Re:SQLite 3.8.0 Release with next-generation query planner
by meshko on Monday August 26, 2013 @11:08PM
(#6281247)
The write up of the NGQP seems (at least superficially?) weird: they say that the reason for the rewrite was poor performance on a particular artificial benchmark (8-way join.. how often do people join 8 tables in practice?). At the same time, I was surprised to learn that the only kind of join algorithm SQLite supports is nested loops. I would think they would benefit more from adding hash and merge joins instead of rewriting the QP from scratch.
[ Parent
| on HN ]
Re:SQLite 3.8.0 Release with next-generation query planner
by hobs on Tuesday August 27, 2013 @12:46AM
(#6281499)
I regularly do 20+ table joins.
Re:SQLite 3.8.0 Release with next-generation query planner
by lsb on Monday August 26, 2013 @11:50PM
(#6281352)
SQLite 3.7.17 already makes automatic indices, if none exists for the join, if the planner recommends it.
Re:SQLite 3.8.0 Release with next-generation query planner
by recursive on Tuesday August 27, 2013 @12:13PM
(#6284323)
> how often do people join 8 tables in practice?
If I were to judge based on my current experience, I'd say more often than they make queries with less than 8 joins. In the stored procedure I'm working on now, there are many queries with up to 20.