Added support for uuid

.github/workflows/build.yml

@@ -28,7 +28,7 @@ jobs:
           dev-files: true
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
+          PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
       - run: |
           export PG_CONFIG=`which pg_config`
           sudo --preserve-env=PG_CONFIG make install
@@ -49,7 +49,7 @@ jobs:
           postgres-version: 14
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter
+          PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter
       - run: make install
       - run: make installcheck
       - if: ${{ failure() }}
@@ -60,9 +60,7 @@ jobs:
           wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_10.tar.gz
           tar xf REL_14_10.tar.gz
       - run: make prove_installcheck PROVE_FLAGS="-I ./postgres-REL_14_10/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
-      - run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make
-        env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING
+      - run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make PG_CFLAGS="-DUSE_ASSERT_CHECKING"
   windows:
     runs-on: windows-latest
     if: ${{ !startsWith(github.ref_name, 'mac') }}
@@ -73,7 +71,6 @@ jobs:
           postgres-version: 14
       - run: |
           call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
-          cd %TEMP% && ^
           nmake /NOLOGO /F Makefile.win && ^
           nmake /NOLOGO /F Makefile.win install && ^
           nmake /NOLOGO /F Makefile.win installcheck && ^
@@ -100,15 +97,4 @@ jobs:
           sudo -u postgres make installcheck
           sudo -u postgres make prove_installcheck
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
-  valgrind:
-    if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: ankane/setup-postgres-valgrind@v1
-        with:
-          postgres-version: 16
-      - run: make
-      - run: sudo --preserve-env=PG_CONFIG make install
-      - run: make installcheck
+          PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare

CHANGELOG.md

@@ -1,12 +1,10 @@
-## 0.6.2 (2024-03-18)
+## 0.7.0 (unreleased)
 
-- Reduced lock contention with parallel HNSW index builds
+- Added support for inline filtering with HNSW
 
-## 0.6.1 (2024-03-04)
+## 0.6.1 (unreleased)
 
 - Fixed error with `ANALYZE` and vectors with different dimensions
-- Fixed segmentation fault with `shared_preload_libraries`
-- Fixed vector subtraction being marked as commutative
 
 ## 0.6.0 (2024-01-29)
 

LICENSE

@@ -1,4 +1,4 @@
-Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
+Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
 
 Portions Copyright (c) 1994, The Regents of the University of California
 

META.json

@@ -2,7 +2,7 @@
 	"name": "vector",
 	"abstract": "Open-source vector similarity search for Postgres",
 	"description": "Supports L2 distance, inner product, and cosine distance",
-	"version": "0.6.2",
+	"version": "0.6.0",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.6.2",
+			"version": "0.6.0",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.6.2
+EXTVERSION = 0.6.0
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)
@@ -71,9 +71,11 @@ PG_MAJOR ?= 16
 .PHONY: docker
 
 docker:
-	docker build --pull --no-cache --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
+	docker build --pull --no-cache --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) .
+	docker build --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
 
 .PHONY: docker-release
 
 docker-release:
-	docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
+	docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) .
+	docker buildx build --push --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.6.2
+EXTVERSION = 0.6.0
 
 OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
 HEADERS = src\vector.h

README.md

@@ -20,13 +20,13 @@ Compile and install the extension (supports Postgres 12+)
 
 ```sh
 cd /tmp
-git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
 ```
 
-See the [installation notes](#installation-notes---linux-and-mac) if you run into issues
+See the [installation notes](#installation-notes) if you run into issues
 
 You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), [pkg](#pkg), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
 
@@ -44,15 +44,12 @@ Then use `nmake` to build:
 
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\16"
-cd %TEMP%
-git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
 ```
 
-See the [installation notes](#installation-notes---windows) if you run into issues
-
 You can also install it with [Docker](#docker) or [conda-forge](#conda-forge).
 
 ## Getting Started
@@ -105,12 +102,6 @@ Insert vectors
 INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
 ```
 
-Or load vectors in bulk using `COPY` ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/bulk_loading.py))
-
-```sql
-COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
-```
-
 Upsert vectors
 
 ```sql
@@ -273,8 +264,6 @@ HINT:  Increase maintenance_work_mem to speed up builds.
 
 Note: Do not set `maintenance_work_mem` so high that it exhausts the memory on the server
 
-Like other index types, it’s faster to create an index after loading your initial data
-
 Starting with 0.6.0, you can also speed up index creation by increasing the number of parallel workers (2 by default)
 
 ```sql
@@ -388,6 +377,12 @@ Create an index on one [or more](https://www.postgresql.org/docs/current/indexes
 CREATE INDEX ON items (category_id);
 ```
 
+Or a composite HNSW index for approximate search (added in 0.7.0)
+
+```sql
+CREATE INDEX ON items USING hnsw (embedding vector_l2_ops, category_id);
+```
+
 Or a [partial index](https://www.postgresql.org/docs/current/indexes-partial.html) on the vector column for approximate search
 
 ```sql
@@ -413,51 +408,13 @@ You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python
 
 ## Performance
 
-### Tuning
-
-Use a tool like [PgTune](https://pgtune.leopard.in.ua/) to set initial values for Postgres server parameters. For instance, `shared_buffers` should typically be 25% of the server’s memory. You can find the config file with:
-
-```sql
-SHOW config_file;
-```
-
-And check individual settings with:
-
-```sql
-SHOW shared_buffers;
-```
-
-Be sure to restart Postgres for changes to take effect.
-
-### Loading
-
-Use `COPY` for bulk loading data ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/bulk_loading.py)).
-
-```sql
-COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
-```
-
-Add any indexes *after* loading the initial data for best performance.
-
-### Indexing
-
-See index build time for [HNSW](#index-build-time) and [IVFFlat](#index-build-time-1).
-
-In production environments, create indexes concurrently to avoid blocking writes.
-
-```sql
-CREATE INDEX CONCURRENTLY ...
-```
-
-### Querying
-
 Use `EXPLAIN ANALYZE` to debug performance.
 
 ```sql
 EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
-#### Exact Search
+### Exact Search
 
 To speed up queries without an index, increase `max_parallel_workers_per_gather`.
 
@@ -471,7 +428,7 @@ If vectors are normalized to length 1 (like [OpenAI embeddings](https://platform
 SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
 ```
 
-#### Approximate Search
+### Approximate Search
 
 To speed up queries with an IVFFlat index, increase the number of inverted lists (at the expense of recall).
 
@@ -479,7 +436,7 @@ To speed up queries with an IVFFlat index, increase the number of inverted lists
 CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 1000);
 ```
 
-### Vacuuming
+## Vacuuming
 
 Vacuuming can take a while for HNSW indexes. Speed it up by reindexing first.
 
@@ -488,41 +445,6 @@ REINDEX INDEX CONCURRENTLY index_name;
 VACUUM table_name;
 ```
 
-## Monitoring
-
-Monitor performance with [pg_stat_statements](https://www.postgresql.org/docs/current/pgstatstatements.html) (be sure to add it to `shared_preload_libraries`).
-
-```sql
-CREATE EXTENSION pg_stat_statements;
-```
-
-Get the most time-consuming queries with:
-
-```sql
-SELECT query, calls, ROUND((total_plan_time + total_exec_time) / calls) AS avg_time_ms,
-    ROUND((total_plan_time + total_exec_time) / 60000) AS total_time_min
-    FROM pg_stat_statements ORDER BY total_plan_time + total_exec_time DESC LIMIT 20;
-```
-
-Note: Replace `total_plan_time + total_exec_time` with `total_time` for Postgres < 13
-
-Monitor recall by comparing results from approximate search with exact search.
-
-```sql
-BEGIN;
-SET LOCAL enable_indexscan = off; -- use exact search
-SELECT ...
-COMMIT;
-```
-
-## Scaling
-
-Scale pgvector the same way you scale Postgres.
-
-Scale vertically by increasing memory, CPU, and storage on a single instance. Use existing tools to [tune parameters](#tuning) and [monitor performance](#monitoring).
-
-Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus) or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus.py)).
-
 ## Languages
 
 Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
@@ -616,18 +538,6 @@ and query with:
 SELECT * FROM items ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
 ```
 
-#### Are binary vectors supported?
-
-You can store binary vectors and perform exact nearest neighbor search by Hamming distance in Postgres without an extension ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py)).
-
-```tsql
-CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
-INSERT INTO items (embedding) VALUES (B'000'), (B'111');
-SELECT * FROM items ORDER BY bit_count(embedding # B'101') LIMIT 5;
-```
-
-Indexing is not currently supported.
-
 #### Do indexes need to fit into memory?
 
 No, but like other index types, you’ll likely see better performance if they do. You can get the size of an index with:
@@ -640,17 +550,7 @@ SELECT pg_size_pretty(pg_relation_size('index_name'));
 
 #### Why isn’t a query using an index?
 
-The query needs to have an `ORDER BY` and `LIMIT`, and the `ORDER BY` must be the result of a distance operator, not an expression.
-
-```sql
--- index
-ORDER BY embedding <=> '[3,1,2]' LIMIT 5;
-
--- no index
-ORDER BY 1 - (embedding <=> '[3,1,2]') DESC LIMIT 5;
-```
-
-You can encourage the planner to use an index for a query with:
+The cost estimation in pgvector < 0.4.3 does not always work well with the planner. You can encourage the planner to use an index for a query with:
 
