Improved cost estimation for IVFFlat [skip ci]

Previously, the cost estimation formula for a HNSW index scan utilized
a methodology that only factored in the entry level for an HNSW scan
and the "m" index parameter, which reflects the number of tuples (or
vectors) to scan at each step of a HNSW graph traversal. While this
would bias the PostgreSQL query planner to choose an HNSW index scan
over other available paths, this could lead to potential suboptimal
index selection, for example, choosing to use a HNSW index instead of
an available B-tree index that has better selectivity.

The number of tuples scanned during HNSW graph traversal is principally
influenced by these factors:

 * The number of tuples stored in the index
 * `m` - the number of tuples that are scanned in each step of the graph
   traversal
 * `hnsw.ef_search` - which influences the total number of steps it
   takes for the scan to converge on the approximated nearest neighbors

Through testing different source models for vectors, we also observed
that the correlation of vectors in mdoels would impact this convergence.
For this first iteration, we've opted to hardcode a constant scaling
factor and set it to `0.55`, though a future commit may turn this into
a configurable parameter.

The high-level formula for estimating the cost of a HNSW index scan is
as such:

```
(entryLevel * m) + (layer0TuplesMax * layer0Selectivity)
```

where

- `(entryLevel * m)` is the lower bound of tuples to scan, as it
accounts for the graph traversal to layer 0 (L0). (L1 and above has an ef=1)
- `layer0TuplesMax` is an estimate of the maximum number of tuples to
scan at L0. This accounts for tuples that may end up being discarded due
to them already being visited. Testing shows that the number of steps
until converge is similar to the value of `hnsw.ef_search`, thus we can
estimate tuples max at `hnsw.ef_search * m * 2`
- `layer0Selectivity` - estimates the percentage of tuples that will
actually be scanned during the index traversal, multipled by the scaling
factor

In addition to the `m` build parameter and `hsnw.ef_search`, costs
estimates can be influenced by standard PostgreSQL costing parameters,
though adjusting those (e.g. `random_page_cost`) should be done with
care.

Co-authored-by: @ankane

.github/workflows/build.yml

@@ -8,17 +8,17 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          # - postgres: 18
+          #   os: ubuntu-24.04
           - postgres: 17
             os: ubuntu-24.04
           - postgres: 16
-            os: ubuntu-24.04
+            os: ubuntu-22.04
           - postgres: 15
             os: ubuntu-22.04
           - postgres: 14
-            os: ubuntu-22.04
-          - postgres: 13
             os: ubuntu-20.04
-          - postgres: 12
+          - postgres: 13
             os: ubuntu-20.04
     steps:
       - uses: actions/checkout@v4

CHANGELOG.md

@@ -1,3 +1,15 @@
+## 0.8.0 (unreleased)
+
+- Added casts for arrays to `sparsevec`
+- Improved cost estimation
+- Reduced memory usage for HNSW index scans
+- Dropped support for Postgres 12
+
+## 0.7.4 (2024-08-05)
+
+- Fixed locking for parallel HNSW index builds
+- Fixed compilation error with GCC 14 on i386 when SSE2 is not enabled
+
 ## 0.7.3 (2024-07-22)
 
 - Fixed `failed to add index item` error with `sparsevec`

Dockerfile

@@ -1,4 +1,4 @@
-ARG PG_MAJOR=16
+ARG PG_MAJOR=17
 FROM postgres:$PG_MAJOR
 ARG PG_MAJOR
 

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": "0.7.3",
+	"version": "0.7.4",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.7.3",
+			"version": "0.7.4",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,8 +1,9 @@
 EXTENSION = vector
-EXTVERSION = 0.7.3
+EXTVERSION = 0.7.4
 
 MODULE_big = vector
-DATA = $(wildcard sql/*--*.sql)
+DATA = $(wildcard sql/*--*--*.sql)
+DATA_built = sql/$(EXTENSION)--$(EXTVERSION).sql
 OBJS = src/bitutils.o src/bitvec.o src/halfutils.o src/halfvec.o src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/sparsevec.o src/vector.o
 HEADERS = src/halfvec.h src/sparsevec.h src/vector.h
 
@@ -42,8 +43,6 @@ all: sql/$(EXTENSION)--$(EXTVERSION).sql
 sql/$(EXTENSION)--$(EXTVERSION).sql: sql/$(EXTENSION).sql
 	cp $< $@
 
-EXTRA_CLEAN = sql/$(EXTENSION)--$(EXTVERSION).sql
-
 PG_CONFIG ?= pg_config
 PGXS := $(shell $(PG_CONFIG) --pgxs)
 include $(PGXS)
@@ -53,7 +52,7 @@ ifeq ($(PROVE),)
 	PROVE = prove
 endif
 
-# for Postgres 15
+# for Postgres < 15
 PROVE_FLAGS += -I ./test/perl
 
 prove_installcheck:
@@ -67,7 +66,7 @@ dist:
 	git archive --format zip --prefix=$(EXTENSION)-$(EXTVERSION)/ --output dist/$(EXTENSION)-$(EXTVERSION).zip master
 
 # for Docker
-PG_MAJOR ?= 16
+PG_MAJOR ?= 17
 
 .PHONY: docker
 

Makefile.win

@@ -1,6 +1,7 @@
 EXTENSION = vector
-EXTVERSION = 0.7.3
+EXTVERSION = 0.7.4
 
+DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql
 OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj 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\sparsevec.obj src\vector.obj
 HEADERS = src\halfvec.h src\sparsevec.h src\vector.h
 
@@ -19,11 +20,6 @@ PG_CFLAGS = $(PG_CFLAGS) $(OPTFLAGS) /O2 /fp:fast
 # https://learn.microsoft.com/en-us/cpp/error-messages/tool-errors/vectorizer-and-parallelizer-messages
 # PG_CFLAGS = $(PG_CFLAGS) /Qvec-report:2
 
-all: sql\$(EXTENSION)--$(EXTVERSION).sql
-
-sql\$(EXTENSION)--$(EXTVERSION).sql: sql\$(EXTENSION).sql
-	copy sql\$(EXTENSION).sql $@
-
 # TODO use pg_config
 !ifndef PGROOT
 !error PGROOT is not set
@@ -43,15 +39,18 @@ SHLIB = $(EXTENSION).dll
 
 LIBS = "$(LIBDIR)\postgres.lib"
 
+all: $(SHLIB) $(DATA_built)
+
 .c.obj:
 	$(CC) $(CFLAGS) /c $< /Fo$@
 
 $(SHLIB): $(OBJS)
 	$(CC) $(CFLAGS) $(OBJS) $(LIBS) /link /DLL /OUT:$(SHLIB)
 
-all: $(SHLIB)
+sql\$(EXTENSION)--$(EXTVERSION).sql: sql\$(EXTENSION).sql
+	copy sql\$(EXTENSION).sql $@
 
-install:
+install: all
 	copy $(SHLIB) "$(PKGLIBDIR)"
 	copy $(EXTENSION).control "$(SHAREDIR)\extension"
 	copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
@@ -70,6 +69,6 @@ uninstall:
 
 clean:
 	del /f $(SHLIB) $(EXTENSION).lib $(EXTENSION).exp
+	del /f $(DATA_built)
 	del /f $(OBJS)
-	del /f sql\$(EXTENSION)--$(EXTVERSION).sql
 	del /f /s /q results regression.diffs regression.out tmp_check tmp_check_iso log output_iso

README.md

@@ -21,7 +21,7 @@ Compile and install the extension (supports Postgres 12+)
 
 ```sh
 cd /tmp
-git clone --branch v0.7.3 https://github.com/pgvector/pgvector.git
+git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -46,12 +46,14 @@ Then use `nmake` to build:
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\16"
 cd %TEMP%
-git clone --branch v0.7.3 https://github.com/pgvector/pgvector.git
+git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
 ```
 
+Note: Postgres 17 is not supported yet due to an upstream issue
+
 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).
@@ -100,13 +102,15 @@ Or add a vector column to an existing table
 ALTER TABLE items ADD COLUMN embedding vector(3);
 ```
 
+Also supports [half-precision](#half-precision-vectors), [binary](#binary-vectors), and [sparse](#sparse-vectors) vectors
+
 Insert vectors
 
 ```sql
 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))
+Or load vectors in bulk using `COPY` ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/loading/example.py))
 
 ```sql
 COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
@@ -145,6 +149,8 @@ Supported distance functions are:
 - `<#>` - (negative) inner product
 - `<=>` - cosine distance
 - `<+>` - L1 distance (added in 0.7.0)
+- `<~>` - Hamming distance (binary vectors, added in 0.7.0)
+- `<%>` - Jaccard distance (binary vectors, added in 0.7.0)
 
 Get the nearest neighbors to a row
 
@@ -202,7 +208,7 @@ You can add an index to use approximate nearest neighbor search, which trades so
 
 Supported index types are:
 
-- [HNSW](#hnsw) - added in 0.5.0
+- [HNSW](#hnsw)
 - [IVFFlat](#ivfflat)
 
 ## HNSW
@@ -473,7 +479,7 @@ SELECT * FROM items ORDER BY embedding::halfvec(3) <-> '[1,2,3]' LIMIT 5;
 
 ## Binary Vectors
 
-Use the `bit` type to store binary vectors ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py))
+Use the `bit` type to store binary vectors ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/imagehash/example.py))
 
 ```sql
 CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
@@ -551,7 +557,7 @@ SELECT id, content FROM items, plainto_tsquery('hello search') query
     WHERE textsearch @@ query ORDER BY ts_rank_cd(textsearch, query) DESC LIMIT 5;
 ```
 
-You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search_rrf.py) or a [cross-encoder](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search.py) to combine results.
+You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search/rrf.py) or a [cross-encoder](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search/cross_encoder.py) to combine results.
 
 ## Indexing Subvectors
 
@@ -597,7 +603,7 @@ 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)).
+Use `COPY` for bulk loading data ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/loading/example.py)).
 
 ```sql
 COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
@@ -687,7 +693,7 @@ 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)).
+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/example.py)).
 
 ## Languages
 
@@ -983,7 +989,7 @@ l2_normalize(sparsevec) → sparsevec | Normalize with Euclidean norm | 0.7.0
 If your machine has multiple Postgres installations, specify the path to [pg_config](https://www.postgresql.org/docs/current/app-pgconfig.html) with:
 
 ```sh
-export PG_CONFIG=/Library/PostgreSQL/16/bin/pg_config
+export PG_CONFIG=/Library/PostgreSQL/17/bin/pg_config
 ```
 
 Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
@@ -994,11 +1000,11 @@ sudo --preserve-env=PG_CONFIG make install
 
 A few common paths on Mac are:
 
-- EDB installer - `/Library/PostgreSQL/16/bin/pg_config`
+- EDB installer - `/Library/PostgreSQL/17/bin/pg_config`
-- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@16/bin/pg_config`
+- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@17/bin/pg_config`
-- Homebrew (x86-64) - `/usr/local/opt/postgresql@16/bin/pg_config`
+- Homebrew (x86-64) - `/usr/local/opt/postgresql@17/bin/pg_config`
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### Missing Header
 
@@ -1007,10 +1013,10 @@ If compilation fails with `fatal error: postgres.h: No such file or directory`,
 For Ubuntu and Debian, use:
 
 ```sh
-sudo apt install postgresql-server-dev-16
+sudo apt install postgresql-server-dev-17
 ```
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### Missing SDK
 
@@ -1043,17 +1049,17 @@ If installation fails with `Access is denied`, re-run the installation instructi
 Get the [Docker image](https://hub.docker.com/r/pgvector/pgvector) with:
 
 ```sh
-docker pull pgvector/pgvector:pg16
+docker pull pgvector/pgvector:pg17
 ```
 
-This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `16` with your Postgres server version, and run it the same way).
+This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `17` with your Postgres server version, and run it the same way).
 
