Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

fix: Correct results for grouping sets when columns contain nulls #12571

Merged
merged 7 commits into from
Oct 7, 2024

Conversation

eejbyfeldt
Copy link
Contributor

@eejbyfeldt eejbyfeldt commented Sep 21, 2024

Which issue does this PR close?

Closes #12570

Rationale for this change

Currently we produce incorrect results when combining grouping sets and columns containing null values.

What changes are included in this PR?

The bug is fixed by introducing an internal column grouping_id when using grouping sets. This extra column makes sure that we create different groups for the nulls from the grouping sets and the data.

This approach is based on how it is implemented in Spark and has previously been proposed here: #5749 Note that this change is smaller in scope and limit the existence of the grouping_id to the physical plan (Added it also to the logical as this will be needed for implementing the grouping function and it simplifies the code significantly). This is done so we end up with a smaller PR that is easier to review. But we might want to follow up and extend it to the logical plan an use it to implement the grouping function (#5647) in a similar manner to what is done in Spark.

Are these changes tested?

Existing and new sqllogictests.

Are there any user-facing changes?

@github-actions github-actions bot added physical-expr Physical Expressions core Core DataFusion crate sqllogictest SQL Logic Tests (.slt) labels Sep 21, 2024
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch 5 times, most recently from f4a220b to 9c840b0 Compare September 22, 2024 10:27
@github-actions github-actions bot added the optimizer Optimizer rules label Sep 22, 2024
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch 2 times, most recently from 8d01437 to fdce177 Compare September 22, 2024 11:19
@eejbyfeldt eejbyfeldt marked this pull request as ready for review September 23, 2024 17:08
@eejbyfeldt eejbyfeldt changed the title fix: Grouping sets when columns contain nulls fix: Correct results for grouping sets when columns contain nulls Sep 23, 2024
@alamb
Copy link
Contributor

alamb commented Sep 23, 2024

Thank you @eejbyfeldt

cc @thinkharderdev as I think you / your team implemented the GROUPING SETS implementation originally

Copy link
Contributor

@thinkharderdev thinkharderdev left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice work! Had a few comments and questions :)

@@ -108,6 +110,8 @@ impl AggregateMode {
}
}

const INTERNAL_GROUPING_ID: &str = "grouping_id";
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What happens if this conflicts with a user-defined field? E.g. if I had a query like:

SELECT grouping_id, count(1) FROM table GROUP BY CUBE(grouping_id)

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seems to just work (which surprised me).

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pushed changes to add it to the logical plan, so now name conflicts are possible. And renamed it __grouping_id, this follows the pattern used for __common_expr which can also conflict with names from user schema. Since it it internal I think it should be possible to follow up and make it fully unique.

datafusion/physical-plan/src/aggregates/mod.rs Outdated Show resolved Hide resolved
datafusion/physical-plan/src/aggregates/mod.rs Outdated Show resolved Hide resolved
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch from fdce177 to fee8bdf Compare September 25, 2024 18:09
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch 2 times, most recently from 230e4ef to 37dc663 Compare September 29, 2024 08:09
@github-actions github-actions bot added logical-expr Logical plan and expressions and removed optimizer Optimizer rules labels Sep 29, 2024
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch 2 times, most recently from 37d6283 to 2914017 Compare October 1, 2024 12:24
@github-actions github-actions bot added the sql SQL Planner label Oct 1, 2024
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch from 2914017 to e9c7e13 Compare October 1, 2024 19:23
@github-actions github-actions bot added optimizer Optimizer rules substrait labels Oct 1, 2024
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch from e9c7e13 to 377e348 Compare October 1, 2024 19:35
@github-actions github-actions bot removed the sql SQL Planner label Oct 1, 2024
@eejbyfeldt
Copy link
Contributor Author

@thinkharderdev I made some changes so that the grouping id exist already in the logical plan. This simplified the logic quite a bit.

I also pushed a follow up PR #12704 that shows how we can use grouping id to implement the grouping function.

Copy link
Contributor

@alamb alamb left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you @eejbyfeldt and @thinkharderdev

I reviewed this PR and the logic makes sense to me, it seems well tested and structured, and commented. Thank you very much for this contribution

datafusion/expr/src/logical_plan/plan.rs Show resolved Hide resolved
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch from 86eb434 to 920b384 Compare October 5, 2024 08:17
Copy link
Contributor

@alamb alamb left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good -- I plan to merge this PR tomorrow unless anyone else would like more time to review.