 ```sql
 BEGIN;
@@ -679,12 +579,6 @@ or choose to store vectors inline:
 ALTER TABLE items ALTER COLUMN embedding SET STORAGE PLAIN;
 ```
 
-#### Why are there less results for a query after adding an HNSW index?
-
-Results are limited by the size of the dynamic candidate list (`hnsw.ef_search`). There may be even less results due to dead tuples or filtering conditions in the query. We recommend setting `hnsw.ef_search` to at least twice the `LIMIT` of the query. If you need more than 500 results, use an IVFFlat index instead.
-
-Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
-
 #### Why are there less results for a query after adding an IVFFlat index?
 
 The index was likely created with too little data for the number of lists. Drop the index until the table has more data.
@@ -693,15 +587,11 @@ The index was likely created with too little data for the number of lists. Drop
 DROP INDEX index_name;
 ```
 
-Results can also be limited by the number of probes (`ivfflat.probes`).
-
-Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
-
 ## Reference
 
 ### Vector Type
 
-Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single-precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
+Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
 
 ### Vector Operators
 
@@ -725,14 +615,14 @@ l1_distance(vector, vector) → double precision | taxicab distance | 0.5.0
 vector_dims(vector) → integer | number of dimensions |
 vector_norm(vector) → double precision | Euclidean norm |
 
-### Vector Aggregate Functions
+### Aggregate Functions
 
 Function | Description | Added
 --- | --- | ---
 avg(vector) → vector | average |
 sum(vector) → vector | sum | 0.5.0
 
-## Installation Notes - Linux and Mac
+## Installation Notes
 
 ### Postgres Location
 
@@ -772,26 +662,6 @@ Note: Replace `16` with your Postgres server version
 
 If compilation fails and the output includes `warning: no such sysroot directory` on Mac, reinstall Xcode Command Line Tools.
 
-### Portability
-
-By default, pgvector compiles with `-march=native` on some platforms for best performance. However, this can lead to `Illegal instruction` errors if trying to run the compiled extension on a different machine.
-
-To compile for portability, use:
-
-```sh
-make OPTFLAGS=""
-```
-
-## Installation Notes - Windows
-
-### Missing Header
-
-If compilation fails with `Cannot open include file: 'postgres.h': No such file or directory`, make sure `PGROOT` is correct.
-
-### Permissions
-
-If installation fails with `Access is denied`, re-run the installation instructions as an administrator.
-
 ## Additional Installation Methods
 
 ### Docker
@@ -807,7 +677,7 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
 cd pgvector
 docker build --build-arg PG_MAJOR=16 -t myuser/pgvector .
 ```
@@ -863,7 +733,7 @@ pkg install postgresql15-pg_vector
 or the port with:
 
 ```sh
-cd /usr/ports/databases/pgvector
+cd /usr/ports/databases/pg_vector
 make install
 ```
 
@@ -976,12 +846,6 @@ make installcheck REGRESS=functions                            # regression test
 make prove_installcheck PROVE_TESTS=test/t/001_ivfflat_wal.pl  # TAP test
 ```
 
-To enable assertions:
-
-```sh
-make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
-```
-
 To enable benchmarking:
 
 ```sh
@@ -994,6 +858,12 @@ To show memory usage:
 make clean && PG_CFLAGS="-DHNSW_MEMORY -DIVFFLAT_MEMORY" make && make install
 ```
 