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.7.3 https://github.com/pgvector/pgvector.git
+git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
 cd pgvector
-docker build --pull --build-arg PG_MAJOR=16 -t myuser/pgvector .
+docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector .
 ```
 
 ### Homebrew
@@ -1064,7 +1070,7 @@ With Homebrew Postgres, you can use:
 brew install pgvector
 ```
 
-Note: This only adds it to the `postgresql@14` formula
+Note: This only adds it to the `postgresql@17` and `postgresql@14` formulas
 
 ### PGXN
 
@@ -1079,22 +1085,22 @@ pgxn install vector
 Debian and Ubuntu packages are available from the [PostgreSQL APT Repository](https://wiki.postgresql.org/wiki/Apt). Follow the [setup instructions](https://wiki.postgresql.org/wiki/Apt#Quickstart) and run:
 
 ```sh
-sudo apt install postgresql-16-pgvector
+sudo apt install postgresql-17-pgvector
 ```
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### Yum
 
 RPM packages are available from the [PostgreSQL Yum Repository](https://yum.postgresql.org/). Follow the [setup instructions](https://www.postgresql.org/download/linux/redhat/) for your distribution and run:
 
 ```sh
-sudo yum install pgvector_16
+sudo yum install pgvector_17
 # or
-sudo dnf install pgvector_16
+sudo dnf install pgvector_17
 ```
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### pkg
 

sql/vector--0.7.3--0.7.4.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.4'" to load this file. \quit

sql/vector--0.7.4--0.8.0.sql

@@ -0,0 +1,26 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.0'" to load this file. \quit
+
+CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(real[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(double precision[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(numeric[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE CAST (integer[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(integer[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (real[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(real[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (double precision[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(double precision[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (numeric[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(numeric[], integer, boolean) AS ASSIGNMENT;

sql/vector.sql

@@ -782,6 +782,18 @@ CREATE FUNCTION halfvec_to_sparsevec(halfvec, integer, boolean) RETURNS sparseve
 CREATE FUNCTION sparsevec_to_halfvec(sparsevec, integer, boolean) RETURNS halfvec
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
+CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(real[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(double precision[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(numeric[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
 -- sparsevec casts
 
 CREATE CAST (sparsevec AS sparsevec)
@@ -799,6 +811,18 @@ CREATE CAST (sparsevec AS halfvec)
 CREATE CAST (halfvec AS sparsevec)
 	WITH FUNCTION halfvec_to_sparsevec(halfvec, integer, boolean) AS IMPLICIT;
 
+CREATE CAST (integer[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(integer[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (real[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(real[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (double precision[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(double precision[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (numeric[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(numeric[], integer, boolean) AS ASSIGNMENT;
+
 -- sparsevec operators
 
 CREATE OPERATOR <-> (

src/bitutils.h

@@ -4,8 +4,8 @@
 #include "postgres.h"
 
 /* Check version in first header */
-#if PG_VERSION_NUM < 120000
+#if PG_VERSION_NUM < 130000
-#error "Requires PostgreSQL 12+"
+#error "Requires PostgreSQL 13+"
 #endif
 
 extern uint64 (*BitHammingDistance) (uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 distance);

src/halfvec.c

@@ -19,11 +19,6 @@
 #include "utils/numeric.h"
 #include "vector.h"
 
-#if PG_VERSION_NUM < 130000
-#define TYPALIGN_DOUBLE 'd'
-#define TYPALIGN_INT 'i'
-#endif
-
 #define STATE_DIMS(x) (ARR_DIMS(x)[0] - 1)
 #define CreateStateDatums(dim) palloc(sizeof(Datum) * (dim + 1))
 
@@ -164,24 +159,6 @@ CheckStateArray(ArrayType *statearray, const char *caller)
 	return (float8 *) ARR_DATA_PTR(statearray);
 }
 
-#if PG_VERSION_NUM < 120003
-static pg_noinline void
-float_overflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: overflow")));
-}
-
-static pg_noinline void
-float_underflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: underflow")));
-}
-#endif
-
 /*
  * Convert textual representation to internal representation
  */

src/halfvec.h

@@ -38,7 +38,7 @@
 /* F16C has better performance than _Float16 (on x86-64) */
 #if defined(__F16C__)
 #define F16C_SUPPORT
-#elif defined(__FLT16_MAX__) && !defined(HALFVEC_DISPATCH) && !defined(__FreeBSD__)
+#elif defined(__FLT16_MAX__) && !defined(HALFVEC_DISPATCH) && !defined(__FreeBSD__) && (!defined(__i386__) || defined(__SSE2__))
 #define FLT16_SUPPORT
 #endif
 

src/hnsw.c

@@ -9,8 +9,10 @@
 #include "commands/vacuum.h"
 #include "hnsw.h"
 #include "miscadmin.h"
+#include "utils/float.h"
 #include "utils/guc.h"
 #include "utils/selfuncs.h"
+#include "utils/spccache.h"
 
 #if PG_VERSION_NUM < 150000
 #define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
@@ -59,17 +61,9 @@ HnswInit(void)
 
 	hnsw_relopt_kind = add_reloption_kind();
 	add_int_reloption(hnsw_relopt_kind, "m", "Max number of connections",
-					  HNSW_DEFAULT_M, HNSW_MIN_M, HNSW_MAX_M
+					  HNSW_DEFAULT_M, HNSW_MIN_M, HNSW_MAX_M, AccessExclusiveLock);
-#if PG_VERSION_NUM >= 130000
-					  ,AccessExclusiveLock
-#endif
-		);
 	add_int_reloption(hnsw_relopt_kind, "ef_construction", "Size of the dynamic candidate list for construction",
-					  HNSW_DEFAULT_EF_CONSTRUCTION, HNSW_MIN_EF_CONSTRUCTION, HNSW_MAX_EF_CONSTRUCTION
+					  HNSW_DEFAULT_EF_CONSTRUCTION, HNSW_MIN_EF_CONSTRUCTION, HNSW_MAX_EF_CONSTRUCTION, AccessExclusiveLock);
-#if PG_VERSION_NUM >= 130000
-					  ,AccessExclusiveLock
-#endif
-		);
 
 	DefineCustomIntVariable("hnsw.ef_search", "Sets the size of the dynamic candidate list for search",
 							"Valid range is 1..1000.", &hnsw_ef_search,
@@ -107,13 +101,17 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	GenericCosts costs;
 	int			m;
 	int			entryLevel;
+	int			layer0TuplesMax;
+	double		layer0Selectivity;
+	double		scalingFactor = 0.55;
+	double		spc_seq_page_cost;
 	Relation	index;
 
 	/* Never use index without order */
 	if (path->indexorderbys == NULL)
 	{
-		*indexStartupCost = DBL_MAX;
+		*indexStartupCost = get_float8_infinity();
-		*indexTotalCost = DBL_MAX;
+		*indexTotalCost = get_float8_infinity();
 		*indexSelectivity = 0;
 		*indexCorrelation = 0;
 		*indexPages = 0;
@@ -126,15 +124,55 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	HnswGetMetaPageInfo(index, &m, NULL);
 	index_close(index, NoLock);
 
-	/* Approximate entry level */
+	/*
-	entryLevel = (int) -log(1.0 / path->indexinfo->tuples) * HnswGetMl(m);
+	 * HNSW cost estimation follows a formula that accounts for the total
+	 * number of tuples indexed combined with the parameters that most
+	 * influence the duration of the index scan, namely: m - the number of
+	 * tuples that are scanned in each step of the HNSW graph traversal
+	 * ef_search - which influences the total number of steps taken at layer 0
+	 *
+	 * The source of the vector data can impact how many steps it takes to
+	 * converge on the set of vectors to return to the executor. Currently, we
+	 * use a hardcoded scaling factor (HNSWScanScalingFactor) to help
+	 * influence that, but this could later become a configurable parameter
+	 * based on the cost estimations.
+	 *
+	 * The tuple estimator formula is below:
+	 *
+	 * numIndexTuples = entryLevel * m + layer0TuplesMax * layer0Selectivity
+	 *
+	 * "entryLevel * m" represents the floor of tuples we need to scan to get
+	 * to layer 0 (L0).
+	 *
+	 * "layer0TuplesMax" is the estimated total number of tuples we'd scan at
+	 * L0 if we weren't discarding already visited tuples as part of the scan.
+	 *
+	 * "layer0Selectivity" estimates the percentage of tuples that are scanned
+	 * at L0, accounting for previously visited tuples, multiplied by the
+	 * "scalingFactor" (currently hardcoded).
+	 */
+	entryLevel = (int) (log(path->indexinfo->tuples + 1) * HnswGetMl(m));
+	layer0TuplesMax = HnswGetLayerM(m, 0) * hnsw_ef_search;
+	layer0Selectivity = (scalingFactor * log(path->indexinfo->tuples + 1)) /
+		(log(m) * (1 + log(hnsw_ef_search)));
 
-	/* TODO Improve estimate of visited tuples (currently underestimates) */
+	costs.numIndexTuples = (entryLevel * m) +
-	/* Account for number of tuples (or entry level), m, and ef_search */
+		(layer0TuplesMax * layer0Selectivity);
-	costs.numIndexTuples = (entryLevel + 2) * m;
 
 	genericcostestimate(root, path, loop_count, &costs);
 
+	get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
+
+	/* Adjust cost if needed since TOAST not included in seq scan cost */
+	if (costs.numIndexPages > path->indexinfo->rel->pages)
+	{
+		/* Change all page cost from random to sequential */
+		costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+
+		/* Remove cost of extra pages */
+		costs.indexTotalCost -= (costs.numIndexPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
+	}
+
 	/* Use total cost since most work happens before first tuple is returned */
 	*indexStartupCost = costs.indexTotalCost;
 	*indexTotalCost = costs.indexTotalCost;
@@ -154,23 +192,10 @@ hnswoptions(Datum reloptions, bool validate)
 		{"ef_construction", RELOPT_TYPE_INT, offsetof(HnswOptions, efConstruction)},
 	};
 
-#if PG_VERSION_NUM >= 130000
 	return (bytea *) build_reloptions(reloptions, validate,
 									  hnsw_relopt_kind,
 									  sizeof(HnswOptions),
 									  tab, lengthof(tab));
-#else
-	relopt_value *options;
-	int			numoptions;
-	HnswOptions *rdopts;
-
-	options = parseRelOptions(reloptions, validate, hnsw_relopt_kind, &numoptions);
-	rdopts = allocateReloptStruct(sizeof(HnswOptions), options, numoptions);
-	fillRelOptions((void *) rdopts, sizeof(HnswOptions), options, numoptions,
-				   validate, tab, lengthof(tab));
-
-	return (bytea *) rdopts;
-#endif
 }
 
 /*
@@ -195,9 +220,7 @@ hnswhandler(PG_FUNCTION_ARGS)
 
 	amroutine->amstrategies = 0;
 	amroutine->amsupport = 3;
-#if PG_VERSION_NUM >= 130000
 	amroutine->amoptsprocnum = 0;
-#endif
 	amroutine->amcanorder = false;
 	amroutine->amcanorderbyop = true;
 	amroutine->amcanbackward = false;	/* can change direction mid-scan */
@@ -210,17 +233,24 @@ hnswhandler(PG_FUNCTION_ARGS)
 	amroutine->amclusterable = false;
 	amroutine->ampredlocks = false;
 	amroutine->amcanparallel = false;
-	amroutine->amcaninclude = false;
+#if PG_VERSION_NUM >= 170000
-#if PG_VERSION_NUM >= 130000
+	amroutine->amcanbuildparallel = true;
-	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
-	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
 #endif
+	amroutine->amcaninclude = false;
+	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
+#if PG_VERSION_NUM >= 160000
+	amroutine->amsummarizing = false;
+#endif
+	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
 	amroutine->amkeytype = InvalidOid;
 
