> I actually think the way to attack this issue is to discuss the kinds
of errors the planner makes, and what tweaks we could do to correct
them.
> Here's the ones I'm aware of:
>
> -- Incorrect selectivity of WHERE clause
> -- Incorrect selectivity of JOIN
> -- Wrong estimate of rows returned from SRF
> -- Incorrect cost estimate for index use
>
> Can you think of any others?
I think your points are too generic, there is no way to get them all
100% correct from statistical
data even with data hints (and it is usually not at all necessary for
good enough plans).
I think we need to more precisely define the problems of our system with
point in time statistics
-- no reaction to degree of other concurrent activity
-- no way to react to abnormal skew that only persists for a very short
duration
-- too late reaction to changing distribution (e.g. current date column
when a new year starts)
and the variant: too late adaption when a table is beeing filled
-- missing cost/selectivity estimates for several parts of the system
Andreas
