Some environments - like OpenShift - have a very limited of file descriptors available. Even in less contrained environments I believe this could make a good performance booster, provided the application can live with the contraints: updating an IndexReader only after an index merge operation means to have queries always slightly out of synch, not exposing the latest changes to the index.
On the other hand this would highly improve the cache hits for IR internal buffers and cache filters.
|