 	/* Interface functions */
 	amroutine->ambuild = hnswbuild;
 	amroutine->ambuildempty = hnswbuildempty;
 	amroutine->aminsert = hnswinsert;
+#if PG_VERSION_NUM >= 170000
+	amroutine->aminsertcleanup = NULL;
+#endif
 	amroutine->ambulkdelete = hnswbulkdelete;
 	amroutine->amvacuumcleanup = hnswvacuumcleanup;
 	amroutine->amcanreturn = NULL;

src/hnsw.h

@@ -76,11 +76,6 @@
 #define SeedRandom(seed) srandom(seed)
 #endif
 
-#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))
-#endif
-
 #define HnswIsElementTuple(tup) ((tup)->type == HNSW_ELEMENT_TUPLE_TYPE)
 #define HnswIsNeighborTuple(tup) ((tup)->type == HNSW_NEIGHBOR_TUPLE_TYPE)
 
@@ -160,11 +155,13 @@ struct HnswNeighborArray
 	HnswCandidate items[FLEXIBLE_ARRAY_MEMBER];
 };
 
-typedef struct HnswPairingHeapNode
+typedef struct HnswSearchCandidate
 {
-	pairingheap_node ph_node;
+	pairingheap_node c_node;
-	HnswCandidate *inner;
+	pairingheap_node w_node;
-}			HnswPairingHeapNode;
+	HnswElementPtr element;
+	float		distance;
+}			HnswSearchCandidate;
 
 /* HNSW index options */
 typedef struct HnswOptions
@@ -385,7 +382,7 @@ void	   *HnswAlloc(HnswAllocator * allocator, Size size);
 HnswElement HnswInitElement(char *base, ItemPointer tid, int m, double ml, int maxLevel, HnswAllocator * alloc);
 HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
 void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing);
-HnswCandidate *HnswEntryCandidate(char *base, HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
+HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
 void		HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building);
 void		HnswSetNeighborTuple(char *base, HnswNeighborTuple ntup, HnswElement e, int m);
 void		HnswAddHeapTid(HnswElement element, ItemPointer heaptid);

src/hnswbuild.c

@@ -60,12 +60,6 @@
 #include "pgstat.h"
 #endif
 
-#if PG_VERSION_NUM >= 130000
-#define CALLBACK_ITEM_POINTER ItemPointer tid
-#else
-#define CALLBACK_ITEM_POINTER HeapTuple hup
-#endif
-
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_status.h"
 #include "utils/wait_event.h"
@@ -75,10 +69,6 @@
 #define PARALLEL_KEY_HNSW_AREA			UINT64CONST(0xA000000000000002)
 #define PARALLEL_KEY_QUERY_TEXT			UINT64CONST(0xA000000000000003)
 
-#if PG_VERSION_NUM < 130000
-#define GENERATIONCHUNK_RAWSIZE (SIZEOF_SIZE_T + SIZEOF_VOID_P * 2)
-#endif
-
 /*
  * Create the metapage
  */
@@ -192,7 +182,9 @@ CreateGraphPages(HnswBuildState * buildstate)
 
 		/* Initial size check */
 		if (etupSize > HNSW_TUPLE_ALLOC_SIZE)
-			elog(ERROR, "index tuple too large");
+			ereport(ERROR,
+					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+					 errmsg("index tuple too large")));
 
 		HnswSetElementTuple(base, etup, element);
 
@@ -379,7 +371,13 @@ UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswEleme
 	for (int lc = e->level; lc >= 0; lc--)
 	{
 		int			lm = HnswGetLayerM(m, lc);
-		HnswNeighborArray *neighbors = HnswGetNeighbors(base, e, lc);
+		Size		neighborsSize = HNSW_NEIGHBOR_ARRAY_SIZE(lm);
+		HnswNeighborArray *neighbors = palloc(neighborsSize);
+
+		/* Copy neighbors to local memory */
+		LWLockAcquire(&e->lock, LW_SHARED);
+		memcpy(neighbors, HnswGetNeighbors(base, e, lc), neighborsSize);
+		LWLockRelease(&e->lock);
 
 		for (int i = 0; i < neighbors->length; i++)
 		{
@@ -389,7 +387,6 @@ UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswEleme
 			/* Keep scan-build happy on Mac x86-64 */
 			Assert(neighborElement);
 
-			/* Use element for lock instead of hc since hc can be replaced */
 			LWLockAcquire(&neighborElement->lock, LW_EXCLUSIVE);
 			HnswUpdateConnection(base, e, hc, lm, lc, NULL, NULL, procinfo, collation);
 			LWLockRelease(&neighborElement->lock);
@@ -578,17 +575,13 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
  * Callback for table_index_build_scan
  */
 static void
-BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
+BuildCallback(Relation index, ItemPointer tid, Datum *values,
 			  bool *isnull, bool tupleIsAlive, void *state)
 {
 	HnswBuildState *buildstate = (HnswBuildState *) state;
 	HnswGraph  *graph = buildstate->graph;
 	MemoryContext oldCtx;
 
-#if PG_VERSION_NUM < 130000
-	ItemPointer tid = &hup->t_self;
-#endif
-
 	/* Skip nulls */
 	if (isnull[0])
 		return;
@@ -651,11 +644,7 @@ HnswMemoryContextAlloc(Size size, void *state)
 	HnswBuildState *buildstate = (HnswBuildState *) state;
 	void	   *chunk = MemoryContextAlloc(buildstate->graphCtx, size);
 
-#if PG_VERSION_NUM >= 130000
 	buildstate->graphData.memoryUsed = MemoryContextMemAllocated(buildstate->graphCtx, false);
-#else
-	buildstate->graphData.memoryUsed += MAXALIGN(size);
-#endif
 
 	return chunk;
 }
@@ -691,17 +680,25 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 
 	/* Disallow varbit since require fixed dimensions */
 	if (TupleDescAttr(index->rd_att, 0)->atttypid == VARBITOID)
-		elog(ERROR, "type not supported for hnsw index");
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("type not supported for hnsw index")));
 
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
-		elog(ERROR, "column does not have dimensions");
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("column does not have dimensions")));
 
 	if (buildstate->dimensions > buildstate->typeInfo->maxDimensions)
-		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", buildstate->typeInfo->maxDimensions);
+		ereport(ERROR,
+				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+				 errmsg("column cannot have more than %d dimensions for hnsw index", buildstate->typeInfo->maxDimensions)));
 
 	if (buildstate->efConstruction < 2 * buildstate->m)
-		elog(ERROR, "ef_construction must be greater than or equal to 2 * m");
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("ef_construction must be greater than or equal to 2 * m")));
 
 	buildstate->reltuples = 0;
 	buildstate->indtuples = 0;

src/hnswinsert.c

@@ -379,8 +379,12 @@ HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, Hns
 			HnswElement neighborElement = HnswPtrAccess(base, hc->element);
 			OffsetNumber offno = neighborElement->neighborOffno;
 
-			/* Get latest neighbors since they may have changed */
+			/*
-			/* Do not lock yet since selecting neighbors can take time */
+			 * Get latest neighbors since they may have changed. Do not lock
+			 * yet since selecting neighbors can take time. Could use
+			 * optimistic locking to retry if another update occurs before
+			 * getting exclusive lock.
+			 */
 			HnswLoadNeighbors(neighborElement, index, m);
 
 			/*

src/hnswscan.c

@@ -160,15 +160,15 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 
 		so->first = false;
 
-#if defined(HNSW_MEMORY) && PG_VERSION_NUM >= 130000
+#if defined(HNSW_MEMORY)
-		elog(INFO, "memory: %zu MB", MemoryContextMemAllocated(so->tmpCtx, false) / (1024 * 1024));
+		elog(INFO, "memory: %zu KB", MemoryContextMemAllocated(so->tmpCtx, false) / 1024);
 #endif
 	}
 
 	while (list_length(so->w) > 0)
 	{
 		char	   *base = NULL;
-		HnswCandidate *hc = llast(so->w);
+		HnswSearchCandidate *hc = llast(so->w);
 		HnswElement element = HnswPtrAccess(base, hc->element);
 		ItemPointer heaptid;
 

src/hnswutils.c

@@ -5,6 +5,7 @@
 #include "access/generic_xlog.h"
 #include "catalog/pg_type.h"
 #include "catalog/pg_type_d.h"
+#include "common/hashfn.h"
 #include "fmgr.h"
 #include "hnsw.h"
 #include "lib/pairingheap.h"
@@ -14,12 +15,6 @@
 #include "utils/memdebug.h"
 #include "utils/rel.h"
 
-#if PG_VERSION_NUM >= 130000
-#include "common/hashfn.h"
-#else
-#include "utils/hashutils.h"
-#endif
-
 #if PG_VERSION_NUM < 170000
 static inline uint64
 murmurhash64(uint64 data)
@@ -112,6 +107,12 @@ typedef union
 	tidhash_hash *tids;
 }			visited_hash;
 
+typedef union
+{
+	HnswElement element;
+	ItemPointerData indextid;
+}			HnswUnvisited;
+
 /*
  * Get the max number of connections in an upper layer for each element in the index
  */
@@ -547,19 +548,19 @@ HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHe
 /*
  * Load an element and optionally get its distance from q
  */
-void
+static void
-HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec, float *maxDistance)
+HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec, float *maxDistance, HnswElement * element)
 {
 	Buffer		buf;
 	Page		page;
 	HnswElementTuple etup;
 
 	/* Read vector */
-	buf = ReadBuffer(index, element->blkno);
+	buf = ReadBuffer(index, blkno);
 	LockBuffer(buf, BUFFER_LOCK_SHARE);
 	page = BufferGetPage(buf);
 
-	etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, element->offno));
+	etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
 
 	Assert(HnswIsElementTuple(etup));
 
@@ -574,19 +575,32 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
 
 	/* Load element */
 	if (distance == NULL || maxDistance == NULL || *distance < *maxDistance)
-		HnswLoadElementFromTuple(element, etup, true, loadVec);
+	{
+		if (*element == NULL)
+			*element = HnswInitElementFromBlock(blkno, offno);
+
+		HnswLoadElementFromTuple(*element, etup, true, loadVec);
+	}
 
 	UnlockReleaseBuffer(buf);
 }
 
 /*
- * Get the distance for a candidate
+ * 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, float *maxDistance)
+{
+	HnswLoadElementImpl(element->blkno, element->offno, distance, q, index, procinfo, collation, loadVec, maxDistance, &element);
+}
+
+/*
+ * Get the distance for an element
  */
 static float
-GetCandidateDistance(char *base, HnswCandidate * hc, Datum q, FmgrInfo *procinfo, Oid collation)
+GetElementDistance(char *base, HnswElement element, Datum q, FmgrInfo *procinfo, Oid collation)
 {
-	HnswElement hce = HnswPtrAccess(base, hc->element);
+	Datum		value = HnswGetValue(base, element);
-	Datum		value = HnswGetValue(base, hce);
 
 	return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q, value));
 }
@@ -594,29 +608,32 @@ GetCandidateDistance(char *base, HnswCandidate * hc, Datum q, FmgrInfo *procinfo
 /*
  * Create a candidate for the entry point
  */
-HnswCandidate *
+HnswSearchCandidate *
 HnswEntryCandidate(char *base, HnswElement entryPoint, Datum q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
 {
-	HnswCandidate *hc = palloc(sizeof(HnswCandidate));
+	HnswSearchCandidate *hc = palloc(sizeof(HnswSearchCandidate));
 
