Currently, the Job Server daemon on startup will index all the history files it finds in its configured location (HISTORY dir). It will also regularly garbage collect any orphaned indices that no longer have a matching history file (by inode name). However, Matt has stated that by design Grid admins could choose to throw away the larger history files in favour of keeping just the smaller index files. Need some clearer design requirements here.
Kicking this as an RFE for a future release. The current code is reported to work and garbage collect the indices. Before removing the garbage collecting functionality from the JS, another defined mechanism should be provided.