+To enable assertions:
+
+```sh
+make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
+```
+
 To get k-means metrics:
 
 ```sh

sql/vector--0.6.0--0.6.1.sql

@@ -1,16 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.1'" to load this file. \quit
-
-DROP OPERATOR - (vector, vector);
-
-CREATE OPERATOR - (
-	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
-);
-
-ALTER OPERATOR <= (vector, vector) SET (
-	RESTRICT = scalarlesel, JOIN = scalarlejoinsel
-);
-
-ALTER OPERATOR >= (vector, vector) SET (
-	RESTRICT = scalargesel, JOIN = scalargejoinsel
-);

sql/vector--0.6.1--0.6.2.sql

@@ -1,2 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.2'" to load this file. \quit

sql/vector--0.6.1--0.7.0.sql

@@ -0,0 +1,34 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.0'" to load this file. \quit
+
+CREATE OPERATOR CLASS vector_bigint_ops
+	DEFAULT FOR TYPE bigint USING hnsw AS
+	OPERATOR 2 < ,
+	OPERATOR 3 <= ,
+	OPERATOR 4 = ,
+	OPERATOR 5 >= ,
+	OPERATOR 6 > ;
+
+CREATE OPERATOR CLASS vector_integer_ops
+	DEFAULT FOR TYPE integer USING hnsw AS
+	OPERATOR 2 < ,
+	OPERATOR 3 <= ,
+	OPERATOR 4 = ,
+	OPERATOR 5 >= ,
+	OPERATOR 6 > ;
+
+CREATE OPERATOR CLASS vector_text_ops
+	DEFAULT FOR TYPE text USING hnsw AS
+	OPERATOR 2 < ,
+	OPERATOR 3 <= ,
+	OPERATOR 4 = ,
+	OPERATOR 5 >= ,
+	OPERATOR 6 > ;
+
+CREATE OPERATOR CLASS vector_uuid_ops
+	DEFAULT FOR TYPE uuid USING hnsw AS
+	OPERATOR 2 < ,
+	OPERATOR 3 <= ,
+	OPERATOR 4 = ,
+	OPERATOR 5 >= ,
+	OPERATOR 6 > ;

sql/vector.sql

@@ -1,7 +1,7 @@
 -- complain if script is sourced in psql, rather than via CREATE EXTENSION
 \echo Use "CREATE EXTENSION vector" to load this file. \quit
 
--- vector type
+-- type
 
 CREATE TYPE vector;
 
@@ -29,7 +29,7 @@ CREATE TYPE vector (
 	STORAGE   = external
 );
 
--- vector functions
+-- functions
 
 CREATE FUNCTION l2_distance(vector, vector) RETURNS float8
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -58,7 +58,7 @@ CREATE FUNCTION vector_sub(vector, vector) RETURNS vector
 CREATE FUNCTION vector_mul(vector, vector) RETURNS vector
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
--- vector private functions
+-- private functions
 
 CREATE FUNCTION vector_lt(vector, vector) RETURNS bool
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -99,7 +99,7 @@ CREATE FUNCTION vector_avg(double precision[]) RETURNS vector
 CREATE FUNCTION vector_combine(double precision[], double precision[]) RETURNS double precision[]
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
--- vector aggregates
+-- aggregates
 
 CREATE AGGREGATE avg(vector) (
 	SFUNC = vector_accum,
@@ -117,7 +117,7 @@ CREATE AGGREGATE sum(vector) (
 	PARALLEL = SAFE
 );
 
--- vector cast functions
+-- cast functions
 
 CREATE FUNCTION vector(vector, integer, boolean) RETURNS vector
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -137,7 +137,7 @@ CREATE FUNCTION array_to_vector(numeric[], integer, boolean) RETURNS vector
 CREATE FUNCTION vector_to_float4(vector, integer, boolean) RETURNS real[]
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
--- vector casts
+-- casts
 
 CREATE CAST (vector AS vector)
 	WITH FUNCTION vector(vector, integer, boolean) AS IMPLICIT;
@@ -157,7 +157,7 @@ CREATE CAST (double precision[] AS vector)
 CREATE CAST (numeric[] AS vector)
 	WITH FUNCTION array_to_vector(numeric[], integer, boolean) AS ASSIGNMENT;
 
--- vector operators
+-- operators
 
 CREATE OPERATOR <-> (
 	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = l2_distance,
@@ -180,7 +180,8 @@ CREATE OPERATOR + (
 );
 
 CREATE OPERATOR - (
-	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
+	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub,
+	COMMUTATOR = -
 );
 
 CREATE OPERATOR * (
@@ -194,10 +195,11 @@ CREATE OPERATOR < (
 	RESTRICT = scalarltsel, JOIN = scalarltjoinsel
 );
 
+-- should use scalarlesel and scalarlejoinsel, but not supported in Postgres < 11
 CREATE OPERATOR <= (
 	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_le,
 	COMMUTATOR = >= , NEGATOR = > ,
-	RESTRICT = scalarlesel, JOIN = scalarlejoinsel
+	RESTRICT = scalarltsel, JOIN = scalarltjoinsel
 );
 
 CREATE OPERATOR = (
@@ -212,10 +214,11 @@ CREATE OPERATOR <> (
 	RESTRICT = eqsel, JOIN = eqjoinsel
 );
 
+-- should use scalargesel and scalargejoinsel, but not supported in Postgres < 11
 CREATE OPERATOR >= (
 	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_ge,
 	COMMUTATOR = <= , NEGATOR = < ,
-	RESTRICT = scalargesel, JOIN = scalargejoinsel
+	RESTRICT = scalargtsel, JOIN = scalargtjoinsel
 );
 
 CREATE OPERATOR > (
@@ -240,7 +243,7 @@ CREATE ACCESS METHOD hnsw TYPE INDEX HANDLER hnswhandler;
 
 COMMENT ON ACCESS METHOD hnsw IS 'hnsw index access method';
 
--- vector opclasses
+-- opclasses
 
 CREATE OPERATOR CLASS vector_ops
 	DEFAULT FOR TYPE vector USING btree AS
@@ -287,3 +290,37 @@ CREATE OPERATOR CLASS vector_cosine_ops
 	OPERATOR 1 <=> (vector, vector) FOR ORDER BY float_ops,
 	FUNCTION 1 vector_negative_inner_product(vector, vector),
 	FUNCTION 2 vector_norm(vector);
+
+-- hnsw attributes
+
+CREATE OPERATOR CLASS vector_bigint_ops
+	DEFAULT FOR TYPE bigint USING hnsw AS
+	OPERATOR 2 < ,
+	OPERATOR 3 <= ,
+	OPERATOR 4 = ,
+	OPERATOR 5 >= ,
+	OPERATOR 6 > ;
+
+CREATE OPERATOR CLASS vector_integer_ops
+	DEFAULT FOR TYPE integer USING hnsw AS
+	OPERATOR 2 < ,
+	OPERATOR 3 <= ,
+	OPERATOR 4 = ,
+	OPERATOR 5 >= ,
+	OPERATOR 6 > ;
+
+CREATE OPERATOR CLASS vector_text_ops
+	DEFAULT FOR TYPE text USING hnsw AS
+	OPERATOR 2 < ,
+	OPERATOR 3 <= ,
+	OPERATOR 4 = ,
+	OPERATOR 5 >= ,
+	OPERATOR 6 > ;
+
+CREATE OPERATOR CLASS vector_uuid_ops
+	DEFAULT FOR TYPE uuid USING hnsw AS
+	OPERATOR 2 < ,
+	OPERATOR 3 <= ,
+	OPERATOR 4 = ,
+	OPERATOR 5 >= ,
+	OPERATOR 6 > ;

src/hnsw.c

@@ -8,7 +8,6 @@
 #include "commands/progress.h"
 #include "commands/vacuum.h"
 #include "hnsw.h"
-#include "miscadmin.h"
 #include "utils/guc.h"
 #include "utils/selfuncs.h"
 
@@ -29,7 +28,7 @@ static relopt_kind hnsw_relopt_kind;
  * this grows bigger, we should use a shmem_request_hook and
  * RequestAddinShmemSpace() to pre-reserve space for this.
  */
-void
+static void
 HnswInitLockTranche(void)
 {
 	int		   *tranche_ids;
@@ -54,8 +53,7 @@ HnswInitLockTranche(void)
 void
 HnswInit(void)
 {
-	if (!process_shared_preload_libraries_in_progress)
-		HnswInitLockTranche();
+	HnswInitLockTranche();
 
 	hnsw_relopt_kind = add_reloption_kind();
 	add_int_reloption(hnsw_relopt_kind, "m", "Max number of connections",
@@ -202,7 +200,7 @@ hnswhandler(PG_FUNCTION_ARGS)
 	amroutine->amcanorderbyop = true;
 	amroutine->amcanbackward = false;	/* can change direction mid-scan */
 	amroutine->amcanunique = false;
-	amroutine->amcanmulticol = false;
+	amroutine->amcanmulticol = true;
 	amroutine->amoptionalkey = true;
 	amroutine->amsearcharray = false;
 	amroutine->amsearchnulls = false;

src/hnsw.h

@@ -55,11 +55,6 @@
 #define HNSW_UPDATE_ENTRY_GREATER 1
 #define HNSW_UPDATE_ENTRY_ALWAYS 2
 
-typedef enum HnswType
-{
-	HNSW_TYPE_VECTOR
-}			HnswType;
-
 /* Build phases */
 /* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
 #define PROGRESS_HNSW_PHASE_LOAD		2
@@ -85,7 +80,7 @@ typedef enum HnswType
 
 #if PG_VERSION_NUM < 130000
 #define list_delete_last(list) list_truncate(list, list_length(list) - 1)
-#define list_sort(list, cmp) ((list) = list_qsort(list, cmp))
+#define list_sort(list, cmp) list_qsort(list, cmp)
 #endif
 
 #define HnswIsElementTuple(tup) ((tup)->type == HNSW_ELEMENT_TUPLE_TYPE)
@@ -133,8 +128,9 @@ HnswPtrDeclare(HnswElementData, HnswElementRelptr, HnswElementPtr);
 HnswPtrDeclare(HnswNeighborArray, HnswNeighborArrayRelptr, HnswNeighborArrayPtr);
 HnswPtrDeclare(HnswNeighborArrayPtr, HnswNeighborsRelptr, HnswNeighborsPtr);
 HnswPtrDeclare(char, DatumRelptr, DatumPtr);
+HnswPtrDeclare(IndexTupleData, IndexTupleRelptr, IndexTuplePtr);
 
-struct HnswElementData
+typedef struct HnswElementData
 {
 	HnswElementPtr next;
 	ItemPointerData heaptids[HNSW_HEAPTIDS];
@@ -148,8 +144,9 @@ struct HnswElementData
 	OffsetNumber neighborOffno;
 	BlockNumber neighborPage;
 	DatumPtr	value;
+	IndexTuplePtr itup;
 	LWLock		lock;
-};
+}			HnswElementData;
 
 typedef HnswElementData * HnswElement;
 
@@ -160,12 +157,12 @@ typedef struct HnswCandidate
 	bool		closer;
 }			HnswCandidate;
 
-struct HnswNeighborArray
+typedef struct HnswNeighborArray
 {
 	int			length;
 	bool		closerSet;
 	HnswCandidate items[FLEXIBLE_ARRAY_MEMBER];
-};
+}			HnswNeighborArray;
 
 typedef struct HnswPairingHeapNode
 {
@@ -190,7 +187,6 @@ typedef struct HnswGraph
 
 	/* Entry state */
 	LWLock		entryLock;
-	LWLock		entryWaitLock;
 	HnswElementPtr entryPoint;
 
 	/* Allocations state */
@@ -247,7 +243,6 @@ typedef struct HnswBuildState
 	Relation	index;
 	IndexInfo  *indexInfo;
 	ForkNumber	forkNum;
-	HnswType	type;
 
 	/* Settings */
 	int			dimensions;
@@ -268,6 +263,7 @@ typedef struct HnswBuildState