Thanks again @eejbyfeldt

@alamb
Copy link
Contributor

alamb commented Oct 6, 2024

🤔 looks like some newly added CI tests are failing

@github-actions github-actions bot added the sql SQL Planner label Oct 6, 2024
@eejbyfeldt eejbyfeldt force-pushed the fix-grouping-sets-with-null-values branch from 543daa3 to 1f61ddf Compare October 6, 2024 19:18
@eejbyfeldt
Copy link
Contributor Author

🤔 looks like some newly added CI tests are failing

Fix the test. But looks like there might be other failures that are also on master.

@alamb
Copy link
Contributor

alamb commented Oct 7, 2024

Gah -- I think CI / clippy will be fixed by #12724

Sorry about this @eejbyfeldt

@alamb
Copy link
Contributor

alamb commented Oct 7, 2024

Merged up from main to pick up #12724 and hopefully get a clean CI run

@alamb alamb merged commit ef227f4 into apache:main Oct 7, 2024
24 checks passed
@alamb
Copy link
Contributor

alamb commented Oct 7, 2024

🚀

alamb added a commit that referenced this pull request Oct 8, 2024
* Add support for external tables with qualified names (#12645)

* Make  support schemas

* Set default name to table

* Remove print statements and stale comment

* Add tests for create table

* Fix typo

* Update datafusion/sql/src/statement.rs

Co-authored-by: Jonah Gao <[email protected]>

* convert create_external_table to objectname

* Add sqllogic tests

* Fix failing tests

---------

Co-authored-by: Jonah Gao <[email protected]>

* Fix Regex signature types (#12690)

* Fix Regex signature types

* Uncomment the shared tests in string_query.slt.part and removed tests copies everywhere else

* Test `LIKE` and `MATCH` with flags; Remove new tests from regexp.slt

* Refactor `ByteGroupValueBuilder` to use `MaybeNullBufferBuilder` (#12681)

* Fix malformed hex string literal in docs (#12708)

* Simplify match patterns in coercion rules (#12711)

Remove conditions where unnecessary.
Refactor to improve readability.

* Remove aggregate functions dependency on frontend (#12715)

* Remove aggregate functions dependency on frontend

DataFusion is a SQL query engine and also a reusable library for
building query engines. The core functionality should not depend on
frontend related functionalities like `sqlparser` or `datafusion-sql`.

* Remove duplicate license header

* Minor: Remove clone in `transform_to_states` (#12707)

* rm clone

Signed-off-by: jayzhan211 <[email protected]>

* fmt

Signed-off-by: jayzhan211 <[email protected]>

---------

Signed-off-by: jayzhan211 <[email protected]>

* Refactor tests for union sorting properties, add tests for unions and constants (#12702)

* Refactor tests for union sorting properties

* update doc test

* Undo import reordering

* remove unecessary static lifetimes

* Fix: support Qualified Wildcard in count aggregate function (#12673)

* Reduce code duplication in `PrimitiveGroupValueBuilder` with const generics (#12703)

* Reduce code duplication in `PrimitiveGroupValueBuilder` with const generics

* Fix docs

* Disallow duplicated qualified field names (#12608)

* Disallow duplicated qualified field names

* Fix tests

* Optimize base64/hex decoding by pre-allocating output buffers (~2x faster) (#12675)

* add bench

* replace macro with generic function

* remove duplicated code

* optimize base64/hex decode

* Allow DynamicFileCatalog support to query partitioned file (#12683)

* support to query partitioned table for dynamic file catalog

* cargo clippy

* split partitions inferring to another function

* Support `LIMIT` Push-down logical plan optimization for `Extension` nodes (#12685)

* Update trait `UserDefinedLogicalNodeCore`

Signed-off-by: Austin Liu <[email protected]>

* Update corresponding interface

Signed-off-by: Austin Liu <[email protected]>

Add rewrite rule for `push-down-limit` for `Extension`

Signed-off-by: Austin Liu <[email protected]>

* Add rewrite rule for `push-down-limit` for `Extension` and tests

Signed-off-by: Austin Liu <[email protected]>

* Update corresponding interface

Signed-off-by: Austin Liu <[email protected]>

* Reorganize to match guard

Signed-off-by: Austin Liu <[email protected]>

* Clena up

Signed-off-by: Austin Liu <[email protected]>

Clean up

Signed-off-by: Austin Liu <[email protected]>

---------

Signed-off-by: Austin Liu <[email protected]>

* Fix AvroReader: Add union resolving for nested struct arrays (#12686)

