Vector Distance Crash Tests Grammar by saikumar-vs · Pull Request #438 · Percona-QA/percona-qa

saikumar-vs · 2026-07-28T17:43:34Z

Introduced init_vectors.sql to correct vector dimensions and ensure all required IDs are represented for testing.
Added README.md to provide guidance on the usage and structure of the new grammar files.
Created vector_distance.yy for extended RQG grammar, covering various error paths and valid queries for VECTOR_DISTANCE.
Included vector_distance.zz as a companion gendata file to support the new grammar.

Steps to Test : perl runall-new.pl --post-gendata-sql=$PWD/conf/vector_1/init_vectors.sql --basedir= --grammar=conf/vector_1/vector_distance.yy --gendata=conf/vector_1/vector_distance.zz --threads=4 --queries=50000 --duration=600 --reporter=Shutdown,Backtrace,ErrorLog,QueryTimeout --debug --sqltrace

- Introduced `init_vectors.sql` to correct vector dimensions and ensure all required IDs are represented for testing. - Added `README.md` to provide guidance on the usage and structure of the new grammar files. - Created `vector_distance.yy` for extended RQG grammar, covering various error paths and valid queries for VECTOR_DISTANCE. - Included `vector_distance.zz` as a companion gendata file to support the new grammar. These changes enhance regression testing capabilities for vector distance calculations in the Percona Server.

This reverts commit e3c8e11.

This reverts commit c20ba2d.

- Introduced `init_vectors.sql` to correct vector dimensions and ensure all required IDs are represented for testing. - Added `README.md` to provide guidance on the usage and structure of the new grammar files. - Created `vector_distance.yy` for extended RQG grammar, covering various error paths and valid queries for VECTOR_DISTANCE. - Included `vector_distance.zz` as a companion gendata file to support the new grammar. These changes enhance regression testing capabilities for vector distance calculations in the Percona Server.

Adds Random Query Generator (RQG) assets to fuzz the new native DISTANCE() and VECTOR_DISTANCE() functions in Percona Server. * vector_distance2.yy: Grammar rules to fuzz function parameters, validate compile-time metric constants, and test runtime dimension enforcement. * vector_distance2.sql: Table definitions and seed data including 3D, 4D, and 16D vectors to trigger both narrow and wide SIMD dispatch routes. Includes specific boundary conditions for NULL propagation, zero vectors, and extreme FLT_MAX float32 values (which intentionally trigger ERROR 1690 due to IEEE 754 limitations during intermediate SIMD subtraction).

Copilot

Pull request overview

This PR adds a new Random Query Generator (RQG) “kit” under randgen/conf/vector_distance/ intended to fuzz-test Percona Server’s VECTOR_DISTANCE/DISTANCE behavior using a dedicated schema/data initializer, a grammar file, and a minimal gendata companion.

Changes:

Added an extended RQG grammar (vector_distance.yy) to exercise valid paths and multiple error paths for vector distance functions.
Added init_vectors.sql to create/populate test.vt with representative IDs/dimensions/anomaly vectors needed by the grammar.
Added documentation (README.md) and a companion gendata file (vector_distance.zz), plus an additional “v2” grammar+SQL pair (vector_distance2.yy/.sql).

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
randgen/conf/vector_distance/vector_distance.yy	New extended grammar for `VECTOR_DISTANCE`/`DISTANCE` fuzz coverage.
randgen/conf/vector_distance/init_vectors.sql	Creates/populates `test.vt` with specific IDs and vector shapes for grammar stability.
randgen/conf/vector_distance/vector_distance.zz	Minimal gendata file; currently documents schema creation incorrectly.
randgen/conf/vector_distance/README.md	Usage guide; currently misstates how `vt` is created and omits `init_vectors.sql` in the “files” list.
randgen/conf/vector_distance/vector_distance2.yy	Additional grammar; contains SQL-generation issues (e.g., missing `FROM`, incorrect `TO_VECTOR` usage).
randgen/conf/vector_distance/vector_distance2.sql	Companion schema/data for `vector_distance2.yy`.

Comments suppressed due to low confidence (4)

randgen/conf/vector_distance/README.md:22

This section claims the vt schema is created in a query_init rule, but vector_distance.yy does not define query_init and the run instructions rely on init_vectors.sql via --post-gendata-sql. The description should match the actual setup mechanism.