 	HnswGraph  *graph;
 	double		ml;
 	int			maxLevel;
+	Vector	   *normvec;
 
 	/* Memory */
 	MemoryContext graphCtx;
@@ -372,12 +368,11 @@ typedef struct HnswVacuumState
 int			HnswGetM(Relation index);
 int			HnswGetEfConstruction(Relation index);
 FmgrInfo   *HnswOptionalProcInfo(Relation index, uint16 procnum);
-HnswType	HnswGetType(Relation index);
-bool		HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type);
+bool		HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
 Buffer		HnswNewBuffer(Relation index, ForkNumber forkNum);
 void		HnswInitPage(Buffer buf, Page page);
 void		HnswInit(void);
-List	   *HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement);
+List	   *HnswSearchLayer(char *base, Datum q, IndexScanDesc scan, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement);
 HnswElement HnswGetEntryPoint(Relation index);
 void		HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
 void	   *HnswAlloc(HnswAllocator * allocator, Size size);
@@ -391,12 +386,14 @@ void		HnswAddHeapTid(HnswElement element, ItemPointer heaptid);
 void		HnswInitNeighbors(char *base, HnswElement element, int m, HnswAllocator * alloc);
 bool		HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull, ItemPointer heap_tid, bool building);
 void		HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building);
-void		HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec);
-void		HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec);
-void		HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element);
+void		HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec, Relation index);
+void		HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec, IndexScanDesc scan, bool *matches);
+void		HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element, bool useIndexTuple);
 void		HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
 void		HnswLoadNeighbors(HnswElement element, Relation index, int m);
-void		HnswInitLockTranche(void);
+TupleDesc	HnswTupleDesc(Relation index);
+IndexTuple	HnswFormIndexTuple(Relation index, TupleDesc tupdesc, Datum value, Datum *values, bool *isnull);
+bool		HnswElementIsDuplicate(char *base, HnswElement a, HnswElement b, Relation index);
 PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);
 
 /* Index access methods */

src/hnswbuild.c

@@ -157,6 +157,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 	Page		page;
 	HnswElementPtr iter = buildstate->graph->head;
 	char	   *base = buildstate->hnswarea;
+	bool		useIndexTuple = IndexRelationGetNumberOfAttributes(index) > 1;
 
 	/* Calculate sizes */
 	maxSize = HNSW_MAX_SIZE;
@@ -177,6 +178,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 		Size		ntupSize;
 		Size		combinedSize;
 		Pointer		valuePtr = HnswPtrAccess(base, element->value);
+		IndexTuple	itup = HnswPtrAccess(base, element->itup);
 
 		/* Update iterator */
 		iter = element->next;
@@ -185,7 +187,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 		MemSet(etup, 0, HNSW_TUPLE_ALLOC_SIZE);
 
 		/* Calculate sizes */
-		etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(valuePtr));
+		etupSize = HNSW_ELEMENT_TUPLE_SIZE(useIndexTuple ? IndexTupleSize(itup) : VARSIZE_ANY(valuePtr));
 		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, buildstate->m);
 		combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
 
@@ -193,7 +195,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 		if (etupSize > HNSW_TUPLE_ALLOC_SIZE)
 			elog(ERROR, "index tuple too large");
 
-		HnswSetElementTuple(base, etup, element);
+		HnswSetElementTuple(base, etup, element, useIndexTuple);
 
 		/* Keep element and neighbors on the same page if possible */
 		if (PageGetFreeSpace(page) < etupSize || (combinedSize <= maxSize && PageGetFreeSpace(page) < combinedSize))
@@ -334,19 +336,17 @@ AddDuplicateInMemory(HnswElement element, HnswElement dup)
  * Find duplicate element
  */
 static bool
-FindDuplicateInMemory(char *base, HnswElement element)
+FindDuplicateInMemory(char *base, HnswElement element, Relation index)
 {
 	HnswNeighborArray *neighbors = HnswGetNeighbors(base, element, 0);
-	Datum		value = HnswGetValue(base, element);
 
 	for (int i = 0; i < neighbors->length; i++)
 	{
 		HnswCandidate *neighbor = &neighbors->items[i];
 		HnswElement neighborElement = HnswPtrAccess(base, neighbor->element);
-		Datum		neighborValue = HnswGetValue(base, neighborElement);
 
 		/* Exit early since ordered by distance */
-		if (!datumIsEqual(value, neighborValue, false, -1))
+		if (!HnswElementIsDuplicate(base, element, neighborElement, index))
 			return false;
 
 		/* Check for space */
@@ -406,7 +406,7 @@ UpdateGraphInMemory(FmgrInfo *procinfo, Oid collation, HnswElement element, int
 	char	   *base = buildstate->hnswarea;
 
 	/* Look for duplicate */
-	if (FindDuplicateInMemory(base, element))
+	if (FindDuplicateInMemory(base, element, buildstate->index))
 		return;
 
 	/* Add element */
@@ -431,15 +431,10 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	HnswGraph  *graph = buildstate->graph;
 	HnswElement entryPoint;
 	LWLock	   *entryLock = &graph->entryLock;
-	LWLock	   *entryWaitLock = &graph->entryWaitLock;
 	int			efConstruction = buildstate->efConstruction;
 	int			m = buildstate->m;
 	char	   *base = buildstate->hnswarea;
 
-	/* Wait if another process needs exclusive lock on entry lock */
-	LWLockAcquire(entryWaitLock, LW_EXCLUSIVE);
-	LWLockRelease(entryWaitLock);
-
 	/* Get entry point */
 	LWLockAcquire(entryLock, LW_SHARED);
 	entryPoint = HnswPtrAccess(base, graph->entryPoint);
@@ -450,10 +445,8 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 		/* Release shared lock */
 		LWLockRelease(entryLock);
 
-		/* Tell other processes to wait and get exclusive lock */
-		LWLockAcquire(entryWaitLock, LW_EXCLUSIVE);
+		/* Get exclusive lock */
 		LWLockAcquire(entryLock, LW_EXCLUSIVE);
-		LWLockRelease(entryWaitLock);
 
 		/* Get latest entry point after lock is acquired */
 		entryPoint = HnswPtrAccess(base, graph->entryPoint);
@@ -478,10 +471,14 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	HnswGraph  *graph = buildstate->graph;
 	HnswElement element;
 	HnswAllocator *allocator = &buildstate->allocator;
-	Size		valueSize;
+	IndexTuple	itup;
+	Size		itupSize;
+	IndexTuple	itupPtr;
 	Pointer		valuePtr;
 	LWLock	   *flushLock = &graph->flushLock;
 	char	   *base = buildstate->hnswarea;
+	TupleDesc	tupdesc = HnswTupleDesc(index);
+	bool		unused;
 
 	/* Detoast once for all calls */
 	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
@@ -489,12 +486,13 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	/* Normalize if needed */
 	if (buildstate->normprocinfo != NULL)
 	{
-		if (!HnswNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->type))
+		if (!HnswNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
 			return false;
 	}
 
 	/* Get datum size */
-	valueSize = VARSIZE_ANY(DatumGetPointer(value));
+	itup = HnswFormIndexTuple(index, tupdesc, value, values, isnull);
+	itupSize = IndexTupleSize(itup);
 
 	/* Ensure graph not flushed when inserting */
 	LWLockAcquire(flushLock, LW_SHARED);
@@ -541,7 +539,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 
 	/* Ok, we can proceed to allocate the element */
 	element = HnswInitElement(base, heaptid, buildstate->m, buildstate->ml, buildstate->maxLevel, allocator);
-	valuePtr = HnswAlloc(allocator, valueSize);
+	itupPtr = HnswAlloc(allocator, itupSize);
 
 	/*
 	 * We have now allocated the space needed for the element, so we don't
@@ -550,8 +548,10 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	 */
 	LWLockRelease(&graph->allocatorLock);
 
-	/* Copy the datum */
-	memcpy(valuePtr, DatumGetPointer(value), valueSize);
+	/* Copy the index tuple */
+	memcpy(itupPtr, itup, itupSize);
+	HnswPtrStore(base, element->itup, itupPtr);
+	valuePtr = DatumGetPointer(index_getattr(itupPtr, 1, tupdesc, &unused));
 	HnswPtrStore(base, element->value, valuePtr);
 
 	/* Create a lock for the element */
@@ -608,9 +608,6 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
 static void
 InitGraph(HnswGraph * graph, char *base, long memoryTotal)
 {
-	/* Initialize the lock tranche if needed */
-	HnswInitLockTranche();
-
 	HnswPtrStore(base, graph->head, (HnswElement) NULL);
 	HnswPtrStore(base, graph->entryPoint, (HnswElement) NULL);
 	graph->memoryUsed = 0;
@@ -619,7 +616,6 @@ InitGraph(HnswGraph * graph, char *base, long memoryTotal)
 	graph->indtuples = 0;
 	SpinLockInit(&graph->lock);
 	LWLockInitialize(&graph->entryLock, hnsw_lock_tranche_id);
-	LWLockInitialize(&graph->entryWaitLock, hnsw_lock_tranche_id);
 	LWLockInitialize(&graph->allocatorLock, hnsw_lock_tranche_id);
 	LWLockInitialize(&graph->flushLock, hnsw_lock_tranche_id);
 }
@@ -671,24 +667,34 @@ HnswSharedMemoryAlloc(Size size, void *state)
 static void
 InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo, ForkNumber forkNum)
