On Fri, Feb 4, 2011 at 5:17 PM, Bruce Momjian <bruce@momjian.us> wrote:
> Mladen Gogala wrote:
>> characteristic of a religious community chastising a sinner. Let me
>> remind you again: all other major databases have that possibility:
>> Oracle, MySQL, DB2, SQL Server and Informix. Requiring burden of proof
>> about hints is equivalent to saying that all these databases are
>> developed by idiots and have a crappy optimizer.
>
> You need to state the case for hints independent of what other databases
> do, and indepdendent of fixing the problems where the optimizer doesn't
> match reatility.
And that kind of limits to an area where we would the ability to nudge
costs instead of just set them for an individual part of a query.
i.e. join b on (a.a=b.b set selectivity=0.01) or (a.a=b.b set
selectivity=1.0) or something like that. i.e. a.a and b.b have a lot
of matches or few, etc. If there's any thought of hinting it should
be something that a DBA, knowing his data model well, WILL know more
than the current planner because the planner can't get cross table
statistics yet.
But then, why not do something to allow cross table indexes and / or
statistics? To me that would go much further to helping fix the
issues where the current planner "flies blind".
--
To understand recursion, one must first understand recursion.