Initial HNSW implementation

This commit includes the work done in collaboration with Hugo Wen from
Amazon:

    MDEV-33408 Alter HNSW graph storage and fix memory leak

    This commit changes the way HNSW graph information is stored in the
    second table. Instead of storing connections as separate records, it now
    stores neighbors for each node, leading to significant performance
    improvements and storage savings.

    Comparing with the previous approach, the insert speed is 5 times faster,
    search speed improves by 23%, and storage usage is reduced by 73%, based
    on ann-benchmark tests with random-xs-20-euclidean and
    random-s-100-euclidean datasets.

    Additionally, in previous code, vector objects were not released after
    use, resulting in excessive memory consumption (over 20GB for building
    the index with 90,000 records), preventing tests with large datasets.
    Now ensure that vectors are released appropriately during the insert and
    search functions. Note there are still some vectors that need to be
    cleaned up after search query completion. Needs to be addressed in a
    future commit.

    All new code of the whole pull request, including one or several files
    that are either new files or modified ones, are contributed under the
    BSD-new license. I am contributing on behalf of my employer Amazon Web
    Services, Inc.

As well as the commit:

    Introduce session variables to manage HNSW index parameters

    Three variables:

    hnsw_max_connection_per_layer
    hnsw_ef_constructor
    hnsw_ef_search

    ann-benchmark tool is also updated to support these variables in commit
    https://github.com/HugoWenTD/ann-benchmarks/commit/e09784e for branch
    https://github.com/HugoWenTD/ann-benchmarks/tree/mariadb-configurable

    All new code of the whole pull request, including one or several files
    that are either new files or modified ones, are contributed under the
    BSD-new license. I am contributing on behalf of my employer Amazon Web
    Services, Inc.
Co-authored-by: Hugo Wen <wenhug@amazon.com>

mysql-test/main/mysqld--help.result

@@ -402,6 +402,11 @@ The following specify which files/extra groups are read (specified before remain
  height-balanced, DOUBLE_PREC_HB - double precision
  height-balanced, JSON_HB - height-balanced, stored as
  JSON
+ --hnsw-ef-constructor 
+ hnsw_ef_constructor
+ --hnsw-ef-search    hnsw_ef_search
+ --hnsw-max-connection-per-layer 
+ hnsw_max_connection_per_layer
  --host-cache-size=# How many host names should be cached to avoid resolving
  (Automatically configured unless set explicitly)
  --idle-readonly-transaction-timeout=# 
@@ -1700,6 +1705,9 @@ gtid-strict-mode FALSE
 help TRUE
 histogram-size 254
 histogram-type JSON_HB
+hnsw-ef-constructor 10
+hnsw-ef-search 10
+hnsw-max-connection-per-layer 50
 host-cache-size 279
 idle-readonly-transaction-timeout 0
 idle-transaction-timeout 0

mysql-test/suite/sys_vars/r/sysvars_server_notembedded.result

@@ -1422,6 +1422,36 @@ NUMERIC_BLOCK_SIZE	NULL
 ENUM_VALUE_LIST	SINGLE_PREC_HB,DOUBLE_PREC_HB,JSON_HB
 READ_ONLY	NO
 COMMAND_LINE_ARGUMENT	REQUIRED
+VARIABLE_NAME	HNSW_EF_CONSTRUCTOR
+VARIABLE_SCOPE	SESSION
+VARIABLE_TYPE	INT UNSIGNED
+VARIABLE_COMMENT	hnsw_ef_constructor
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	4294967295
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	NONE
+VARIABLE_NAME	HNSW_EF_SEARCH
+VARIABLE_SCOPE	SESSION
+VARIABLE_TYPE	INT UNSIGNED
+VARIABLE_COMMENT	hnsw_ef_search
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	4294967295
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	NONE
+VARIABLE_NAME	HNSW_MAX_CONNECTION_PER_LAYER
+VARIABLE_SCOPE	SESSION
+VARIABLE_TYPE	INT UNSIGNED
+VARIABLE_COMMENT	hnsw_max_connection_per_layer
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	4294967295
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	NONE
 VARIABLE_NAME	HOSTNAME
 VARIABLE_SCOPE	GLOBAL
 VARIABLE_TYPE	VARCHAR

sql/item.h

@@ -6620,7 +6620,6 @@ class Item_int_with_ref :public Item_int
 #include "item_subselect.h"
 #include "item_xmlfunc.h"
 #include "item_jsonfunc.h"
-#include "item_vectorfunc.h"
 #include "item_create.h"
 #include "item_vers.h"
 #endif

sql/item_create.cc

@@ -36,6 +36,7 @@
 #include "sp.h"
 #include "sql_time.h"
 #include "sql_type_geom.h"
+#include "item_vectorfunc.h"
 #include <mysql/plugin_function.h>
 
