[
https://issues.apache.org/jira/browse/FLINK-5406?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15853576#comment-15853576
]
ASF GitHub Bot commented on FLINK-5406:
---------------------------------------
Github user godfreyhe commented on a diff in the pull request:
https://github.com/apache/flink/pull/3101#discussion_r99522172
--- Diff:
flink-libraries/flink-table/src/test/scala/org/apache/flink/table/AggregationTest.scala
---
@@ -165,14 +165,20 @@ class AggregationTest extends TableTestBase {
val util = batchTestUtil()
val sourceTable = util.addTable[(Int, Long, Int)]("MyTable", 'a, 'b,
'c)
- val resultTable = sourceTable.groupBy('a)
+ // Move "where" before "groupBy" for the former query would generate
+ // nondeterministic plans with same cost. If we change
FlinkRuleSets.DATASET_OPT_RULES,
+ // the importance of relNode may change, and the test may fail. This
issue is mentioned
+ // in FLINK-5394, we could move "where" to the end when FLINK-5394 is
fixed.
+ val resultTable = sourceTable.where('a === 1)
--- End diff --
OK
> add normalization phase for predicate logical plan rewriting between
> decorrelate query phase and volcano optimization phase
> ---------------------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-5406
> URL: https://issues.apache.org/jira/browse/FLINK-5406
> Project: Flink
> Issue Type: New Feature
> Components: Table API & SQL
> Reporter: godfrey he
> Assignee: godfrey he
>
> Normalization phase is for predicate logical plan rewriting and is
> independent of cost module. The rules in normalization phase do not need to
> repeatedly applied to different logical plan which is different to volcano
> optimization phase. And the benefit of normalization phase is to reduce the
> running time of volcano planner.
> *ReduceExpressionsRule* can apply various simplifying transformations on
> RexNode trees. Currently, there are two transformations:
> 1) Constant reduction, which evaluates constant subtrees, replacing them with
> a corresponding RexLiteral
> 2) Removal of redundant casts, which occurs when the argument into the cast
> is the same as the type of the resulting cast expression
> the above transformations do not depend on the cost module, so we can move
> the rules in *ReduceExpressionsRule* from
> DATASET_OPT_RULES/DATASTREAM_OPT_RULES to DataSet/DataStream Normalization
> Rules.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