 {
-	int			maxDimensions = HNSW_MAX_DIM;
-
 	buildstate->heap = heap;
 	buildstate->index = index;
 	buildstate->indexInfo = indexInfo;
 	buildstate->forkNum = forkNum;
-	buildstate->type = HnswGetType(index);
 
 	buildstate->m = HnswGetM(index);
 	buildstate->efConstruction = HnswGetEfConstruction(index);
 	buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
 
+	/* For now */
+	if (IndexRelationGetNumberOfKeyAttributes(index) > 3)
+		elog(ERROR, "index cannot have more than three columns");
+
+	if (!OidIsValid(index_getprocid(index, 1, HNSW_DISTANCE_PROC)))
+		elog(ERROR, "first column must be a vector");
+
+	for (int i = 1; i < IndexRelationGetNumberOfKeyAttributes(index); i++)
+	{
+		if (OidIsValid(index_getprocid(index, i + 1, HNSW_DISTANCE_PROC)))
+			elog(ERROR, "column %d cannot be a vector", i + 1);
+	}
+
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
 		elog(ERROR, "column does not have dimensions");
 
-	if (buildstate->dimensions > maxDimensions)
-		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
+	if (buildstate->dimensions > HNSW_MAX_DIM)
+		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", HNSW_MAX_DIM);
 
 	if (buildstate->efConstruction < 2 * buildstate->m)
 		elog(ERROR, "ef_construction must be greater than or equal to 2 * m");
@@ -706,6 +712,9 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 	buildstate->ml = HnswGetMl(buildstate->m);
 	buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
 
+	/* Reuse for each tuple */
+	buildstate->normvec = InitVector(buildstate->dimensions);
+
 	buildstate->graphCtx = GenerationContextCreate(CurrentMemoryContext,
 												   "Hnsw build graph context",
 #if PG_VERSION_NUM >= 150000
@@ -729,6 +738,7 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 static void
 FreeBuildState(HnswBuildState * buildstate)
 {
+	pfree(buildstate->normvec);
 	MemoryContextDelete(buildstate->graphCtx);
 	MemoryContextDelete(buildstate->tmpCtx);
 }
@@ -984,14 +994,6 @@ HnswBeginParallel(HnswBuildState * buildstate, bool isconcurrent, int request)
 	/* Report less than allocated so never fails */
 	InitGraph(&hnswshared->graphData, hnswarea, esthnswarea - 1024 * 1024);
 
-	/*
-	 * Avoid base address for relptr for Postgres < 14.5
-	 * https://github.com/postgres/postgres/commit/7201cd18627afc64850537806da7f22150d1a83b
-	 */
-#if PG_VERSION_NUM < 140005
-	hnswshared->graphData.memoryUsed += MAXALIGN(1);
-#endif
-
 	shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_SHARED, hnswshared);
 	shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_AREA, hnswarea);
 

src/hnswinsert.c

@@ -136,9 +136,12 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 	OffsetNumber freeNeighborOffno = InvalidOffsetNumber;
 	BlockNumber newInsertPage = InvalidBlockNumber;
 	char	   *base = NULL;
+	bool		useIndexTuple = IndexRelationGetNumberOfAttributes(index) > 1;
+	Pointer		valuePtr = HnswPtrAccess(base, e->value);
+	IndexTuple	itup = HnswPtrAccess(base, e->itup);
 
 	/* Calculate sizes */
-	etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(HnswPtrAccess(base, e->value)));
+	etupSize = HNSW_ELEMENT_TUPLE_SIZE(useIndexTuple ? IndexTupleSize(itup) : VARSIZE_ANY(valuePtr));
 	ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
 	combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
 	maxSize = HNSW_MAX_SIZE;
@@ -146,7 +149,7 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 
 	/* Prepare element tuple */
 	etup = palloc0(etupSize);
-	HnswSetElementTuple(base, etup, e);
+	HnswSetElementTuple(base, etup, e, useIndexTuple);
 
 	/* Prepare neighbor tuple */
 	ntup = palloc0(ntupSize);
@@ -504,16 +507,14 @@ FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
 {
 	char	   *base = NULL;
 	HnswNeighborArray *neighbors = HnswGetNeighbors(base, element, 0);
-	Datum		value = HnswGetValue(base, element);
 
 	for (int i = 0; i < neighbors->length; i++)
 	{
 		HnswCandidate *neighbor = &neighbors->items[i];
 		HnswElement neighborElement = HnswPtrAccess(base, neighbor->element);
-		Datum		neighborValue = HnswGetValue(base, neighborElement);
 
 		/* Exit early since ordered by distance */
-		if (!datumIsEqual(value, neighborValue, false, -1))
+		if (!HnswElementIsDuplicate(base, element, neighborElement, index))
 			return false;
 
 		if (AddDuplicateOnDisk(index, element, neighborElement, building))
@@ -564,6 +565,10 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
 	Oid			collation = index->rd_indcollation[0];
 	LOCKMODE	lockmode = ShareLock;
 	char	   *base = NULL;
+	TupleDesc	tupdesc = HnswTupleDesc(index);
+	bool		unused;
+	IndexTuple	itup;
+	Pointer		valuePtr;
 
 	/*
 	 * Get a shared lock. This allows vacuum to ensure no in-flight inserts
@@ -577,7 +582,10 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
 
 	/* Create an element */
 	element = HnswInitElement(base, heap_tid, m, HnswGetMl(m), HnswGetMaxLevel(m), NULL);
-	HnswPtrStore(base, element->value, DatumGetPointer(value));
+	itup = HnswFormIndexTuple(index, tupdesc, value, values, isnull);
+	HnswPtrStore(base, element->itup, itup);
+	valuePtr = DatumGetPointer(index_getattr(itup, 1, tupdesc, &unused));
+	HnswPtrStore(base, element->value, valuePtr);
 
 	/* Prevent concurrent inserts when likely updating entry point */
 	if (entryPoint == NULL || element->level > entryPoint->level)
@@ -622,7 +630,7 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
 	normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
 	if (normprocinfo != NULL)
 	{
-		if (!HnswNormValue(normprocinfo, collation, &value, HnswGetType(index)))
+		if (!HnswNormValue(normprocinfo, collation, &value, NULL))
 			return;
 	}
 

src/hnswscan.c

@@ -33,11 +33,35 @@ GetScanItems(IndexScanDesc scan, Datum q)
 
 	for (int lc = entryPoint->level; lc >= 1; lc--)
 	{
-		w = HnswSearchLayer(base, q, ep, 1, lc, index, procinfo, collation, m, false, NULL);
+		w = HnswSearchLayer(base, q, NULL, ep, 1, lc, index, procinfo, collation, m, false, NULL);
 		ep = w;
 	}
 
-	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
+	/* Only pass scan to check matches for layer 0 */
+	return HnswSearchLayer(base, q, scan, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
+}
+
+/*
+ * Get dimensions from metapage
+ */
+static int
+GetDimensions(Relation index)
+{
+	Buffer		buf;
+	Page		page;
+	HnswMetaPage metap;
+	int			dimensions;
+
+	buf = ReadBuffer(index, HNSW_METAPAGE_BLKNO);
+	LockBuffer(buf, BUFFER_LOCK_SHARE);
+	page = BufferGetPage(buf);
+	metap = HnswPageGetMeta(page);
+
+	dimensions = metap->dimensions;
+
+	UnlockReleaseBuffer(buf);
+
+	return dimensions;
 }
 
 /*
@@ -50,7 +74,7 @@ GetScanValue(IndexScanDesc scan)
 	Datum		value;
 
 	if (scan->orderByData->sk_flags & SK_ISNULL)
-		value = PointerGetDatum(NULL);
+		value = PointerGetDatum(InitVector(GetDimensions(scan->indexRelation)));
 	else
 	{
 		value = scan->orderByData->sk_argument;
@@ -61,7 +85,7 @@ GetScanValue(IndexScanDesc scan)
 
 		/* Fine if normalization fails */
 		if (so->normprocinfo != NULL)
-			HnswNormValue(so->normprocinfo, so->collation, &value, HnswGetType(scan->indexRelation));
+			HnswNormValue(so->normprocinfo, so->collation, &value, NULL);
 	}
 
 	return value;
@@ -179,7 +203,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 		MemoryContextSwitchTo(oldCtx);
 
 		scan->xs_heaptid = *heaptid;
-		scan->xs_recheck = false;
+		scan->xs_recheck = scan->numberOfKeys > 0;
 		scan->xs_recheckorderby = false;
 		return true;
 	}

src/hnswutils.c

@@ -3,6 +3,7 @@
 #include <math.h>
 
 #include "access/generic_xlog.h"
+#include "access/relscan.h"
 #include "hnsw.h"
 #include "lib/pairingheap.h"
 #include "storage/bufmgr.h"
@@ -11,6 +12,10 @@
 #include "utils/rel.h"
 #include "vector.h"
 
+#if PG_VERSION_NUM < 130000
+#define TYPSTORAGE_PLAIN 'p'
+#endif
+
 #if PG_VERSION_NUM >= 130000
 #include "common/hashfn.h"
 #else
@@ -149,15 +154,6 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
 	return index_getprocinfo(index, 1, procnum);
 }
 
-/*
- * Get vector type
- */
-HnswType
-HnswGetType(Relation index)
-{
-	return HNSW_TYPE_VECTOR;
-}
-
 /*
  * Divide by the norm
  *
@@ -167,25 +163,21 @@ HnswGetType(Relation index)
  * if it's different than the original value
  */
 bool
-HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
+HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
 {
 	double		norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
 
 	if (norm > 0)
 	{
-		/* TODO Remove vector-specific code */
-		if (type == HNSW_TYPE_VECTOR)
-		{
-			Vector	   *v = DatumGetVector(*value);
-			Vector	   *result = InitVector(v->dim);
+		Vector	   *v = DatumGetVector(*value);
 
-			for (int i = 0; i < v->dim; i++)
-				result->x[i] = v->x[i] / norm;
+		if (result == NULL)
+			result = InitVector(v->dim);
 
-			*value = PointerGetDatum(result);
-		}
-		else
-			elog(ERROR, "Unsupported type");
+		for (int i = 0; i < v->dim; i++)
