zhidongqu-db opened a new pull request, #54011:
URL: https://github.com/apache/spark/pull/54011
<!--
Thanks for sending a pull request! Here are some tips for you:
1. If this is your first time, please read our contributor guidelines:
https://spark.apache.org/contributing.html
2. Ensure you have added or run the appropriate tests for your PR:
https://spark.apache.org/developer-tools.html
3. If the PR is unfinished, add '[WIP]' in your PR title, e.g.,
'[WIP][SPARK-XXXX] Your PR title ...'.
4. Be sure to keep the PR description updated to reflect all changes.
5. Please write your PR title to summarize what this PR proposes.
6. If possible, provide a concise example to reproduce the issue for a
faster review.
7. If you want to add a new configuration, please read the guideline first
for naming configurations in
'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'.
8. If you want to add or modify an error type or message, please read the
guideline first in
'common/utils/src/main/resources/error/README.md'.
-->
### What changes were proposed in this pull request?
<!--
Please clarify what changes you are proposing. The purpose of this section
is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR. See the examples below.
1. If you refactor some codes with changing classes, showing the class
hierarchy will help reviewers.
2. If you fix some SQL features, you can provide some references of other
DBMSes.
3. If there is design documentation, please add the link.
4. If there is a discussion in the mailing list, please add the link.
-->
This PR adds support for vector aggregation functions to Spark SQL, enabling
element-wise sum and average computations across groups of vectors.
- vector_sum(vectors) - Returns the element-wise sum of float vectors in a
group. Each element in the result is the sum of the corresponding elements
across all input vectors.
- vector_avg(vectors) - Returns the element-wise average of float vectors in
a group. Each element in the result is the arithmetic mean of the corresponding
elements across all input vectors.
Key implementation details:
- Type Safety: Functions accept only ARRAY<FLOAT> for vectors. No implicit
type casting is performed - passing ARRAY<DOUBLE> or ARRAY<INT> results in a
DATATYPE_MISMATCH.UNEXPECTED_INPUT_TYPE error.
- Dimension Validation: All vectors in a group must have the same dimension;
throws VECTOR_DIMENSION_MISMATCH error if dimensions do not match.
- NULL Handling: NULL vectors are skipped in aggregation. Non-NULL vectors
containing NULL elements are also treated as NULL and skipped.
- Edge Cases: Returns NULL if all values in the group are invalid. Returns
an empty array [] if all input vectors are empty.
- Compact Buffer Storage: Aggregate state uses BINARY format (dim * 4 bytes)
instead of ARRAY<FLOAT> for more efficient storage without null field overhead.
This PR only includes SQL language support; DataFrame API will be added in a
separate PR.
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
Vector aggregation functions are fundamental operations for:
- Clustering workloads: Computing cluster centroids by averaging member
vectors
- RAG applications: Aggregating embeddings across document chunks
- Distributed ML: Gradient accumulation and combining pre-aggregated vectors
across partitions
- Recommendation systems: Computing user preference vectors from interaction
history
These functions complement the vector distance/similarity functions and are
commonly available in other systems (Snowflake's VECTOR_SUM/VECTOR_AVG,
PostgreSQL pgvector's SUM/AVG over vectors).
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as
new features, bug fixes, or other behavior changes. Documentation-only updates
are not considered user-facing changes.
If yes, please clarify the previous behavior and the change this PR proposes
- provide the console output, description and/or an example to show the
behavior difference if possible.
If possible, please also clarify if this is a user-facing change compared to
the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
Yes, this PR introduces 2 new SQL aggregate functions:
```
-- Setup example table
CREATE TABLE vector_data (category STRING, embedding ARRAY<FLOAT>);
INSERT INTO vector_data VALUES
('A', array(1.0F, 2.0F, 3.0F)),
('A', array(4.0F, 5.0F, 6.0F)),
('B', array(2.0F, 1.0F, 4.0F)),
('B', array(3.0F, 2.0F, 1.0F));
-- Element-wise sum per category
SELECT category, vector_sum(embedding) AS sum_vector
FROM vector_data
GROUP BY category
ORDER BY category;
-- category: A, sum_vector: [5.0, 7.0, 9.0]
-- category: B, sum_vector: [5.0, 3.0, 5.0]
-- Element-wise average per category (centroid computation)
SELECT category, vector_avg(embedding) AS centroid
FROM vector_data
GROUP BY category
ORDER BY category;
-- category: A, centroid: [2.5, 3.5, 4.5]
-- category: B, centroid: [2.5, 1.5, 2.5]
-- Scalar aggregation (no GROUP BY)
SELECT vector_sum(embedding) AS total_sum, vector_avg(embedding) AS
overall_centroid
FROM vector_data;
-- total_sum: [10.0, 10.0, 14.0]
-- overall_centroid: [2.5, 2.5, 3.5]
-- NULL vectors are skipped
INSERT INTO vector_data VALUES ('A', NULL);
SELECT category, vector_avg(embedding) FROM vector_data WHERE category = 'A'
GROUP BY category;
-- Returns: [2.5, 3.5, 4.5] (unchanged, NULL skipped)
-- Vectors with NULL elements are skipped
INSERT INTO vector_data VALUES ('A', array(100.0F, NULL, 100.0F));
SELECT category, vector_avg(embedding) FROM vector_data WHERE
category = 'A' GROUP BY category;
-- Returns: [2.5, 3.5, 4.5] (unchanged, vector with NULL element skipped)
```
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some
test cases that check the changes thoroughly including negative and positive
cases if possible.
If it was tested in a way different from regular unit tests, please clarify
how you tested step by step, ideally copy and paste-able, so that other
reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why
it was difficult to add.
If benchmark tests were added, please run the benchmarks in GitHub Actions
for the consistent environment, and the instructions could accord to:
https://spark.apache.org/developer-tools.html#github-workflow-benchmarks.
-->
SQL Golden File Tests: Added vector-agg.sql with test coverage:
- Basic functionality tests for both vector_sum and vector_avg
- GROUP BY aggregation and scalar aggregation (no GROUP BY)
- Mathematical correctness validation
- Empty vector handling (returns empty array)
- NULL vector handling (skipped in aggregation)
- NULL element within vector handling (treated as NULL, skipped)
- All-NULL group handling (returns NULL)
- Dimension mismatch error cases
- Type mismatch error cases (ARRAY<DOUBLE>, ARRAY<INT>)
- Multi-partition tests to exercise merge logic
### Was this patch authored or co-authored using generative AI tooling?
<!--
If generative AI tooling has been used in the process of authoring this
patch, please include the
phrase: 'Generated-by: ' followed by the name of the tool and its version.
If no, write 'No'.
Please refer to the [ASF Generative Tooling
Guidance](https://www.apache.org/legal/generative-tooling.html) for details.
-->
Yes, code assistance with Claude Opus 4.5 in combination with manual editing
by the author.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