 	HnswPtrStore(base, hc->element, entryPoint);
 	if (index == NULL)
-		hc->distance = GetCandidateDistance(base, hc, q, procinfo, collation);
+		hc->distance = GetElementDistance(base, entryPoint, q, procinfo, collation);
 	else
 		HnswLoadElement(entryPoint, &hc->distance, &q, index, procinfo, collation, loadVec, NULL);
 	return hc;
 }
 
+#define HnswGetSearchCandidate(membername, ptr) pairingheap_container(HnswSearchCandidate, membername, ptr)
+#define HnswGetSearchCandidateConst(membername, ptr) pairingheap_const_container(HnswSearchCandidate, membername, ptr)
+
 /*
  * Compare candidate distances
  */
 static int
 CompareNearestCandidates(const pairingheap_node *a, const pairingheap_node *b, void *arg)
 {
-	if (((const HnswPairingHeapNode *) a)->inner->distance < ((const HnswPairingHeapNode *) b)->inner->distance)
+	if (HnswGetSearchCandidateConst(c_node, a)->distance < HnswGetSearchCandidateConst(c_node, b)->distance)
 		return 1;
 
-	if (((const HnswPairingHeapNode *) a)->inner->distance > ((const HnswPairingHeapNode *) b)->inner->distance)
+	if (HnswGetSearchCandidateConst(c_node, a)->distance > HnswGetSearchCandidateConst(c_node, b)->distance)
 		return -1;
 
 	return 0;
@@ -628,27 +645,15 @@ CompareNearestCandidates(const pairingheap_node *a, const pairingheap_node *b, v
 static int
 CompareFurthestCandidates(const pairingheap_node *a, const pairingheap_node *b, void *arg)
 {
-	if (((const HnswPairingHeapNode *) a)->inner->distance < ((const HnswPairingHeapNode *) b)->inner->distance)
+	if (HnswGetSearchCandidateConst(w_node, a)->distance < HnswGetSearchCandidateConst(w_node, b)->distance)
 		return -1;
 
-	if (((const HnswPairingHeapNode *) a)->inner->distance > ((const HnswPairingHeapNode *) b)->inner->distance)
+	if (HnswGetSearchCandidateConst(w_node, a)->distance > HnswGetSearchCandidateConst(w_node, b)->distance)
 		return 1;
 
 	return 0;
 }
 
-/*
- * Create a pairing heap node for a candidate
- */
-static HnswPairingHeapNode *
-CreatePairingHeapNode(HnswCandidate * c)
-{
-	HnswPairingHeapNode *node = palloc(sizeof(HnswPairingHeapNode));
-
-	node->inner = c;
-	return node;
-}
-
 /*
  * Init visited
  */
@@ -667,11 +672,11 @@ InitVisited(char *base, visited_hash * v, Relation index, int ef, int m)
  * Add to visited
  */
 static inline void
-AddToVisited(char *base, visited_hash * v, HnswCandidate * hc, Relation index, bool *found)
+AddToVisited(char *base, visited_hash * v, HnswElementPtr elementPtr, Relation index, bool *found)
 {
 	if (index != NULL)
 	{
-		HnswElement element = HnswPtrAccess(base, hc->element);
+		HnswElement element = HnswPtrAccess(base, elementPtr);
 		ItemPointerData indextid;
 
 		ItemPointerSet(&indextid, element->blkno, element->offno);
@@ -679,23 +684,15 @@ AddToVisited(char *base, visited_hash * v, HnswCandidate * hc, Relation index, b
 	}
 	else if (base != NULL)
 	{
-#if PG_VERSION_NUM >= 130000
+		HnswElement element = HnswPtrAccess(base, elementPtr);
-		HnswElement element = HnswPtrAccess(base, hc->element);
 
-		offsethash_insert_hash(v->offsets, HnswPtrOffset(hc->element), element->hash, found);
+		offsethash_insert_hash(v->offsets, HnswPtrOffset(elementPtr), element->hash, found);
-#else
-		offsethash_insert(v->offsets, HnswPtrOffset(hc->element), found);
-#endif
 	}
 	else
 	{
-#if PG_VERSION_NUM >= 130000
+		HnswElement element = HnswPtrAccess(base, elementPtr);
-		HnswElement element = HnswPtrAccess(base, hc->element);
 
-		pointerhash_insert_hash(v->pointers, (uintptr_t) HnswPtrPointer(hc->element), element->hash, found);
+		pointerhash_insert_hash(v->pointers, (uintptr_t) HnswPtrPointer(elementPtr), element->hash, found);
-#else
-		pointerhash_insert(v->pointers, (uintptr_t) HnswPtrPointer(hc->element), found);
-#endif
 	}
 }
 
@@ -703,20 +700,96 @@ AddToVisited(char *base, visited_hash * v, HnswCandidate * hc, Relation index, b
  * Count element towards ef
  */
 static inline bool
-CountElement(char *base, HnswElement skipElement, HnswCandidate * hc)
+CountElement(HnswElement skipElement, HnswElement e)
 {
-	HnswElement e;
-
 	if (skipElement == NULL)
 		return true;
 
 	/* Ensure does not access heaptidsLength during in-memory build */
 	pg_memory_barrier();
 
-	e = HnswPtrAccess(base, hc->element);
+	/* Keep scan-build happy on Mac x86-64 */
+	Assert(e);
+
 	return e->heaptidsLength != 0;
 }
 
+/*
+ * Load unvisited neighbors from memory
+ */
+static void
+HnswLoadUnvisitedFromMemory(char *base, HnswElement element, HnswUnvisited * unvisited, int *unvisitedLength, visited_hash * v, int lc, HnswNeighborArray * localNeighborhood, Size neighborhoodSize)
+{
+	/* Get the neighborhood at layer lc */
+	HnswNeighborArray *neighborhood = HnswGetNeighbors(base, element, lc);
+
+	/* Copy neighborhood to local memory */
+	LWLockAcquire(&element->lock, LW_SHARED);
+	memcpy(localNeighborhood, neighborhood, neighborhoodSize);
+	LWLockRelease(&element->lock);
+
+	*unvisitedLength = 0;
+
+	for (int i = 0; i < localNeighborhood->length; i++)
+	{
+		HnswCandidate *hc = &localNeighborhood->items[i];
+		bool		found;
+
+		AddToVisited(base, v, hc->element, NULL, &found);
+
+		if (!found)
+			unvisited[(*unvisitedLength)++].element = HnswPtrAccess(base, hc->element);
+	}
+}
+
+/*
+ * Load unvisited neighbors from disk
+ */
+static void
+HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *unvisitedLength, visited_hash * v, Relation index, int m, int lm, int lc)
+{
+	Buffer		buf;
+	Page		page;
+	HnswNeighborTuple ntup;
+	int			start;
+	ItemPointerData indextids[HNSW_MAX_M * 2];
+
+	buf = ReadBuffer(index, element->neighborPage);
+	LockBuffer(buf, BUFFER_LOCK_SHARE);
+	page = BufferGetPage(buf);
+
+	ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, element->neighborOffno));
+
+	/* Ensure expected neighbors */
+	if (ntup->count != (element->level + 2) * m)
+	{
+		UnlockReleaseBuffer(buf);
+		return;
+	}
+
+	/* Copy to minimize lock time */
+	start = (element->level - lc) * m;
+	memcpy(&indextids, ntup->indextids + start, lm * sizeof(ItemPointerData));
+
+	UnlockReleaseBuffer(buf);
+
+	*unvisitedLength = 0;
+
+	for (int i = 0; i < lm; i++)
+	{
+		ItemPointer indextid = &indextids[i];
+		bool		found;
+
+		if (!ItemPointerIsValid(indextid))
+			break;
+
+		tidhash_insert(v->tids, *indextid, &found);
+
+		if (!found)
+			unvisited[(*unvisitedLength)++].indextid = *indextid;
+	}
+}
+
 /*
  * Algorithm 2 from paper
  */
@@ -729,43 +802,45 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
 	int			wlen = 0;
 	visited_hash v;
 	ListCell   *lc2;
-	HnswNeighborArray *neighborhoodData = NULL;
+	HnswNeighborArray *localNeighborhood = NULL;
-	Size		neighborhoodSize;
+	Size		neighborhoodSize = 0;
+	int			lm = HnswGetLayerM(m, lc);
+	HnswUnvisited *unvisited = palloc(lm * sizeof(HnswUnvisited));
+	int			unvisitedLength;
 
 	InitVisited(base, &v, index, ef, m);
 
 	/* Create local memory for neighborhood if needed */
 	if (index == NULL)
 	{
-		neighborhoodSize = HNSW_NEIGHBOR_ARRAY_SIZE(HnswGetLayerM(m, lc));
+		neighborhoodSize = HNSW_NEIGHBOR_ARRAY_SIZE(lm);
-		neighborhoodData = palloc(neighborhoodSize);
+		localNeighborhood = palloc(neighborhoodSize);
 	}
 
 	/* Add entry points to v, C, and W */
 	foreach(lc2, ep)
 	{
-		HnswCandidate *hc = (HnswCandidate *) lfirst(lc2);
+		HnswSearchCandidate *hc = (HnswSearchCandidate *) lfirst(lc2);
 		bool		found;
 
-		AddToVisited(base, &v, hc, index, &found);
+		AddToVisited(base, &v, hc->element, index, &found);
 
-		pairingheap_add(C, &(CreatePairingHeapNode(hc)->ph_node));
+		pairingheap_add(C, &hc->c_node);
-		pairingheap_add(W, &(CreatePairingHeapNode(hc)->ph_node));
+		pairingheap_add(W, &hc->w_node);
 
 		/*
 		 * Do not count elements being deleted towards ef when vacuuming. It
 		 * would be ideal to do this for inserts as well, but this could
 		 * affect insert performance.
 		 */
-		if (CountElement(base, skipElement, hc))
+		if (CountElement(skipElement, HnswPtrAccess(base, hc->element)))
 			wlen++;
 	}
 
 	while (!pairingheap_is_empty(C))
 	{
-		HnswNeighborArray *neighborhood;
+		HnswSearchCandidate *c = HnswGetSearchCandidate(c_node, pairingheap_remove_first(C));
-		HnswCandidate *c = ((HnswPairingHeapNode *) pairingheap_remove_first(C))->inner;
+		HnswSearchCandidate *f = HnswGetSearchCandidate(w_node, pairingheap_first(W));
-		HnswCandidate *f = ((HnswPairingHeapNode *) pairingheap_first(W))->inner;
 		HnswElement cElement;
 
 		if (c->distance > f->distance)
@@ -773,73 +848,67 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
 
 		cElement = HnswPtrAccess(base, c->element);
 
-		if (HnswPtrIsNull(base, cElement->neighbors))
-			HnswLoadNeighbors(cElement, index, m);
-
-		/* Get the neighborhood at layer lc */
-		neighborhood = HnswGetNeighbors(base, cElement, lc);
-
-		/* Copy neighborhood to local memory if needed */
 		if (index == NULL)
+			HnswLoadUnvisitedFromMemory(base, cElement, unvisited, &unvisitedLength, &v, lc, localNeighborhood, neighborhoodSize);
+		else
+			HnswLoadUnvisitedFromDisk(cElement, unvisited, &unvisitedLength, &v, index, m, lm, lc);
+
+		for (int i = 0; i < unvisitedLength; i++)
 		{
-			LWLockAcquire(&cElement->lock, LW_SHARED);
+			HnswElement eElement;
-			memcpy(neighborhoodData, neighborhood, neighborhoodSize);
+			HnswSearchCandidate *e;
-			LWLockRelease(&cElement->lock);
+			float		eDistance;
-			neighborhood = neighborhoodData;
+			bool		alwaysAdd = wlen < ef;
-		}
 
