it has been seen in deployments that with large number of containers on datanode. we have seen cache misses on the rocksDB cache.
This jira proposes
a) to have separate caches for open and closed container
b) rocksDB containers in open container cache are never evicted.
c) Open containers are moved from open to close container cache it they are closed/quasi closed.
d) Closed container are cached in LRU fashion as they are cached right now