Archives of the TeradataForum

Message Posted: Thu, 06 Mar 2008 @ 10:53:39 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

I know that bad stats can results in worse query performance than no stats, but I have never seen a case where no stats were better then recent stats until today.

We have a report that�s been in production for a long time now. We received recent complaints that when filtering for a certain value, the report was taking over 30 minutes.

I captured the SQL the report was using and ran explain on it both with and without the offending Where clause. What I discovered was that the addition of this particular Where clause resulted in a product join. The behavior was different with different values in the Where clause, so I suspected something with the stats.

I dropped the stats on that column and the query ran in about 10 seconds.

I collected them again, thinking that maybe they were just old, and ended up killing the query after about 10 minutes.

I was surprised that good stats would result in a bad plan. Does anyone have any insight on this? I always feel that identifying the stats to collect is a crap shoot.

James

	<Prev	Next>		<<First	<Prev	Next>	Last>>