+			result->x[i] = v->x[i] / norm;
+
+		*value = PointerGetDatum(result);
 
 		return true;
 	}
@@ -308,6 +300,82 @@ HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno)
 	return element;
 }
 
+/*
+ * Get the tuple descriptor
+ */
+TupleDesc
+HnswTupleDesc(Relation index)
+{
+	TupleDesc	tupdesc = CreateTupleDescCopyConstr(RelationGetDescr(index));
+
+	/* Prevent compression */
+	TupleDescAttr(tupdesc, 0)->attstorage = TYPSTORAGE_PLAIN;
+
+	return tupdesc;
+}
+
+/*
+ * Form an index tuple
+ */
+IndexTuple
+HnswFormIndexTuple(Relation index, TupleDesc tupdesc, Datum value, Datum *values, bool *isnull)
+{
+	Size		size = sizeof(Datum) * IndexRelationGetNumberOfAttributes(index);
+	Datum	   *newValues = palloc(size);
+
+	memcpy(newValues, values, size);
+	newValues[0] = value;
+
+	return index_form_tuple(tupdesc, newValues, isnull);
+}
+
+/*
+ * Check if elements are duplicates
+ */
+bool
+HnswElementIsDuplicate(char *base, HnswElement a, HnswElement b, Relation index)
+{
+	if (IndexRelationGetNumberOfAttributes(index) == 1)
+	{
+		Datum		value = HnswGetValue(base, a);
+		Datum		value2 = HnswGetValue(base, b);
+
+		return datumIsEqual(value, value2, false, -1);
+	}
+	else
+	{
+		TupleDesc	tupdesc = RelationGetDescr(index);
+		IndexTuple	itup = HnswPtrAccess(base, a->itup);
+		IndexTuple	itup2 = HnswPtrAccess(base, b->itup);
+
+		for (int i = 0; i < tupdesc->natts; i++)
+		{
+			Datum		value;
+			Datum		value2;
+			bool		isnull;
+			bool		isnull2;
+
+			value = index_getattr(itup, i + 1, tupdesc, &isnull);
+			value2 = index_getattr(itup2, i + 1, tupdesc, &isnull2);
+
+			if (isnull || isnull2)
+			{
+				if (isnull != isnull2)
+					return false;
+			}
+			else
+			{
+				Form_pg_attribute att = TupleDescAttr(tupdesc, i);
+
+				if (!datumIsEqual(value, value2, att->attbyval, att->attlen))
+					return false;
+			}
+		}
+
+		return true;
+	}
+}
+
 /*
  * Get the metapage info
  */
@@ -417,10 +485,8 @@ HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, Bloc
  * Set element tuple, except for neighbor info
  */
 void
-HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element)
+HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element, bool useIndexTuple)
 {
-	Pointer		valuePtr = HnswPtrAccess(base, element->value);
-
 	etup->type = HNSW_ELEMENT_TUPLE_TYPE;
 	etup->level = element->level;
 	etup->deleted = 0;
@@ -431,7 +497,19 @@ HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element)
 		else
 			ItemPointerSetInvalid(&etup->heaptids[i]);
 	}
-	memcpy(&etup->data, valuePtr, VARSIZE_ANY(valuePtr));
+
+	if (useIndexTuple)
+	{
+		IndexTuple	itup = HnswPtrAccess(base, element->itup);
+
+		memcpy(&etup->data, itup, IndexTupleSize(itup));
+	}
+	else
+	{
+		Pointer		valuePtr = HnswPtrAccess(base, element->value);
+
+		memcpy(&etup->data, valuePtr, VARSIZE_ANY(valuePtr));
+	}
 }
 
 /*
@@ -535,7 +613,7 @@ HnswLoadNeighbors(HnswElement element, Relation index, int m)
  * Load an element from a tuple
  */
 void
-HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec)
+HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec, Relation index)
 {
 	element->level = etup->level;
 	element->deleted = etup->deleted;
@@ -558,17 +636,64 @@ HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHe
 	if (loadVec)
 	{
 		char	   *base = NULL;
-		Datum		value = datumCopy(PointerGetDatum(&etup->data), false, -1);
 
-		HnswPtrStore(base, element->value, DatumGetPointer(value));
+		if (IndexRelationGetNumberOfAttributes(index) > 1)
+		{
+			TupleDesc	tupdesc = RelationGetDescr(index);
+			bool		unused;
+			IndexTuple	itup = CopyIndexTuple((IndexTuple) &etup->data);
+			Datum		value = index_getattr(itup, 1, tupdesc, &unused);
+
+			HnswPtrStore(base, element->itup, itup);
+			HnswPtrStore(base, element->value, DatumGetPointer(value));
+		}
+		else
+		{
+			Datum		value = datumCopy(PointerGetDatum(&etup->data), false, -1);
+
+			HnswPtrStore(base, element->value, DatumGetPointer(value));
+		}
 	}
 }
 
+/*
+ * Check if an element matches
+ */
+static bool
+HnswCheckMatches(Relation index, HnswElementTuple etup, IndexScanDesc scan)
+{
+	if (scan == NULL)
+		return true;
+
+	for (int i = 0; i < scan->numberOfKeys; i++)
+	{
+		IndexTuple	itup = (IndexTuple) &etup->data;
+		TupleDesc	tupdesc = RelationGetDescr(index);
+		ScanKey		key = &scan->keyData[i];
+		bool		isnull;
+		Datum		value = index_getattr(itup, key->sk_attno, tupdesc, &isnull);
+		bool		attnull = key->sk_flags & SK_ISNULL;
+
+		if (isnull || attnull)
+		{
+			if (isnull != attnull)
+				return false;
+		}
+		else
+		{
+			if (!DatumGetBool(FunctionCall2Coll(&key->sk_func, key->sk_collation, value, key->sk_argument)))
+				return false;
+		}
+	}
+
+	return true;
+}
+
 /*
  * Load an element and optionally get its distance from q
  */
 void
-HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
+HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec, IndexScanDesc scan, bool *matches)
 {
 	Buffer		buf;
 	Page		page;
@@ -584,17 +709,30 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
 	Assert(HnswIsElementTuple(etup));
 
 	/* Load element */
-	HnswLoadElementFromTuple(element, etup, true, loadVec);
+	HnswLoadElementFromTuple(element, etup, true, loadVec, index);
 
 	/* Calculate distance */
 	if (distance != NULL)
 	{
-		if (DatumGetPointer(*q) == NULL)
-			*distance = 0;
+		Datum		value;
+
+		if (IndexRelationGetNumberOfAttributes(index) > 1)
+		{
+			IndexTuple	itup = (IndexTuple) &etup->data;
+			TupleDesc	tupdesc = RelationGetDescr(index);
+			bool		unused;
+
+			value = index_getattr(itup, 1, tupdesc, &unused);
+		}
 		else
-			*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->data)));
+			value = PointerGetDatum(&etup->data);
+
+		*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, value));
 	}
 
+	if (matches != NULL)
+		*matches = HnswCheckMatches(index, etup, scan);
+
 	UnlockReleaseBuffer(buf);
 }
 
@@ -622,7 +760,7 @@ HnswEntryCandidate(char *base, HnswElement entryPoint, Datum q, Relation index,
 	if (index == NULL)
 		hc->distance = GetCandidateDistance(base, hc, q, procinfo, collation);
 	else
-		HnswLoadElement(entryPoint, &hc->distance, &q, index, procinfo, collation, loadVec);
+		HnswLoadElement(entryPoint, &hc->distance, &q, index, procinfo, collation, loadVec, NULL, NULL);
 	return hc;
 }
 
@@ -740,12 +878,14 @@ CountElement(char *base, HnswElement skipElement, HnswCandidate * hc)
  * Algorithm 2 from paper
  */
 List *
-HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement)
+HnswSearchLayer(char *base, Datum q, IndexScanDesc scan, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement)
 {
 	List	   *w = NIL;
 	pairingheap *C = pairingheap_allocate(CompareNearestCandidates, NULL);
 	pairingheap *W = pairingheap_allocate(CompareFurthestCandidates, NULL);
 	int			wlen = 0;
+	uint64		additional = 0;
+	uint64		maxAdditional = (scan != NULL && scan->numberOfKeys > 0) ? 100 * ef : 0;
 	visited_hash v;
 	ListCell   *lc2;
 	HnswNeighborArray *neighborhoodData = NULL;
@@ -817,14 +957,18 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
 			if (!visited)
 			{
 				float		eDistance;
+				bool		eMatches;
 				HnswElement eElement = HnswPtrAccess(base, e->element);
 
 				f = ((HnswPairingHeapNode *) pairingheap_first(W))->inner;
 
 				if (index == NULL)
+				{
 					eDistance = GetCandidateDistance(base, e, q, procinfo, collation);
+					eMatches = true;
+				}
 				else
-					HnswLoadElement(eElement, &eDistance, &q, index, procinfo, collation, inserting);
+					HnswLoadElement(eElement, &eDistance, &q, index, procinfo, collation, inserting, scan, &eMatches);
 
 				Assert(!eElement->deleted);
 
@@ -843,6 +987,16 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
 					pairingheap_add(C, &(CreatePairingHeapNode(ec)->ph_node));
 					pairingheap_add(W, &(CreatePairingHeapNode(ec)->ph_node));
 
+					/*
+					 * Do not count elements that do not match filter towards
+					 * ef
+					 */
+					if (!eMatches)
+					{
+						if ((++additional) <= maxAdditional)
+							continue;
+					}
+
 					/*
 					 * Do not count elements being deleted towards ef when
 					 * vacuuming. It would be ideal to do this for inserts as
@@ -878,15 +1032,12 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
 static int
 #if PG_VERSION_NUM >= 130000
 CompareCandidateDistances(const ListCell *a, const ListCell *b)
-{
-	HnswCandidate *hca = lfirst(a);
-	HnswCandidate *hcb = lfirst(b);
 #else
 CompareCandidateDistances(const void *a, const void *b)
-{
-	HnswCandidate *hca = lfirst(*(ListCell **) a);
-	HnswCandidate *hcb = lfirst(*(ListCell **) b);
 #endif
+{
+	HnswCandidate *hca = lfirst((ListCell *) a);
+	HnswCandidate *hcb = lfirst((ListCell *) b);
 
 	if (hca->distance < hcb->distance)
 		return 1;
@@ -909,15 +1060,12 @@ CompareCandidateDistances(const void *a, const void *b)
 static int
 #if PG_VERSION_NUM >= 130000
