Details
-
Type:
Bug
-
Status: Closed
-
Priority:
Major
-
Resolution: Fixed
-
Affects Version/s: 1.3.0-incubating
-
Fix Version/s: 1.3.0-incubating
-
Component/s: None
-
Labels:None
Description
In the following query, the subquery is producing a scalar value but the logical plan still creates a SqlSingleValueAggFunction.
select r_regionkey from region where r_regionkey > (select min(n_regionkey) * 2 from nation);
If the aggregate is just min(n_regionkey) instead of the expression min(n_regionkey) * 2 then no SqlSingleValueAggFunction is created. Ideally, both should behave the same.
This is not necessarily a bug but it does create a burden on the underlying execution engine to support this function even in cases where it may not be needed.
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
I am working on this issue.
Sent a pull request:
https://github.com/apache/incubator-calcite/pull/80
Util.java has a new method to determine if the output for a subquery is scalar. Essentially, when this method suggests scalar output, no necessity to add SqlSingleValueAggFunction.
It looks like Julian has reviewed this. I agree with his comments. Sean Hsuan-Yi Chu I still think that if someone was to accidentally use the new Util.isSingleValue() function for a subquery that had an aggregate function and a group-by and the caller did not check for the grouping list, they would incorrectly get True.
I added some tests for scalar sub-queries in https://git1-us-west.apache.org/repos/asf?p=incubator-calcite.git;a=blob;f=core/src/test/resources/sql/scalar.oq;h=103df2926309e18d380b2bff1e89875ba6287086;hb=f107218b956a1b25d11c8d06c6378bdd22c75378 so I think we're good.
Resolved in release 1.3.0-incubating (2015-05-30).
I'd say it's a bug, because we are generating a more complex, and perhaps less efficient, expression than we need.
It could be fixed (a) by recognizing the pattern at sql-to-rel time, of (b) using a planner rule. For (b) it is not safe to rely on estimated row count, but you could rely on RelMetadataQuery.areColumnsUnique or RelMetadataQuery.getUniqueKeys for an empty list of columns.