[
https://issues.apache.org/jira/browse/FLINK-5266?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15730904#comment-15730904
]
ASF GitHub Bot commented on FLINK-5266:
---------------------------------------
GitHub user KurtYoung opened a pull request:
https://github.com/apache/flink/pull/2961
[FLINK-5266] [table] eagerly project unused fields when selecting
aggregation fields
This PR is based on #2926 , only the second commit is related.
I add a "plan" test directory to hold all the plan level tests. And i also
did a small refactory for ProjectionTranslator, thought it's better to keep
each method only do one thing.
@fhueske As we discussed earlier in the jira:
https://issues.apache.org/jira/browse/FLINK-5266 about where the logics should
be added. I decided to add them when we selecting fields from a normal or
grouped table. Since this kind of logics involves some fields references
rewrite, if we choose to add the needed projection node when we convert the
LogicalPlan to Calcite's RelNode, we should also take care the whole rewrite
thing.
However, if we add the project node in the first place, we only need to
extract all the field references used in all selecting expressions, and treat
them as UnresolvedFieldReferences. The validation part will take care of the
rewrite thing. I think this will be easier and more consistent with other
procedures. (Noticed all the "construct" logic are fairly simple)
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/KurtYoung/flink flink-5266
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/2961.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #2961
----
commit 374d231d44f84ae385d8f8adb2353685e1214ff6
Author: Kurt Young <ykt...@gmail.com>
Date: 2016-12-08T01:27:55Z
[FLINK-5226] [table] Use correct DataSetCostFactory and improve DataSetCalc
costs.
commit 8a3ecf8e6362acd9370b11f08018a0143fc9be18
Author: Kurt Young <ykt...@gmail.com>
Date: 2016-12-08T02:35:43Z
[FLINK-5266] [table] eagerly project unused fields when selecting
aggregation fields
Add a "plan" test dir to hold all the plan level unit tests
Small refactory with ProjectionTranslator, keep each method handle one
single thing
----
> Eagerly project unused fields when selecting aggregation fields
> ---------------------------------------------------------------
>
> Key: FLINK-5266
> URL: https://issues.apache.org/jira/browse/FLINK-5266
> Project: Flink
> Issue Type: Improvement
> Components: Table API & SQL
> Reporter: Kurt Young
> Assignee: Kurt Young
>
> When we call table's {{select}} method and if it contains some aggregations,
> we will project fields after the aggregation. Would be better to project
> unused fields before the aggregation, and can furthermore leave the
> opportunity to push the project into scan.
> For example, the current logical plan of a simple query:
> {code}
> table.select('a.sum as 's, 'a.max)
> {code}
> is
> {code}
> LogicalProject(s=[$0], TMP_2=[$1])
> LogicalAggregate(group=[{}], TMP_0=[SUM($5)], TMP_1=[MAX($5)])
> LogicalTableScan(table=[[supplier]])
> {code}
> Would be better if we can project unused fields right after scan, and looks
> like this:
> {code}
> LogicalProject(s=[$0], EXPR$1=[$0])
> LogicalAggregate(group=[{}], EXPR$1=[SUM($0)])
> LogicalProject(a=[$5])
> LogicalTableScan(table=[[supplier]])
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
