Merge branch 'master' into hqann2

Co-authored-by: "Jonathan S. Katz" <jkatz@amazon.com>

.github/workflows/build.yml

@@ -49,7 +49,7 @@ jobs:
           - postgres: 16
             os: macos-14
           - postgres: 14
-            os: macos-12
+            os: macos-13
     steps:
       - uses: actions/checkout@v4
       - uses: ankane/setup-postgres@v1

CHANGELOG.md

@@ -1,6 +1,10 @@
 ## 0.8.0 (unreleased)
 
+- Added support for inline filtering with HNSW
+- Added support for iterative index scans
 - Added casts for arrays to `sparsevec`
+- Improved cost estimation
+- Improved performance of HNSW inserts and on-disk index builds
 - Reduced memory usage for HNSW index scans
 - Dropped support for Postgres 12
 

Dockerfile

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

Makefile

@@ -66,7 +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
 

README.md

@@ -52,6 +52,8 @@ 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).
@@ -100,6 +102,8 @@ 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
@@ -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
 
@@ -433,6 +439,12 @@ Create an index on one [or more](https://www.postgresql.org/docs/current/indexes
 CREATE INDEX ON items (category_id);
 ```
 
+Or a composite HNSW index for approximate search (added in 0.8.0)
+
+```sql
+CREATE INDEX ON items USING hnsw (embedding vector_l2_ops, category_id);
+```
+
 Or a [partial index](https://www.postgresql.org/docs/current/indexes-partial.html) on the vector column for approximate search
 
 ```sql
@@ -983,7 +995,7 @@ 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 +1006,11 @@ 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 +1019,10 @@ 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 +1055,17 @@ 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 v0.7.4 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 +1076,7 @@ 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 +1091,22 @@ 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
 
@@ -1183,6 +1195,7 @@ Thanks to:
 - [k-means++: The Advantage of Careful Seeding](https://theory.stanford.edu/~sergei/papers/kMeansPP-soda.pdf)
 - [Concept Decompositions for Large Sparse Text Data using Clustering](https://www.cs.utexas.edu/users/inderjit/public_papers/concept_mlj.pdf)
 - [Efficient and Robust Approximate Nearest Neighbor Search using Hierarchical Navigable Small World Graphs](https://arxiv.org/ftp/arxiv/papers/1603/1603.09320.pdf)
+- [HQANN: Efficient and Robust Similarity Search for Hybrid Queries with Structured and Unstructured Constraints](https://arxiv.org/pdf/2207.07940.pdf)
 
 ## History
 

sql/vector--0.7.4--0.8.0.sql

@@ -24,3 +24,11 @@ CREATE CAST (double precision[] AS sparsevec)
 
 CREATE CAST (numeric[] AS sparsevec)
 	WITH FUNCTION array_to_sparsevec(numeric[], integer, boolean) AS ASSIGNMENT;
+
+CREATE FUNCTION hnsw_attribute_distance(integer, integer) RETURNS float8
+	AS 'MODULE_PATHNAME', 'hnsw_int4_attribute_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE OPERATOR CLASS vector_integer_ops
+	DEFAULT FOR TYPE integer USING hnsw AS
+	OPERATOR 2 = (integer, integer),
+	FUNCTION 4 hnsw_attribute_distance(integer, integer);

sql/vector.sql

@@ -916,3 +916,13 @@ CREATE OPERATOR CLASS sparsevec_l1_ops
 	OPERATOR 1 <+> (sparsevec, sparsevec) FOR ORDER BY float_ops,
 	FUNCTION 1 l1_distance(sparsevec, sparsevec),
 	FUNCTION 3 hnsw_sparsevec_support(internal);
+
+-- hnsw attributes
+
+CREATE FUNCTION hnsw_attribute_distance(integer, integer) RETURNS float8
+	AS 'MODULE_PATHNAME', 'hnsw_int4_attribute_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE OPERATOR CLASS vector_integer_ops
+	DEFAULT FOR TYPE integer USING hnsw AS
+	OPERATOR 2 = (integer, integer),
+	FUNCTION 4 hnsw_attribute_distance(integer, integer);

src/halfvec.c

@@ -159,24 +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/hnsw.c

@@ -12,12 +12,22 @@
 #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)
 #endif
 
+static const struct config_enum_entry hnsw_iterative_search_options[] = {
+	{"off", HNSW_ITERATIVE_SEARCH_OFF, false},
+	{"on", HNSW_ITERATIVE_SEARCH_RELAXED, false},
+	{"strict", HNSW_ITERATIVE_SEARCH_STRICT, false},
+	{NULL, 0, false}
+};
+
 int			hnsw_ef_search;
+int			hnsw_iterative_search_max_tuples;
+int			hnsw_iterative_search;
 int			hnsw_lock_tranche_id;
 static relopt_kind hnsw_relopt_kind;
 
@@ -68,6 +78,15 @@ HnswInit(void)
 							"Valid range is 1..1000.", &hnsw_ef_search,
 							HNSW_DEFAULT_EF_SEARCH, HNSW_MIN_EF_SEARCH, HNSW_MAX_EF_SEARCH, PGC_USERSET, 0, NULL, NULL, NULL);
 
+	DefineCustomEnumVariable("hnsw.iterative_search", "Sets iterative search",
+							 NULL, &hnsw_iterative_search,
+							 HNSW_ITERATIVE_SEARCH_OFF, hnsw_iterative_search_options, PGC_USERSET, 0, NULL, NULL, NULL);
+
+	/* TODO Ensure ivfflat.max_probes uses same value for "all" */
+	DefineCustomIntVariable("hnsw.iterative_search_max_tuples", "Sets the max number of candidates to visit for iterative search",
+							"-1 means all", &hnsw_iterative_search_max_tuples,
+							-1, -1, INT_MAX, PGC_USERSET, 0, NULL, NULL, NULL);
+
 	MarkGUCPrefixReserved("hnsw");
 }
 
@@ -99,7 +118,9 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 {
 	GenericCosts costs;
 	int			m;
-	int			entryLevel;
+	double		ratio;
+	double		startupPages;
+	double		spc_seq_page_cost;
 	Relation	index;
 
 	/* Never use index without order */
@@ -115,21 +136,71 @@ hnswcostestimate(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);
 	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).
+	 */
+	if (path->indexinfo->tuples > 0)
+	{
+		double		scalingFactor = 0.55;
+		int			entryLevel = (int) (log(path->indexinfo->tuples) * HnswGetMl(m));
+		int			layer0TuplesMax = HnswGetLayerM(m, 0) * hnsw_ef_search;
+		double		layer0Selectivity = scalingFactor * log(path->indexinfo->tuples) / (log(m) * (1 + log(hnsw_ef_search)));
 
-	/* TODO Improve estimate of visited tuples (currently underestimates) */
+		ratio = (entryLevel * m + layer0TuplesMax * layer0Selectivity) / path->indexinfo->tuples;
-	/* Account for number of tuples (or entry level), m, and ef_search */
-	costs.numIndexTuples = (entryLevel + 2) * m;
 
-	genericcostestimate(root, path, loop_count, &costs);
+		if (ratio > 1)
+			ratio = 1;
+	}
+	else
+		ratio = 1;
 
-	/* Use total cost since most work happens before first tuple is returned */
+	get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
-	*indexStartupCost = costs.indexTotalCost;
+
+	/* Startup cost is cost before returning the first row */
+	costs.indexStartupCost = costs.indexTotalCost * ratio;
+
+	/* Adjust cost if needed since TOAST not included in seq scan cost */
+	startupPages = costs.numIndexPages * ratio;
+	if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
+	{
+		/* Change all page cost from random to sequential */
+		costs.indexStartupCost -= startupPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+
+		/* Remove cost of extra pages */
+		costs.indexStartupCost -= (startupPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
+	}
+
+	*indexStartupCost = costs.indexStartupCost;
 	*indexTotalCost = costs.indexTotalCost;
 	*indexSelectivity = costs.indexSelectivity;
 	*indexCorrelation = costs.indexCorrelation;
@@ -174,13 +245,13 @@ hnswhandler(PG_FUNCTION_ARGS)
 	IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
 
 	amroutine->amstrategies = 0;
-	amroutine->amsupport = 3;
+	amroutine->amsupport = 4;
 	amroutine->amoptsprocnum = 0;
 	amroutine->amcanorder = false;
 	amroutine->amcanorderbyop = true;
 	amroutine->amcanbackward = false;	/* can change direction mid-scan */
 	amroutine->amcanunique = false;
-	amroutine->amcanmulticol = false;
+	amroutine->amcanmulticol = true;
 	amroutine->amoptionalkey = true;
 	amroutine->amsearcharray = false;
 	amroutine->amsearchnulls = false;
@@ -232,3 +303,17 @@ hnswhandler(PG_FUNCTION_ARGS)
 
 	PG_RETURN_POINTER(amroutine);
 }
+
+/*
+ * Get the distance between two int4 attributes
+ */
+PGDLLEXPORT PG_FUNCTION_INFO_V1(hnsw_int4_attribute_distance);
+Datum
+hnsw_int4_attribute_distance(PG_FUNCTION_ARGS)
+{
+	int32		a = PG_GETARG_INT32(0);
+	int32		b = PG_GETARG_INT32(1);
+	double		distance = ((double) a) - ((double) b);
+
+	PG_RETURN_FLOAT8(distance);
+}

src/hnsw.h

@@ -19,6 +19,7 @@
 #define HNSW_DISTANCE_PROC 1
 #define HNSW_NORM_PROC 2
 #define HNSW_TYPE_INFO_PROC 3
+#define HNSW_ATTRIBUTE_DISTANCE_PROC 4
 
 #define HNSW_VERSION	1
 #define HNSW_MAGIC_NUMBER 0xA953A953
@@ -88,6 +89,9 @@
 /* Ensure fits on page and in uint8 */
 #define HnswGetMaxLevel(m) Min(((BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - offsetof(HnswNeighborTupleData, indextids) - sizeof(ItemIdData)) / (sizeof(ItemPointerData)) / (m)) - 2, 255)
 
+#define HnswGetSearchCandidate(membername, ptr) pairingheap_container(HnswSearchCandidate, membername, ptr)
+#define HnswGetSearchCandidateConst(membername, ptr) pairingheap_const_container(HnswSearchCandidate, membername, ptr)
+
 #define HnswGetValue(base, element) PointerGetDatum(HnswPtrAccess(base, (element)->value))
 
 #if PG_VERSION_NUM < 140005
@@ -104,10 +108,21 @@
 #define HnswPtrPointer(hp) (hp).ptr
 #define HnswPtrOffset(hp) relptr_offset((hp).relptr)
 
+#define HnswUseIndexTuple(index) (IndexRelationGetNumberOfAttributes(index) > 1)
+
 /* Variables */
 extern int	hnsw_ef_search;
+extern int	hnsw_iterative_search;
+extern int	hnsw_iterative_search_max_tuples;
 extern int	hnsw_lock_tranche_id;
 
+typedef enum HnswIterativeSearchType
+{
+	HNSW_ITERATIVE_SEARCH_OFF,
+	HNSW_ITERATIVE_SEARCH_RELAXED,
+	HNSW_ITERATIVE_SEARCH_STRICT
+}			HnswIterativeSearchType;
+
 typedef struct HnswElementData HnswElementData;
 typedef struct HnswNeighborArray HnswNeighborArray;
 
@@ -121,6 +136,7 @@ HnswPtrDeclare(HnswElementData, HnswElementRelptr, HnswElementPtr);
 HnswPtrDeclare(HnswNeighborArray, HnswNeighborArrayRelptr, HnswNeighborArrayPtr);
 HnswPtrDeclare(HnswNeighborArrayPtr, HnswNeighborsRelptr, HnswNeighborsPtr);
 HnswPtrDeclare(char, DatumRelptr, DatumPtr);
+HnswPtrDeclare(IndexTupleData, IndexTupleRelptr, IndexTuplePtr);
 
 struct HnswElementData
 {
@@ -129,6 +145,7 @@ struct HnswElementData
 	uint8		heaptidsLength;
 	uint8		level;
 	uint8		deleted;
+	uint8		version;
 	uint32		hash;
 	HnswNeighborsPtr neighbors;
 	BlockNumber blkno;
@@ -136,6 +153,7 @@ struct HnswElementData
 	OffsetNumber neighborOffno;
 	BlockNumber neighborPage;
 	DatumPtr	value;
+	IndexTuplePtr itup;
 	LWLock		lock;
 };
 
