On Mon, May 27, 2013 at 2:36 PM, Wolfgang Enzinger <sql...@enzinger.net>wrote:
> Hallo list, > > I detected a dramatic performance drop with a certain query which occurred > somewhere between versions 3.7.09.0 and 3.7.17.0. Now this is not a real > showstopper because I also noticed than an ANALYZE cures the issue in the > latter version immediately. However, the drop without ANALYZE is really > dramatic (~5 secs. in 3.7.09.0, but no result even after a couple of hours > in 3.7.17.0). > > Is this of any interest? What it means is that the query optimizer in 3.7.17 is choosing a different query plan from 3.7.9 and that the new plan is slower than the old one. The query optimizer is now in the process of being completely rewritten for the next release. So we are not really interested in patching 3.7.17 for this. However, once we get the new query planner running and ready to beta-test, we really, really would appreciate you giving it a spin and letting us know if you encounter any problems with it. > If so, I can make a showcase which demonstrates the > problem (with obfuscated data). EXPLAIN QUERY PLAN shows essential > differences also between the two versions, although every SEARCH TABLE > operation uses an index in both cases. > > While we're at it: is there a place where I can download "historic" > precompiled binaries? If so, I could probably tell more exactly where the > drop was introduced. > > Wolfgang > > _______________________________________________ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users > -- D. Richard Hipp d...@sqlite.org _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