-		for (int i = 0; i < neighborhood->length; i++)
+			f = HnswGetSearchCandidate(w_node, pairingheap_first(W));
-		{
-			HnswCandidate *e = &neighborhood->items[i];
-			bool		visited;
 
-			AddToVisited(base, &v, e, index, &visited);
+			if (index == NULL)
-
-			if (!visited)
 			{
-				float		eDistance;
+				eElement = unvisited[i].element;
-				HnswElement eElement = HnswPtrAccess(base, e->element);
+				eDistance = GetElementDistance(base, eElement, q, procinfo, collation);
-				bool		alwaysAdd = wlen < ef;
+			}
+			else
+			{
+				ItemPointer indextid = &unvisited[i].indextid;
+				BlockNumber blkno = ItemPointerGetBlockNumber(indextid);
+				OffsetNumber offno = ItemPointerGetOffsetNumber(indextid);
 
-				f = ((HnswPairingHeapNode *) pairingheap_first(W))->inner;
+				/* Avoid any allocations if not adding */
+				eElement = NULL;
+				HnswLoadElementImpl(blkno, offno, &eDistance, &q, index, procinfo, collation, inserting, alwaysAdd ? NULL : &f->distance, &eElement);
 
-				if (index == NULL)
+				if (eElement == NULL)
-					eDistance = GetCandidateDistance(base, e, q, procinfo, collation);
+					continue;
-				else
+			}
-					HnswLoadElement(eElement, &eDistance, &q, index, procinfo, collation, inserting, alwaysAdd ? NULL : &f->distance);
 
-				if (eDistance < f->distance || alwaysAdd)
+			if (!(eDistance < f->distance || alwaysAdd))
-				{
+				continue;
-					HnswCandidate *ec;
 
-					Assert(!eElement->deleted);
+			Assert(!eElement->deleted);
 
-					/* Make robust to issues */
+			/* Make robust to issues */
-					if (eElement->level < lc)
+			if (eElement->level < lc)
-						continue;
+				continue;
 
-					/* Copy e */
+			/* Create a new candidate */
-					ec = palloc(sizeof(HnswCandidate));
+			e = palloc(sizeof(HnswSearchCandidate));
-					HnswPtrStore(base, ec->element, eElement);
+			HnswPtrStore(base, e->element, eElement);
-					ec->distance = eDistance;
+			e->distance = eDistance;
+			pairingheap_add(C, &e->c_node);
+			pairingheap_add(W, &e->w_node);
 
-					pairingheap_add(C, &(CreatePairingHeapNode(ec)->ph_node));
+			/*
-					pairingheap_add(W, &(CreatePairingHeapNode(ec)->ph_node));
+			 * Do not count elements being deleted towards ef when vacuuming.
+			 * It would be ideal to do this for inserts as well, but this
+			 * could affect insert performance.
+			 */
+			if (CountElement(skipElement, eElement))
+			{
+				wlen++;
 
-					/*
+				/* No need to decrement wlen */
-					 * Do not count elements being deleted towards ef when
+				if (wlen > ef)
-					 * vacuuming. It would be ideal to do this for inserts as
+					pairingheap_remove_first(W);
-					 * well, but this could affect insert performance.
-					 */
-					if (CountElement(base, skipElement, e))
-					{
-						wlen++;
-
-						/* No need to decrement wlen */
-						if (wlen > ef)
-							pairingheap_remove_first(W);
-					}
-				}
 			}
 		}
 	}
@@ -847,7 +916,7 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
 	/* Add each element of W to w */
 	while (!pairingheap_is_empty(W))
 	{
-		HnswCandidate *hc = ((HnswPairingHeapNode *) pairingheap_remove_first(W))->inner;
+		HnswSearchCandidate *hc = HnswGetSearchCandidate(w_node, pairingheap_remove_first(W));
 
 		w = lappend(w, hc);
 	}
@@ -859,17 +928,10 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
  * Compare candidate distances with pointer tie-breaker
  */
 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
 
 	if (hca->distance < hcb->distance)
 		return 1;
@@ -890,17 +952,10 @@ CompareCandidateDistances(const void *a, const void *b)
  * Compare candidate distances with offset tie-breaker
  */
 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
 
 	if (hca->distance < hcb->distance)
 		return 1;
@@ -1110,7 +1165,7 @@ HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm
 				if (HnswPtrIsNull(base, hc3Element->value))
 					HnswLoadElement(hc3Element, &hc3->distance, &q, index, procinfo, collation, true, NULL);
 				else
-					hc3->distance = GetCandidateDistance(base, hc3, q, procinfo, collation);
+					hc3->distance = GetElementDistance(base, hc3Element, q, procinfo, collation);
 
 				/* Prune element if being deleted */
 				if (hc3Element->heaptidsLength == 0)
@@ -1182,7 +1237,6 @@ RemoveElements(char *base, List *w, HnswElement skipElement)
 	return w2;
 }
 
-#if PG_VERSION_NUM >= 130000
 /*
  * Precompute hash
  */
@@ -1198,7 +1252,6 @@ PrecomputeHash(char *base, HnswElement element)
 	else
 		element->hash = hash_offset(HnswPtrOffset(ptr));
 }
-#endif
 
 /*
  * Algorithm 1 from paper
@@ -1213,11 +1266,9 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
 	Datum		q = HnswGetValue(base, element);
 	HnswElement skipElement = existing ? element : NULL;
 
-#if PG_VERSION_NUM >= 130000
 	/* Precompute hash */
 	if (index == NULL)
 		PrecomputeHash(base, element);
-#endif
 
 	/* No neighbors if no entry point */
 	if (entryPoint == NULL)
@@ -1246,16 +1297,27 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
 	{
 		int			lm = HnswGetLayerM(m, lc);
 		List	   *neighbors;
-		List	   *lw;
+		List	   *lw = NIL;
+		ListCell   *lc2;
 
 		w = HnswSearchLayer(base, q, ep, efConstruction, lc, index, procinfo, collation, m, true, skipElement);
 
+		/* Convert search candidates to candidates */
+		foreach(lc2, w)
+		{
+			HnswSearchCandidate *sc = lfirst(lc2);
+			HnswCandidate *hc = palloc(sizeof(HnswCandidate));
+
+			hc->element = sc->element;
+			hc->distance = sc->distance;
+
+			lw = lappend(lw, hc);
+		}
+
 		/* Elements being deleted or skipped can help with search */
 		/* but should be removed before selecting neighbors */
 		if (index != NULL)
-			lw = RemoveElements(base, w, skipElement);
+			lw = RemoveElements(base, lw, skipElement);
-		else
-			lw = w;
 
 		/*
 		 * Candidates are sorted, but not deterministically. Could set
@@ -1280,7 +1342,9 @@ SparsevecCheckValue(Pointer v)
 	SparseVector *vec = (SparseVector *) v;
 
 	if (vec->nnz > HNSW_MAX_NNZ)
-		elog(ERROR, "sparsevec cannot have more than %d non-zero elements for hnsw index", HNSW_MAX_NNZ);
+		ereport(ERROR,
+				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+				 errmsg("sparsevec cannot have more than %d non-zero elements for hnsw index", HNSW_MAX_NNZ)));
 }
 
 /*

src/ivfbuild.c

@@ -26,12 +26,6 @@
 #include "pgstat.h"
 #endif
 
-#if PG_VERSION_NUM >= 130000
-#define CALLBACK_ITEM_POINTER ItemPointer tid
-#else
-#define CALLBACK_ITEM_POINTER HeapTuple hup
-#endif
-
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_status.h"
 #include "utils/wait_event.h"
@@ -96,7 +90,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
  * Callback for sampling
  */
 static void
-SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
+SampleCallback(Relation index, ItemPointer tid, Datum *values,
 			   bool *isnull, bool tupleIsAlive, void *state)
 {
 	IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
@@ -207,16 +201,12 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
  * Callback for table_index_build_scan
  */
 static void
-BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
+BuildCallback(Relation index, ItemPointer tid, Datum *values,
 			  bool *isnull, bool tupleIsAlive, void *state)
 {
 	IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
 	MemoryContext oldCtx;
 
-#if PG_VERSION_NUM < 130000
-	ItemPointer tid = &hup->t_self;
-#endif
-
 	/* Skip nulls */
 	if (isnull[0])
 		return;
@@ -335,14 +325,20 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 
 	/* Disallow varbit since require fixed dimensions */
 	if (TupleDescAttr(index->rd_att, 0)->atttypid == VARBITOID)
-		elog(ERROR, "type not supported for ivfflat index");
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("type not supported for ivfflat index")));
 
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
-		elog(ERROR, "column does not have dimensions");
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("column does not have dimensions")));
 
 	if (buildstate->dimensions > buildstate->typeInfo->maxDimensions)
-		elog(ERROR, "column cannot have more than %d dimensions for ivfflat index", buildstate->typeInfo->maxDimensions);
+		ereport(ERROR,
+				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+				 errmsg("column cannot have more than %d dimensions for ivfflat index", buildstate->typeInfo->maxDimensions)));
 
 	buildstate->reltuples = 0;
 	buildstate->indtuples = 0;
@@ -355,7 +351,9 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 
 	/* Require more than one dimension for spherical k-means */
 	if (buildstate->kmeansnormprocinfo != NULL && buildstate->dimensions == 1)
-		elog(ERROR, "dimensions must be greater than one for this opclass");
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("dimensions must be greater than one for this opclass")));
 
 	/* Create tuple description for sorting */
 	buildstate->tupdesc = CreateTemplateTupleDesc(3);
@@ -562,6 +560,20 @@ PrintKmeansMetrics(IvfflatBuildState * buildstate)
 }
 #endif
 
+/*
+ * Initialize build sort state
+ */
+static Tuplesortstate *
+InitBuildSortState(TupleDesc tupdesc, int memory, SortCoordinate coordinate)
+{
+	AttrNumber	attNums[] = {1};
+	Oid			sortOperators[] = {Int4LessOperator};
+	Oid			sortCollations[] = {InvalidOid};
+	bool		nullsFirstFlags[] = {false};
+
+	return tuplesort_begin_heap(tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, memory, coordinate, false);
+}
+
 /*
  * Within leader, wait for end of heap scan
  */
@@ -609,12 +621,6 @@ IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, S
 	double		reltuples;
 	IndexInfo  *indexInfo;
 
-	/* Sort options, which must match AssignTuples */
-	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Int4LessOperator};
-	Oid			sortCollations[] = {InvalidOid};
-	bool		nullsFirstFlags[] = {false};
-
 	/* Initialize local tuplesort coordination state */
 	coordinate = palloc0(sizeof(SortCoordinateData));
 	coordinate->isWorker = true;
@@ -627,7 +633,7 @@ IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, S
 	InitBuildState(&buildstate, ivfspool->heap, ivfspool->index, indexInfo);
 	memcpy(buildstate.centers->items, ivfcenters, buildstate.centers->itemsize * buildstate.centers->maxlen);
 	buildstate.centers->length = buildstate.centers->maxlen;
-	ivfspool->sortstate = tuplesort_begin_heap(buildstate.tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, sortmem, coordinate, false);
+	ivfspool->sortstate = InitBuildSortState(buildstate.tupdesc, sortmem, coordinate);
 	buildstate.sortstate = ivfspool->sortstate;
 	scan = table_beginscan_parallel(ivfspool->heap,
 									ParallelTableScanFromIvfflatShared(ivfshared));
@@ -924,12 +930,6 @@ AssignTuples(IvfflatBuildState * buildstate)
 	int			parallel_workers = 0;
 	SortCoordinate coordinate = NULL;
 
