Kevin Hunter <hunteke@earlham.edu> writes:
> At 1:17p -0400 on 01 Jun 2007, Tyler Durden wrote:
>> I find strange that a simple SELECT COUNT(...) is so slow with only
>> 700 000 records.
> The much more knowledgable will correct me, but the abbr. version is
> that it is for data integrity and correctness reasons and the
> inherent way in which the MVCC model works.
The bottom line is that a "correct" implementation (ie, one that fully
respects MVCC behavior) would create enormous overhead, as well as bad
contention bottlenecks for concurrent updates. It doesn't seem worth it.
If you want a cheap approximate answer, there are a couple of ways to
get one, but SELECT COUNT(*) is not that.
regards, tom lane