@@ -160,7 +178,8 @@ typedef struct HnswSearchCandidate
 	pairingheap_node c_node;
 	pairingheap_node w_node;
 	HnswElementPtr element;
-	float		distance;
+	double		distance;
+	bool		matches;
 }			HnswSearchCandidate;
 
 /* HNSW index options */
@@ -185,8 +204,8 @@ typedef struct HnswGraph
 
 	/* Allocations state */
 	LWLock		allocatorLock;
-	long		memoryUsed;
+	Size		memoryUsed;
-	long		memoryTotal;
+	Size		memoryTotal;
 
 	/* Flushed state */
 	LWLock		flushLock;
@@ -237,6 +256,20 @@ typedef struct HnswTypeInfo
 	void		(*checkValue) (Pointer v);
 }			HnswTypeInfo;
 
+typedef struct HnswSupport
+{
+	FmgrInfo   *procinfo[2];
+	FmgrInfo   *normprocinfo;
+	Oid		   *collation;
+}			HnswSupport;
+
+typedef struct HnswQuery
+{
+	Datum		value;
+	IndexTuple	itup;
+	ScanKeyData *keyData;
+}			HnswQuery;
+
 typedef struct HnswBuildState
 {
 	/* Info */
@@ -256,15 +289,15 @@ typedef struct HnswBuildState
 	double		reltuples;
 
 	/* Support functions */
-	FmgrInfo   *procinfo;
+	HnswSupport support;
-	FmgrInfo   *normprocinfo;
-	Oid			collation;
 
 	/* Variables */
 	HnswGraph	graphData;
 	HnswGraph  *graph;
 	double		ml;
 	int			maxLevel;
+	bool		useIndexTuple;
+	TupleDesc	tupdesc;
 
 	/* Memory */
 	MemoryContext graphCtx;
@@ -306,10 +339,10 @@ typedef struct HnswElementTupleData
 	uint8		type;
 	uint8		level;
 	uint8		deleted;
-	uint8		unused;
+	uint8		version;
 	ItemPointerData heaptids[HNSW_HEAPTIDS];
 	ItemPointerData neighbortid;
-	uint16		unused2;
+	uint16		unused;
 	Vector		data;
 }			HnswElementTupleData;
 
@@ -318,24 +351,41 @@ typedef HnswElementTupleData * HnswElementTuple;
 typedef struct HnswNeighborTupleData
 {
 	uint8		type;
-	uint8		unused;
+	uint8		version;
 	uint16		count;
 	ItemPointerData indextids[FLEXIBLE_ARRAY_MEMBER];
 }			HnswNeighborTupleData;
 
 typedef HnswNeighborTupleData * HnswNeighborTuple;
 
+typedef union
+{
+	struct pointerhash_hash *pointers;
+	struct offsethash_hash *offsets;
+	struct tidhash_hash *tids;
+}			visited_hash;
+
+typedef union
+{
+	HnswElement element;
+	ItemPointerData indextid;
+}			HnswUnvisited;
+
 typedef struct HnswScanOpaqueData
 {
 	const		HnswTypeInfo *typeInfo;
 	bool		first;
 	List	   *w;
+	visited_hash v;
+	pairingheap *discarded;
+	HnswQuery	q;
+	int			m;
+	int64		tuples;
+	double		previousDistance;
 	MemoryContext tmpCtx;
 
 	/* Support functions */
-	FmgrInfo   *procinfo;
+	HnswSupport support;
-	FmgrInfo   *normprocinfo;
-	Oid			collation;
 }			HnswScanOpaqueData;
 
 typedef HnswScanOpaqueData * HnswScanOpaque;
@@ -353,8 +403,7 @@ typedef struct HnswVacuumState
 	int			efConstruction;
 
 	/* Support functions */
-	FmgrInfo   *procinfo;
+	HnswSupport support;
-	Oid			collation;
 
 	/* Variables */
 	struct tidhash_hash *deleted;
@@ -370,33 +419,38 @@ typedef struct HnswVacuumState
 int			HnswGetM(Relation index);
 int			HnswGetEfConstruction(Relation index);
 FmgrInfo   *HnswOptionalProcInfo(Relation index, uint16 procnum);
+void		HnswInitSupport(HnswSupport * support, Relation index);
 Datum		HnswNormValue(const HnswTypeInfo * typeInfo, Oid collation, Datum value);
-bool		HnswCheckNorm(FmgrInfo *procinfo, Oid collation, Datum value);
+bool		HnswCheckNorm(HnswSupport * support, Datum value);
 Buffer		HnswNewBuffer(Relation index, ForkNumber forkNum);
 void		HnswInitPage(Buffer buf, Page page);
 void		HnswInit(void);
-List	   *HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement);
+List	   *HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, bool inMemory, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples);
 HnswElement HnswGetEntryPoint(Relation index);
 void		HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
 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);
+void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing, bool inMemory);
-HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
+HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, HnswQuery * q, Relation rel, HnswSupport * support, bool loadVec, bool inMemory);
 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);
+HnswNeighborArray *HnswInitNeighborArray(int lm, HnswAllocator * allocator);
 void		HnswInitNeighbors(char *base, HnswElement element, int m, HnswAllocator * alloc);
-bool		HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull, ItemPointer heap_tid, bool building);
+bool		HnswInsertTupleOnDisk(Relation index, HnswSupport * support, IndexTuple itup, ItemPointer heaptid, bool building, TupleDesc tupdesc);
-void		HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building);
+void		HnswUpdateNeighborsOnDisk(Relation index, HnswSupport * support, HnswElement e, int m, bool checkExisting, bool building);
-void		HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec);
+void		HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec, Relation index);
-void		HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec, float *maxDistance);
+void		HnswLoadElement(HnswElement element, double *distance, bool *matches, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance);
-void		HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element);
+void		HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element, bool useIndexTuple);
-void		HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
+void		HnswUpdateConnection(char *base, HnswNeighborArray * neighbors, HnswElement newElement, float distance, int lm, int *updateIdx, Relation index, HnswSupport * support);
-void		HnswLoadNeighbors(HnswElement element, Relation index, int m);
+bool		HnswFormIndexTuple(IndexTuple *out, Datum *values, bool *isnull, const HnswTypeInfo * typeInfo, HnswSupport * support, TupleDesc tupdesc);
+bool		HnswLoadNeighborTids(HnswElement element, ItemPointerData *indextids, Relation index, int m, int lm, int lc);
 void		HnswInitLockTranche(void);
 const		HnswTypeInfo *HnswGetTypeInfo(Relation index);
 PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);
+Size		HnswGetElementTupleSize(char *base, HnswElement element, bool useIndexTuple);
+bool		HnswIndexTupleIsEqual(IndexTuple a, IndexTuple b, TupleDesc tupdesc);
 
 /* Index access methods */
 IndexBuildResult *hnswbuild(Relation heap, Relation index, IndexInfo *indexInfo);

src/hnswbuild.c

@@ -148,6 +148,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 	Page		page;
 	HnswElementPtr iter = buildstate->graph->head;
 	char	   *base = buildstate->hnswarea;
+	bool		useIndexTuple = buildstate->useIndexTuple;
 
 	/* Calculate sizes */
 	maxSize = HNSW_MAX_SIZE;
@@ -167,7 +168,6 @@ CreateGraphPages(HnswBuildState * buildstate)
 		Size		etupSize;
 		Size		ntupSize;
 		Size		combinedSize;
-		Pointer		valuePtr = HnswPtrAccess(base, element->value);
 
 		/* Update iterator */
 		iter = element->next;
@@ -176,7 +176,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 		MemSet(etup, 0, HNSW_TUPLE_ALLOC_SIZE);
 
 		/* Calculate sizes */
-		etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(valuePtr));
+		etupSize = HnswGetElementTupleSize(base, element, useIndexTuple);
 		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, buildstate->m);
 		combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
 
@@ -186,7 +186,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
 					 errmsg("index tuple too large")));
 
-		HnswSetElementTuple(base, etup, element);
+		HnswSetElementTuple(base, etup, element, useIndexTuple);
 
 		/* Keep element and neighbors on the same page if possible */
 		if (PageGetFreeSpace(page) < etupSize || (combinedSize <= maxSize && PageGetFreeSpace(page) < combinedSize))
@@ -327,19 +327,18 @@ AddDuplicateInMemory(HnswElement element, HnswElement dup)
  * Find duplicate element
  */
 static bool
-FindDuplicateInMemory(char *base, HnswElement element)
+FindDuplicateInMemory(char *base, HnswElement element, bool useIndexTuple, TupleDesc tupdesc)
 {
 	HnswNeighborArray *neighbors = HnswGetNeighbors(base, element, 0);
-	Datum		value = HnswGetValue(base, element);
+	IndexTuple	itup = HnswPtrAccess(base, element->itup);
 
 	for (int i = 0; i < neighbors->length; i++)
 	{
 		HnswCandidate *neighbor = &neighbors->items[i];
 		HnswElement neighborElement = HnswPtrAccess(base, neighbor->element);
-		Datum		neighborValue = HnswGetValue(base, neighborElement);
 
 		/* Exit early since ordered by distance */
-		if (!datumIsEqual(value, neighborValue, false, -1))
+		if (!HnswIndexTupleIsEqual(itup, HnswPtrAccess(base, neighborElement->itup), tupdesc))
 			return false;
 
 		/* Check for space */
@@ -366,7 +365,7 @@ AddElementInMemory(char *base, HnswGraph * graph, HnswElement element)
  * Update neighbors
  */
 static void
-UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswElement e, int m)
+UpdateNeighborsInMemory(char *base, Relation index, HnswSupport * support, HnswElement e, int m)
 {
 	for (int lc = e->level; lc >= 0; lc--)
 	{
@@ -388,7 +387,7 @@ UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswEleme
 			Assert(neighborElement);
 
 			LWLockAcquire(&neighborElement->lock, LW_EXCLUSIVE);
-			HnswUpdateConnection(base, e, hc, lm, lc, NULL, NULL, procinfo, collation);
+			HnswUpdateConnection(base, HnswGetNeighbors(base, neighborElement, lc), e, hc->distance, lm, NULL, index, support);
 			LWLockRelease(&neighborElement->lock);
 		}
 	}
@@ -398,20 +397,20 @@ UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswEleme
  * Update graph in memory
  */
 static void
-UpdateGraphInMemory(FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, HnswBuildState * buildstate)
+UpdateGraphInMemory(HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, HnswBuildState * buildstate)
 {
 	HnswGraph  *graph = buildstate->graph;
 	char	   *base = buildstate->hnswarea;
 
 	/* Look for duplicate */
-	if (FindDuplicateInMemory(base, element))
+	if (FindDuplicateInMemory(base, element, buildstate->useIndexTuple, buildstate->tupdesc))
 		return;
 
 	/* Add element */
 	AddElementInMemory(base, graph, element);
 
 	/* Update neighbors */
-	UpdateNeighborsInMemory(base, procinfo, collation, element, m);
+	UpdateNeighborsInMemory(base, buildstate->index, support, element, m);
 
 	/* Update entry point if needed (already have lock) */
 	if (entryPoint == NULL || element->level > entryPoint->level)
@@ -424,9 +423,9 @@ UpdateGraphInMemory(FmgrInfo *procinfo, Oid collation, HnswElement element, int
 static void
 InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 {
-	FmgrInfo   *procinfo = buildstate->procinfo;
+	Relation	index = buildstate->index;
-	Oid			collation = buildstate->collation;
 	HnswGraph  *graph = buildstate->graph;
+	HnswSupport *support = &buildstate->support;
 	HnswElement entryPoint;
 	LWLock	   *entryLock = &graph->entryLock;
 	LWLock	   *entryWaitLock = &graph->entryWaitLock;
@@ -458,10 +457,10 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, NULL, procinfo, collation, m, efConstruction, false);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false, true);
 