 

sql/item_vectorfunc.cc

@@ -23,6 +23,7 @@
 
 #include <my_global.h>
 #include "item.h"
+#include "item_vectorfunc.h"
 
 key_map Item_func_vec_distance::part_of_sortkey() const
 {
@@ -48,8 +49,18 @@ double Item_func_vec_distance::val_real()
     return 0;
   float *v1= (float*)r1->ptr();
   float *v2= (float*)r2->ptr();
+  return euclidean_vec_distance(v1, v2, (r1->length()) / sizeof(float));
+}
+
+double euclidean_vec_distance(float *v1, float *v2, size_t v_len)
+{
+  float *p1= v1;
+  float *p2= v2;
   double d= 0;
-  for (uint i=0; i < r1->length() / sizeof(float); i++)
-    d+= (v1[i] - v2[i])*(v1[i] - v2[i]);
+  for (size_t i= 0; i < v_len; p1++, p2++, i++)
+  {
+    float dist= *p1 - *p2;
+    d+= dist * dist;
+  }
   return sqrt(d);
 }

sql/item_vectorfunc.h

@@ -17,6 +17,8 @@
    Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1335  USA */
 
 /* This file defines all vector functions */
+#include <my_global.h>
+#include "item.h"
 #include "lex_string.h"
 #include "item_func.h"
 
@@ -34,6 +36,7 @@ class Item_func_vec_distance: public Item_real_func
   {
     return check_argument_types_or_binary(NULL, 0, arg_count);
   }
+
 public:
   Item_func_vec_distance(THD *thd, Item *a, Item *b)
    :Item_real_func(thd, a, b) {}
@@ -51,6 +54,9 @@ class Item_func_vec_distance: public Item_real_func
   key_map part_of_sortkey() const override;
   Item *do_get_copy(THD *thd) const override
   { return get_item_copy<Item_func_vec_distance>(thd, this); }
+  virtual ~Item_func_vec_distance() {};
 };
 
+
+double euclidean_vec_distance(float *v1, float *v2, size_t v_len);
 #endif

sql/sql_base.cc

@@ -9876,7 +9876,7 @@ int TABLE::hlindex_open(uint nr)
       mysql_mutex_unlock(&s->LOCK_share);
     TABLE *table= (TABLE*)alloc_root(&mem_root, sizeof(*table));
     if (!table ||
-        open_table_from_share(in_use, s->hlindex, &empty_clex_str, db_stat, 0,
+        open_table_from_share(in_use, s->hlindex, &empty_clex_str, db_stat, EXTRA_RECORD,
                               in_use->open_options, table, 0))
       return 1;
     hlindex= table;
@@ -9931,7 +9931,7 @@ int TABLE::hlindex_first(uint nr, Item *item, ulonglong limit)
 
   DBUG_ASSERT(hlindex->in_use == in_use);
 
-  return mhnsw_first(this, item, limit);
+  return mhnsw_first(this, key_info + s->keys, item, limit);
 }
 
 int TABLE::hlindex_next()

sql/sql_class.h

@@ -918,6 +918,11 @@ typedef struct system_variables
   my_bool binlog_alter_two_phase;
 
   Charset_collation_map_st character_set_collations;
+
+  /* Temporary for HNSW tests */
+  uint hnsw_max_connection_per_layer;
+  uint hnsw_ef_constructor;
+  uint hnsw_ef_search;
 } SV;
 
 /**

sql/sys_vars.cc

@@ -7383,3 +7383,23 @@ static Sys_var_enum Sys_block_encryption_mode(
   "AES_ENCRYPT() and AES_DECRYPT() functions",
   SESSION_VAR(block_encryption_mode), CMD_LINE(REQUIRED_ARG),
   block_encryption_mode_values, DEFAULT(0));
+
+/* Temporary for HNSW tests */
+static Sys_var_uint Sys_hnsw_ef_search(
+       "hnsw_ef_search",
+       "hnsw_ef_search",
+       SESSION_VAR(hnsw_ef_search), CMD_LINE(NO_ARG),
+       VALID_RANGE(0, UINT_MAX), DEFAULT(10),
+       BLOCK_SIZE(1));
+static Sys_var_uint Sys_hnsw_ef_constructor(
+       "hnsw_ef_constructor",
+       "hnsw_ef_constructor",
+       SESSION_VAR(hnsw_ef_constructor), CMD_LINE(NO_ARG),
+       VALID_RANGE(0, UINT_MAX), DEFAULT(10),
+       BLOCK_SIZE(1));
+static Sys_var_uint Sys_hnsw_max_connection_per_layer(
+       "hnsw_max_connection_per_layer",
+       "hnsw_max_connection_per_layer",
+       SESSION_VAR(hnsw_max_connection_per_layer), CMD_LINE(NO_ARG),
+       VALID_RANGE(0, UINT_MAX), DEFAULT(50),
+       BLOCK_SIZE(1));

sql/vector_mhnsw.h

@@ -15,10 +15,14 @@
    Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1335  USA
 */
 
+#include <my_global.h>
+#include "item.h"
+#include "m_string.h"
+#include "structs.h"
 #include "table.h"
 
 extern const LEX_CSTRING mhnsw_hlindex_table;
 
 int mhnsw_insert(TABLE *table, KEY *keyinfo);
-int mhnsw_first(TABLE *table, Item *dist, ulonglong limit);
+int mhnsw_first(TABLE *table, KEY *keyinfo, Item *dist, ulonglong limit);
 int mhnsw_next(TABLE *table);