Configuring L2 + Query Cache — Providers, Regions, and Tuning

Hibernate L2 is provider-agnostic. The modern path is JCache plus a provider — Caffeine for single JVMs, Hazelcast or Infinispan for clusters.

Configuration steps:

• Enable use_second_level_cache and use_query_cache in application.yml
• Set region.factory_class to JCacheRegionFactory
• Add the provider on the classpath and point javax.cache.provider at it
• Tune regions in the provider's own config — size, TTL, eviction policy

Annotations:

• Use @org.hibernate.annotations.Cache(usage = READ_WRITE, region = "...") on entities
• JPA's @Cacheable alone does nothing — the Hibernate SPI annotation is required
• Annotate @OneToMany collections separately
• Opt queries in with HINT_CACHEABLE and a named region

Choosing a provider:

• Single instance — Caffeine; fastest in-process cache on the JVM
• Multiple pods — Hazelcast for easy embedded clustering, Infinispan for transactional consistency
• Reference data larger than heap — Ehcache 3 with offheap tier

Verification:

• Enable SQL logging; second call to the same query should not fire SQL
• Turn on generate_statistics and expose hit/miss counts via Actuator
• Target > 80% hit ratio for cached entities; below that, retune or remove the cache