* Add union resolving for nested struct arrays

* Add test

* Change test

* Reproduce index error

* fmt

---------

Co-authored-by: Andrew Lamb <[email protected]>

* Adds macros for creating `WindowUDF` and `WindowFunction` expression (#12693)

* Adds macro for udwf singleton

* Adds a doc comment parameter to macro

* Add doc comment for `create_udwf` macro

* Uses default constructor

* Update `Cargo.lock` in `datafusion-cli`

* Fixes: expand `$FN_NAME` in doc strings

* Adds example for macro usage

* Renames macro

* Improve doc comments

* Rename udwf macro

* Minor: doc copy edits

* Adds macro for creating fluent-style expression API

* Adds support for 1 or more parameters in expression function

* Rewrite doc comments

* Rename parameters

* Minor: formatting

* Adds doc comment for `create_udwf_expr` macro

* Improve example docs

* Hides extraneous code in doc comments

* Add a one-line readme

* Adds doc test assertions + minor formatting fixes

* Adds common macro for defining user-defined window functions

* Adds doc comment for `define_udwf_and_expr`

* Defines `RowNumber` using common macro

* Add usage example for common macro

* Adds usage for custom constructor

* Add examples for remaining patterns

* Improve doc comments for usage examples

* Rewrite inner line docs

* Rewrite `create_udwf_expr!` doc comments

* Minor doc improvements

* Fix doc test and usage example

* Add inline comments for macro patterns

* Minor: change doc comment in example

* Support unparsing plans with both Aggregation and Window functions (#12705)

* Support unparsing plans with both Aggregation and Window functions (#35)

* Fix unparsing for aggregation grouping sets

* Add test for grouping set unparsing

* Update datafusion/sql/src/unparser/utils.rs

Co-authored-by: Jax Liu <[email protected]>

* Update datafusion/sql/src/unparser/utils.rs

Co-authored-by: Jax Liu <[email protected]>

* Update

* More tests

---------

Co-authored-by: Jax Liu <[email protected]>

* Fix strpos invocation with dictionary and null (#12712)

In 1b3608d `strpos` signature was
modified to indicate it supports dictionary as input argument, but the
invoke method doesn't support them.

* docs: Update DataFusion introduction to clarify that DataFusion does provide an "out of the box" query engine (#12666)

* Update DataFusion introduction to show that DataFusion offers packaged versions for end users

* change order

* Update README.md

Co-authored-by: Andrew Lamb <[email protected]>

* refine wording and update user guide for consistency

* prettier

---------

Co-authored-by: Andrew Lamb <[email protected]>

* Framework for generating function docs from embedded code documentation (#12668)

* Initial work on #12432 to allow for generation of udf docs from embedded documentation in the code

* Add missing license header.

* Fixed examples.

* Fixing a really weird RustRover/wsl ... something. No clue what happened there.

* permission change

* Cargo fmt update.

* Refactored Documentation to allow it to be used in a const.

* Add documentation for syntax_example

* Refactoring Documentation based on PR feedback.

* Cargo fmt update.

* Doc update

* Fixed copy/paste error.

* Minor text updates.

---------

Co-authored-by: Andrew Lamb <[email protected]>

* Add IMDB(JOB) Benchmark [2/N] (imdb queries) (#12529)

* imdb dataset

* cargo fmt

* Add 113 queries for IMDB(JOB)