Since this grammar needs `VECTOR`-typed columns (which `--gendata`'s built-in generator doesn't produce), it defines its own literal table (`vt`) with explicit columns in `query_init`, rather than using `_table`/`_field`. This is consistent with how `example.yy` itself falls back to literal SQL identifiers where the built-in generators don't fit.

randgen/conf/vector_distance/vector_distance2.yy:4

CREATE TABLE t_vec AS SELECT distance_func AS dist ; has no FROM clause. If distance_func expands to v3/v4/v16 (via vector_expr -> vector_col), this will fail with an unknown-column error rather than exercising VECTOR_DISTANCE/DISTANCE behavior.

      | CREATE TABLE t_vec AS SELECT distance_func AS dist ;

randgen/conf/vector_distance/vector_distance2.yy:21

TO_VECTOR ( 'vector_string' ) uses a literal string "vector_string" rather than expanding the vector_string rule, so it won’t generate the intended vector literals.

      | TO_VECTOR ( 'vector_string' )

randgen/conf/vector_distance/vector_distance2.yy:51

vector_string alternatives are not quoted, so after switching to TO_VECTOR(vector_string) they would expand to TO_VECTOR([0.1,...]), which is invalid SQL. These should be SQL string literals like TO_VECTOR('[0.1,...]').

        [0.1, 0.2, 0.3]
      | [1.5, 2.5, 3.5, 4.5]
      | [0.0, 0.0, 0.0]
      | [3.4028235E38, 3.4028235E38, 3.4028235E38]
      | [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7] ;

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

+Two files, meant to be dropped into your own RQG checkout:
+
+- `vector_distance.yy` — the query grammar
+- `vector_distance.zz` — a minimal companion gendata file


+# The actual VECTOR-typed schema (table `vt`) is built inside the grammar's
+# query_init rule, since RQG's built-in data generator predates support for
+# the VECTOR column type and cannot declare VECTOR(n) columns natively.


+
+-- ids 1, 2: original rows from the uploaded file, UNCHANGED, including the
+-- under-width v384 (3 elements in a VECTOR(384) column). Kept deliberately as
+-- a documented edge case: STRING_TO_VECTOR(v384) here vs v3 will NOT raise a


+query:
+        SELECT distance_func FROM vector_test LIMIT _digit ;


vinperothas

Hi Sai, Thanks for putting this together! The RQG grammar coverage for the VECTOR_DISTANCE edge cases looks really solid, and I appreciate the thoroughness of the boundary tests.

I did notice that the README.md and some of the block comments in init_vectors.sql (like the notes about 'our manual testing session' and 'your actual checkout') read a bit like a conversational chat rather than standard documentation.

Could we polish the README and the SQL comments to be more objective? Stripping out the conversational framing will make it a much cleaner guide

vinperothas · 2026-07-29T13:32:11Z

+_bad_byte_literal:
+	0x0102030405 | 0x010203 | 0x0102030405060708090A | 0x01 ;
+
+# Base Standard Vector Generators


Impressive use of inline perl 👍

vinperothas · 2026-07-29T13:39:18Z

+- Alternatives are `|`-separated, and may span multiple lines
+- Underscore-prefixed tokens (`_table`, `_field`, `_digit`) are RQG's built-in generators, tied to whatever `--gendata` builds
+
+Since this grammar needs `VECTOR`-typed columns (which `--gendata`'s built-in generator doesn't produce), it defines its own literal table (`vt`) with explicit columns in `query_init`, rather than using `_table`/`_field`. This is consistent with how `example.yy` itself falls back to literal SQL identifiers where the built-in generators don't fit.


Shall be updated that the schema is created inside init_vectors.sql

saikumar-vs added 6 commits June 23, 2026 18:56

kmip defaults and some fixs for typos

c20ba2d

update latest url

e3c8e11

Revert "update latest url"

8eb132a

This reverts commit e3c8e11.

Revert "kmip defaults and some fixs for typos"

495b738

This reverts commit c20ba2d.

saikumar-vs requested review from mchawla16 and vinperothas July 28, 2026 17:43

vinperothas requested a review from Copilot July 29, 2026 12:33

Copilot started reviewing on behalf of vinperothas July 29, 2026 12:34 View session

Copilot AI reviewed Jul 29, 2026

View reviewed changes

vinperothas reviewed Jul 29, 2026

View reviewed changes

Navigation Menu

Uh oh!

Vector Distance Crash Tests Grammar - #438

Vector Distance Crash Tests Grammar#438
saikumar-vs wants to merge 7 commits into
masterfrom
vector_branch1

saikumar-vs commented Jul 28, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

vinperothas left a comment

Uh oh!

vinperothas Jul 29, 2026

Uh oh!

vinperothas Jul 29, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

saikumar-vs commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

vinperothas left a comment

Choose a reason for hiding this comment

Uh oh!

vinperothas Jul 29, 2026

Choose a reason for hiding this comment

Uh oh!

vinperothas Jul 29, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

saikumar-vs commented Jul 28, 2026 •

edited

Loading