MDEV-35327: Add VEC_DISTANCE_MANHATTAN (L1) distance function by TQSH-Dev · Pull Request #4654 · MariaDB/server

TQSH-Dev · 2026-02-15T09:57:08Z

Implemented the Manhattan distance function as described in the Jira task. Added item_vectorfunc implementation and registered the function in item_create

CLAassistant · 2026-02-15T09:57:14Z

All committers have signed the CLA.

vuvova

please add tests. with and without a vector index

TQSH-Dev · 2026-02-15T13:57:21Z

please add tests. with and without a vector index

Thanks for the review,will do it by tommorow.

sql/item_vectorfunc.cc

grooverdan · 2026-02-16T00:29:52Z

sql/item_vectorfunc.cc

  switch (kind) {
  case EUCLIDEAN: calc_distance= calc_distance_euclidean; break;
  case COSINE:    calc_distance= calc_distance_cosine; break;
+  case MANHATTAN: calc_distance= calc_distance_manhattan;break;


nit: additional space before break

grooverdan · 2026-02-16T00:30:13Z

sql/item_vectorfunc.h


 public:
-  enum distance_kind { EUCLIDEAN, COSINE, AUTO } kind;
+  enum distance_kind { EUCLIDEAN, COSINE, MANHATTAN,AUTO } kind;


nit: sapce before AUTO

grooverdan · 2026-02-16T00:32:03Z

sql/item_create.cc

+};
+
+
+Create_func_vec_distance_manhattan Create_func_vec_distance_manhattan::s_singleton;


nit: 1 blank line before and 2 blank lines after for consistency.

gkodinov

This is a preliminary review: I have no new remarks aside from what others already did.

TQSH-Dev · 2026-02-16T16:23:56Z

please add tests. with and without a vector index

added

TQSH-Dev · 2026-02-16T18:40:08Z

The build failure in rpl.rpl_row_foreign_key_mdl appears to be a random timeout unrelated to my changes, I am unsure as to why this is happening. My added test main.mdev_35327 passed successfully.

vuvova · 2026-02-17T16:13:54Z

mysql-test/main/mdev_35327.test

+--echo #
+--echo # With Vector Index
+--echo #
+CREATE VECTOR INDEX idx ON t1(v);


This is an index created with the euclidean distance. It cannot be used for searches with manhattan distance.

You didn't implement support for manhattan distance at all. It's rather more complex, than adding a new Item class.

Thanks for the feedback. You are absolutely correct,I dug into the EXPLAIN plans and realized that while the VEC_DISTANCE_MANHATTAN function returns the correct values, the optimizer is still performing a full table scan. I'll work on this.

TQSH-Dev · 2026-02-18T19:38:54Z

Fixed the part_of_sortkey loop to correctly identify Manhattan distance for index utilization.New test cases in mdev_35327.test confirm the mhnsw index is now being utilized for Manhattan distance queries.

vuvova · 2026-02-20T16:57:44Z

sql/item_vectorfunc.cc

    Field *f= item->field;
    KEY *keyinfo= f->table->s->key_info;
-    for (uint i= f->table->s->keys; i < f->table->s->total_keys; i++)
+    for (uint i= 0; i < f->table->s->total_keys; i++)


why is that?

vuvova · 2026-02-20T17:04:43Z

mysql-test/main/mdev_35327.test

+--echo # Without Vector Index
+--echo #
+CREATE TABLE t1 (id INT, v VECTOR(3) NOT NULL);
+INSERT INTO t1 VALUES (1, VEC_FromText('[0,0,0]')), (2, VEC_FromText('[1,1,1]')), (3, VEC_FromText('[5,5,5]'));


this isn't a very good dataset. L1 distance between your vectors is 0,3,12, as your tests show.
L2 distance would be 0, 1.732, 6.9282.

That is, you cannot know if the index works correctly, Euclidean and Manhattan provide exactly the same ordering. Try vectors where L1 and L2 produce different results.

vuvova · 2026-02-21T16:47:27Z

mysql-test/main/mdev_35327.result

+#
+CREATE VECTOR INDEX idx ON t1(v) DISTANCE=manhattan;
+# Output should be 3,5 and 6 again
+SELECT id, VEC_DISTANCE_MANHATTAN(v, VEC_FromText('[0,0,0]')) as dist FROM t1 ORDER BY dist;


you don't have LIMIT here, so the index is not used. You have LIMIT in EXPLAIN below, but not here.

vuvova · 2026-02-21T21:20:50Z

mysql-test/main/mdev_35327.result

+1	6
+# Checking if the vector index is actually implemented using manhattan distance
+EXPLAIN SELECT id FROM t1 FORCE INDEX (idx) 
+ORDER BY VEC_DISTANCE_MANHATTAN(v, VEC_FromText('[0,0,0]')) LIMIT 1;


why you don't use EXPLAIN of exactly the same query as above? You cannot see whether the query above uses the index, you only know that the one here with LIMIT 1 does.

please, do a query with EXPLAIN and exactly the same query without EXPLAIN.

vuvova requested changes Feb 15, 2026

View reviewed changes

grooverdan requested changes Feb 16, 2026

View reviewed changes

grooverdan added the External Contribution All PRs from entities outside of MariaDB Foundation, Corporation, Codership agreements. label Feb 16, 2026

gkodinov approved these changes Feb 16, 2026

View reviewed changes

TQSH-Dev requested a review from vuvova February 17, 2026 04:35

vuvova requested changes Feb 17, 2026

View reviewed changes

TQSH-Dev force-pushed the mdev-35327-manhattan branch 2 times, most recently from 6af8462 to 5131eae Compare February 18, 2026 16:52

TQSH-Dev changed the base branch from main to bb-10.11-new-innodb-julius February 18, 2026 18:17

TQSH-Dev requested a review from ottok as a code owner February 18, 2026 18:17

TQSH-Dev changed the base branch from bb-10.11-new-innodb-julius to main February 18, 2026 18:17

TQSH-Dev force-pushed the mdev-35327-manhattan branch from 359e824 to 36d3ef8 Compare February 18, 2026 18:50

TQSH-Dev requested a review from vuvova February 18, 2026 19:39

vuvova requested changes Feb 20, 2026

View reviewed changes

TQSH-Dev force-pushed the mdev-35327-manhattan branch from a24ad01 to 28407cb Compare February 21, 2026 12:02

TQSH-Dev requested review from grooverdan and vuvova February 21, 2026 13:50

vuvova requested changes Feb 21, 2026

View reviewed changes

MDEV-35327: Implement Manhattan distance for HNSW vector index

ed38e47

TQSH-Dev force-pushed the mdev-35327-manhattan branch from 28407cb to ed38e47 Compare February 21, 2026 17:54

TQSH-Dev requested a review from vuvova February 21, 2026 17:56

vuvova requested changes Feb 21, 2026

View reviewed changes

		};


		Create_func_vec_distance_manhattan Create_func_vec_distance_manhattan::s_singleton;

Uh oh!

Comments

Conversation

TQSH-Dev commented Feb 15, 2026

Uh oh!

CLAassistant commented Feb 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vuvova left a comment

Choose a reason for hiding this comment

Uh oh!

TQSH-Dev commented Feb 15, 2026

Uh oh!

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!

TQSH-Dev commented Feb 16, 2026

Uh oh!

TQSH-Dev commented Feb 16, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

TQSH-Dev commented Feb 18, 2026

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!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

5 participants

CLAassistant commented Feb 15, 2026 •

edited

Loading