Signed-off-by: Austin Liu <[email protected]>

* Add `get_query_sql` from `query_id` string

Signed-off-by: Austin Liu <[email protected]>

* Fix CSV reader & Remove Parquet partition

Signed-off-by: Austin Liu <[email protected]>

* Add benchmark IMDB runner

Signed-off-by: Austin Liu <[email protected]>

* Add `run_imdb` script

Signed-off-by: Austin Liu <[email protected]>

* Add checker for imdb option

Signed-off-by: Austin Liu <[email protected]>

* Add SLT for IMDB

Signed-off-by: Austin Liu <[email protected]>

* Fix `get_query_sql()` for CI roundtrip test

Signed-off-by: Austin Liu <[email protected]>

Fix `get_query_sql()` for CI roundtrip test

Signed-off-by: Austin Liu <[email protected]>

Fix `get_query_sql()` for CI roundtrip test

Signed-off-by: Austin Liu <[email protected]>

* Clean up

Signed-off-by: Austin Liu <[email protected]>

* Add missing license

Signed-off-by: Austin Liu <[email protected]>

* Add IMDB(JOB) queries `2b` to `5c`

Signed-off-by: Austin Liu <[email protected]>

* Add `INCLUDE_IMDB` in CI verify-benchmark-results

Signed-off-by: Austin Liu <[email protected]>

* Prepare IMDB dataset

Signed-off-by: Austin Liu <[email protected]>

Prepare IMDB dataset

Signed-off-by: Austin Liu <[email protected]>

* use uint as id type

* format

* Seperate `tpch` and `imdb` benchmarking CI jobs

Signed-off-by: Austin Liu <[email protected]>

Fix path

Signed-off-by: Austin Liu <[email protected]>

Fix path

Signed-off-by: Austin Liu <[email protected]>

Remove `tpch` in `imdb` benchmark

Signed-off-by: Austin Liu <[email protected]>

* Remove IMDB(JOB) slt in CI

Signed-off-by: Austin Liu <[email protected]>

Remove IMDB(JOB) slt in CI

Signed-off-by: Austin Liu <[email protected]>

---------

Signed-off-by: Austin Liu <[email protected]>
Co-authored-by: DouPache <[email protected]>

* Minor: avoid clone while calculating union equivalence properties (#12722)

* Minor: avoid clone while calculating union equivalence properties

* Update datafusion/physical-expr/src/equivalence/properties.rs

* fmt

* Simplify streaming_merge function parameters (#12719)

* simplify streaming_merge function parameters

* revert test change

* change StreamingMergeConfig into builder pattern

* Fix links on docs index page (#12750)

* Provide field and schema metadata missing on cross joins, and union with null fields. (#12729)

* test: reproducer for missing schema metadata on cross join

* fix: pass thru schema metadata on cross join

* fix: preserve metadata when transforming to view types

* test: reproducer for missing field metadata in left hand NULL field of union

* fix: preserve field metadata from right side of union

* chore: safe indexing

* Minor: Update string tests for strpos (#12739)

* Apply `type_union_resolution` to array and values (#12753)

* cleanup make array coercion rule

Signed-off-by: jayzhan211 <[email protected]>

* change to type union resolution

Signed-off-by: jayzhan211 <[email protected]>

* change value too

Signed-off-by: jayzhan211 <[email protected]>

* fix tpyo

Signed-off-by: jayzhan211 <[email protected]>

---------

Signed-off-by: jayzhan211 <[email protected]>

* Add `DocumentationBuilder::with_standard_argument` to reduce copy/paste (#12747)

* Add `DocumentationBuilder::with_standard_expression` to reduce copy/paste

* fix doc

* fix standard argument

* Update docs

* Improve documentation to explain what is different

* fix `equal_to` in `PrimitiveGroupValueBuilder` (#12758)

* fix `equal_to` in `PrimitiveGroupValueBuilder`.

* fix typo.

* add uts.

* reduce calling of `is_null`.

* Minor: doc how field name is to be set (#12757)

* Fix `equal_to` in `ByteGroupValueBuilder` (#12770)

* Fix `equal_to` in `ByteGroupValueBuilder`

* refactor null_equal_to

* Update datafusion/physical-plan/src/aggregates/group_values/group_column.rs

* Allow simplification even when nullable (#12746)

The nullable requirement seem to have been added in #1401 but as far as
I can tell they are not needed for these 2 cases.

I think this can be shown using this truth table: (generated using
datafusion-cli without this patch)
```
> CREATE TABLE t (v BOOLEAN) as values (true), (false), (NULL);
> select t.v, t2.v, t.v AND (t.v OR t2.v), t.v OR (t.v AND t2.v) from t cross join t as t2;
+-------+-------+---------------------+---------------------+
| v     | v     | t.v AND t.v OR t2.v | t.v OR t.v AND t2.v |
+-------+-------+---------------------+---------------------+
| true  | true  | true                | true                |
| true  | false | true                | true                |
| true  |       | true                | true                |
| false | true  | false               | false               |
| false | false | false               | false               |
| false |       | false               | false               |
|       | true  |                     |                     |
|       | false |                     |                     |
|       |       |                     |                     |
+-------+-------+---------------------+---------------------+
```

And it seems Spark applies both of these and DuckDB applies only the
first one.

