I don't know how good a test this one query is, but using the same query as in the expert search bug, I see performance decrease.
http://pastebin.com/wRhKTPWu Doug. > Hello all, > > I was puzzling today about our planner's apparently inability to properly > use some of the views (e.g. record_attr, rec_descriptor). The hstore > passthru step seems to be the most likely source of opacity, so I wondered > if we might try to use crosstab() instead. > > Here is a (fairly ugly) attempt to do so: > > http://pastebin.com/aEvqzXAc > > I have very little experience using crosstabs, so this can probably be > tidied up somehow. That said, I see promising results on the test data (a > sample query falls from ~50ms to ~1.5ms), but we need to try it on a > production set to really see if it helps. > > Anyone care to try it out, or weigh in on this approach? A good test > might be the "expert search" without Mike's latest patch, or perhaps just > do some selects of rec_descriptor and see what happens. > > Thanks! > Dan > > > Daniel Wells > Library Programmer/Analyst > Hekman Library, Calvin College > 616.526.7133 > > > >
