[
https://issues.apache.org/jira/browse/TRAFODION-1562?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15652762#comment-15652762
]
ASF GitHub Bot commented on TRAFODION-1562:
-------------------------------------------
Github user zellerh commented on a diff in the pull request:
https://github.com/apache/incubator-trafodion/pull/826#discussion_r87316978
--- Diff: core/sql/optimizer/Inlining.cpp ---
@@ -618,25 +619,58 @@ RETDesc
*GenericUpdate::createOldAndNewCorrelationNames(BindWA *bindWA, NABoolea
}
if ((getOperatorType() != REL_UNARY_INSERT) ||
- getUpdateCKorUniqueIndexKey())
+ getUpdateCKorUniqueIndexKey() ||
+ ((getOperatorType() == REL_UNARY_INSERT) &&((Insert
*)this)->isMerge()) ||
+ ((getOperatorType() == REL_UNARY_INSERT) && ((Insert
*)this)->isUpsert() && (CmpCommon::getDefault(TRAF_UPSERT_TO_EFF_TREE) == DF_ON
)))
{
// DELETE or UPDATE --
// Now merge the old/target/before valueid's (the Scan child RETDesc)
// into this RETDesc such that these cols are all named "OLD@.<col>"
- //
+ //f
Scan *scan ;
if (getOperatorType() != REL_UNARY_INSERT)
scan = getScanNode();
else
scan = getLeftmostScanNode();
+ if ((getOperatorType() == REL_UNARY_INSERT) && ((Insert
*)this)->isUpsert() && (CmpCommon::getDefault(TRAF_UPSERT_TO_EFF_TREE) == DF_ON
))
+ {
+ RelSequence *olapChild = getOlapChild();
+ CorrName
corrName(getTableDesc()->getCorrNameObj().getQualifiedNameObj(),
+ bindWA->wHeap(),
+ OLDCorr);
+
+ // ColumnDescList *colList =
(olapChild->getRETDesc())->getColumnList();
+ for (short i = 0; i< olapChild->getRETDesc()->getDegree();i++)
+ {
+ // we remembered if the original columns was from the right side of
+ // this olap node so add those to the RetDesc since those are the
+ //ones we want to delete from the dependent indexes.
+ if
((olapChild->getRETDesc()->getValueId(i)).getItemExpr()->origOpType() ==
ITM_INSTANTIATE_NULL)
--- End diff --
The left side of the left join could also produce some ITM_INSTANTIATE_NULL
expressions, couldn't it? Wouldn't that happen if the select part of the upsert
is itself a left join? Maybe get to the left join somehow and use its
nullInstantiatedOutput() method, then map that through the sequence node's map?
> Changes in query tree when the upsert command is transformed into merge
> -----------------------------------------------------------------------
>
> Key: TRAFODION-1562
> URL: https://issues.apache.org/jira/browse/TRAFODION-1562
> Project: Apache Trafodion
> Issue Type: Sub-task
> Components: sql-cmp
> Reporter: Selvaganesan Govindarajan
> Assignee: Sandhya Sundaresan
> Attachments: BatchUpsertTransformation.pdf
>
>
> to improve the performance as explained in the main JIRA
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