 CompareCandidateDistancesOffset(const ListCell *a, const ListCell *b)
-{
-	HnswCandidate *hca = lfirst(a);
-	HnswCandidate *hcb = lfirst(b);
 #else
 CompareCandidateDistancesOffset(const void *a, const void *b)
-{
-	HnswCandidate *hca = lfirst(*(ListCell **) a);
-	HnswCandidate *hcb = lfirst(*(ListCell **) b);
 #endif
+{
+	HnswCandidate *hca = lfirst((ListCell *) a);
+	HnswCandidate *hcb = lfirst((ListCell *) b);
 
 	if (hca->distance < hcb->distance)
 		return 1;
@@ -976,9 +1124,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
 {
 	List	   *r = NIL;
 	List	   *w = list_copy(c);
-	HnswCandidate **wd;
-	int			wdlen = 0;
-	int			wdoff = 0;
+	pairingheap *wd;
 	HnswNeighborArray *neighbors = HnswGetNeighbors(base, e2, lc);
 	bool		mustCalculate = !neighbors->closerSet;
 	List	   *added = NIL;
@@ -987,7 +1133,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
 	if (list_length(w) <= lm)
 		return w;
 
-	wd = palloc(sizeof(HnswCandidate *) * list_length(w));
+	wd = pairingheap_allocate(CompareNearestCandidates, NULL);
 
 	/* Ensure order of candidates is deterministic for closer caching */
 	if (sortCandidates)
@@ -1053,21 +1199,21 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
 		if (e->closer)
 			r = lappend(r, e);
 		else
-			wd[wdlen++] = e;
+			pairingheap_add(wd, &(CreatePairingHeapNode(e)->ph_node));
 	}
 
 	/* Cached value can only be used in future if sorted deterministically */
 	neighbors->closerSet = sortCandidates;
 
 	/* Keep pruned connections */
-	while (wdoff < wdlen && list_length(r) < lm)
-		r = lappend(r, wd[wdoff++]);
+	while (!pairingheap_is_empty(wd) && list_length(r) < lm)
+		r = lappend(r, ((HnswPairingHeapNode *) pairingheap_remove_first(wd))->inner);
 
 	/* Return pruned for update connections */
 	if (pruned != NULL)
 	{
-		if (wdoff < wdlen)
-			*pruned = wd[wdoff];
+		if (!pairingheap_is_empty(wd))
+			*pruned = ((HnswPairingHeapNode *) pairingheap_first(wd))->inner;
 		else
 			*pruned = linitial(w);
 	}
@@ -1125,7 +1271,7 @@ HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm
 				HnswElement hc3Element = HnswPtrAccess(base, hc3->element);
 
 				if (HnswPtrIsNull(base, hc3Element->value))
-					HnswLoadElement(hc3Element, &hc3->distance, &q, index, procinfo, collation, true);
+					HnswLoadElement(hc3Element, &hc3->distance, &q, index, procinfo, collation, true, NULL, NULL);
 				else
 					hc3->distance = GetCandidateDistance(base, hc3, q, procinfo, collation);
 
@@ -1247,7 +1393,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
 	/* 1st phase: greedy search to insert level */
 	for (int lc = entryLevel; lc >= level + 1; lc--)
 	{
-		w = HnswSearchLayer(base, q, ep, 1, lc, index, procinfo, collation, m, true, skipElement);
+		w = HnswSearchLayer(base, q, NULL, ep, 1, lc, index, procinfo, collation, m, true, skipElement);
 		ep = w;
 	}
 
@@ -1265,7 +1411,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
 		List	   *neighbors;
 		List	   *lw;
 
-		w = HnswSearchLayer(base, q, ep, efConstruction, lc, index, procinfo, collation, m, true, skipElement);
+		w = HnswSearchLayer(base, q, NULL, ep, efConstruction, lc, index, procinfo, collation, m, true, skipElement);
 
 		/* Elements being deleted or skipped can help with search */
 		/* but should be removed before selecting neighbors */

src/hnswvacuum.c

@@ -256,7 +256,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 		LockPage(index, HNSW_UPDATE_LOCK, ShareLock);
 
 		/* Load element */
-		HnswLoadElement(highestPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true);
+		HnswLoadElement(highestPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL, NULL);
 
 		/* Repair if needed */
 		if (NeedsUpdated(vacuumstate, highestPoint))
@@ -294,7 +294,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 			 * is outdated, this can remove connections at higher levels in
 			 * the graph until they are repaired, but this should be fine.
 			 */
-			HnswLoadElement(entryPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true);
+			HnswLoadElement(entryPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL, NULL);
 
 			if (NeedsUpdated(vacuumstate, entryPoint))
 			{
@@ -370,7 +370,7 @@ RepairGraph(HnswVacuumState * vacuumstate)
 
 			/* Create an element */
 			element = HnswInitElementFromBlock(blkno, offno);
-			HnswLoadElementFromTuple(element, etup, false, true);
+			HnswLoadElementFromTuple(element, etup, false, true, index);
 
 			elements = lappend(elements, element);
 		}
@@ -440,6 +440,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 	BlockNumber insertPage = InvalidBlockNumber;
 	Relation	index = vacuumstate->index;
 	BufferAccessStrategy bas = vacuumstate->bas;
+	bool		useIndexTuple = IndexRelationGetNumberOfAttributes(index);
 
 	/*
 	 * Wait for index scans to complete. Scans before this point may contain
@@ -521,7 +522,18 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 
 			/* Overwrite element */
 			etup->deleted = 1;
-			MemSet(&etup->data, 0, VARSIZE_ANY(&etup->data));
+			if (useIndexTuple)
+			{
+				IndexTuple	itup = (IndexTuple) &etup->data;
+
+				MemSet(itup, 0, IndexTupleSize(itup));
+			}
+			else
+			{
+				Vector	   *vec = (Vector *) (&etup->data);
+
+				MemSet(vec, 0, VARSIZE_ANY(vec));
+			}
 
 			/* Overwrite neighbors */
 			for (int i = 0; i < ntup->count; i++)

src/ivfbuild.c

@@ -57,7 +57,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 	 */
 	if (buildstate->kmeansnormprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value))
+		if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value, buildstate->normvec))
 			return;
 	}
 
@@ -105,7 +105,7 @@ SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
 	oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
 
 	/* Add sample */
-	AddSample(values, buildstate);
+	AddSample(values, state);
 
 	/* Reset memory context */
 	MemoryContextSwitchTo(oldCtx);
@@ -153,7 +153,7 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
 	/* Normalize if needed */
 	if (buildstate->normprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value))
+		if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
 			return;
 	}
 
@@ -356,6 +356,9 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 	buildstate->centers = VectorArrayInit(buildstate->lists, buildstate->dimensions);
 	buildstate->listInfo = palloc(sizeof(ListInfo) * buildstate->lists);
 
+	/* Reuse for each tuple */
+	buildstate->normvec = InitVector(buildstate->dimensions);
+
 	buildstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
 											   "Ivfflat build temporary context",
 											   ALLOCSET_DEFAULT_SIZES);
@@ -377,6 +380,7 @@ FreeBuildState(IvfflatBuildState * buildstate)
 {
 	VectorArrayFree(buildstate->centers);
 	pfree(buildstate->listInfo);
+	pfree(buildstate->normvec);
 
 #ifdef IVFFLAT_KMEANS_DEBUG
 	pfree(buildstate->listSums);

src/ivfflat.h

@@ -172,6 +172,7 @@ typedef struct IvfflatBuildState
 	VectorArray samples;
 	VectorArray centers;
 	ListInfo   *listInfo;
+	Vector	   *normvec;
 
 #ifdef IVFFLAT_KMEANS_DEBUG
 	double		inertia;
@@ -266,7 +267,7 @@ void		VectorArrayFree(VectorArray arr);
 void		PrintVectorArray(char *msg, VectorArray arr);
 void		IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
 FmgrInfo   *IvfflatOptionalProcInfo(Relation index, uint16 procnum);
-bool		IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value);
+bool		IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
 int			IvfflatGetLists(Relation index);
 void		IvfflatGetMetaPageInfo(Relation index, int *lists, int *dimensions);
 void		IvfflatUpdateList(Relation index, ListInfo listInfo, BlockNumber insertPage, BlockNumber originalInsertPage, BlockNumber startPage, ForkNumber forkNum);

src/ivfinsert.c

@@ -85,7 +85,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 	normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
 	if (normprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value))
+		if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, NULL))
 			return;
 	}
 

src/ivfscan.c

@@ -293,7 +293,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 
 			/* Fine if normalization fails */
 			if (so->normprocinfo != NULL)
-				IvfflatNormValue(so->normprocinfo, so->collation, &value);
+				IvfflatNormValue(so->normprocinfo, so->collation, &value, NULL);
 		}
 
 		IvfflatBench("GetScanLists", GetScanLists(scan, value));

src/ivfutils.c

@@ -75,14 +75,16 @@ IvfflatOptionalProcInfo(Relation index, uint16 procnum)
  * if it's different than the original value
  */
 bool
-IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value)
+IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
 {
 	double		norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
 
 	if (norm > 0)
 	{
 		Vector	   *v = DatumGetVector(*value);
-		Vector	   *result = InitVector(v->dim);
+
+		if (result == NULL)
