MDEV-35283: support nullable indexed vector columns by annbaek · Pull Request #5212 · MariaDB/server

annbaek · 2026-06-10T07:06:58Z

Allow creating VECTOR INDEX on nullable columns. NULL values are simply not indexed - rows with NULL vectors won't be found via the vector index, but the CREATE TABLE/ALTER TABLE is permitted.

Changes:

Remove NOT_NULL_FLAG check for vector index creation in mysql_prepare_create_table_finalize()
Add NULL checks in hlindexes_on_insert/update/delete to skip vector index operations when the vector field is NULL
Handle UPDATE transitions: NULL->non-NULL (insert only), non-NULL->NULL (invalidate only), NULL->NULL (skip)
Remove obsolete DBUG_ASSERTs in mhnsw_insert() that assumed non-NULL was enforced by ER_INDEX_CANNOT_HAVE_NULL

CLAassistant · 2026-06-10T07:07:04Z

All committers have signed the CLA.

gemini-code-assist

Code Review

This pull request implements support for nullable indexed vector columns (MDEV-35283) by removing the NOT NULL constraint check during table creation and updating the index insert, update, and delete hooks to handle null values. The review feedback recommends removing an unused table parameter from the helper function vec_field_is_null and restoring a safety assertion in sql/vector_mhnsw.cc that validates the vector data length and alignment.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

gemini-code-assist · 2026-06-10T07:08:42Z

+static bool vec_field_is_null(const TABLE *table, const uchar *rec,
+                              const KEY *keyinfo)


The table parameter is unused in vec_field_is_null. It should be removed to simplify the function signature and avoid unused parameter warnings.

static bool vec_field_is_null(const uchar *rec, const KEY *keyinfo)

I second the AI here.

gemini-code-assist · 2026-06-10T07:08:42Z

-    if (int err= mhnsw_insert(this, key_info + s->keys))
+  {
+    KEY *keyinfo= key_info + s->keys;
+    if (vec_field_is_null(this, record[0], keyinfo))


Update the call to vec_field_is_null to match the simplified signature without the unused table parameter.

if (vec_field_is_null(record[0], keyinfo))

gemini-code-assist · 2026-06-10T07:08:42Z

+    bool old_is_null= vec_field_is_null(this, record[1], keyinfo);
+    bool new_is_null= vec_field_is_null(this, record[0], keyinfo);


Update the calls to vec_field_is_null to match the simplified signature without the unused table parameter.

bool old_is_null= vec_field_is_null(record[1], keyinfo); bool new_is_null= vec_field_is_null(record[0], keyinfo);

gemini-code-assist · 2026-06-10T07:08:42Z

-    if (int err= mhnsw_invalidate(this, buf, key_info + s->keys))
+  {
+    KEY *keyinfo= key_info + s->keys;
+    if (vec_field_is_null(this, buf, keyinfo))


Update the call to vec_field_is_null to match the simplified signature without the unused table parameter.

if (vec_field_is_null(buf, keyinfo))

gemini-code-assist · 2026-06-10T07:08:42Z

+  DBUG_ASSERT(res);
  DBUG_ASSERT(table->file->ref_length <= graph->field[FIELD_TREF]->field_length);


The assertion DBUG_ASSERT(res->length() > 0 && res->length() % 4 == 0); was a valuable safety check to ensure that the vector data is not empty and is properly aligned (since each float dimension is 4 bytes). Since res is guaranteed to be non-NULL by the preceding DBUG_ASSERT(res); (as NULL vectors are skipped before calling mhnsw_insert), this assertion is safe to keep and should not be removed.

DBUG_ASSERT(res); DBUG_ASSERT(res->length() > 0 && res->length() % 4 == 0); DBUG_ASSERT(table->file->ref_length <= graph->field[FIELD_TREF]->field_length);

+1. No need to remove the assert.

gkodinov

Thank you for your contribution! This is a preliminary review.

The regression tests are failing because of your added tests.
Please fix. There are also some optional stylistic suggestions below.

gkodinov · 2026-06-10T07:42:11Z

+--echo # MDEV-35283 - support nullable indexed vector columns
+--echo #
+
+if ($MTR_COMBINATION_INNODB) {


you will also need to update vector,*.diff: right now this test is failing when run with Aria and MyISAM.

