[
https://issues.apache.org/jira/browse/TRAFODION-1562?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15863257#comment-15863257
]
ASF GitHub Bot commented on TRAFODION-1562:
-------------------------------------------
Github user sandhyasun commented on a diff in the pull request:
https://github.com/apache/incubator-trafodion/pull/946#discussion_r100732218
--- Diff: core/sql/optimizer/BindRelExpr.cpp ---
@@ -10600,37 +10600,67 @@ Upsert is also converted into merge when
TRAF_UPSERT_MODE is set to MERGE and
there are omitted cols with default values in case of aligned format table
or
omitted current timestamp cols in case of non-aligned row format
*/
-NABoolean Insert::isUpsertThatNeedsMerge(NABoolean isAlignedRowFormat,
NABoolean omittedDefaultCols,
- NABoolean
omittedCurrentDefaultClassCols) const
+NABoolean Insert::isUpsertThatNeedsTransformation(NABoolean
isAlignedRowFormat,
+ NABoolean
omittedDefaultCols,
+ NABoolean
omittedCurrentDefaultClassCols,
+ NABoolean &toMerge) const
{
- // The necessary conditions to convert upsert to merge and
+ toMerge = FALSE;
+ // If the the table has an identity column in clustering key or has a
syskey
+ // we dont need to do this transformation.The incoming row will always
be
+ // unique. So first check if we any of the conditions are satisfied to
+ //even try the transform
+ NABoolean mustTryTransform = FALSE;
if (isUpsert() &&
- (NOT getIsTrafLoadPrep()) &&
- (NOT (getTableDesc()->isIdentityColumnGeneratedAlways() &&
getTableDesc()->hasIdentityColumnInClusteringKey())) &&
- (NOT
(getTableDesc()->getClusteringIndex()->getNAFileSet()->hasSyskey())) &&
- // table has secondary indexes or
- (getTableDesc()->hasSecondaryIndexes() ||
- // CQD is set to MERGE
- ((CmpCommon::getDefault(TRAF_UPSERT_MODE) == DF_MERGE) &&
- // omitted current default columns with non-aligned row format
tables
- // or omitted default columns with aligned row format tables
- (((NOT isAlignedRowFormat) && omittedCurrentDefaultClassCols)
||
- (isAlignedRowFormat && omittedDefaultCols))) ||
- // CQD is set to Optimal, for non-aligned row format with
omitted
- // current columns, it is converted into merge though it is not
- // optimal for performance - This is done to ensure that when
the
- // CQD is set to optimal, non-aligned format would behave like
- // merge when any column is omitted
- ((CmpCommon::getDefault(TRAF_UPSERT_MODE) == DF_OPTIMAL) &&
- ((NOT isAlignedRowFormat) && omittedCurrentDefaultClassCols))
- )
- )
- return TRUE;
- else
- return FALSE;
+ ( NOT getIsTrafLoadPrep()) &&
+ (NOT (getTableDesc()->isIdentityColumnGeneratedAlways() &&
+ getTableDesc()->hasIdentityColumnInClusteringKey())) &&
+
(NOT(getTableDesc()->getClusteringIndex()->getNAFileSet()->hasSyskey())))
+ {
+ mustTryTransform = TRUE;
+ }
+
+ // Transform upsert to merge in case of special modes and
+ // omitted default columns
+ // Case 1 : CQD is set to MERGE, omitted current(timestamp) default
+ // columns with non-aligned row format table or omitted
+ // default columns with aligned row format tables
+
+ // Case 2 : CQD is set to Optimal, for non-aligned row format with
omitted
+ // current(timestamp) columns, it is converted into merge
+ // though it is not optimal for performance. This is done to
ensure
+ // that when the CQD is set to optimal, non-aligned format
would
+ // behave like merge when any column is omitted
+ if (isUpsert() &&
+ mustTryTransform &&
+ ((CmpCommon::getDefault(TRAF_UPSERT_MODE) == DF_MERGE) &&
+ (((NOT isAlignedRowFormat) && omittedCurrentDefaultClassCols) ||
+ (isAlignedRowFormat && omittedDefaultCols)))
+ ||
+ ((CmpCommon::getDefault(TRAF_UPSERT_MODE) == DF_OPTIMAL) &&
+ ((NOT isAlignedRowFormat) && omittedCurrentDefaultClassCols)))
+ {
+ toMerge = TRUE;
--- End diff --
If we choose the new tree, we won't get merge behaviour. We will get
"replace" behaviour. I do think we need to make DF_OPTIMAL the default
behaviour so the more efficient one is the default behaviour and if the user
wants merge behaviour they can choose the slower method. We can discuss that
separately.
> 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
(v6.3.15#6346)