 	/* Update graph in memory */
-	UpdateGraphInMemory(procinfo, collation, element, m, efConstruction, entryPoint, buildstate);
+	UpdateGraphInMemory(support, element, m, efConstruction, entryPoint, buildstate);
 
 	/* Release entry lock */
 	LWLockRelease(entryLock);
@@ -473,33 +472,24 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 static bool
 InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, HnswBuildState * buildstate)
 {
-	const		HnswTypeInfo *typeInfo = buildstate->typeInfo;
 	HnswGraph  *graph = buildstate->graph;
 	HnswElement element;
 	HnswAllocator *allocator = &buildstate->allocator;
-	Size		valueSize;
+	HnswSupport *support = &buildstate->support;
-	Pointer		valuePtr;
 	LWLock	   *flushLock = &graph->flushLock;
 	char	   *base = buildstate->hnswarea;
+	TupleDesc	tupdesc = buildstate->tupdesc;
+	IndexTuple	itup;
+	Size		itupSize;
+	IndexTuple	itupShared;
+	bool		unused;
 
-	/* Detoast once for all calls */
+	/* Form index tuple */
-	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
+	if (!HnswFormIndexTuple(&itup, values, isnull, buildstate->typeInfo, support, tupdesc))
+		return false;
 
-	/* Check value */
+	/* Get tuple size */
-	if (typeInfo->checkValue != NULL)
+	itupSize = IndexTupleSize(itup);
-		typeInfo->checkValue(DatumGetPointer(value));
-
-	/* Normalize if needed */
-	if (buildstate->normprocinfo != NULL)
-	{
-		if (!HnswCheckNorm(buildstate->normprocinfo, buildstate->collation, value))
-			return false;
-
-		value = HnswNormValue(typeInfo, buildstate->collation, value);
-	}
-
-	/* Get datum size */
-	valueSize = VARSIZE_ANY(DatumGetPointer(value));
 
 	/* Ensure graph not flushed when inserting */
 	LWLockAcquire(flushLock, LW_SHARED);
@@ -509,7 +499,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	{
 		LWLockRelease(flushLock);
 
-		return HnswInsertTupleOnDisk(index, value, values, isnull, heaptid, true);
+		return HnswInsertTupleOnDisk(index, support, itup, heaptid, true, tupdesc);
 	}
 
 	/*
@@ -541,12 +531,12 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 
 		LWLockRelease(flushLock);
 
-		return HnswInsertTupleOnDisk(index, value, values, isnull, heaptid, true);
+		return HnswInsertTupleOnDisk(index, support, itup, heaptid, true, tupdesc);
 	}
 
 	/* Ok, we can proceed to allocate the element */
 	element = HnswInitElement(base, heaptid, buildstate->m, buildstate->ml, buildstate->maxLevel, allocator);
-	valuePtr = HnswAlloc(allocator, valueSize);
+	itupShared = HnswAlloc(allocator, itupSize);
 
 	/*
 	 * We have now allocated the space needed for the element, so we don't
@@ -555,9 +545,10 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	 */
 	LWLockRelease(&graph->allocatorLock);
 
-	/* Copy the datum */
+	/* Copy the tuple */
-	memcpy(valuePtr, DatumGetPointer(value), valueSize);
+	memcpy(itupShared, itup, itupSize);
-	HnswPtrStore(base, element->value, valuePtr);
+	HnswPtrStore(base, element->itup, itupShared);
+	HnswPtrStore(base, element->value, DatumGetPointer(index_getattr(itupShared, 1, tupdesc, &unused)));
 
 	/* Create a lock for the element */
 	LWLockInitialize(&element->lock, hnsw_lock_tranche_id);
@@ -607,7 +598,7 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
  * Initialize the graph
  */
 static void
-InitGraph(HnswGraph * graph, char *base, long memoryTotal)
+InitGraph(HnswGraph * graph, char *base, Size memoryTotal)
 {
 	/* Initialize the lock tranche if needed */
 	HnswInitLockTranche();
@@ -684,6 +675,19 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
 				 errmsg("type not supported for hnsw index")));
 
+	/* TODO See if needed */
+	if (IndexRelationGetNumberOfKeyAttributes(index) > 2)
+		elog(ERROR, "index cannot have more than two columns");
+
+	if (!OidIsValid(index_getprocid(index, 1, HNSW_DISTANCE_PROC)))
+		elog(ERROR, "first column must be a vector");
+
+	for (int i = 1; i < IndexRelationGetNumberOfKeyAttributes(index); i++)
+	{
+		if (!OidIsValid(index_getprocid(index, i + 1, HNSW_ATTRIBUTE_DISTANCE_PROC)))
+			elog(ERROR, "column %d cannot be a vector", i + 1);
+	}
+
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
 		ereport(ERROR,
@@ -704,14 +708,14 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 	buildstate->indtuples = 0;
 
 	/* Get support functions */
-	buildstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
+	HnswInitSupport(&buildstate->support, index);
-	buildstate->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
-	buildstate->collation = index->rd_indcollation[0];
 
-	InitGraph(&buildstate->graphData, NULL, maintenance_work_mem * 1024L);
+	InitGraph(&buildstate->graphData, NULL, (Size) maintenance_work_mem * 1024L);
 	buildstate->graph = &buildstate->graphData;
 	buildstate->ml = HnswGetMl(buildstate->m);
 	buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
+	buildstate->useIndexTuple = HnswUseIndexTuple(index);
+	buildstate->tupdesc = RelationGetDescr(index);
 
 	buildstate->graphCtx = GenerationContextCreate(CurrentMemoryContext,
 												   "Hnsw build graph context",

src/hnswinsert.c

@@ -36,7 +36,7 @@ GetInsertPage(Relation index)
  * Check for a free offset
  */
 static bool
-HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size etupSize, Size ntupSize, Buffer *nbuf, Page *npage, OffsetNumber *freeOffno, OffsetNumber *freeNeighborOffno, BlockNumber *newInsertPage)
+HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size etupSize, Size ntupSize, Buffer *nbuf, Page *npage, OffsetNumber *freeOffno, OffsetNumber *freeNeighborOffno, BlockNumber *newInsertPage, uint8 *tupleVersion)
 {
 	OffsetNumber offno;
 	OffsetNumber maxoffno = PageGetMaxOffsetNumber(page);
@@ -98,6 +98,7 @@ HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size
 			{
 				*freeOffno = offno;
 				*freeNeighborOffno = neighborOffno;
+				*tupleVersion = etup->version;
 				return true;
 			}
 			else if (*nbuf != buf)
@@ -153,10 +154,12 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 	OffsetNumber freeOffno = InvalidOffsetNumber;
 	OffsetNumber freeNeighborOffno = InvalidOffsetNumber;
 	BlockNumber newInsertPage = InvalidBlockNumber;
+	uint8		tupleVersion;
 	char	   *base = NULL;
+	bool		useIndexTuple = HnswUseIndexTuple(index);
 
 	/* Calculate sizes */
-	etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(HnswPtrAccess(base, e->value)));
+	etupSize = HnswGetElementTupleSize(base, e, useIndexTuple);
 	ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
 	combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
 	maxSize = HNSW_MAX_SIZE;
@@ -164,7 +167,7 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 
 	/* Prepare element tuple */
 	etup = palloc0(etupSize);
-	HnswSetElementTuple(base, etup, e);
+	HnswSetElementTuple(base, etup, e, useIndexTuple);
 
 	/* Prepare neighbor tuple */
 	ntup = palloc0(ntupSize);
@@ -202,7 +205,7 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 		}
 
 		/* Next, try space from a deleted element */
-		if (HnswFreeOffset(index, buf, page, e, etupSize, ntupSize, &nbuf, &npage, &freeOffno, &freeNeighborOffno, &newInsertPage))
+		if (HnswFreeOffset(index, buf, page, e, etupSize, ntupSize, &nbuf, &npage, &freeOffno, &freeNeighborOffno, &newInsertPage, &tupleVersion))
 		{
 			if (nbuf != buf)
 			{
@@ -212,6 +215,10 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 					npage = GenericXLogRegisterBuffer(state, nbuf, 0);
 			}
 
+			/* Set tuple version */
+			etup->version = tupleVersion;
+			ntup->version = tupleVersion;
+
 			break;
 		}
 
@@ -334,6 +341,110 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 		*updatedInsertPage = newInsertPage;
 }
 
+/*
+ * Load neighbors
+ */
+static HnswNeighborArray *
+HnswLoadNeighbors(HnswElement element, Relation index, int m, int lm, int lc)
+{
+	char	   *base = NULL;
+	HnswNeighborArray *neighbors = HnswInitNeighborArray(lm, NULL);
+	ItemPointerData indextids[HNSW_MAX_M * 2];
+
+	if (!HnswLoadNeighborTids(element, indextids, index, m, lm, lc))
+		return neighbors;
+
+	for (int i = 0; i < lm; i++)
+	{
+		ItemPointer indextid = &indextids[i];
+		HnswElement e;
+		HnswCandidate *hc;
+
+		if (!ItemPointerIsValid(indextid))
+			break;
+
+		e = HnswInitElementFromBlock(ItemPointerGetBlockNumber(indextid), ItemPointerGetOffsetNumber(indextid));
+		hc = &neighbors->items[neighbors->length++];
+		HnswPtrStore(base, hc->element, e);
+	}
+
+	return neighbors;
+}
+
+/*
+ * Load elements for insert
+ */
+static void
+LoadElementsForInsert(HnswNeighborArray * neighbors, HnswQuery * q, int *idx, Relation index, HnswSupport * support)
+{
+	char	   *base = NULL;
+
+	for (int i = 0; i < neighbors->length; i++)
+	{
+		HnswCandidate *hc = &neighbors->items[i];
+		HnswElement element = HnswPtrAccess(base, hc->element);
+		double		distance;
+		bool		matches;
+
+		HnswLoadElement(element, &distance, &matches, q, index, support, true, NULL);
+		hc->distance = distance;
+
+		/* Prune element if being deleted */
+		if (element->heaptidsLength == 0)
+		{
+			*idx = i;
+			break;
+		}
+	}
+}
+
+/*
+ * Get update index
+ */
+static int
+GetUpdateIndex(HnswElement element, HnswElement newElement, float distance, int m, int lm, int lc, Relation index, HnswSupport * support, MemoryContext updateCtx)
+{
+	char	   *base = NULL;
+	int			idx = -1;
+	HnswNeighborArray *neighbors;
+	MemoryContext oldCtx = MemoryContextSwitchTo(updateCtx);
+
+	/*
+	 * 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.
+	 */
+	neighbors = HnswLoadNeighbors(element, index, m, lm, lc);
+
+	/*
+	 * Could improve performance for vacuuming by checking neighbors against
+	 * list of elements being deleted to find index. It's important to exclude
+	 * already deleted elements for this since they can be replaced at any
+	 * time.
+	 */
+
+	if (neighbors->length < lm)
+		idx = -2;
+	else
+	{
+		HnswQuery	q;
+
+		q.value = HnswGetValue(base, element);
+		q.itup = HnswPtrAccess(base, element->itup);
+		q.keyData = NULL;
+
+		LoadElementsForInsert(neighbors, &q, &idx, index, support);
+
+		if (idx == -1)
+			HnswUpdateConnection(base, neighbors, newElement, distance, lm, &idx, index, support);
+	}
+
+	MemoryContextSwitchTo(oldCtx);
+	MemoryContextReset(updateCtx);
+
+	return idx;
+}
+
 /*
  * Check if connection already exists
  */
@@ -354,14 +465,94 @@ ConnectionExists(HnswElement e, HnswNeighborTuple ntup, int startIdx, int lm)
 	return false;
 }
 