-	/* Sort options, which must match IvfflatParallelScanAndSort */
-	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Int4LessOperator};
-	Oid			sortCollations[] = {InvalidOid};
-	bool		nullsFirstFlags[] = {false};
-
 	pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_ASSIGN);
 
 	/* Calculate parallel workers */
@@ -950,7 +950,7 @@ AssignTuples(IvfflatBuildState * buildstate)
 	}
 
 	/* Begin serial/leader tuplesort */
-	buildstate->sortstate = tuplesort_begin_heap(buildstate->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, maintenance_work_mem, coordinate, false);
+	buildstate->sortstate = InitBuildSortState(buildstate->tupdesc, maintenance_work_mem, coordinate);
 
 	/* Add tuples to sort */
 	if (buildstate->heap != NULL)

src/ivfflat.c

@@ -7,6 +7,7 @@
 #include "commands/progress.h"
 #include "commands/vacuum.h"
 #include "ivfflat.h"
+#include "utils/float.h"
 #include "utils/guc.h"
 #include "utils/selfuncs.h"
 #include "utils/spccache.h"
@@ -26,11 +27,7 @@ IvfflatInit(void)
 {
 	ivfflat_relopt_kind = add_reloption_kind();
 	add_int_reloption(ivfflat_relopt_kind, "lists", "Number of inverted lists",
-					  IVFFLAT_DEFAULT_LISTS, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS
+					  IVFFLAT_DEFAULT_LISTS, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, AccessExclusiveLock);
-#if PG_VERSION_NUM >= 130000
-					  ,AccessExclusiveLock
-#endif
-		);
 
 	DefineCustomIntVariable("ivfflat.probes", "Sets the number of probes",
 							"Valid range is 1..lists.", &ivfflat_probes,
@@ -78,8 +75,8 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	/* Never use index without order */
 	if (path->indexorderbys == NULL)
 	{
-		*indexStartupCost = DBL_MAX;
+		*indexStartupCost = get_float8_infinity();
-		*indexTotalCost = DBL_MAX;
+		*indexTotalCost = get_float8_infinity();
 		*indexSelectivity = 0;
 		*indexCorrelation = 0;
 		*indexPages = 0;
@@ -88,6 +85,8 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 
 	MemSet(&costs, 0, sizeof(costs));
 
+	genericcostestimate(root, path, loop_count, &costs);
+
 	index = index_open(path->indexinfo->indexoid, NoLock);
 	IvfflatGetMetaPageInfo(index, &lists, NULL);
 	index_close(index, NoLock);
@@ -97,14 +96,9 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	if (ratio > 1.0)
 		ratio = 1.0;
 
-	/*
+	/* Set startup cost since most work happens before first tuple is returned */
-	 * This gives us the subset of tuples to visit. This value is passed into
+	costs.indexStartupCost = costs.indexTotalCost * ratio;
-	 * the generic cost estimator to determine the number of pages to visit
+	costs.numIndexPages *= ratio;
-	 * during the index scan.
-	 */
-	costs.numIndexTuples = path->indexinfo->tuples * ratio;
-
-	genericcostestimate(root, path, loop_count, &costs);
 
 	get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
 
@@ -112,30 +106,25 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	if (costs.numIndexPages > path->indexinfo->rel->pages && ratio < 0.5)
 	{
 		/* Change all page cost from random to sequential */
-		costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+		costs.indexStartupCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
 
 		/* Remove cost of extra pages */
-		costs.indexTotalCost -= (costs.numIndexPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
+		costs.indexStartupCost -= (costs.numIndexPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
 	}
 	else
 	{
 		/* Change some page cost from random to sequential */
-		costs.indexTotalCost -= 0.5 * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+		costs.indexStartupCost -= 0.5 * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
 	}
 
-	/*
+	*indexStartupCost = costs.indexStartupCost;
-	 * If the list selectivity is lower than what is returned from the generic
-	 * cost estimator, use that.
-	 */
-	if (ratio < costs.indexSelectivity)
-		costs.indexSelectivity = ratio;
-
-	/* Use total cost since most work happens before first tuple is returned */
-	*indexStartupCost = costs.indexTotalCost;
 	*indexTotalCost = costs.indexTotalCost;
 	*indexSelectivity = costs.indexSelectivity;
 	*indexCorrelation = costs.indexCorrelation;
 	*indexPages = costs.numIndexPages;
+
+	Assert(*indexStartupCost > 0);
+	Assert(*indexTotalCost > 0);
 }
 
 /*
@@ -148,23 +137,10 @@ ivfflatoptions(Datum reloptions, bool validate)
 		{"lists", RELOPT_TYPE_INT, offsetof(IvfflatOptions, lists)},
 	};
 
-#if PG_VERSION_NUM >= 130000
 	return (bytea *) build_reloptions(reloptions, validate,
 									  ivfflat_relopt_kind,
 									  sizeof(IvfflatOptions),
 									  tab, lengthof(tab));
-#else
-	relopt_value *options;
-	int			numoptions;
-	IvfflatOptions *rdopts;
-
-	options = parseRelOptions(reloptions, validate, ivfflat_relopt_kind, &numoptions);
-	rdopts = allocateReloptStruct(sizeof(IvfflatOptions), options, numoptions);
-	fillRelOptions((void *) rdopts, sizeof(IvfflatOptions), options, numoptions,
-				   validate, tab, lengthof(tab));
-
-	return (bytea *) rdopts;
-#endif
 }
 
 /*
@@ -189,9 +165,7 @@ ivfflathandler(PG_FUNCTION_ARGS)
 
 	amroutine->amstrategies = 0;
 	amroutine->amsupport = 5;
-#if PG_VERSION_NUM >= 130000
 	amroutine->amoptsprocnum = 0;
-#endif
 	amroutine->amcanorder = false;
 	amroutine->amcanorderbyop = true;
 	amroutine->amcanbackward = false;	/* can change direction mid-scan */
@@ -204,17 +178,24 @@ ivfflathandler(PG_FUNCTION_ARGS)
 	amroutine->amclusterable = false;
 	amroutine->ampredlocks = false;
 	amroutine->amcanparallel = false;
-	amroutine->amcaninclude = false;
+#if PG_VERSION_NUM >= 170000
-#if PG_VERSION_NUM >= 130000
+	amroutine->amcanbuildparallel = true;
-	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
-	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
 #endif
+	amroutine->amcaninclude = false;
+	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
+#if PG_VERSION_NUM >= 160000
+	amroutine->amsummarizing = false;
+#endif
+	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
 	amroutine->amkeytype = InvalidOid;
 
 	/* Interface functions */
 	amroutine->ambuild = ivfflatbuild;
 	amroutine->ambuildempty = ivfflatbuildempty;
 	amroutine->aminsert = ivfflatinsert;
+#if PG_VERSION_NUM >= 170000
+	amroutine->aminsertcleanup = NULL;
+#endif
 	amroutine->ambulkdelete = ivfflatbulkdelete;
 	amroutine->amvacuumcleanup = ivfflatvacuumcleanup;
 	amroutine->amcanreturn = NULL;	/* tuple not included in heapsort */

src/ivfflat.h

@@ -253,8 +253,9 @@ typedef struct IvfflatScanOpaqueData
 	/* Sorting */
 	Tuplesortstate *sortstate;
 	TupleDesc	tupdesc;
-	TupleTableSlot *slot;
+	TupleTableSlot *vslot;
-	bool		isnull;
+	TupleTableSlot *mslot;
+	BufferAccessStrategy bas;
 
 	/* Support functions */
 	FmgrInfo   *procinfo;

src/ivfkmeans.c

@@ -151,12 +151,8 @@ RandomCenters(Relation index, VectorArray centers, const IvfflatTypeInfo * typeI
 static void
 ShowMemoryUsage(MemoryContext context, Size estimatedSize)
 {
-#if PG_VERSION_NUM >= 130000
 	elog(INFO, "total memory: %zu MB",
 		 MemoryContextMemAllocated(context, true) / (1024 * 1024));
-#else
-	MemoryContextStats(context);
-#endif
 	elog(INFO, "estimated memory: %zu MB", estimatedSize / (1024 * 1024));
 }
 #endif
@@ -327,7 +323,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
 	newCenters->length = numCenters;
 
 #ifdef IVFFLAT_MEMORY
-	ShowMemoryUsage(MemoryContextGetParent(CurrentMemoryContext));
+	ShowMemoryUsage(MemoryContextGetParent(CurrentMemoryContext), totalSize);
 #endif
 
 	/* Pick initial centers */

src/ivfscan.c

@@ -11,16 +11,23 @@
 #include "pgstat.h"
 #include "storage/bufmgr.h"
 
+#ifdef IVFFLAT_MEMORY
+#include "utils/memutils.h"
+#endif
+
+#define GetScanList(ptr) pairingheap_container(IvfflatScanList, ph_node, ptr)
+#define GetScanListConst(ptr) pairingheap_const_container(IvfflatScanList, ph_node, ptr)
+
 /*
  * Compare list distances
  */
 static int
 CompareLists(const pairingheap_node *a, const pairingheap_node *b, void *arg)
 {
-	if (((const IvfflatScanList *) a)->distance > ((const IvfflatScanList *) b)->distance)
+	if (GetScanListConst(a)->distance > GetScanListConst(b)->distance)
 		return 1;
 
-	if (((const IvfflatScanList *) a)->distance < ((const IvfflatScanList *) b)->distance)
+	if (GetScanListConst(a)->distance < GetScanListConst(b)->distance)
 		return -1;
 
 	return 0;
@@ -72,14 +79,14 @@ GetScanLists(IndexScanDesc scan, Datum value)
 
 				/* Calculate max distance */
 				if (listCount == so->probes)
-					maxDistance = ((IvfflatScanList *) pairingheap_first(so->listQueue))->distance;
+					maxDistance = GetScanList(pairingheap_first(so->listQueue))->distance;
 			}
 			else if (distance < maxDistance)
 			{
 				IvfflatScanList *scanlist;
 
 				/* Remove */
-				scanlist = (IvfflatScanList *) pairingheap_remove_first(so->listQueue);
+				scanlist = GetScanList(pairingheap_remove_first(so->listQueue));
 
 				/* Reuse */
 				scanlist->startPage = list->startPage;
@@ -87,7 +94,7 @@ GetScanLists(IndexScanDesc scan, Datum value)
 				pairingheap_add(so->listQueue, &scanlist->ph_node);
 
 				/* Update max distance */
-				maxDistance = ((IvfflatScanList *) pairingheap_first(so->listQueue))->distance;
+				maxDistance = GetScanList(pairingheap_first(so->listQueue))->distance;
 			}
 		}
 
@@ -106,19 +113,12 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
 	TupleDesc	tupdesc = RelationGetDescr(scan->indexRelation);
 	double		tuples = 0;
-	TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
+	TupleTableSlot *slot = so->vslot;
-
-	/*
-	 * Reuse same set of shared buffers for scan
-	 *
-	 * See postgres/src/backend/storage/buffer/README for description
-	 */
-	BufferAccessStrategy bas = GetAccessStrategy(BAS_BULKREAD);
 
