You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@trafodion.apache.org by "David Wayne Birdsall (JIRA)" <ji...@apache.org> on 2015/09/18 22:41:04 UTC
[jira] [Created] (TRAFODION-1494) UPDATE STATS sample table
population query gets serial plan
David Wayne Birdsall created TRAFODION-1494:
-----------------------------------------------
Summary: UPDATE STATS sample table population query gets serial plan
Key: TRAFODION-1494
URL: https://issues.apache.org/jira/browse/TRAFODION-1494
Project: Apache Trafodion
Issue Type: Bug
Components: sql-cmp
Affects Versions: 1.1 (pre-incubation)
Environment: All
Reporter: David Wayne Birdsall
Assignee: David Wayne Birdsall
Fix For: 2.0-incubating
If a table is large enough so that sample data does not fit in memory, UPDATE STATS will create a table and use an UPSERT or LOAD statement to load it with a sample of the table of interest. The query plan being generated for this UPSERT/LOAD is a serial plan which is quite slow. It should generate a parallel plan.
The problem is in CmpSeabaseDDL::restoreAllControlsAndFlags. We set a bunch of CQDs for metadata queries so their plans are not affected by any user CQDs. The procedure CmpSeabaseDDL::restoreAllControlsAndFlags is in charge of restoring the CQDs afterward, but fails to.
One of the CQDs that fails to get reset is "attempt_esp_parallelism". It is turned OFF for the metadata queries. Since it does not get reset, all plans afterward (including sample table population) are serial.
The bug is in the following statement in CmpSeabaseDDL::restoreAllControlsAndFlags:
if (CmpCommon::context()->getCntlCount() > 0 &&
CmpCommon::context()->getCntlCount() < CQD_SENT_MAX)
The "<" should be a "<=". (Off-by-one bug.)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)