+/*
+ * Update neighbor
+ */
+static void
+UpdateNeighborOnDisk(HnswElement element, HnswElement newElement, int idx, int m, int lm, int lc, Relation index, bool checkExisting, bool building)
+{
+	Buffer		buf;
+	Page		page;
+	GenericXLogState *state;
+	HnswNeighborTuple ntup;
+	int			startIdx;
+	OffsetNumber offno = element->neighborOffno;
+
+	/* Register page */
+	buf = ReadBuffer(index, element->neighborPage);
+	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
+	if (building)
+	{
+		state = NULL;
+		page = BufferGetPage(buf);
+	}
+	else
+	{
+		state = GenericXLogStart(index);
+		page = GenericXLogRegisterBuffer(state, buf, 0);
+	}
+
+	/* Get tuple */
+	ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, offno));
+
+	/* Calculate index for update */
+	startIdx = (element->level - lc) * m;
+
+	/* Check for existing connection */
+	if (checkExisting && ConnectionExists(newElement, ntup, startIdx, lm))
+		idx = -1;
+	else if (idx == -2)
+	{
+		/* Find free offset if still exists */
+		/* TODO Retry updating connections if not */
+		for (int j = 0; j < lm; j++)
+		{
+			if (!ItemPointerIsValid(&ntup->indextids[startIdx + j]))
+			{
+				idx = startIdx + j;
+				break;
+			}
+		}
+	}
+	else
+		idx += startIdx;
+
+	/* Make robust to issues */
+	if (idx >= 0 && idx < ntup->count)
+	{
+		ItemPointer indextid = &ntup->indextids[idx];
+
+		/* Update neighbor on the buffer */
+		ItemPointerSet(indextid, newElement->blkno, newElement->offno);
+
+		/* Commit */
+		if (building)
+			MarkBufferDirty(buf);
+		else
+			GenericXLogFinish(state);
+	}
+	else if (!building)
+		GenericXLogAbort(state);
+
+	UnlockReleaseBuffer(buf);
+}
+
 /*
  * Update neighbors
  */
 void
-HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building)
+HnswUpdateNeighborsOnDisk(Relation index, HnswSupport * support, HnswElement e, int m, bool checkExisting, bool building)
 {
 	char	   *base = NULL;
 
+	/* Use separate memory context to improve performance for larger vectors */
+	MemoryContext updateCtx = GenerationContextCreate(CurrentMemoryContext,
+													  "Hnsw insert update context",
+#if PG_VERSION_NUM >= 150000
+													  128 * 1024, 128 * 1024,
+#endif
+													  128 * 1024);
+
 	for (int lc = e->level; lc >= 0; lc--)
 	{
 		int			lm = HnswGetLayerM(m, lc);
@@ -370,96 +561,20 @@ HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, Hns
 		for (int i = 0; i < neighbors->length; i++)
 		{
 			HnswCandidate *hc = &neighbors->items[i];
-			Buffer		buf;
-			Page		page;
-			GenericXLogState *state;
-			HnswNeighborTuple ntup;
-			int			idx = -1;
-			int			startIdx;
 			HnswElement neighborElement = HnswPtrAccess(base, hc->element);
-			OffsetNumber offno = neighborElement->neighborOffno;
+			int			idx;
 
-			/*
+			idx = GetUpdateIndex(neighborElement, e, hc->distance, m, lm, lc, index, support, updateCtx);
-			 * 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);
-
-			/*
-			 * Could improve performance for vacuuming by checking neighbors
-			 * against list of elements being deleted to find index. It's
-			 * important to exclude already deleted elements for this since
-			 * they can be replaced at any time.
-			 */
-
-			/* Select neighbors */
-			HnswUpdateConnection(NULL, e, hc, lm, lc, &idx, index, procinfo, collation);
 
 			/* New element was not selected as a neighbor */
 			if (idx == -1)
 				continue;
 
-			/* Register page */
+			UpdateNeighborOnDisk(neighborElement, e, idx, m, lm, lc, index, checkExisting, building);
-			buf = ReadBuffer(index, neighborElement->neighborPage);
-			LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-			if (building)
-			{
-				state = NULL;
-				page = BufferGetPage(buf);
-			}
-			else
-			{
-				state = GenericXLogStart(index);
-				page = GenericXLogRegisterBuffer(state, buf, 0);
-			}
-
-			/* Get tuple */
-			ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, offno));
-
-			/* Calculate index for update */
-			startIdx = (neighborElement->level - lc) * m;
-
-			/* Check for existing connection */
-			if (checkExisting && ConnectionExists(e, ntup, startIdx, lm))
-				idx = -1;
-			else if (idx == -2)
-			{
-				/* Find free offset if still exists */
-				/* TODO Retry updating connections if not */
-				for (int j = 0; j < lm; j++)
-				{
-					if (!ItemPointerIsValid(&ntup->indextids[startIdx + j]))
-					{
-						idx = startIdx + j;
-						break;
-					}
-				}
-			}
-			else
-				idx += startIdx;
-
-			/* Make robust to issues */
-			if (idx >= 0 && idx < ntup->count)
-			{
-				ItemPointer indextid = &ntup->indextids[idx];
-
-				/* Update neighbor on the buffer */
-				ItemPointerSet(indextid, e->blkno, e->offno);
-
-				/* Commit */
-				if (building)
-					MarkBufferDirty(buf);
-				else
-					GenericXLogFinish(state);
-			}
-			else if (!building)
-				GenericXLogAbort(state);
-
-			UnlockReleaseBuffer(buf);
 		}
 	}
+
+	MemoryContextDelete(updateCtx);
 }
 
 /*
@@ -522,21 +637,30 @@ AddDuplicateOnDisk(Relation index, HnswElement element, HnswElement dup, bool bu
  * Find duplicate element
  */
 static bool
-FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
+FindDuplicateOnDisk(Relation index, HnswElement element, bool building, TupleDesc tupdesc)
 {
 	char	   *base = NULL;
 	HnswNeighborArray *neighbors = HnswGetNeighbors(base, element, 0);
 	Datum		value = HnswGetValue(base, element);
+	IndexTuple	itup = HnswPtrAccess(base, element->itup);
 
 	for (int i = 0; i < neighbors->length; i++)
 	{
 		HnswCandidate *neighbor = &neighbors->items[i];
 		HnswElement neighborElement = HnswPtrAccess(base, neighbor->element);
-		Datum		neighborValue = HnswGetValue(base, neighborElement);
 
-		/* Exit early since ordered by distance */
+		if (HnswUseIndexTuple(index))
-		if (!datumIsEqual(value, neighborValue, false, -1))
+		{
-			return false;
+			/* Exit early since ordered by distance */
+			if (!HnswIndexTupleIsEqual(itup, HnswPtrAccess(base, neighborElement->itup), tupdesc))
+				return false;
+		}
+		else
+		{
+			/* Exit early since ordered by distance */
+			if (!datumIsEqual(value, HnswGetValue(base, neighborElement), false, -1))
+				return false;
+		}
 
 		if (AddDuplicateOnDisk(index, element, neighborElement, building))
 			return true;
@@ -549,12 +673,12 @@ FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
  * Update graph on disk
  */
 static void
-UpdateGraphOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
+UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building, TupleDesc tupdesc)
 {
 	BlockNumber newInsertPage = InvalidBlockNumber;
 
 	/* Look for duplicate */
-	if (FindDuplicateOnDisk(index, element, building))
+	if (FindDuplicateOnDisk(index, element, building, tupdesc))
 		return;
 
 	/* Add element */
@@ -565,7 +689,7 @@ UpdateGraphOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement
 		HnswUpdateMetaPage(index, 0, NULL, newInsertPage, MAIN_FORKNUM, building);
 
 	/* Update neighbors */
-	HnswUpdateNeighborsOnDisk(index, procinfo, collation, element, m, false, building);
+	HnswUpdateNeighborsOnDisk(index, support, element, m, false, building);
 
 	/* Update entry point if needed */
 	if (entryPoint == NULL || element->level > entryPoint->level)
@@ -576,16 +700,15 @@ UpdateGraphOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement
  * Insert a tuple into the index
  */
 bool
-HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull, ItemPointer heap_tid, bool building)
+HnswInsertTupleOnDisk(Relation index, HnswSupport * support, IndexTuple itup, ItemPointer heaptid, bool building, TupleDesc tupdesc)
 {
 	HnswElement entryPoint;
 	HnswElement element;
 	int			m;
 	int			efConstruction = HnswGetEfConstruction(index);
-	FmgrInfo   *procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
-	Oid			collation = index->rd_indcollation[0];
 	LOCKMODE	lockmode = ShareLock;
 	char	   *base = NULL;
+	bool		unused;
 
 	/*
 	 * Get a shared lock. This allows vacuum to ensure no in-flight inserts
@@ -598,8 +721,9 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
 	HnswGetMetaPageInfo(index, &m, &entryPoint);
 
 	/* Create an element */
-	element = HnswInitElement(base, heap_tid, m, HnswGetMl(m), HnswGetMaxLevel(m), NULL);
+	element = HnswInitElement(base, heaptid, m, HnswGetMl(m), HnswGetMaxLevel(m), NULL);
-	HnswPtrStore(base, element->value, DatumGetPointer(value));
+	HnswPtrStore(base, element->itup, itup);
+	HnswPtrStore(base, element->value, DatumGetPointer(index_getattr(itup, 1, tupdesc, &unused)));
 
 	/* Prevent concurrent inserts when likely updating entry point */
 	if (entryPoint == NULL || element->level > entryPoint->level)
@@ -616,10 +740,10 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, false);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false, false);
 
 	/* Update graph on disk */
-	UpdateGraphOnDisk(index, procinfo, collation, element, m, efConstruction, entryPoint, building);
+	UpdateGraphOnDisk(index, support, element, m, efConstruction, entryPoint, building, tupdesc);
 
 	/* Release lock */
 	UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);
@@ -631,31 +755,20 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
  * Insert a tuple into the index
  */
 static void
-HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid)
+HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid)
 {
-	Datum		value;
+	IndexTuple	itup;
 	const		HnswTypeInfo *typeInfo = HnswGetTypeInfo(index);
-	FmgrInfo   *normprocinfo;
+	TupleDesc	tupdesc = RelationGetDescr(index);
-	Oid			collation = index->rd_indcollation[0];
+	HnswSupport support;
 
-	/* Detoast once for all calls */
+	HnswInitSupport(&support, index);
-	value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
 
-	/* Check value */
+	/* Form index tuple */
-	if (typeInfo->checkValue != NULL)
+	if (!HnswFormIndexTuple(&itup, values, isnull, typeInfo, &support, tupdesc))
-		typeInfo->checkValue(DatumGetPointer(value));
+		return;
 
-	/* Normalize if needed */
+	HnswInsertTupleOnDisk(index, &support, itup, heaptid, false, tupdesc);
-	normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
-	if (normprocinfo != NULL)
-	{
-		if (!HnswCheckNorm(normprocinfo, collation, value))
-			return;
-
-		value = HnswNormValue(typeInfo, collation, value);
-	}
-
-	HnswInsertTupleOnDisk(index, value, values, isnull, heap_tid, false);
 }
 
 /*

src/hnswscan.c

@@ -5,39 +5,78 @@
 #include "pgstat.h"
 #include "storage/bufmgr.h"
 #include "storage/lmgr.h"
+#include "utils/float.h"
 #include "utils/memutils.h"
 
 /*
  * Algorithm 5 from paper
  */
 static List *
-GetScanItems(IndexScanDesc scan, Datum q)
+GetScanItems(IndexScanDesc scan, Datum value)
 {
 	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
 	Relation	index = scan->indexRelation;
-	FmgrInfo   *procinfo = so->procinfo;
+	HnswSupport *support = &so->support;
-	Oid			collation = so->collation;
 	List	   *ep;
 	List	   *w;
 	int			m;
 	HnswElement entryPoint;
 	char	   *base = NULL;
+	bool		inMemory = false;
+	HnswQuery  *q = &so->q;
+
+	q->value = value;
+	q->itup = NULL;
+	q->keyData = scan->keyData;
 
 	/* Get m and entry point */
 	HnswGetMetaPageInfo(index, &m, &entryPoint);
 