 	/* Search closest probes lists */
 	while (!pairingheap_is_empty(so->listQueue))
 	{
-		BlockNumber searchPage = ((IvfflatScanList *) pairingheap_remove_first(so->listQueue))->startPage;
+		BlockNumber searchPage = GetScanList(pairingheap_remove_first(so->listQueue))->startPage;
 
 		/* Search all entry pages for list */
 		while (BlockNumberIsValid(searchPage))
@@ -127,7 +127,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
 			Page		page;
 			OffsetNumber maxoffno;
 
-			buf = ReadBufferExtended(scan->indexRelation, MAIN_FORKNUM, searchPage, RBM_NORMAL, bas);
+			buf = ReadBufferExtended(scan->indexRelation, MAIN_FORKNUM, searchPage, RBM_NORMAL, so->bas);
 			LockBuffer(buf, BUFFER_LOCK_SHARE);
 			page = BufferGetPage(buf);
 			maxoffno = PageGetMaxOffsetNumber(page);
@@ -166,8 +166,6 @@ GetScanItems(IndexScanDesc scan, Datum value)
 		}
 	}
 
-	FreeAccessStrategy(bas);
-
 	if (tuples < 100)
 		ereport(DEBUG1,
 				(errmsg("index scan found few tuples"),
@@ -217,6 +215,20 @@ GetScanValue(IndexScanDesc scan)
 	return value;
 }
 
+/*
+ * Initialize scan sort state
+ */
+static Tuplesortstate *
+InitScanSortState(TupleDesc tupdesc)
+{
+	AttrNumber	attNums[] = {1};
+	Oid			sortOperators[] = {Float8LessOperator};
+	Oid			sortCollations[] = {InvalidOid};
+	bool		nullsFirstFlags[] = {false};
+
+	return tuplesort_begin_heap(tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
+}
+
 /*
  * Prepare for an index scan
  */
@@ -227,10 +239,6 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	IvfflatScanOpaque so;
 	int			lists;
 	int			dimensions;
-	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Float8LessOperator};
-	Oid			sortCollations[] = {InvalidOid};
-	bool		nullsFirstFlags[] = {false};
 	int			probes = ivfflat_probes;
 
 	scan = RelationGetIndexScan(index, nkeys, norderbys);
@@ -258,9 +266,18 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "heaptid", TIDOID, -1, 0);
 
 	/* Prep sort */
-	so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
+	so->sortstate = InitScanSortState(so->tupdesc);
 
-	so->slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsMinimalTuple);
+	/* Need separate slots for puttuple and gettuple */
+	so->vslot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
+	so->mslot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsMinimalTuple);
+
+	/*
+	 * Reuse same set of shared buffers for scan
+	 *
+	 * See postgres/src/backend/storage/buffer/README for description
+	 */
+	so->bas = GetAccessStrategy(BAS_BULKREAD);
 
 	so->listQueue = pairingheap_allocate(CompareLists, scan);
 
@@ -277,10 +294,8 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
 {
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
 
-#if PG_VERSION_NUM >= 130000
 	if (!so->first)
 		tuplesort_reset(so->sortstate);
-#endif
 
 	so->first = true;
 	pairingheap_reset(so->listQueue);
@@ -327,14 +342,19 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		IvfflatBench("GetScanItems", GetScanItems(scan, value));
 		so->first = false;
 
+#if defined(IVFFLAT_MEMORY)
+		elog(INFO, "memory: %zu MB", MemoryContextMemAllocated(CurrentMemoryContext, true) / (1024 * 1024));
+#endif
+
 		/* Clean up if we allocated a new value */
 		if (value != scan->orderByData->sk_argument)
 			pfree(DatumGetPointer(value));
 	}
 
-	if (tuplesort_gettupleslot(so->sortstate, true, false, so->slot, NULL))
+	if (tuplesort_gettupleslot(so->sortstate, true, false, so->mslot, NULL))
 	{
-		ItemPointer heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->slot, 2, &so->isnull));
+		bool		isnull;
+		ItemPointer heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->mslot, 2, &isnull));
 
 		scan->xs_heaptid = *heaptid;
 		scan->xs_recheck = false;
@@ -355,6 +375,10 @@ ivfflatendscan(IndexScanDesc scan)
 
 	pairingheap_free(so->listQueue);
 	tuplesort_end(so->sortstate);
+	FreeAccessStrategy(so->bas);
+	FreeTupleDesc(so->tupdesc);
+
+	/* TODO Free vslot and mslot without freeing TupleDesc */
 
 	pfree(so);
 	scan->opaque = NULL;

src/sparsevec.c

@@ -3,6 +3,7 @@
 #include <limits.h>
 #include <math.h>
 
+#include "catalog/pg_type.h"
 #include "common/string.h"
 #include "fmgr.h"
 #include "halfutils.h"
@@ -11,6 +12,7 @@
 #include "sparsevec.h"
 #include "utils/array.h"
 #include "utils/builtins.h"
+#include "utils/lsyscache.h"
 #include "vector.h"
 
 #if PG_VERSION_NUM >= 120000
@@ -670,6 +672,137 @@ halfvec_to_sparsevec(PG_FUNCTION_ARGS)
 	PG_RETURN_POINTER(result);
 }
 
+/*
+ * Convert array to sparse vector
+ */
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(array_to_sparsevec);
+Datum
+array_to_sparsevec(PG_FUNCTION_ARGS)
+{
+	ArrayType  *array = PG_GETARG_ARRAYTYPE_P(0);
+	int32		typmod = PG_GETARG_INT32(1);
+	SparseVector *result;
+	int16		typlen;
+	bool		typbyval;
+	char		typalign;
+	Datum	   *elemsp;
+	int			nelemsp;
+	int			nnz = 0;
+	float	   *values;
+	int			j = 0;
+
+	if (ARR_NDIM(array) > 1)
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("array must be 1-D")));
+
+	if (ARR_HASNULL(array) && array_contains_nulls(array))
+		ereport(ERROR,
+				(errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
+				 errmsg("array must not contain nulls")));
+
+	get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
+	deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, NULL, &nelemsp);
+
+	CheckDim(nelemsp);
+	CheckExpectedDim(typmod, nelemsp);
+
+#ifdef _MSC_VER
+/* /fp:fast may not propagate +/-Infinity or NaN */
+#define IS_NOT_ZERO(v) (isnan((float) (v)) || isinf((float) (v)) || ((float) (v)) != 0)
+#else
+#define IS_NOT_ZERO(v) (((float) (v)) != 0)
+#endif
+
+	if (ARR_ELEMTYPE(array) == INT4OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			nnz += IS_NOT_ZERO(DatumGetInt32(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == FLOAT8OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			nnz += IS_NOT_ZERO(DatumGetFloat8(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == FLOAT4OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			nnz += IS_NOT_ZERO(DatumGetFloat4(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == NUMERICOID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			nnz += IS_NOT_ZERO(DirectFunctionCall1(numeric_float4, elemsp[i]));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("unsupported array type")));
+	}
+
+	result = InitSparseVector(nelemsp, nnz);
+	values = SPARSEVEC_VALUES(result);
+
+#define PROCESS_ARRAY_ELEM(elem) \
+	do { \
+		float v = (float) (elem); \
+		if (IS_NOT_ZERO(v)) { \
+			/* Safety check */ \
+			if (j >= result->nnz) \
+				elog(ERROR, "safety check failed"); \
+			result->indices[j] = i; \
+			values[j] = v; \
+			j++; \
+		} \
+	} while (0)
+
+	if (ARR_ELEMTYPE(array) == INT4OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			PROCESS_ARRAY_ELEM(DatumGetInt32(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == FLOAT8OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			PROCESS_ARRAY_ELEM(DatumGetFloat8(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == FLOAT4OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			PROCESS_ARRAY_ELEM(DatumGetFloat4(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == NUMERICOID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			PROCESS_ARRAY_ELEM(DatumGetFloat4(DirectFunctionCall1(numeric_float4, elemsp[i])));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("unsupported array type")));
+	}
+
+#undef PROCESS_ARRAY_ELEM
+#undef IS_NOT_ZERO
+
+	/*
+	 * Free allocation from deconstruct_array. Do not free individual elements
+	 * when pass-by-reference since they point to original array.
+	 */
+	pfree(elemsp);
+
+	if (j != result->nnz)
+		elog(ERROR, "correctness check failed");
+
+	/* Check elements */
+	for (int i = 0; i < result->nnz; i++)
+		CheckElement(values[i]);
+
+	PG_RETURN_POINTER(result);
+}
+
 /*
  * Get the L2 squared distance between sparse vectors
  */

src/vector.c

@@ -26,11 +26,6 @@
 #include "varatt.h"
 #endif
 
-#if PG_VERSION_NUM < 130000
-#define TYPALIGN_DOUBLE 'd'
-#define TYPALIGN_INT 'i'
-#endif
-
 #define STATE_DIMS(x) (ARR_DIMS(x)[0] - 1)
 #define CreateStateDatums(dim) palloc(sizeof(Datum) * (dim + 1))
 
@@ -160,24 +155,6 @@ CheckStateArray(ArrayType *statearray, const char *caller)
 	return (float8 *) ARR_DATA_PTR(statearray);
 }
 
-#if PG_VERSION_NUM < 120003
-static pg_noinline void
-float_overflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: overflow")));
-}
-
-static pg_noinline void
-float_underflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: underflow")));
-}
-#endif
-
 /*
  * Convert textual representation to internal representation
  */

test/expected/cast.out

@@ -208,6 +208,62 @@ SELECT '{1:1e-8}/1'::sparsevec::halfvec;
  [0]
 (1 row)
 
+SELECT ARRAY[1,0,2,0,3,0]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT ARRAY[1.0,0.0,2.0,0.0,3.0,0.0]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT ARRAY[1,0,2,0,3,0]::float4[]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT ARRAY[1,0,2,0,3,0]::float8[]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT ARRAY[1,0,2,0,3,0]::numeric[]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec;
+    sparsevec    
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(6);
+    sparsevec    
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(5);
+ERROR:  expected 5 dimensions, not 6
+SELECT '{NULL}'::real[]::sparsevec;
+ERROR:  array must not contain nulls
+SELECT '{NaN}'::real[]::sparsevec;
+ERROR:  NaN not allowed in sparsevec
+SELECT '{Infinity}'::real[]::sparsevec;
+ERROR:  infinite value not allowed in sparsevec
+SELECT '{-Infinity}'::real[]::sparsevec;
+ERROR:  infinite value not allowed in sparsevec
+SELECT '{}'::real[]::sparsevec;
+ERROR:  sparsevec must have at least 1 dimension
+SELECT '{{1}}'::real[]::sparsevec;
+ERROR:  array must be 1-D
 SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
 ERROR:  vector cannot have more than 16000 dimensions
 SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;

test/perl/PostgreSQL/Test/Cluster.pm

@@ -0,0 +1,11 @@
+package PostgreSQL::Test::Cluster;
+
+use PostgresNode;
+
+sub new
+{
+	my ($class, $name) = @_;
+	return get_new_node($name);
+}
+
+1;

test/perl/PostgreSQL/Test/Utils.pm

@@ -0,0 +1,5 @@
+package PostgreSQL::Test::Utils;
+
+use TestLib;
+
+1;

test/perl/PostgresNode.pm

@@ -1,8 +0,0 @@
-use PostgreSQL::Test::Cluster;
-
-sub get_new_node
-{
-	return PostgreSQL::Test::Cluster->new(@_);
-}
-
-1;

test/perl/TestLib.pm

@@ -1,3 +0,0 @@
-use PostgreSQL::Test::Utils;
-
-1;

test/sql/cast.sql

@@ -58,6 +58,22 @@ SELECT '{}/16001'::sparsevec::halfvec;
 SELECT '{1:65520}/1'::sparsevec::halfvec;
 SELECT '{1:1e-8}/1'::sparsevec::halfvec;
 
