I've written about my problem ~2 years ago:
http://wikitech-l.wikimedia.narkive.com/6G0YPmWQ/need-a-way-to-modify-text-…
It seems I've lost the latest message, so I want to answer to it now:
With lsearchd and Elasticsearch, we absolutely
wouldn't want to munge
file text into page content (with sql-backed search, you might maybe).
Why?? Aren't these also just the fulltext search backends? As I
understand they're much faster than sql-backed search engines. What
would prevent them to store file texts?
Personally I use Sphinx (
http://sphinxsearch.com) with TikaMW, and of
course everything is fine.