Currently the JiraLuceneFieldCache has a cache that does nothing (this is one purpose). The cache used to be in place (but I am sure it was broken back then) bu it was disabled.
It used by Both Lucene and 1/2DimensionalermHitCollectors and anything else that uses the StatisticAccessorBean.searchCountMap().
This creates an index the size of the index that sticks around for the request and does not cache it.
There was an old cache in place but Scott removed this around 3.8.
Options:
- Fix cache - add caching to getMatches() but not for getCustom().
- Switch everything that use this to use a DocHitCollector (a LOT more memory efficient but may need to do performance hits).
Improvement
Open
Major
One/TwoDimensionalTermHitCollectors use StatsJiraLuceneFieldCache with no cacheing