+	so->m = m;
+
 	if (entryPoint == NULL)
 		return NIL;
 
-	ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, procinfo, collation, false));
+	ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, support, false, inMemory));
 
 	for (int lc = entryPoint->level; lc >= 1; lc--)
 	{
-		w = HnswSearchLayer(base, q, ep, 1, lc, index, procinfo, collation, m, false, NULL);
+		w = HnswSearchLayer(base, q, ep, 1, lc, index, support, m, false, NULL, inMemory, NULL, NULL, true, NULL);
 		ep = w;
 	}
 
-	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
+	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, inMemory, &so->v, hnsw_iterative_search != HNSW_ITERATIVE_SEARCH_OFF ? &so->discarded : NULL, true, &so->tuples);
+}
+
+/*
+ * Resume scan at ground level with discarded candidates
+ */
+static List *
+ResumeScanItems(IndexScanDesc scan)
+{
+	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
+	Relation	index = scan->indexRelation;
+	List	   *ep = NIL;
+	char	   *base = NULL;
+	int			batch_size = hnsw_ef_search;
+
+	if (pairingheap_is_empty(so->discarded))
+		return NIL;
+
+	/* Get next batch of candidates */
+	for (int i = 0; i < batch_size; i++)
+	{
+		HnswSearchCandidate *sc;
+
+		if (pairingheap_is_empty(so->discarded))
+			break;
+
+		sc = HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded));
+
+		ep = lappend(ep, sc);
+	}
+
+	return HnswSearchLayer(base, &so->q, ep, batch_size, 0, index, &so->support, so->m, false, NULL, false, &so->v, &so->discarded, false, &so->tuples);
 }
 
 /*
@@ -60,8 +99,8 @@ GetScanValue(IndexScanDesc scan)
 		Assert(!VARATT_IS_EXTENDED(DatumGetPointer(value)));
 
 		/* Normalize if needed */
-		if (so->normprocinfo != NULL)
+		if (so->support.normprocinfo != NULL)
-			value = HnswNormValue(so->typeInfo, so->collation, value);
+			value = HnswNormValue(so->typeInfo, so->support.collation[0], value);
 	}
 
 	return value;
@@ -81,14 +120,14 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
 	so = (HnswScanOpaque) palloc(sizeof(HnswScanOpaqueData));
 	so->typeInfo = HnswGetTypeInfo(index);
 	so->first = true;
+	so->v.tids = NULL;
+	so->discarded = NULL;
 	so->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
 									   "Hnsw scan temporary context",
 									   ALLOCSET_DEFAULT_SIZES);
 
 	/* Set support functions */
-	so->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
+	HnswInitSupport(&so->support, index);
-	so->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
-	so->collation = index->rd_indcollation[0];
 
 	scan->opaque = so;
 
@@ -103,7 +142,15 @@ hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int no
 {
 	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
 
+	if (so->v.tids != NULL)
+		tidhash_reset(so->v.tids);
+
+	if (so->discarded != NULL)
+		pairingheap_reset(so->discarded);
+
 	so->first = true;
+	so->tuples = 0;
+	so->previousDistance = -get_float8_infinity();
 	MemoryContextReset(so->tmpCtx);
 
 	if (keys && scan->numberOfKeys > 0)
@@ -165,22 +212,100 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 #endif
 	}
 
-	while (list_length(so->w) > 0)
+	for (;;)
 	{
 		char	   *base = NULL;
-		HnswSearchCandidate *hc = llast(so->w);
+		HnswSearchCandidate *sc;
-		HnswElement element = HnswPtrAccess(base, hc->element);
+		HnswElement element;
 		ItemPointer heaptid;
 
+		if (list_length(so->w) == 0)
+		{
+			if (hnsw_iterative_search == HNSW_ITERATIVE_SEARCH_OFF)
+				break;
+
+			/* Empty index */
+			if (so->discarded == NULL)
+				break;
+
+			/* Reached max number of additional tuples */
+			if (hnsw_iterative_search_max_tuples != -1 && so->tuples >= hnsw_iterative_search_max_tuples)
+			{
+				if (pairingheap_is_empty(so->discarded))
+					break;
+
+				/* Return remaining tuples */
+				so->w = lappend(so->w, HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded)));
+			}
+			/* Prevent scans from consuming too much memory */
+			else if (MemoryContextMemAllocated(so->tmpCtx, false) > (Size) work_mem * 1024L)
+			{
+				if (pairingheap_is_empty(so->discarded))
+				{
+					ereport(DEBUG1,
+							(errmsg("hnsw index scan exceeded work_mem after " INT64_FORMAT " tuples", so->tuples),
+							 errhint("Increase work_mem to scan more tuples.")));
+
+					break;
+				}
+
+				/* Return remaining tuples */
+				so->w = lappend(so->w, HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded)));
+			}
+			else
+			{
+				/*
+				 * Locking ensures when neighbors are read, the elements they
+				 * reference will not be deleted (and replaced) during the
+				 * iteration.
+				 *
+				 * Elements loaded into memory on previous iterations may have
+				 * been deleted (and replaced), so when reading neighbors, the
+				 * element version must be checked.
+				 */
+				LockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
+
+				so->w = ResumeScanItems(scan);
+
+				UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
+
+#if defined(HNSW_MEMORY)
+				elog(INFO, "memory: %zu KB", MemoryContextMemAllocated(so->tmpCtx, false) / 1024);
+#endif
+			}
+
+			if (list_length(so->w) == 0)
+				break;
+		}
+
+		sc = llast(so->w);
+		element = HnswPtrAccess(base, sc->element);
+
 		/* Move to next element if no valid heap TIDs */
-		if (element->heaptidsLength == 0)
+		if (!sc->matches || element->heaptidsLength == 0)
 		{
 			so->w = list_delete_last(so->w);
+
+			/* Mark memory as free for next iteration */
+			if (hnsw_iterative_search != HNSW_ITERATIVE_SEARCH_OFF)
+			{
+				pfree(element);
+				pfree(sc);
+			}
+
 			continue;
 		}
 
 		heaptid = &element->heaptids[--element->heaptidsLength];
 
+		if (hnsw_iterative_search == HNSW_ITERATIVE_SEARCH_STRICT)
+		{
+			if (sc->distance < so->previousDistance)
+				continue;
+
+			so->previousDistance = sc->distance;
+		}
+
 		MemoryContextSwitchTo(oldCtx);
 
 		scan->xs_heaptid = *heaptid;

src/hnswvacuum.c

@@ -184,13 +184,12 @@ static void
 RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswElement entryPoint)
 {
 	Relation	index = vacuumstate->index;
+	HnswSupport *support = &vacuumstate->support;
 	Buffer		buf;
 	Page		page;
 	GenericXLogState *state;
 	int			m = vacuumstate->m;
 	int			efConstruction = vacuumstate->efConstruction;
-	FmgrInfo   *procinfo = vacuumstate->procinfo;
-	Oid			collation = vacuumstate->collation;
 	BufferAccessStrategy bas = vacuumstate->bas;
 	HnswNeighborTuple ntup = vacuumstate->ntup;
 	Size		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, m);
@@ -205,7 +204,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 	element->heaptidsLength = 0;
 
 	/* Find neighbors for element, skipping itself */
-	HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, true);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, true, false);
 
 	/* Zero memory for each element */
 	MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);
@@ -229,7 +228,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 	UnlockReleaseBuffer(buf);
 
 	/* Update neighbors */
-	HnswUpdateNeighborsOnDisk(index, procinfo, collation, element, m, true, false);
+	HnswUpdateNeighborsOnDisk(index, support, element, m, true, false);
 }
 
 /*
@@ -239,6 +238,7 @@ static void
 RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 {
 	Relation	index = vacuumstate->index;
+	HnswSupport *support = &vacuumstate->support;
 	HnswElement highestPoint = &vacuumstate->highestPoint;
 	HnswElement entryPoint;
 	MemoryContext oldCtx = MemoryContextSwitchTo(vacuumstate->tmpCtx);
@@ -256,7 +256,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 		LockPage(index, HNSW_UPDATE_LOCK, ShareLock);
 
 		/* Load element */
-		HnswLoadElement(highestPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL);
+		HnswLoadElement(highestPoint, NULL, NULL, NULL, index, support, true, NULL);
 
 		/* Repair if needed */
 		if (NeedsUpdated(vacuumstate, highestPoint))
@@ -294,7 +294,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 			 * is outdated, this can remove connections at higher levels in
 			 * the graph until they are repaired, but this should be fine.
 			 */
-			HnswLoadElement(entryPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL);
+			HnswLoadElement(entryPoint, NULL, NULL, NULL, index, support, true, NULL);
 
 			if (NeedsUpdated(vacuumstate, entryPoint))
 			{
@@ -370,7 +370,7 @@ RepairGraph(HnswVacuumState * vacuumstate)
 
 			/* Create an element */
 			element = HnswInitElementFromBlock(blkno, offno);
-			HnswLoadElementFromTuple(element, etup, false, true);
+			HnswLoadElementFromTuple(element, etup, false, true, index);
 
 			elements = lappend(elements, element);
 		}
@@ -440,6 +440,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 	BlockNumber insertPage = InvalidBlockNumber;
 	Relation	index = vacuumstate->index;
 	BufferAccessStrategy bas = vacuumstate->bas;
+	bool		useIndexTuple = HnswUseIndexTuple(index);
 
 	/*
 	 * Wait for index scans to complete. Scans before this point may contain
@@ -521,12 +522,27 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 
 			/* Overwrite element */
 			etup->deleted = 1;
-			MemSet(&etup->data, 0, VARSIZE_ANY(&etup->data));
+			if (useIndexTuple)
+			{
+				IndexTuple	itup = (IndexTuple) &etup->data;
+
+				MemSet(itup, 0, IndexTupleSize(itup));
+			}
+			else
+				MemSet(&etup->data, 0, VARSIZE_ANY(&etup->data));
 
 			/* Overwrite neighbors */
 			for (int i = 0; i < ntup->count; i++)
 				ItemPointerSetInvalid(&ntup->indextids[i]);
 
+			/* Increment version */
+			/* This is used to avoid incorrect reads for iterative scans */
+			/* Reserve some bits for future use */
+			etup->version++;
+			if (etup->version > 15)
+				etup->version = 1;
+			ntup->version = etup->version;
+
 			/*
 			 * We modified the tuples in place, no need to call
 			 * PageIndexTupleOverwrite
@@ -573,13 +589,13 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
 	vacuumstate->callback_state = callback_state;
 	vacuumstate->efConstruction = HnswGetEfConstruction(index);
 	vacuumstate->bas = GetAccessStrategy(BAS_BULKREAD);
-	vacuumstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
-	vacuumstate->collation = index->rd_indcollation[0];
 	vacuumstate->ntup = palloc0(HNSW_TUPLE_ALLOC_SIZE);
 	vacuumstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
 												"Hnsw vacuum temporary context",
 												ALLOCSET_DEFAULT_SIZES);
 
+	HnswInitSupport(&vacuumstate->support, index);
+
 	/* Get m from metapage */
 	HnswGetMetaPageInfo(index, &vacuumstate->m, NULL);
 

src/ivfbuild.c

@@ -138,7 +138,7 @@ SampleRows(IvfflatBuildState * buildstate)
  * Add tuple to sort
  */
 static void
-AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
+AddTupleToSort(Relation index, ItemPointer tid, Datum *values, bool *isnull, IvfflatBuildState * buildstate)
 {
 	double		distance;
 	double		minDistance = DBL_MAX;
@@ -184,6 +184,11 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
 	slot->tts_isnull[1] = false;
 	slot->tts_values[2] = value;
 	slot->tts_isnull[2] = false;
+	for (int i = 1; i < buildstate->tupdesc->natts; i++)
+	{
+		slot->tts_values[2 + i] = values[i];
+		slot->tts_isnull[2 + i] = isnull[i];
+	}
 	ExecStoreVirtualTuple(slot);
 
 	/*
@@ -215,7 +220,7 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
 	oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
 
