The amazon Elasticache offers a Redis 2.8 features.
But under memory pressure the default configuration act like a cache and evicts some keys to free memory, without raising any errors (look at the Evictions and Freeable memory monitoring charts activity).
This is not acceptable when using Redis as a job queue. The results is corrupted work states (scheduled or completed job counter that never go down)
The default configuration is set to maxmemory-policy, volatile-lru it can be turned to noevictions.
Test must be done to be sure we handle properly the out of memory case.