAfterWrite	-	Time after which entries are refreshed asynchronously
maximumSize	0 (unlimited)	Maximum number of entries
recordStats	false	Enable statistics recording

Accessing Statistics

CaffeineCachingSnapshotProvider provider = ...;
CacheStats stats = provider.stats();

System.out.println("Hit rate: " + stats.hitRate());
System.out.println("Miss rate: " + stats.missRate());
System.out.println("Load count: " + stats.loadCount());

Evaluation Cache

Definition caching keeps evaluation sensitive to the current request context. Only add a result cache if stale answers are acceptable for its TTL. Its key uses the feature, identity and context hash (including claims/request/entity), but no definition revision or current time. A refresh or TimeWindow boundary therefore does not automatically invalidate an answer.

Use a short TTL and explicitly invalidate from application code when you know definitions have changed. TogglyClient.onRefresh does not currently invoke listeners; it cannot wire automatic invalidation. The following cache must be long-lived, and every evaluation must pass the complete context:

CaffeineEvaluationCache evaluationCache = CaffeineEvaluationCache.builder()
    .expireAfterWrite(Duration.ofSeconds(10))
    .maximumSize(10000)
    .recordStats()
    .build();

// Use cached evaluation
boolean enabled = evaluationCache.getOrCompute(
    "my-feature",
    context,
    () -> togglyClient.isEnabled("my-feature", context)
);

// Invalidate when features change
evaluationCache.invalidateFeature("my-feature");
evaluationCache.invalidateAll();

Redis Cache

Redis provides distributed caching, ideal for multi-instance deployments.

Installation

<dependency>
    <groupId>io.toggly</groupId>
    <artifactId>toggly-cache-redis</artifactId>
    <version>1.7.0</version>
</dependency>

Configuration

Basic

SnapshotProvider redisProvider = new RedisCachingSnapshotProvider(
    new HttpSnapshotProvider(config),
    RedisCacheConfig.builder()
        .host("localhost")
        .port(6379)
        .ttl(Duration.ofMinutes(5))
        .build()
);

TogglyClient client = new TogglyClient(config, redisProvider);

With Authentication

RedisCacheConfig redisConfig = RedisCacheConfig.builder()
    .host("redis.example.com")
    .port(6379)
    .password("your-password")
    .database(0)
    .ssl()
    .ttl(Duration.ofMinutes(5))
    .keyPrefix("myapp:toggly:")
    .timeout(2000)
    .build();

SnapshotProvider redisProvider = new RedisCachingSnapshotProvider(
    new HttpSnapshotProvider(config),
    redisConfig
);

With Existing Pool

// Use an existing JedisPool
JedisPool existingPool = ...;

SnapshotProvider redisProvider = new RedisCachingSnapshotProvider(
    new HttpSnapshotProvider(config),
    existingPool,
    "toggly:",                    // key prefix
    Duration.ofMinutes(5)         // TTL
);

Configuration Options

Option	Default	Description
host	localhost	Redis host
port	6379	Redis port
password	null	Redis password
database	0	Redis database number
keyPrefix	toggly:	Prefix for cache keys
ttl	5 minutes	Time-to-live for cached entries
timeout	2000	Connection timeout in milliseconds
ssl	false	Enable SSL/TLS

Close the client before redisProvider at shutdown. The Redis wrapper closes its HTTP delegate and any Jedis pool it created itself. An existing pool remains caller-owned. Do not close a shared pool from each client.

Cache Invalidation

RedisCachingSnapshotProvider provider = ...;

// Invalidate the cache
provider.invalidate();

Spring Boot Integration

Choose one of these configurations in your component-scanned package. It replaces the plain HTTP provider bean from the Spring setup, not the client. TogglyConfig is the SDK type; FeatureCacheConfiguration is your class. Spring closes the outer provider, which closes its HTTP delegate. The Redis tab requires application properties spring.redis.host and spring.redis.port.

Caffeine

import io.toggly.core.config.TogglyConfig;
import io.toggly.core.snapshot.HttpSnapshotProvider;
import io.toggly.core.snapshot.SnapshotProvider;
import io.toggly.cache.caffeine.CaffeineCacheConfig;
import io.toggly.cache.caffeine.CaffeineCachingSnapshotProvider;
import java.time.Duration;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;

@Configuration
public class FeatureCacheConfiguration {

    @Bean(destroyMethod = "close")
    public SnapshotProvider snapshotProvider(TogglyConfig config) {
        return new CaffeineCachingSnapshotProvider(
            new HttpSnapshotProvider(config),
            CaffeineCacheConfig.builder()
                .expireAfterWrite(Duration.ofMinutes(5))
                .refreshAfterWrite(Duration.ofMinutes(1))
                .recordStats()
                .build()
        );
    }
}

Redis

import io.toggly.core.config.TogglyConfig;
import io.toggly.core.snapshot.HttpSnapshotProvider;
import io.toggly.core.snapshot.SnapshotProvider;
import io.toggly.cache.redis.RedisCacheConfig;
import io.toggly.cache.redis.RedisCachingSnapshotProvider;
import java.time.Duration;
import org.springframework.beans.factory.annotation.Value;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;

@Configuration
public class FeatureCacheConfiguration {

    @Bean(destroyMethod = "close")
    public SnapshotProvider snapshotProvider(
            TogglyConfig config,
            @Value("${spring.redis.host}") String redisHost,
            @Value("${spring.redis.port}") int redisPort) {

        return new RedisCachingSnapshotProvider(
            new HttpSnapshotProvider(config),
            RedisCacheConfig.builder()
                .host(redisHost)
                .port(redisPort)
                .ttl(Duration.ofMinutes(5))
                .build()
        );
    }
}

Caching Strategies

1. Cache-Aside (Default)

The SDK checks cache first, fetches from API on miss:

// Automatic cache-aside behavior
boolean enabled = client.isEnabled("my-feature");
// 1. Check cache
// 2. If miss, fetch from API
// 3. Store in cache
// 4. Return result

2. Write-Through

Cache is updated when refreshing:

// Fetch definitions and update this local cache (does not edit dashboard flags)
client.refresh();

3. Background Refresh

Caffeine can refresh entries asynchronously:

CaffeineCacheConfig.builder()
    .expireAfterWrite(Duration.ofMinutes(10))
    .refreshAfterWrite(Duration.ofMinutes(1))  // Async refresh
    .build();

Best Practices

1. Use appropriate TTL: Balance freshness vs. performance
2. Monitor cache stats: Track hit rates to optimize settings
3. Use Redis for distributed: When running multiple instances
4. Prefer definition caching: Result caches can retain stale targeting decisions
5. Handle cache failures gracefully: SDK falls back to API

Reliability (signed snapshots)

When signed definitions are enabled, snapshots store exact server defs JSON (signedDefsJson) for verification — never re-serialize models to verify. Configure onError on TogglyConfig.Builder, and call clearCache() to drop persisted feature/JWKS snapshots after corruption or key rotation.

WebSocket signing-key-updated clears JWKS and forces a definitions refresh (see Live updates).

Full signed-definitions setup: Signed definitions.

See Server-side reliability.

Cache Warming

Attempt to pre-fetch definitions at startup. The callback in your TogglyConfig reports failures; completion of this listener is not proof of a fresh snapshot:

@Component
public class CacheWarmer implements ApplicationListener<ApplicationReadyEvent> {

    @Autowired
    private TogglyClient togglyClient;

    @Override
    public void onApplicationEvent(ApplicationReadyEvent event) {
        // Attempt to warm the cache
        togglyClient.refresh();
    }
}