Cache following a "Most Recently Used" (MRU) algorithm for maintaining a bounded in-memory size; the "Least Recently Used" (LRU) entry is the first available for removal from the cache.
This implementation uses a "soft limit" to the in-memory size of the cache, meaning that all cache entries are kept within a completely {@link java.lang.ref.SoftReference}-based map with the most recently utilized entries additionally kept in a hard-reference manner to prevent those cache entries soft references from becoming enqueued by the garbage collector. Thus the actual size of this cache impl can actually grow beyond the stated max size bound as long as GC is not actively seeking soft references for enqueuement.
The soft-size is bounded and configurable. This allows controlling memory usage which can grow out of control under some circumstances, especially when very large heaps are in use. Although memory usage per se should not be a problem with soft references, which are cleared when necessary, this can trigger extremely slow stop-the-world GC pauses when nearing full heap usage, even with CMS concurrent GC (i.e. concurrent mode failure). This is most evident when ad-hoc HQL queries are produced by the application, leading to poor soft-cache hit ratios. This can also occur with heavy use of SQL IN clauses, which will generate multiples SQL queries (even if parameterized), one for each collection/array size passed to the IN clause. Many slightly different queries will eventually fill the heap and trigger a full GC to reclaim space, leading to unacceptable pauses in some cases.
Note: This class is serializable, however all entries are discarded on serialization.
@author Steve Ebersole
@author Manuel Dominguez Sarmiento
@see org.hibernate.cfg.Environment#QUERY_PLAN_CACHE_MAX_STRONG_REFERENCES
@see org.hibernate.cfg.Environment#QUERY_PLAN_CACHE_MAX_SOFT_REFERENCES