The Default In-Memory Cache With More Than One Instance

With no cache library on the classpath, Spring Boot's cache is a ConcurrentHashMap held in the memory of the running process. On one instance of the application that is a real, working cache. With two or more instances behind a load balancer, each has its own separate map, and nothing keeps them in step.

The symptom is a value that is right on some requests and old on others. A fare updated through instance A is refreshed in A's map, while instance B keeps serving the old fare from its own map until that entry is evicted, and which one a request reaches depends on the load balancer.

The default needs nothing but @EnableCaching, which is why every local test passes: a laptop runs one instance. The failure only exists in a deployment with several, and nothing reports a conflict, because each map has no way to know the others exist.

The fix is a shared or coordinated cache provider: Redis, which every instance reads from the same place, or Caffeine together with a way to invalidate entries on every instance when a local cache is still wanted. spring-boot-starter-cache does not pick one for you; it wires up whichever provider it finds on the classpath, and falls back to the map when it finds none.