+			result = InitVector(v->dim);
 
 		for (int i = 0; i < v->dim; i++)
 			result->x[i] = v->x[i] / norm;

src/vector.c

@@ -197,8 +197,6 @@ vector_in(PG_FUNCTION_ARGS)
 
 	while (pt != NULL && *stringEnd != ']')
 	{
-		float		val;
-
 		if (dim == VECTOR_MAX_DIM)
 			ereport(ERROR,
 					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
@@ -214,23 +212,15 @@ vector_in(PG_FUNCTION_ARGS)
 					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
 
 		/* Use strtof like float4in to avoid a double-rounding problem */
-		errno = 0;
-		val = strtof(pt, &stringEnd);
+		x[dim] = strtof(pt, &stringEnd);
+		CheckElement(x[dim]);
+		dim++;
 
 		if (stringEnd == pt)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
 
-		/* Check for range error like float4in */
-		if (errno == ERANGE && (val == 0 || isinf(val)))
-			ereport(ERROR,
-					(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-					 errmsg("\"%s\" is out of range for type vector", pt)));
-
-		CheckElement(val);
-		x[dim++] = val;
-
 		while (vector_isspace(*stringEnd))
 			stringEnd++;
 

test/expected/hnsw_l2.out

@@ -12,11 +12,14 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
  [0,0,0]
 (4 rows)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
- count 
--------
-     4
-(1 row)
+SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
+   val   
+---------
+ [0,0,0]
+ [1,1,1]
+ [1,2,3]
+ [1,2,4]
+(4 rows)
 
 SELECT COUNT(*) FROM t;
  count 

test/expected/input.out

@@ -63,21 +63,9 @@ SELECT '[1.5e-38,-1.5e-38]'::vector;
 (1 row)
 
 SELECT '[4e38,1]'::vector;
-ERROR:  "4e38" is out of range for type vector
+ERROR:  infinite value not allowed in vector
 LINE 1: SELECT '[4e38,1]'::vector;
                ^
-SELECT '[-4e38,1]'::vector;
-ERROR:  "-4e38" is out of range for type vector
-LINE 1: SELECT '[-4e38,1]'::vector;
-               ^
-SELECT '[1e-46,1]'::vector;
-ERROR:  "1e-46" is out of range for type vector
-LINE 1: SELECT '[1e-46,1]'::vector;
-               ^
-SELECT '[-1e-46,1]'::vector;
-ERROR:  "-1e-46" is out of range for type vector
-LINE 1: SELECT '[-1e-46,1]'::vector;
-               ^
 SELECT '[1,2,3'::vector;
 ERROR:  malformed vector literal: "[1,2,3"
 LINE 1: SELECT '[1,2,3'::vector;
@@ -128,30 +116,8 @@ SELECT '[1, ,3]'::vector;
 ERROR:  invalid input syntax for type vector: "[1, ,3]"
 LINE 1: SELECT '[1, ,3]'::vector;
                ^
-SELECT '[1,2,3]'::vector(3);
- vector  
----------
- [1,2,3]
-(1 row)
-
 SELECT '[1,2,3]'::vector(2);
 ERROR:  expected 2 dimensions, not 3
-SELECT '[1,2,3]'::vector(3, 2);
-ERROR:  invalid type modifier
-LINE 1: SELECT '[1,2,3]'::vector(3, 2);
-                          ^
-SELECT '[1,2,3]'::vector('a');
-ERROR:  invalid input syntax for type integer: "a"
-LINE 1: SELECT '[1,2,3]'::vector('a');
-                          ^
-SELECT '[1,2,3]'::vector(0);
-ERROR:  dimensions for type vector must be at least 1
-LINE 1: SELECT '[1,2,3]'::vector(0);
-                          ^
-SELECT '[1,2,3]'::vector(16001);
-ERROR:  dimensions for type vector cannot exceed 16000
-LINE 1: SELECT '[1,2,3]'::vector(16001);
-                          ^
 SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
  unnest  
 ---------

test/sql/hnsw_l2.sql

@@ -7,7 +7,7 @@ CREATE INDEX ON t USING hnsw (val vector_l2_ops);
 INSERT INTO t (val) VALUES ('[1,2,4]');
 
 SELECT * FROM t ORDER BY val <-> '[3,3,3]';
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
+SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
 SELECT COUNT(*) FROM t;
 
 TRUNCATE t;

test/sql/input.sql

@@ -11,9 +11,6 @@ SELECT '[1.5e38,-1.5e38]'::vector;
 SELECT '[1.5e+38,-1.5e+38]'::vector;
 SELECT '[1.5e-38,-1.5e-38]'::vector;
 SELECT '[4e38,1]'::vector;
-SELECT '[-4e38,1]'::vector;
-SELECT '[1e-46,1]'::vector;
-SELECT '[-1e-46,1]'::vector;
 SELECT '[1,2,3'::vector;
 SELECT '[1,2,3]9'::vector;
 SELECT '1,2,3'::vector;
@@ -25,13 +22,7 @@ SELECT '[1,]'::vector;
 SELECT '[1a]'::vector;
 SELECT '[1,,3]'::vector;
 SELECT '[1, ,3]'::vector;
-
-SELECT '[1,2,3]'::vector(3);
 SELECT '[1,2,3]'::vector(2);
-SELECT '[1,2,3]'::vector(3, 2);
-SELECT '[1,2,3]'::vector('a');
-SELECT '[1,2,3]'::vector(0);
-SELECT '[1,2,3]'::vector(16001);
 
 SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
 SELECT '{"[1,2,3]"}'::vector(2)[];

test/t/016_hnsw_vacuum_recall.pl

@@ -86,7 +86,7 @@ foreach (@queries)
 	push(@expected, $res);
 }
 
-test_recall(0.19, $limit, "before vacuum");
+test_recall(0.20, $limit, "before vacuum");
 test_recall(0.95, 100, "before vacuum");
 
 # TODO Test concurrent inserts with vacuum

test/t/020_hnsw_filtering.pl

@@ -0,0 +1,154 @@
+use strict;
+use warnings;
+use PostgresNode;
+use TestLib;
+use Test::More;
+
+my $node;
+my @queries = ();
+my @cs = ();
+my @expected;
+my $limit = 20;
+my $dim = 3;
+my $array_sql = join(",", ('random()') x $dim);
+my $nc = 50;
+my @types = ("int4", "int8", "text", "uuid", "varchar");
+my $type = $types[rand(@types)];
+
+sub cast_c
+{
+	my ($c) = @_;
+
+	if ($type eq "uuid")
+	{
+		return "('00000000-0000-0000-0000-0000000000' || LPAD(($c)::text, 2, '0'))::uuid";
+	}
+	else
+	{
+		return "($c)::$type";
+	}
+}
+
+sub test_recall
+{
+	my ($min, $operator) = @_;
+	my $correct = 0;
+	my $total = 0;
+
+	my $explain = $node->safe_psql("postgres", qq(
+		SET enable_seqscan = off;
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE c = @{[cast_c($cs[0])]} ORDER BY v $operator '$queries[0]' LIMIT $limit;
+	));
+	like($explain, qr/Index Cond/);
+
+	for my $i (0 .. $#queries)
+	{
+		my $actual = $node->safe_psql("postgres", qq(
+			SET enable_seqscan = off;
+			SELECT i FROM tst WHERE c = @{[cast_c($cs[$i])]} ORDER BY v $operator '$queries[$i]' LIMIT $limit;
+		));
+		my @actual_ids = split("\n", $actual);
+		my %actual_set = map { $_ => 1 } @actual_ids;
+
+		is(scalar(@actual_ids), $limit);
+
+		my @expected_ids = split("\n", $expected[$i]);
+
+		foreach (@expected_ids)
+		{
+			if (exists($actual_set{$_}))
+			{
+				$correct++;
+			}
+			$total++;
+		}
+	}
+
+	cmp_ok($correct / $total, ">=", $min, $operator);
+}
+
+# Initialize node
+$node = get_new_node('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c $type, c2 $type);");
+$node->safe_psql("postgres", qq(
+	INSERT INTO tst SELECT i, ARRAY[$array_sql], @{[cast_c("i % $nc")]}, @{[cast_c("i % $nc")]} FROM generate_series(1, 10000) i;
+));
+$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, c, c2);");
+$node->safe_psql("postgres", qq(
+	INSERT INTO tst SELECT i, ARRAY[$array_sql], @{[cast_c("i % $nc")]}, @{[cast_c("i % $nc")]} FROM generate_series(1, 10000) i;
+));
+$node->safe_psql("postgres", "ANALYZE tst;");
+
+# Generate queries
+for (1 .. 20)
+{
+	my @r = ();
+	for (1 .. $dim)
+	{
+		push(@r, rand());
+	}
+	push(@queries, "[" . join(",", @r) . "]");
+	push(@cs, int(rand() * $nc));
+}
+
+# Get exact results
+@expected = ();
+for my $i (0 .. $#queries)
+{
+	my $res = $node->safe_psql("postgres", qq(
+		SET enable_indexscan = off;
+		SELECT i FROM tst WHERE c = @{[cast_c($cs[$i])]} ORDER BY v <-> '$queries[$i]' LIMIT $limit;
+	));
+	push(@expected, $res);
+}
+
+# Test recall
+test_recall(0.99, '<->');
+
+# Test no conditions
+my $explain = $node->safe_psql("postgres", qq(
+	EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v <-> '$queries[0]' LIMIT $limit;
+));
+like($explain, qr/Index Scan/);
+
+# Test range
+$explain = $node->safe_psql("postgres", qq(
+	EXPLAIN ANALYZE SELECT i FROM tst WHERE c >= @{[cast_c(1)]} AND c <= @{[cast_c(3)]} ORDER BY v <-> '$queries[0]' LIMIT $limit;
+));
+like($explain, qr/Index Cond: \(\(\S+ >= \S+\) AND \(\S+ <= \S+\)\)/);
+
+# Test multiple conditions
+$explain = $node->safe_psql("postgres", qq(
+	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = @{[cast_c($cs[0])]} AND c2 = @{[cast_c($cs[0])]} ORDER BY v <-> '$queries[0]' LIMIT $limit;
+));
+like($explain, qr/Index Cond: \(\(\S+ = \S+\) AND \(\S+ = \S+\)\)/);
+
+# Test no order
+$explain = $node->safe_psql("postgres", qq(
+	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = @{[cast_c($cs[0])]} LIMIT $limit;
+));
+like($explain, qr/Seq Scan/);
+
+# Test vacuum
+$node->safe_psql("postgres", "DELETE FROM tst WHERE c > @{[cast_c(5)]};");
+$node->safe_psql("postgres", "VACUUM tst;");
+
+# Test columns
+my ($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (c);");
+like($stderr, qr/first column must be a vector/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (c, v vector_l2_ops);");
+like($stderr, qr/first column must be a vector/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, c, c, c);");
+like($stderr, qr/index cannot have more than three columns/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, v vector_l2_ops);");
+like($stderr, qr/column 2 cannot be a vector/);
+
+done_testing();

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.6.2'
+default_version = '0.6.0'
 module_pathname = '$libdir/vector'
 relocatable = true