At 00:24 08/03/19, Matthew wrote:
On Tue, 18 Mar 2008, Chris Kratz wrote:
In moderately complex to very complex ad hoc queries in our system,
we were consistently having the system massively underestimate the
number of rows coming out of join at a low level making these
queries very slow and inefficient.
I have long thought that perhaps Postgres should be a little more
cautious about its estimates, and assume the worst case scenario
sometimes, rather than blindly following the estimates from the
statistics. The problem is that Postgres uses the statistics to
generate best estimates of the cost. However, it does not take into
account the consequences of being wrong. If it was more clever, then
it may be able to decide to use a non-optimal algorithm according to
the best estimate, if the optimal algorithm has the possibility of
blowing up to 1000 times the work if the estimates are off by a bit.
Such cleverness would be very cool, but (I understand) a lot of
work. It would hopefully solve this problem.
Matthew
Just a crazy thought. If Postgres could check its own estimates or
set some limits while executing the query and, if it found that the
estimates were way off, fall back to a less optimal plan immediately
or the next time, that would be cool.
KC
--
Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
