Local cached maps and Pub/Sub invalidation

Latest stable Based on Bluetape4k release 1.11.0

Near Cache removes round trips and adds copies

RLocalCachedMap can answer hot reads from JVM memory. It also creates one copy per application node, so invalidation delivery and reconnect behavior become part of consistency. A Redis interruption can leave stale local values rather than causing only remote lookup failures.

val users = localCachedMap<String, User>("users", client) {
    cacheSize(10_000)
    evictionPolicy(LocalCachedMapOptions.EvictionPolicy.LRU)
    timeToLive(Duration.ofMinutes(5))
    syncStrategy(LocalCachedMapOptions.SyncStrategy.INVALIDATE)
    reconnectionStrategy(LocalCachedMapOptions.ReconnectionStrategy.LOAD)
}

All nodes must use the same map name and codec to share remote data and invalidation traffic.

INVALIDATE or UPDATE

SyncStrategy.INVALIDATE removes another node’s local entry and makes its next read fetch Redis. UPDATE sends the changed value to local caches. Choose from update frequency, message size, and acceptable stale time.

Invalidation Pub/Sub is not a durable business event log. ReconnectionStrategy and TTL define recovery for messages missed while disconnected.

Reconnection policy

In 1.11.0, RedissonNearCache.defaultLocalCacheOptions uses LFU, a 60-second local TTL, 120-second max idle, ReconnectionStrategy.LOAD, and SyncStrategy.UPDATE. These are usable defaults, not workload-independent answers.

Test the miss spike after reconnect, invalidation traffic under write load, and the maximum stale interval after an outage.

The exact scope of destroy

RedissonNearCache.destroy() calls frontCache.destroy() only. It leaves the Redis map intact, separating instance shutdown from shared-data deletion. Explicit map deletion belongs to a separate administrative policy, never a normal shutdown hook.

Destroy near-cache instances that are replaced during runtime so local listeners and resources do not accumulate.

Pattern invalidation cost

RedisCacheInvalidationStrategy.invalidateByPattern finds matching keys with keySet(pattern) and removes them with fastRemove. Broad patterns can scan and delete many keys. Use narrow namespaces and observe target count and latency.