Archives of the TeradataForum

Message Posted: Thu, 31 Jul 2003 @ 20:52:40 GMT

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

Srini,

Do you know if the failed job suffered from skewed spool file? Were you running with the same data as during the batch job?

I've seen this happen when the SQL is reading from a work table which didn't have stats collected on it and it was often skewed. Most of the time it ran fine, but occasionally you'd get skewed data, the optimiser would happen to pick particular AMP to generate it's row counts from, the join plan was 'not optimal' and the processing when awol. The thing that we noticed was that on occassions when the job failed, someone on support would copy the work table to their own database and re-run the SQL, it worked fine ! (random amp sample really is 'random' !)

Collecting stats on the work table cured this problem.

Check that out and see what you find.

Cheers,

Dave

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