 	/* Add tuple to sort */
-	AddTupleToSort(index, tid, values, buildstate);
+	AddTupleToSort(index, tid, values, isnull, buildstate);
 
 	/* Reset memory context */
 	MemoryContextSwitchTo(oldCtx);
@@ -226,19 +231,20 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
  * Get index tuple from sort state
  */
 static inline void
-GetNextTuple(Tuplesortstate *sortstate, TupleDesc tupdesc, TupleTableSlot *slot, IndexTuple *itup, int *list)
+GetNextTuple(Tuplesortstate *sortstate, TupleDesc tupdesc, TupleTableSlot *slot, Datum *values, bool *isnull, IndexTuple *itup, int *list)
 {
-	Datum		value;
-	bool		isnull;
-
 	if (tuplesort_gettupleslot(sortstate, true, false, slot, NULL))
 	{
-		*list = DatumGetInt32(slot_getattr(slot, 1, &isnull));
+		bool		unused;
-		value = slot_getattr(slot, 3, &isnull);
+
+		*list = DatumGetInt32(slot_getattr(slot, 1, &unused));
+
+		for (int i = 0; i < tupdesc->natts; i++)
+			values[i] = slot_getattr(slot, 3 + i, &isnull[i]);
 
 		/* Form the index tuple */
-		*itup = index_form_tuple(tupdesc, &value, &isnull);
+		*itup = index_form_tuple(tupdesc, values, isnull);
-		(*itup)->t_tid = *((ItemPointer) DatumGetPointer(slot_getattr(slot, 2, &isnull)));
+		(*itup)->t_tid = *((ItemPointer) DatumGetPointer(slot_getattr(slot, 2, &unused)));
 	}
 	else
 		*list = -1;
@@ -254,14 +260,16 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
 	IndexTuple	itup = NULL;	/* silence compiler warning */
 	int64		inserted = 0;
 
-	TupleTableSlot *slot = MakeSingleTupleTableSlot(buildstate->tupdesc, &TTSOpsMinimalTuple);
+	TupleTableSlot *slot = MakeSingleTupleTableSlot(buildstate->sortdesc, &TTSOpsMinimalTuple);
-	TupleDesc	tupdesc = RelationGetDescr(index);
+	TupleDesc	tupdesc = buildstate->tupdesc;
+	Datum	   *values = palloc(tupdesc->natts * sizeof(Datum));
+	bool	   *isnull = palloc(tupdesc->natts * sizeof(bool));
 
 	pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_LOAD);
 
 	pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_TOTAL, buildstate->indtuples);
 
-	GetNextTuple(buildstate->sortstate, tupdesc, slot, &itup, &list);
+	GetNextTuple(buildstate->sortstate, tupdesc, slot, values, isnull, &itup, &list);
 
 	for (int i = 0; i < buildstate->centers->length; i++)
 	{
@@ -297,7 +305,7 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
 
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_DONE, ++inserted);
 
-			GetNextTuple(buildstate->sortstate, tupdesc, slot, &itup, &list);
+			GetNextTuple(buildstate->sortstate, tupdesc, slot, values, isnull, &itup, &list);
 		}
 
 		insertPage = BufferGetBlockNumber(buf);
@@ -307,6 +315,9 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
 		/* Set the start and insert pages */
 		IvfflatUpdateList(index, buildstate->listInfo[i], insertPage, InvalidBlockNumber, startPage, forkNum);
 	}
+
+	pfree(values);
+	pfree(isnull);
 }
 
 /*
@@ -319,6 +330,7 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 	buildstate->index = index;
 	buildstate->indexInfo = indexInfo;
 	buildstate->typeInfo = IvfflatGetTypeInfo(index);
+	buildstate->tupdesc = RelationGetDescr(index);
 
 	buildstate->lists = IvfflatGetLists(index);
 	buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
@@ -356,12 +368,13 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 				 errmsg("dimensions must be greater than one for this opclass")));
 
 	/* Create tuple description for sorting */
-	buildstate->tupdesc = CreateTemplateTupleDesc(3);
+	buildstate->sortdesc = CreateTemplateTupleDesc(2 + buildstate->tupdesc->natts);
-	TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
+	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
-	TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
+	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
-	TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 3, "vector", RelationGetDescr(index)->attrs[0].atttypid, -1, 0);
+	for (int i = 0; i < buildstate->tupdesc->natts; i++)
+		TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) (3 + i), NULL, buildstate->tupdesc->attrs[i].atttypid, -1, 0);
 
-	buildstate->slot = MakeSingleTupleTableSlot(buildstate->tupdesc, &TTSOpsVirtual);
+	buildstate->slot = MakeSingleTupleTableSlot(buildstate->sortdesc, &TTSOpsVirtual);
 
 	buildstate->centers = VectorArrayInit(buildstate->lists, buildstate->dimensions, buildstate->typeInfo->itemSize(buildstate->dimensions));
 	buildstate->listInfo = palloc(sizeof(ListInfo) * buildstate->lists);
@@ -633,7 +646,7 @@ 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 = InitBuildSortState(buildstate.tupdesc, sortmem, coordinate);
+	ivfspool->sortstate = InitBuildSortState(buildstate.sortdesc, sortmem, coordinate);
 	buildstate.sortstate = ivfspool->sortstate;
 	scan = table_beginscan_parallel(ivfspool->heap,
 									ParallelTableScanFromIvfflatShared(ivfshared));
@@ -950,7 +963,7 @@ AssignTuples(IvfflatBuildState * buildstate)
 	}
 
 	/* Begin serial/leader tuplesort */
-	buildstate->sortstate = InitBuildSortState(buildstate->tupdesc, maintenance_work_mem, coordinate);
+	buildstate->sortstate = InitBuildSortState(buildstate->sortdesc, maintenance_work_mem, coordinate);
 
 	/* Add tuples to sort */
 	if (buildstate->heap != NULL)

src/ivfflat.c

@@ -17,8 +17,16 @@
 #endif
 
 int			ivfflat_probes;
+int			ivfflat_iterative_search;
+int			ivfflat_iterative_search_max_probes;
 static relopt_kind ivfflat_relopt_kind;
 
+static const struct config_enum_entry ivfflat_iterative_search_options[] = {
+	{"off", IVFFLAT_ITERATIVE_SEARCH_OFF, false},
+	{"on", IVFFLAT_ITERATIVE_SEARCH_RELAXED, false},
+	{NULL, 0, false}
+};
+
 /*
  * Initialize index options and variables
  */
@@ -33,6 +41,14 @@ IvfflatInit(void)
 							"Valid range is 1..lists.", &ivfflat_probes,
 							IVFFLAT_DEFAULT_PROBES, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
 
+	DefineCustomEnumVariable("ivfflat.iterative_search", "Sets whether to use iterative search",
+							 NULL, &ivfflat_iterative_search,
+							 IVFFLAT_ITERATIVE_SEARCH_OFF, ivfflat_iterative_search_options, PGC_USERSET, 0, NULL, NULL, NULL);
+
+	DefineCustomIntVariable("ivfflat.iterative_search_max_probes", "Sets the max number of probes for iterative search",
+							"Zero sets to the number of lists", &ivfflat_iterative_search_max_probes,
+							0, 0, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
+
 	MarkGUCPrefixReserved("ivfflat");
 }
 
@@ -57,39 +73,6 @@ ivfflatbuildphasename(int64 phasenum)
 	}
 }
 
-/*
- * Estimate the number of probes needed to get requested tuples
- */
-static int
-EstimateProbes(PlannerInfo *root, IndexPath *path, int lists)
-{
-	double		selectivity = 1;
-	ListCell   *lc;
-	double		tuplesPerList;
-
-	/* Cannot estimate without limit */
-	/* limit_tuples includes offset */
-	if (root->limit_tuples < 0)
-		return 0;
-
-	/* Get the selectivity of non-index conditions */
-	foreach(lc, path->indexinfo->indrestrictinfo)
-	{
-		RestrictInfo *rinfo = lfirst(lc);
-
-		/* Skip DEFAULT_INEQ_SEL since it may be a distance filter */
-		if (rinfo->norm_selec >= 0 && rinfo->norm_selec <= 1 && rinfo->norm_selec != (Selectivity) DEFAULT_INEQ_SEL)
-			selectivity *= rinfo->norm_selec;
-	}
-
-	tuplesPerList = path->indexinfo->tuples * selectivity / (double) lists;
-	if (tuplesPerList == 0)
-		return lists;
-
-	/* No need to cap at this point */
-	return root->limit_tuples / tuplesPerList;
-}
-
 /*
  * Estimate the cost of an index scan
  */
@@ -101,8 +84,9 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 {
 	GenericCosts costs;
 	int			lists;
-	int			probes;
 	double		ratio;
+	double		sequentialRatio = 0.5;
+	double		startupPages;
 	double		spc_seq_page_cost;
 	Relation	index;
 
@@ -119,52 +103,37 @@ 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);
 
-	probes = Max(ivfflat_probes, EstimateProbes(root, path, lists));
-
 	/* Get the ratio of lists that we need to visit */
-	ratio = ((double) probes) / lists;
+	ratio = ((double) ivfflat_probes) / lists;
 	if (ratio > 1.0)
 		ratio = 1.0;
 
-	/*
-	 * This gives us the subset of tuples to visit. This value is passed into
-	 * the generic cost estimator to determine the number of pages to visit
-	 * 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);
 
+	/* Change some page cost from random to sequential */
+	costs.indexTotalCost -= sequentialRatio * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+
+	/* Startup cost is cost before returning the first row */
+	costs.indexStartupCost = costs.indexTotalCost * ratio;
+
 	/* Adjust cost if needed since TOAST not included in seq scan cost */
-	if (costs.numIndexPages > path->indexinfo->rel->pages && ratio < 0.5)
+	startupPages = costs.numIndexPages * ratio;
+	if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
 	{
-		/* Change all page cost from random to sequential */
+		/* Change rest of page cost from random to sequential */
-		costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+		costs.indexStartupCost -= (1 - sequentialRatio) * startupPages * (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 -= (startupPages - 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);
 	}
 
-	/*
+	*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;

src/ivfflat.h

@@ -80,6 +80,14 @@
 
 /* Variables */
 extern int	ivfflat_probes;
+extern int	ivfflat_iterative_search;
+extern int	ivfflat_iterative_search_max_probes;
+
+typedef enum IvfflatIterativeSearchType
+{
+	IVFFLAT_ITERATIVE_SEARCH_OFF,
+	IVFFLAT_ITERATIVE_SEARCH_RELAXED
+}			IvfflatIterativeSearchType;
 
 typedef struct VectorArrayData
 {
@@ -165,6 +173,7 @@ typedef struct IvfflatBuildState
 	Relation	index;
 	IndexInfo  *indexInfo;
 	const		IvfflatTypeInfo *typeInfo;
+	TupleDesc	tupdesc;
 
 	/* Settings */
 	int			dimensions;
@@ -198,7 +207,7 @@ typedef struct IvfflatBuildState
 
 	/* Sorting */
 	Tuplesortstate *sortstate;
-	TupleDesc	tupdesc;
+	TupleDesc	sortdesc;
 	TupleTableSlot *slot;
 
 	/* Memory */
@@ -247,8 +256,10 @@ typedef struct IvfflatScanOpaqueData
 {
 	const		IvfflatTypeInfo *typeInfo;
 	int			probes;
+	int			maxProbes;
 	int			dimensions;
 	bool		first;
+	Datum		value;
 
 	/* Sorting */
 	Tuplesortstate *sortstate;
@@ -265,6 +276,8 @@ typedef struct IvfflatScanOpaqueData
 
 	/* Lists */
 	pairingheap *listQueue;
+	BlockNumber *listPages;
+	int			listIndex;
 	IvfflatScanList lists[FLEXIBLE_ARRAY_MEMBER];	/* must come last */
 }			IvfflatScanOpaqueData;
 

src/ivfinsert.c

@@ -78,6 +78,8 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 	BlockNumber insertPage = InvalidBlockNumber;
 	ListInfo	listInfo;
 	BlockNumber originalInsertPage;
+	TupleDesc	tupdesc = RelationGetDescr(index);
+	Datum	   *newValues = palloc(tupdesc->natts * sizeof(Datum));
 
