Hey :), On Wed, Dec 21, 2016 at 2:49 PM, Philip Van Hoof <[email protected]> wrote: > On Wed, 2016-12-21 at 14:29 +0100, Carlos Garnacho wrote:
<snip> > >> I need testing with further nodes though, as I'd expect things to get >> linearly worse with data, both in size with the additional indexes and >> performance w/o indexes. I kinda expected we'd have to pick our poison >> here though :). > > > Might be that just counting references and incrementing them in a column > in the Resource table is cheaper than letting SQLite do all the extra > required indexes? Yeah, could be... the workload is spread between insertions and deletes then, but at least the extra updates would poke already indexed columns. A downside is that it also makes the harvesting of broken triples rather manual. > > But right. Without testing, it's hard to know. And if SQLite does it for > us with CASCADING Delete and RESTRICT: that's less code for us write. That is my line of thought, the more we get the database to work for us the better. Anyhow, I'll try to play with your explicit refcounting idea, could be made to be automatically maintained with triggers. Cheers, Carlos _______________________________________________ tracker-list mailing list [email protected] https://mail.gnome.org/mailman/listinfo/tracker-list