However, I do not see anything InnoDB specific here in this fix. Why limit testing to InnoDB only to start with?

Maybe just remove the if() altogether?

gkodinov · 2026-06-10T07:46:41Z

+static bool vec_field_is_null(const TABLE *table, const uchar *rec,
+                              const KEY *keyinfo)


I second the AI here.

gkodinov · 2026-06-10T07:46:56Z

-    if (int err= mhnsw_insert(this, key_info + s->keys))
+  {
+    KEY *keyinfo= key_info + s->keys;
+    if (vec_field_is_null(this, record[0], keyinfo))


gkodinov · 2026-06-10T07:47:05Z

+    bool old_is_null= vec_field_is_null(this, record[1], keyinfo);
+    bool new_is_null= vec_field_is_null(this, record[0], keyinfo);


gkodinov · 2026-06-10T07:48:46Z

+  DBUG_ASSERT(res);
  DBUG_ASSERT(table->file->ref_length <= graph->field[FIELD_TREF]->field_length);


+1. No need to remove the assert.

gkodinov · 2026-06-10T07:49:42Z

  return 0;
 }

+static bool vec_field_is_null(const TABLE *table, const uchar *rec,


I'd even do inline

gkodinov · 2026-06-10T07:51:15Z

+                              const KEY *keyinfo)
+{
+  Field *field= keyinfo->key_part->field;
+  return field->is_null_in_record(rec);


I'd put an assert that keyinfo->keypart is a vector column type.

Allow creating VECTOR INDEX on nullable columns. NULL values are simply not indexed - rows with NULL vectors won't be found via the vector index, but the CREATE TABLE/ALTER TABLE is permitted. Changes: - Remove NOT_NULL_FLAG check for vector index creation in mysql_prepare_create_table_finalize() - Add inline vec_field_is_null() helper with vector key type assert - Add NULL checks in hlindexes_on_insert/update/delete to skip vector index operations when the vector field is NULL - Handle UPDATE transitions: NULL->non-NULL (insert only), non-NULL->NULL (invalidate only), NULL->NULL (skip) - Clean up DBUG_ASSERT comment in mhnsw_insert() that referenced the now-removed ER_INDEX_CANNOT_HAVE_NULL constraint

gkodinov

LGTM. One additional optimization idea below. Please stand by for the final review.

gkodinov · 2026-06-10T08:43:38Z

+    KEY *keyinfo= key_info + s->keys;
+    if (vec_field_is_null(record[0], keyinfo))
+      return 0;
+    if (int err= mhnsw_insert(this, keyinfo))


Final optional suggestion: it looks like you could just fold the null-ness check (and return success) into mhnsw_insert() and mhnsw_invalidate(). You will need to react on NULLs anyway everyplace these are called. This will further localize the check and ensure there's no duplicated code :

if (is_null) skip mhnsw_xxx()

gemini-code-assist Bot reviewed Jun 10, 2026

View reviewed changes

gkodinov added the External Contribution All PRs from entities outside of MariaDB Foundation, Corporation, Codership agreements. label Jun 10, 2026

gkodinov self-assigned this Jun 10, 2026

annbaek force-pushed the MDEV-35283-nullable-vector-index branch from 8eb4d79 to ec611a0 Compare June 10, 2026 07:47

gkodinov requested changes Jun 10, 2026

View reviewed changes

annbaek force-pushed the MDEV-35283-nullable-vector-index branch from ec611a0 to 12f09b1 Compare June 10, 2026 08:12

gkodinov approved these changes Jun 10, 2026

View reviewed changes

gkodinov assigned vuvova and unassigned gkodinov Jun 10, 2026

gkodinov requested a review from vuvova June 10, 2026 08:46

		static bool vec_field_is_null(const TABLE table, const uchar rec,
		const KEY *keyinfo)

		bool old_is_null= vec_field_is_null(this, record[1], keyinfo);
		bool new_is_null= vec_field_is_null(this, record[0], keyinfo);

		DBUG_ASSERT(res);
		DBUG_ASSERT(table->file->ref_length <= graph->field[FIELD_TREF]->field_length);

Uh oh!

Conversation

annbaek commented Jun 10, 2026

Uh oh!

CLAassistant commented Jun 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gkodinov left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gkodinov left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

4 participants

CLAassistant commented Jun 10, 2026 •

edited

Loading