+SELECT ARRAY[1,0,2,0,3,0]::sparsevec;
+SELECT ARRAY[1.0,0.0,2.0,0.0,3.0,0.0]::sparsevec;
+SELECT ARRAY[1,0,2,0,3,0]::float4[]::sparsevec;
+SELECT ARRAY[1,0,2,0,3,0]::float8[]::sparsevec;
+SELECT ARRAY[1,0,2,0,3,0]::numeric[]::sparsevec;
+
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec;
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(6);
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(5);
+SELECT '{NULL}'::real[]::sparsevec;
+SELECT '{NaN}'::real[]::sparsevec;
+SELECT '{Infinity}'::real[]::sparsevec;
+SELECT '{-Infinity}'::real[]::sparsevec;
+SELECT '{}'::real[]::sparsevec;
+SELECT '{{1}}'::real[]::sparsevec;
+
 SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
 SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;
 

test/t/001_ivfflat_wal.pl

@@ -2,9 +2,9 @@
 
 # Test generic xlog record work for ivfflat index replication.
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 32;
@@ -49,7 +49,7 @@ sub test_index_replay
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize primary node
-$node_primary = get_new_node('primary');
+$node_primary = PostgreSQL::Test::Cluster->new('primary');
 $node_primary->init(allows_streaming => 1);
 if ($dim > 32)
 {
@@ -67,7 +67,7 @@ my $backup_name = 'my_backup';
 $node_primary->backup($backup_name);
 
 # Create streaming replica linking to primary
-$node_replica = get_new_node('replica');
+$node_replica = PostgreSQL::Test::Cluster->new('replica');
 $node_replica->init_from_backup($node_primary, $backup_name, has_streaming => 1);
 $node_replica->start;
 

test/t/002_ivfflat_vacuum.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
@@ -15,7 +15,7 @@ for (1 .. $dim)
 my $array_sql = join(", ", @r);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/003_ivfflat_vector_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -49,7 +49,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/004_ivfflat_vector_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/005_ivfflat_query_recall.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/006_ivfflat_lists.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/007_ivfflat_inserts.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 768;
@@ -9,7 +9,7 @@ my $dim = 768;
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/008_ivfflat_centers.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/009_ivfflat_filtering.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
@@ -11,7 +11,7 @@ my $limit = 20;
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 
@@ -94,8 +94,7 @@ like($explain, qr/Seq Scan/);
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query';
 ));
-# TODO Do not use index
+like($explain, qr/Seq Scan/);
-like($explain, qr/Index Scan using idx/);
 
 # Test attribute index
 $node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);");
@@ -110,7 +109,6 @@ $node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING ivfflat (v v
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
 ));
-# TODO Use partial index
+like($explain, qr/Index Scan using partial_idx/);
-like($explain, qr/Index Scan using idx/);
 
 done_testing();

test/t/010_hnsw_wal.pl

@@ -2,9 +2,9 @@
 
 # Test generic xlog record work for hnsw index replication.
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 32;
@@ -49,7 +49,7 @@ sub test_index_replay
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize primary node
-$node_primary = get_new_node('primary');
+$node_primary = PostgreSQL::Test::Cluster->new('primary');
 $node_primary->init(allows_streaming => 1);
 if ($dim > 32)
 {
@@ -67,7 +67,7 @@ my $backup_name = 'my_backup';
 $node_primary->backup($backup_name);
 
 # Create streaming replica linking to primary
-$node_replica = get_new_node('replica');
+$node_replica = PostgreSQL::Test::Cluster->new('replica');
 $node_replica->init_from_backup($node_primary, $backup_name, has_streaming => 1);
 $node_replica->start;
 

test/t/011_hnsw_vacuum.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
@@ -15,7 +15,7 @@ for (1 .. $dim)
 my $array_sql = join(", ", @r);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/012_hnsw_vector_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -47,7 +47,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/013_hnsw_vector_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -47,7 +47,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/014_hnsw_vector_vacuum_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/015_hnsw_vector_duplicates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/016_hnsw_inserts.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Ensures elements and neighbors on both same and different pages
@@ -10,7 +10,7 @@ my $dim = 1900;
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/017_hnsw_filtering.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
@@ -11,16 +11,20 @@ my $limit = 20;
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 
 # Create table and index
 $node->safe_psql("postgres", "CREATE EXTENSION vector;");
 $node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int4, t text);");
+$node->safe_psql("postgres", "CREATE TABLE cat (i int4 PRIMARY KEY, t text, b boolean);");
 $node->safe_psql("postgres",
 	"INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc, 'test ' || i FROM generate_series(1, 10000) i;"
 );
+$node->safe_psql("postgres",
+	"INSERT INTO cat SELECT i, 'cat ' || i, i % 5 = 0 FROM generate_series(1, $nc) i;"
+);
 $node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
 $node->safe_psql("postgres", "ANALYZE tst;");
 
@@ -96,13 +100,25 @@ $explain = $node->safe_psql("postgres", qq(
 ));
 like($explain, qr/Seq Scan/);
 
+# Test join
+$explain = $node->safe_psql("postgres", qq(
+	EXPLAIN ANALYZE SELECT cat.t FROM cat INNER JOIN tst ON cat.i = tst.c ORDER BY v <-> '$query' LIMIT $limit;
+));
+like($explain, qr/Index Scan using idx/);
+
+# Test join with attribute filtering
+$explain = $node->safe_psql("postgres", qq(
+	EXPLAIN ANALYZE SELECT cat.t FROM cat INNER JOIN tst ON cat.i = tst.c WHERE cat.b = 't' ORDER BY v <-> '$query' LIMIT $limit;
+));
+like($explain, qr/Index Scan using idx/);
+
 # Test attribute index
 $node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);");
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
 ));
-# TODO Use attribute index
+# Use attribute index
-like($explain, qr/Index Scan using idx/);
+like($explain, qr/Bitmap Index Scan on attribute_idx/);
 
 # Test partial index
 $node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING hnsw (v vector_l2_ops) WHERE (c = $c);");

test/t/018_aggregates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 
@@ -53,7 +53,7 @@ sub test_aggregate
 	else
 	{
 		# Does not raise overflow error in this instance due to loss of precision
-		is($res, "[24576,24576,49152]")
+		is($res, "[24576,24576,49152]");
 	}
 }
 

test/t/019_storage.pl

@@ -1,13 +1,13 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 1024;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/020_hnsw_bit_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/021_hnsw_bit_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/022_hnsw_bit_vacuum_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/023_hnsw_bit_duplicates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/024_hnsw_halfvec_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/025_hnsw_halfvec_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/026_hnsw_halfvec_vacuum_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/027_hnsw_halfvec_duplicates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/028_hnsw_sparsevec_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -47,7 +47,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/029_hnsw_sparsevec_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -47,7 +47,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/030_hnsw_sparsevec_vacuum_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/031_hnsw_sparsevec_duplicates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/032_ivfflat_halfvec_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/033_comparison.pl

@@ -1,14 +1,14 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
 my $array_sql = join(",", ('floor(random() * 2)::int - 1') x 3);
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/034_distance_functions.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -10,7 +10,7 @@ my $dim = 5;
 my $array_sql = join(",", ('floor(random() * 4)::int - 2') x $dim);
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/035_ivfflat_bit_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/036_ivfflat_bit_centers.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/037_inputs.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/038_hnsw_sparsevec_vacuum_insert.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
+use warnings FATAL => 'all';
-use PostgresNode;
+use PostgreSQL::Test::Cluster;
-use TestLib;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 
@@ -14,25 +14,29 @@ $node->safe_psql("postgres", "CREATE EXTENSION vector;");
 $node->safe_psql("postgres", "CREATE TABLE tst (i serial, v sparsevec(100000));");
 $node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v sparsevec_l2_ops);");
 
-for (1 .. 3) {
+for (1 .. 3)
-    for (1 .. 100) {
+{
-        my @elements;
+	for (1 .. 100)
-        my %indices;
+	{
-        for (1 .. int(rand() * 100)) {
+		my @elements;
-            my $index = int(rand() * (100000 - 1)) + 1;
+		my %indices;
-            if (!exists($indices{$index})) {
+		for (1 .. int(rand() * 100))
-                my $value = rand();
+		{
-                push(@elements, "$index:$value");
+			my $index = int(rand() * (100000 - 1)) + 1;
-                $indices{$index} = 1;
+			if (!exists($indices{$index}))
-            }
+			{
-        }
+				my $value = rand();
-        my $embedding = "{" . join(",", @elements) . "}/100000";
+				push(@elements, "$index:$value");
-        $node->safe_psql("postgres", "INSERT INTO tst (v) VALUES ('$embedding');");
+				$indices{$index} = 1;
-    }
+			}
+		}
+		my $embedding = "{" . join(",", @elements) . "}/100000";
+		$node->safe_psql("postgres", "INSERT INTO tst (v) VALUES ('$embedding');");
+	}
 
-    $node->safe_psql("postgres", "DELETE FROM tst WHERE i % 2 = 0;");
+	$node->safe_psql("postgres", "DELETE FROM tst WHERE i % 2 = 0;");
-    $node->safe_psql("postgres", "VACUUM tst;");
+	$node->safe_psql("postgres", "VACUUM tst;");
-    is(1, 1);
+	is(1, 1);
 }
 
 done_testing();

test/t/039_hnsw_cost.pl

@@ -0,0 +1,51 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my @dims = (384, 1536);
+my $limit = 10;
+
+# Initialize node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+
+for my $dim (@dims)
+{
+	my $array_sql = join(",", ('random()') x $dim);
+	my $n = $dim == 384 ? 3000 : 1000;
+
+	# Create table and index
+	$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
+	$node->safe_psql("postgres",
+		"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, $n) i;"
+	);
+	$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
+	$node->safe_psql("postgres", "ANALYZE tst;");
+
+	# Generate query
+	my @r = ();
+	for (1 .. $dim)
+	{
+		push(@r, rand());
+	}
+	my $query = "[" . join(",", @r) . "]";
+
+	my $explain = $node->safe_psql("postgres", qq(
+		EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v <-> '$query' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx/);
+
+	$explain = $node->safe_psql("postgres", qq(
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx/);
+
+	$node->safe_psql("postgres", "DROP TABLE tst;");
+}
+
+done_testing();

test/t/040_ivfflat_cost.pl

@@ -0,0 +1,50 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my @dims = (384, 1536);
+my $limit = 10;
+
+# Initialize node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+
+for my $dim (@dims)
+{
+	my $array_sql = join(",", ('random()') x $dim);
+
+	# Create table and index
+	$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
+	$node->safe_psql("postgres",
+		"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 6000) i;"
+	);
+	$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING ivfflat (v vector_l2_ops) WITH (lists = 5);");
+	$node->safe_psql("postgres", "ANALYZE tst;");
+
+	# Generate query
+	my @r = ();
+	for (1 .. $dim)
+	{
+		push(@r, rand());
+	}
+	my $query = "[" . join(",", @r) . "]";
+
+	my $explain = $node->safe_psql("postgres", qq(
+		EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v <-> '$query' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx/);
+
+	$explain = $node->safe_psql("postgres", qq(
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx/);
+
+	$node->safe_psql("postgres", "DROP TABLE tst;");
+}
+
+done_testing();

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.7.3'
+default_version = '0.7.4'
 module_pathname = '$libdir/vector'
 relocatable = true