Failures, cancellation, lifecycle, and operations

Failure contracts by path

Failure point	1.11.0 behavior	Operational decision
JCache CRUD	await() propagates the Redisson future failure	Keep timeout and retry ownership in one layer.
Sync evaluator	Fails the promise and removes in-flight state	A later request can evaluate again.
Async Redis put	Can log a warning and still return the evaluated result	Observe calculation success separately from cache storage.
Suspend cancellation	Fails/removes the deferred and rethrows cancellation	Never convert cancellation into a cached fallback.
Sync near-cache close	Suppresses destroy failure	Do not report it as successful data deletion.
Suspend close	Logs ordinary errors and rethrows cancellation	Separate shutdown timeout from close errors.

Failure amplification

In cache-aside, Redis misses or failures shift reads to the source database. A broad miss wave can saturate the source pool. More retries can pressure both Redis and the database, turning the cache into a failure amplifier. Set timeout, retry budget, stale fallback, and source bulkheads together. Never hide write failure behind a read fallback.

Cancellation and owner scope

A cancelled suspend evaluator is not stored. Its in-flight entry is removed, so a later call can recover. Cancelling one waiter differs from cancelling the owner evaluation; structured concurrency should make that owner scope explicit.

Shutdown order

1. Stop accepting new cache work.
2. Await or cancel in-flight evaluators according to policy.
3. Close wrappers.
4. Let the final owner close the Redisson client and JCache provider.

Run clear() or clearAll() separately when data must be removed. Normal shutdown should not erase shared Redis state.

Signals to monitor

Track Redis latency, timeout and reconnect; invalidation lag; evaluator count/latency/failure/cancellation; local size and Redis map size; eviction and decode failures; and source-database pool pressure during cache failure. Native stats() alone cannot prove network savings.

The test task uses Redis Testcontainers and should run sequentially with other heavyweight backend suites:

./gradlew :bluetape4k-cache-redisson:test --no-build-cache --no-configuration-cache

Source and tests

• RedissonSuspendMemoizerTest.kt
• ResilientRedissonNearCacheTest.kt
• ResilientRedissonSuspendNearCacheTest.kt
• RedisServers.kt