* Fix unnest conjunction with selecting wildcard expression (#12760)

* fix unnest statement with wildcard expression

* add commnets

* Improve `round` scalar function unparsing for Postgres (#12744)

* Postgres: enforce required `NUMERIC` type for `round` scalar function (#34)

Includes initial support for dialects to override scalar functions unparsing

* Document scalar_function_to_sql_overrides fn

* Fix stack overflow calculating projected orderings (#12759)

* Fix stack overflow calculating projected orderings

* fix docs

* Port / Add Documentation for `VarianceSample` and `VariancePopulation` (#12742)

* Upgrade arrow/parquet to `53.1.0` / fix clippy (#12724)

* Update to arrow/parquet 53.1.0

* Update some API

* update for changed file sizes

* Use non deprecated APIs

* Use ParquetMetadataReader from @etseidl

* remove upstreamed implementation

* Update CSV schema

* Use upstream is_null and is_not_null kernels

* feat: add support for Substrait ExtendedExpression (#12728)

* Add support for serializing and deserializing Substrait ExtendedExpr message

* Address clippy reviews

* Reuse existing rename method

* Transformed::new_transformed: Fix documentation formatting (#12787)

Co-authored-by: Andrew Lamb <[email protected]>

* fix: Correct results for grouping sets when columns contain nulls (#12571)

* Fix grouping sets behavior when data contains nulls

* PR suggestion comment

* Update new test case

* Add grouping_id to the logical plan

* Add doc comment next to INTERNAL_GROUPING_ID

* Fix unparsing of Aggregate with grouping sets

---------

Co-authored-by: Andrew Lamb <[email protected]>

* Migrate documentation for all string functions from scalar_functions.md to code  (#12775)

* Added documentation for string and unicode functions.

* Fixed issues with aliases.

* Cargo fmt.

* Minor doc fixes.

* Update docs for var_pop/samp

---------

Co-authored-by: Andrew Lamb <[email protected]>

* Account for constant equivalence properties in union, tests (#12562)

* Minor: clarify comment about empty dependencies (#12786)

* Introduce Signature::String and return error if  input of `strpos` is integer (#12751)

* fix sig

Signed-off-by: jayzhan211 <[email protected]>

* fix

Signed-off-by: jayzhan211 <[email protected]>

* fix error

Signed-off-by: jayzhan211 <[email protected]>

* fix all signature

Signed-off-by: jayzhan211 <[email protected]>

* fix all signature

Signed-off-by: jayzhan211 <[email protected]>

* change default type

Signed-off-by: jayzhan211 <[email protected]>

* clippy

Signed-off-by: jayzhan211 <[email protected]>

* fix docs

Signed-off-by: jayzhan211 <[email protected]>

* rm deadcode

Signed-off-by: jayzhan211 <[email protected]>

* cleanup

Signed-off-by: jayzhan211 <[email protected]>

* cleanup

Signed-off-by: jayzhan211 <[email protected]>

* rm test

Signed-off-by: jayzhan211 <[email protected]>

---------

Signed-off-by: jayzhan211 <[email protected]>

* Minor: improve docs on MovingMin/MovingMax (#12790)

* Add slt tests (#12721)

---------

Signed-off-by: jayzhan211 <[email protected]>
Signed-off-by: Austin Liu <[email protected]>
Co-authored-by: OussamaSaoudi <[email protected]>
Co-authored-by: Jonah Gao <[email protected]>
Co-authored-by: Dmitrii Blaginin <[email protected]>
Co-authored-by: Andrew Lamb <[email protected]>
Co-authored-by: Tomoaki Kawada <[email protected]>
Co-authored-by: Piotr Findeisen <[email protected]>
Co-authored-by: Jay Zhan <[email protected]>
Co-authored-by: HuSen <[email protected]>
Co-authored-by: Emil Ejbyfeldt <[email protected]>
Co-authored-by: Simon Vandel Sillesen <[email protected]>
Co-authored-by: Jax Liu <[email protected]>
Co-authored-by: Austin Liu <[email protected]>
Co-authored-by: JonasDev1 <[email protected]>
Co-authored-by: jcsherin <[email protected]>
Co-authored-by: Sergei Grebnov <[email protected]>
Co-authored-by: Andy Grove <[email protected]>
Co-authored-by: Bruce Ritchie <[email protected]>
Co-authored-by: DouPache <[email protected]>
Co-authored-by: mertak-synnada <[email protected]>
Co-authored-by: Bryce Mecum <[email protected]>
Co-authored-by: wiedld <[email protected]>
Co-authored-by: kamille <[email protected]>
Co-authored-by: Weston Pace <[email protected]>
Co-authored-by: Val Lorentz <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
core Core DataFusion crate logical-expr Logical plan and expressions optimizer Optimizer rules physical-expr Physical Expressions sql SQL Planner sqllogictest SQL Logic Tests (.slt) substrait
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Incorrect results when using grouping sets with data containing nulls
3 participants