Archives of the TeradataForum
Message Posted: Fri, 27 Sep 2002 @ 16:39:22 GMT
We have a query that normally runs in 30 seconds or less when the system has a low-to-moderate workload and does not skew. It does a straightforward 3-table join and returns about 5000 rows.
We have another query that does an INSERT SELECT into an empty table and skews onto 2 (virtual) AMPs.
If the normally quick query is submitted when the INSERT SELECT is running then it becomes a never-ending skewed query onto 2 or, for one test run, 1 AMP(s). In addition PMON indicates that one of our 4 nodes becomes either CPU- or I/O-bound when this occurs.
When the problem happens the 3-table join will not finish even if the skewed query is cancelled long before. On one occasion we allowed it to run for about 2 more hours before cancelling it. We have checked on both PMON and HUTCNS and neither indicate that the cancelled INSERT SELECT is aborting - we don't expect it to do so because the target table is empty at the start of the query. When run without the other query running the INSERT SELECT completes within 10 minutes.
We can replicate this problem at will but it is currently puzzling the Teradata tech-support people.
Has anybody experienced similar and if so were you able to prevent the problem by changing any system parameters?
|Copyright 2016 - All Rights Reserved|
|Last Modified: 23 Jun 2019|