 	/* Detoast once for all calls */
 	value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
@@ -98,12 +100,16 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 	IvfflatGetMetaPageInfo(index, NULL, NULL);
 
 	/* Find the insert page - sets the page and list info */
-	FindInsertPage(index, values, &insertPage, &listInfo);
+	FindInsertPage(index, &value, &insertPage, &listInfo);
 	Assert(BlockNumberIsValid(insertPage));
 	originalInsertPage = insertPage;
 
+	newValues[0] = value;
+	for (int i = 1; i < tupdesc->natts; i++)
+		newValues[i] = values[i];
+
 	/* Form tuple */
-	itup = index_form_tuple(RelationGetDescr(index), &value, isnull);
+	itup = index_form_tuple(tupdesc, newValues, isnull);
 	itup->t_tid = *heap_tid;
 
 	/* Get tuple size */

src/ivfscan.c

@@ -65,7 +65,7 @@ GetScanLists(IndexScanDesc scan, Datum value)
 			/* Use procinfo from the index instead of scan key for performance */
 			distance = DatumGetFloat8(so->distfunc(so->procinfo, so->collation, PointerGetDatum(&list->center), value));
 
-			if (listCount < so->probes)
+			if (listCount < so->maxProbes)
 			{
 				IvfflatScanList *scanlist;
 
@@ -78,7 +78,7 @@ GetScanLists(IndexScanDesc scan, Datum value)
 				pairingheap_add(so->listQueue, &scanlist->ph_node);
 
 				/* Calculate max distance */
-				if (listCount == so->probes)
+				if (listCount == so->maxProbes)
 					maxDistance = GetScanList(pairingheap_first(so->listQueue))->distance;
 			}
 			else if (distance < maxDistance)
@@ -102,6 +102,11 @@ GetScanLists(IndexScanDesc scan, Datum value)
 
 		UnlockReleaseBuffer(cbuf);
 	}
+
+	for (int i = listCount - 1; i >= 0; i--)
+		so->listPages[i] = GetScanList(pairingheap_remove_first(so->listQueue))->startPage;
+
+	Assert(pairingheap_is_empty(so->listQueue));
 }
 
 /*
@@ -114,11 +119,14 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	TupleDesc	tupdesc = RelationGetDescr(scan->indexRelation);
 	double		tuples = 0;
 	TupleTableSlot *slot = so->vslot;
+	int			batchProbes = 0;
+
+	tuplesort_reset(so->sortstate);
 
 	/* Search closest probes lists */
-	while (!pairingheap_is_empty(so->listQueue))
+	while (so->listIndex < so->maxProbes && (++batchProbes) <= so->probes)
 	{
-		BlockNumber searchPage = GetScanList(pairingheap_remove_first(so->listQueue))->startPage;
+		BlockNumber searchPage = so->listPages[so->listIndex++];
 
 		/* Search all entry pages for list */
 		while (BlockNumberIsValid(searchPage))
@@ -166,13 +174,17 @@ GetScanItems(IndexScanDesc scan, Datum value)
 		}
 	}
 
-	if (tuples < 100)
+	if (tuples < 100 && ivfflat_iterative_search == IVFFLAT_ITERATIVE_SEARCH_OFF)
 		ereport(DEBUG1,
 				(errmsg("index scan found few tuples"),
 				 errdetail("Index may have been created with little data."),
 				 errhint("Recreate the index and possibly decrease lists.")));
 
 	tuplesort_performsort(so->sortstate);
+
+#if defined(IVFFLAT_MEMORY)
+	elog(INFO, "memory: %zu MB", MemoryContextMemAllocated(CurrentMemoryContext, true) / (1024 * 1024));
+#endif
 }
 
 /*
@@ -240,6 +252,7 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	int			lists;
 	int			dimensions;
 	int			probes = ivfflat_probes;
+	int			maxProbes;
 
 	scan = RelationGetIndexScan(index, nkeys, norderbys);
 
@@ -249,10 +262,21 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	if (probes > lists)
 		probes = lists;
 
-	so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + probes * sizeof(IvfflatScanList));
+	if (ivfflat_iterative_search != IVFFLAT_ITERATIVE_SEARCH_OFF)
+	{
+		if (ivfflat_iterative_search_max_probes == 0)
+			maxProbes = lists;
+		else
+			maxProbes = Min(ivfflat_iterative_search_max_probes, lists);
+	}
+	else
+		maxProbes = probes;
+
+	so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + maxProbes * sizeof(IvfflatScanList));
 	so->typeInfo = IvfflatGetTypeInfo(index);
 	so->first = true;
 	so->probes = probes;
+	so->maxProbes = maxProbes;
 	so->dimensions = dimensions;
 
 	/* Set support functions */
@@ -280,6 +304,8 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	so->bas = GetAccessStrategy(BAS_BULKREAD);
 
 	so->listQueue = pairingheap_allocate(CompareLists, scan);
+	so->listPages = palloc(maxProbes * sizeof(BlockNumber));
+	so->listIndex = 0;
 
 	scan->opaque = so;
 
@@ -294,11 +320,9 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
 {
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
 
-	if (!so->first)
-		tuplesort_reset(so->sortstate);
-
 	so->first = true;
 	pairingheap_reset(so->listQueue);
+	so->listIndex = 0;
 
 	if (keys && scan->numberOfKeys > 0)
 		memmove(scan->keyData, keys, scan->numberOfKeys * sizeof(ScanKeyData));
@@ -314,6 +338,8 @@ bool
 ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 {
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
+	ItemPointer heaptid;
+	bool		isnull;
 
 	/*
 	 * Index can be used to scan backward, but Postgres doesn't support
@@ -341,28 +367,25 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		IvfflatBench("GetScanLists", GetScanLists(scan, value));
 		IvfflatBench("GetScanItems", GetScanItems(scan, value));
 		so->first = false;
+		so->value = value;
 
-#if defined(IVFFLAT_MEMORY)
+		/* TODO clean up if we allocated a new value */
-		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->mslot, NULL))
+	while (!tuplesort_gettupleslot(so->sortstate, true, false, so->mslot, NULL))
 	{
-		bool		isnull;
+		if (so->listIndex == so->maxProbes)
-		ItemPointer heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->mslot, 2, &isnull));
+			return false;
 
-		scan->xs_heaptid = *heaptid;
+		IvfflatBench("GetScanItems", GetScanItems(scan, so->value));
-		scan->xs_recheck = false;
-		scan->xs_recheckorderby = false;
-		return true;
 	}
 
-	return false;
+	heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->mslot, 2, &isnull));
+
+	scan->xs_heaptid = *heaptid;
+	scan->xs_recheck = false;
+	scan->xs_recheckorderby = false;
+	return true;
 }
 
 /*
@@ -374,6 +397,7 @@ ivfflatendscan(IndexScanDesc scan)
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
 
 	pairingheap_free(so->listQueue);
+	pfree(so->listPages);
 	tuplesort_end(so->sortstate);
 	FreeAccessStrategy(so->bas);
 	FreeTupleDesc(so->tupdesc);

src/ivfvacuum.c

@@ -26,7 +26,7 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 		Page		cpage;
 		OffsetNumber coffno;
 		OffsetNumber cmaxoffno;
-		BlockNumber startPages[MaxOffsetNumber];
+		BlockNumber listPages[MaxOffsetNumber];
 		ListInfo	listInfo;
 
 		cbuf = ReadBuffer(index, blkno);
@@ -40,7 +40,7 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 		{
 			IvfflatList list = (IvfflatList) PageGetItem(cpage, PageGetItemId(cpage, coffno));
 
-			startPages[coffno - FirstOffsetNumber] = list->startPage;
+			listPages[coffno - FirstOffsetNumber] = list->startPage;
 		}
 
 		listInfo.blkno = blkno;
@@ -50,7 +50,7 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 
 		for (coffno = FirstOffsetNumber; coffno <= cmaxoffno; coffno = OffsetNumberNext(coffno))
 		{
-			BlockNumber searchPage = startPages[coffno - FirstOffsetNumber];
+			BlockNumber searchPage = listPages[coffno - FirstOffsetNumber];
 			BlockNumber insertPage = InvalidBlockNumber;
 
 			/* Iterate over entry pages */

src/vector.c

@@ -155,24 +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/t/002_ivfflat_vacuum.pl

@@ -6,13 +6,7 @@ use Test::More;
 
 my $dim = 3;
 
-my @r = ();
+my $array_sql = join(",", ('random()') x $dim);
-for (1 .. $dim)
-{
-	my $v = int(rand(1000)) + 1;
-	push(@r, "i % $v");
-}
-my $array_sql = join(", ", @r);
 
 # Initialize node
 my $node = PostgreSQL::Test::Cluster->new('node');
@@ -23,19 +17,20 @@ $node->start;
 $node->safe_psql("postgres", "CREATE EXTENSION vector;");
 $node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
 $node->safe_psql("postgres",
-	"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
+	"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
 );
 $node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v vector_l2_ops);");
 
 # Get size
 my $size = $node->safe_psql("postgres", "SELECT pg_total_relation_size('tst_v_idx');");
 
+# Store values
+$node->safe_psql("postgres", "CREATE TABLE tmp AS SELECT * FROM tst;");
+
 # Delete all, vacuum, and insert same data
 $node->safe_psql("postgres", "DELETE FROM tst;");
 $node->safe_psql("postgres", "VACUUM tst;");
-$node->safe_psql("postgres",
+$node->safe_psql("postgres", "INSERT INTO tst SELECT * FROM tmp;");
-	"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
-);
 
 # Check size
 my $new_size = $node->safe_psql("postgres", "SELECT pg_total_relation_size('tst_v_idx');");

test/t/009_ivfflat_filtering.pl

@@ -5,7 +5,7 @@ use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
-my $nc = 100;
+my $nc = 50;
 my $limit = 20;
 
 my $array_sql = join(",", ('random()') x $dim);
@@ -94,15 +94,15 @@ like($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);");
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
 ));
-like($explain, qr/Index Scan on attribute_idx/);
+# TODO Use attribute index
+like($explain, qr/Index Scan using idx/);
 
 # Test partial index
 $node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING ivfflat (v vector_l2_ops) WITH (lists = 5) WHERE (c = $c);");

test/t/017_hnsw_filtering.pl

@@ -18,9 +18,13 @@ $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;");
 
@@ -37,8 +41,7 @@ my $c = int(rand() * $nc);
 my $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
 ));
-# TODO Do not use index
+like($explain, qr/Seq Scan/);
-like($explain, qr/Index Scan using idx/);
 
 # Test attribute filtering with few rows removed
 $explain = $node->safe_psql("postgres", qq(
@@ -56,8 +59,7 @@ like($explain, qr/Index Scan using idx/);
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c < 1 ORDER BY v <-> '$query' LIMIT $limit;
 ));
-# TODO Do not use index
+like($explain, qr/Seq Scan/);
-like($explain, qr/Index Scan using idx/);
 
 # Test attribute filtering with few rows removed like
 $explain = $node->safe_psql("postgres", qq(
@@ -96,13 +98,25 @@ $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/039_hnsw_cost.pl

