On Thu, Apr 9, 2009 at 9:24 PM, Mark Miller <[email protected]> wrote: > Grant Ingersoll wrote: >> >> https://issues.apache.org/jira/browse/LUCENE-1516 >> >> Has anyone looked yet at how >> https://issues.apache.org/jira/browse/LUCENE-1516 might be used in Solr? >> Seems like it might allow us to speed up searcher turnaround. >> >> WDYT? > > Can't wait for it to happen myself. Part of the problem with it is that it > counts on all of the benefits of LUCENE-1483 (per segment caches for > reopen), and Solr has some tricky issues ahead to get there I think.
Including the fact that Solr filters, function queries, and facets are all top-level multi-reader based. For some things like uninverted-field faceting, it's completely non-trivial to switch to an incremental per-segment algorithm. Another issue with per-segment caching is that it helps the average case, but not the worst case... because a big segment merge can mean throwing away most of your per-segment caches anyway, right? Anyway, I think part of the groundwork has been laid for per-segment caching in Solr via SolrIndexReader. -Yonik http://www.lucidimagination.com
