In response to André Volpato <andre.volpato@xxxxxxxxxxxxxxxxxxxxx>: > Richard Huxton escreveu: > > Reg Me Please wrote: > >> While I would not spend resources in fine tuning the count(*), I would > >> spend some to underastand why and how the other ones do it better. > >> > >> Just to be better. > > > > The problem is well understood, and there is extensive discussion in > > the mailing lists archives. The basic problem is that with PG's > > implementation of MVCC the indexes don't have row visibility > > information. The simple solution of adding it to every index entry > > would increase index size substantially imposing costs on every index > > access and update. > > > > There's a thread in -hackers called "Visibility map thoughts" that is > > looking at the situation again and if/how to implement visibility > > information in a compact form. > > > Remember that you can always use serial fields to count a table, like: > > alter table foo add id serial; > select id from foo order by id desc limit 1; > > This should return the same value than count(*), in a few msecs. I don't think so. What kind of accuracy do you have when rows are deleted? Also, sequences are not transactional, so rolled-back transactions will increment the sequence without actually adding rows. -- Bill Moran http://www.potentialtech.com ---------------------------(end of broadcast)--------------------------- TIP 4: Have you searched our list archives? http://archives.postgresql.org/