@@ -0,0 +1,60 @@
+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, 2000) 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/);
+
+	# 3x the rows are needed for distance filters
+	# since the planner uses DEFAULT_INEQ_SEL for the selectivity (should be 1)
+	# Recreate index for performance
+	$node->safe_psql("postgres", "DROP INDEX idx;");
+	$node->safe_psql("postgres",
+		"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(2001, 6000) i;"
+	);
+	$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
+	$node->safe_psql("postgres", "ANALYZE tst;");
+
+	$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, 5000) 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();

test/t/041_ivfflat_iterative_search.pl

@@ -0,0 +1,54 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my $dim = 3;
+my $array_sql = join(",", ('random()') x $dim);
+
+# Initialize node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i int4 PRIMARY KEY, v vector($dim));");
+$node->safe_psql("postgres",
+	"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
+);
+$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v vector_l2_ops);");
+
+my $count = $node->safe_psql("postgres", qq(
+	SET enable_seqscan = off;
+	SET ivfflat.probes = 10;
+	SET ivfflat.iterative_search = on;
+	SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
+));
+is($count, 10);
+
+foreach ((30, 50, 70))
+{
+	my $max_probes = $_;
+	my $expected = $max_probes / 10;
+	my $sum = 0;
+
+	for my $i (1 .. 20)
+	{
+		$count = $node->safe_psql("postgres", qq(
+			SET enable_seqscan = off;
+			SET ivfflat.probes = 10;
+			SET ivfflat.iterative_search = on;
+			SET ivfflat.iterative_search_max_probes = $max_probes;
+			SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst WHERE i = $i) LIMIT 11) t;
+		));
+		$sum += $count;
+	}
+
+	my $avg = $sum / 20;
+	cmp_ok($avg, '>', $expected - 2);
+	cmp_ok($avg, '<', $expected + 2);
+}
+
+done_testing();

test/t/042_ivfflat_iterative_search_recall.pl

@@ -0,0 +1,125 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my $node;
+my @queries = ();
+my @expected;
+my $limit = 20;
+my @cs = (100, 1000);
+
+sub test_recall
+{
+	my ($c, $probes, $min, $operator) = @_;
+	my $correct = 0;
+	my $total = 0;
+
+	my $explain = $node->safe_psql("postgres", qq(
+		SET enable_seqscan = off;
+		SET ivfflat.probes = $probes;
+		SET ivfflat.iterative_search = on;
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[0]' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx on tst/);
+
+	for my $i (0 .. $#queries)
+	{
+		my $actual = $node->safe_psql("postgres", qq(
+			SET enable_seqscan = off;
+			SET ivfflat.probes = $probes;
+			SET ivfflat.iterative_search = on;
+			SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[$i]' LIMIT $limit;
+		));
+		my @actual_ids = split("\n", $actual);
+
+		my @expected_ids = split("\n", $expected[$i]);
+		my %expected_set = map { $_ => 1 } @expected_ids;
+
+		foreach (@actual_ids)
+		{
+			if (exists($expected_set{$_}))
+			{
+				$correct++;
+			}
+		}
+
+		$total += $limit;
+	}
+
+	cmp_ok($correct / $total, ">=", $min, $operator);
+}
+
+# Initialize node
+$node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));");
+$node->safe_psql("postgres",
+	"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 100000) i;"
+);
+
+# Generate queries
+for (1 .. 20)
+{
+	my $r1 = rand();
+	my $r2 = rand();
+	my $r3 = rand();
+	push(@queries, "[$r1,$r2,$r3]");
+}
+
+# Check each index type
+my @operators = ("<->", "<=>");
+my @opclasses = ("vector_l2_ops", "vector_cosine_ops");
+
+for my $i (0 .. $#operators)
+{
+	my $operator = $operators[$i];
+	my $opclass = $opclasses[$i];
+
+	$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING ivfflat (v $opclass);");
+
+	foreach (@cs)
+	{
+		my $c = $_;
+
+		# Get exact results
+		@expected = ();
+		foreach (@queries)
+		{
+			my $res = $node->safe_psql("postgres", qq(
+				SET enable_indexscan = off;
+				WITH top AS (
+					SELECT v $operator '$_' AS distance FROM tst WHERE i % $c = 0 ORDER BY distance LIMIT $limit
+				)
+				SELECT i FROM tst WHERE (v $operator '$_') <= (SELECT MAX(distance) FROM top)
+			));
+			push(@expected, $res);
+		}
+
+		if ($c == 100)
+		{
+			test_recall($c, 1, 0.58, $operator);
+			test_recall($c, 10, 0.98, $operator);
+		}
+		else
+		{
+			if ($operator eq "<->")
+			{
+				test_recall($c, 1, 0.80, $operator);
+			}
+			else
+			{
+				test_recall($c, 1, 0.88, $operator);
+			}
+		}
+	}
+
+	$node->safe_psql("postgres", "DROP INDEX idx;");
+}
+
+done_testing();

test/t/043_hnsw_iterative_search.pl

@@ -0,0 +1,67 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my $dim = 3;
+my $array_sql = join(",", ('random()') x $dim);
+
+# Initialize node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i int4 PRIMARY KEY, v vector($dim));");
+$node->safe_psql("postgres",
+	"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
+);
+$node->safe_psql("postgres", qq(
+	SET maintenance_work_mem = '128MB';
+	SET max_parallel_maintenance_workers = 2;
+	CREATE INDEX ON tst USING hnsw (v vector_l2_ops)
+));
+
+my $count = $node->safe_psql("postgres", qq(
+	SET enable_seqscan = off;
+	SET hnsw.iterative_search = on;
+	SET work_mem = '8MB';
+	SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
+));
+is($count, 10);
+
+foreach ((30000, 50000, 70000))
+{
+	my $max_tuples = $_;
+	my $expected = $max_tuples / 10000;
+	my $sum = 0;
+
+	for my $i (1 .. 20)
+	{
+		$count = $node->safe_psql("postgres", qq(
+			SET enable_seqscan = off;
+			SET hnsw.iterative_search = on;
+			SET hnsw.iterative_search_max_tuples = $max_tuples;
+			SET work_mem = '8MB';
+			SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst WHERE i = $i) LIMIT 11) t;
+		));
+		$sum += $count;
+	}
+
+	my $avg = $sum / 20;
+	cmp_ok($avg, '>', $expected - 2);
+	cmp_ok($avg, '<', $expected + 2);
+}
+
+my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
+	SET enable_seqscan = off;
+	SET hnsw.iterative_search = on;
+	SET client_min_messages = debug1;
+	SET work_mem = '2MB';
+	SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
+));
+like($stderr, qr/hnsw index scan exceeded work_mem after \d+ tuples/);
+
+done_testing();

test/t/044_hnsw_iterative_search_recall.pl

@@ -0,0 +1,131 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my $node;
+my @queries = ();
+my @expected;
+my $limit = 20;
+my $dim = 3;
+my $array_sql = join(",", ('random()') x $dim);
+my @cs = (100, 1000);
+
+sub test_recall
+{
+	my ($c, $ef_search, $min, $operator) = @_;
+	my $correct = 0;
+	my $total = 0;
+
+	my $explain = $node->safe_psql("postgres", qq(
+		SET enable_seqscan = off;
+		SET hnsw.ef_search = $ef_search;
+		SET hnsw.iterative_search = on;
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[0]' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx on tst/);
+
+	for my $i (0 .. $#queries)
+	{
+		my $actual = $node->safe_psql("postgres", qq(
+			SET enable_seqscan = off;
+			SET hnsw.ef_search = $ef_search;
+			SET hnsw.iterative_search = on;
+			SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[$i]' LIMIT $limit;
+		));
+		my @actual_ids = split("\n", $actual);
+
+		my @expected_ids = split("\n", $expected[$i]);
+		my %expected_set = map { $_ => 1 } @expected_ids;
+
+		foreach (@actual_ids)
+		{
+			if (exists($expected_set{$_}))
+			{
+				$correct++;
+			}
+		}
+
+		$total += $limit;
+	}
+
+	cmp_ok($correct / $total, ">=", $min, $operator);
+}
+
+# Initialize node
+$node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
+$node->safe_psql("postgres",
+	"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
+);
+
+# Generate queries
+for (1 .. 20)
+{
+	my @r = ();
+	for (1 .. $dim)
+	{
+		push(@r, rand());
+	}
+	push(@queries, "[" . join(",", @r) . "]");
+}
+
+# Check each index type
+my @operators = ("<->", "<=>");
+my @opclasses = ("vector_l2_ops", "vector_cosine_ops");
+
+for my $i (0 .. $#operators)
+{
+	my $operator = $operators[$i];
+	my $opclass = $opclasses[$i];
+
+	$node->safe_psql("postgres", qq(
+		SET maintenance_work_mem = '128MB';
+		CREATE INDEX idx ON tst USING hnsw (v $opclass);
+	));
+
+	foreach (@cs)
+	{
+		my $c = $_;
+
+		# Get exact results
+		@expected = ();
+		foreach (@queries)
+		{
+			my $res = $node->safe_psql("postgres", qq(
+				SET enable_indexscan = off;
+				WITH top AS (
+					SELECT v $operator '$_' AS distance FROM tst WHERE i % $c = 0 ORDER BY distance LIMIT $limit
+				)
+				SELECT i FROM tst WHERE (v $operator '$_') <= (SELECT MAX(distance) FROM top)
+			));
+			push(@expected, $res);
+		}
+
+		if ($c == 100)
+		{
+			test_recall($c, 40, 0.99, $operator);
+		}
+		else
+		{
+			if ($operator eq "<->")
+			{
+				test_recall($c, 40, 0.99, $operator);
+			}
+			else
+			{
+				test_recall($c, 40, 0.99, $operator);
+			}
+		}
+	}
+
+	$node->safe_psql("postgres", "DROP INDEX idx;");
+}
+
+done_testing();

test/t/045_hnsw_hqann.pl

@@ -0,0 +1,113 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my $node;
+my @queries = ();
+my @cs = ();
+my @expected;
+my $limit = 20;
+my $dim = 3;
+my $array_sql = join(",", ('random()') x $dim);
+my $nc = 1000;
+
+sub test_recall
+{
+	my ($min, $operator) = @_;
+	my $correct = 0;
+	my $total = 0;
+
+	my $explain = $node->safe_psql("postgres", qq(
+		SET enable_seqscan = off;
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $cs[0] ORDER BY v $operator '$queries[0]' LIMIT $limit;
+	));
+	like($explain, qr/Index Cond/);
+
+	for my $i (0 .. $#queries)
+	{
+		my $actual = $node->safe_psql("postgres", qq(
+			SET enable_seqscan = off;
+			SELECT i FROM tst WHERE c = $cs[$i] ORDER BY v $operator '$queries[$i]' LIMIT $limit;
+		));
+		my @actual_ids = split("\n", $actual);
+		my %actual_set = map { $_ => 1 } @actual_ids;
+
+		is(scalar(@actual_ids), $limit);
+
+		my @expected_ids = split("\n", $expected[$i]);
+
+		foreach (@expected_ids)
+		{
+			if (exists($actual_set{$_}))
+			{
+				$correct++;
+			}
+			$total++;
+		}
+	}
+
+	cmp_ok($correct / $total, ">=", $min, $operator);
+}
+
+# Initialize node
+$node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int4);");
+$node->safe_psql("postgres",
+	"INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc FROM generate_series(1, 20000) i;"
+);
+
+# Generate queries
+for (1 .. 20)
+{
+	my @r = ();
+	for (1 .. $dim)
+	{
+		push(@r, rand());
+	}
+	push(@queries, "[" . join(",", @r) . "]");
+	push(@cs, int(rand() * $nc));
+}
+
+# Get exact results
+@expected = ();
+for my $i (0 .. $#queries)
+{
+	my $res = $node->safe_psql("postgres", "SELECT i FROM tst WHERE c = $cs[$i] ORDER BY v <=> '$queries[$i]' LIMIT $limit;");
+	push(@expected, $res);
+}
+
+# Add index
+$node->safe_psql("postgres", qq(
+	SET maintenance_work_mem = '256MB';
+	SET max_parallel_maintenance_workers = 2;
+	CREATE INDEX ON tst USING hnsw (v vector_cosine_ops, c);
+));
+
+# Test recall
+test_recall(0.99, '<=>');
+
+# Test vacuum
+$node->safe_psql("postgres", "DELETE FROM tst WHERE c > 5;");
+$node->safe_psql("postgres", "VACUUM tst;");
+
+# Test columns
+my ($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (c);");
+like($stderr, qr/first column must be a vector/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (c, v vector_cosine_ops);");
+like($stderr, qr/first column must be a vector/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_cosine_ops, c, c);");
+like($stderr, qr/index cannot have more than two columns/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_cosine_ops, v vector_cosine_ops);");
+like($stderr, qr/column 2 cannot be a vector/);
+
+done_testing();