Improved performance of HNSW index scans from disk [skip ci]

.github/workflows/build.yml

@@ -30,7 +30,7 @@ jobs:
           dev-files: true
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }}
+          PG_CFLAGS: ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }} -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
       - run: |
           export PG_CONFIG=`which pg_config`
           sudo --preserve-env=PG_CONFIG make install
@@ -48,7 +48,7 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - postgres: 18
+          - postgres: 17
             os: macos-26
           - postgres: 14
             os: macos-15-intel
@@ -59,7 +59,7 @@ jobs:
           postgres-version: ${{ matrix.postgres }}
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-unknown-warning-option ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }}
+          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-unknown-warning-option
       - run: make install
       - run: make installcheck
       - if: ${{ failure() }}
@@ -72,7 +72,7 @@ jobs:
           tar xf $TAG.tar.gz
           mv postgres-$TAG postgres
         env:
-          TAG: ${{ matrix.postgres == 18 && 'REL_18_0' || 'REL_14_19' }}
+          TAG: ${{ matrix.postgres == 17 && 'REL_17_6' || 'REL_14_19' }}
       - run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl"
         env:
           PERL5LIB: /Users/runner/perl5/lib/perl5
@@ -136,7 +136,7 @@ jobs:
       - uses: actions/checkout@v5
       - uses: ankane/setup-postgres-valgrind@v1
         with:
-          postgres-version: 18
+          postgres-version: 17
           check-ub: yes
       - run: make OPTFLAGS=""
       - run: sudo --preserve-env=PG_CONFIG make install

CHANGELOG.md

@@ -1,7 +1,6 @@
 ## 0.8.2 (unreleased)
 
-- Improved `install` target on Windows
-- Fixed `Index Searches` in `EXPLAIN` output for Postgres 18
+- Improved performance of HNSW index scans from disk
 
 ## 0.8.1 (2025-09-04)
 

Makefile.win

@@ -57,7 +57,7 @@ install: all
 	copy $(SHLIB) "$(PKGLIBDIR)"
 	copy $(EXTENSION).control "$(SHAREDIR)\extension"
 	copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
-	if not exist "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)" mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
+	mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
 	for %f in ($(HEADERS)) do copy %f "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
 
 installcheck:

README.md

@@ -36,7 +36,7 @@ You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#p
 Ensure [C++ support in Visual Studio](https://learn.microsoft.com/en-us/cpp/build/building-on-the-command-line?view=msvc-170#download-and-install-the-tools) is installed and run `x64 Native Tools Command Prompt for VS [version]` as administrator. Then use `nmake` to build:
 
 ```cmd
-set "PGROOT=C:\Program Files\PostgreSQL\18"
+set "PGROOT=C:\Program Files\PostgreSQL\17"
 cd %TEMP%
 git clone --branch v0.8.1 https://github.com/pgvector/pgvector.git
 cd pgvector
@@ -695,10 +695,10 @@ CREATE INDEX CONCURRENTLY ...
 
 ### Querying
 
-Use `EXPLAIN (ANALYZE, BUFFERS)` to debug performance.
+Use `EXPLAIN ANALYZE` to debug performance.
 
 ```sql
-EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
+EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
 #### Exact Search
@@ -771,7 +771,6 @@ Use pgvector from any language with a Postgres client. You can even generate and
 
 Language | Libraries / Examples
 --- | ---
-Ada | [pgvector-ada](https://github.com/pgvector/pgvector-ada)
 C | [pgvector-c](https://github.com/pgvector/pgvector-c)
 C++ | [pgvector-cpp](https://github.com/pgvector/pgvector-cpp)
 C#, F#, Visual Basic | [pgvector-dotnet](https://github.com/pgvector/pgvector-dotnet)
@@ -791,7 +790,6 @@ Lisp | [pgvector-lisp](https://github.com/pgvector/pgvector-lisp)
 Lua | [pgvector-lua](https://github.com/pgvector/pgvector-lua)
 Nim | [pgvector-nim](https://github.com/pgvector/pgvector-nim)
 OCaml | [pgvector-ocaml](https://github.com/pgvector/pgvector-ocaml)
-Pascal | [pgvector-pascal](https://github.com/pgvector/pgvector-pascal)
 Perl | [pgvector-perl](https://github.com/pgvector/pgvector-perl)
 PHP | [pgvector-php](https://github.com/pgvector/pgvector-php)
 Python | [pgvector-python](https://github.com/pgvector/pgvector-python)
@@ -814,7 +812,7 @@ Yes, pgvector uses the write-ahead log (WAL), which allows for replication and p
 
 #### What if I want to index vectors with more than 2,000 dimensions?
 
-You can use [half-precision vectors](#half-precision-vectors) or [half-precision indexing](#half-precision-indexing) to index up to 4,000 dimensions or [binary quantization](#binary-quantization) to index up to 64,000 dimensions. Other options are [indexing subvectors](#indexing-subvectors) (for models that support it) or [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction).
+You can use [half-precision indexing](#half-precision-indexing) to index up to 4,000 dimensions or [binary quantization](#binary-quantization) to index up to 64,000 dimensions. Another option is [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction).
 
 #### Can I store vectors with different dimensions in the same column?
 
@@ -1066,7 +1064,7 @@ 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/18/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:
@@ -1077,11 +1075,11 @@ sudo --preserve-env=PG_CONFIG make install
 
 A few common paths on Mac are:
 
-- EDB installer - `/Library/PostgreSQL/18/bin/pg_config`
-- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@18/bin/pg_config`
-- Homebrew (x86-64) - `/usr/local/opt/postgresql@18/bin/pg_config`
+- EDB installer - `/Library/PostgreSQL/17/bin/pg_config`
+- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@17/bin/pg_config`
+- Homebrew (x86-64) - `/usr/local/opt/postgresql@17/bin/pg_config`
 
-Note: Replace `18` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### Missing Header
 
@@ -1090,10 +1088,10 @@ 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-18
+sudo apt install postgresql-server-dev-17
 ```
 
-Note: Replace `18` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### Missing SDK
 
@@ -1140,15 +1138,13 @@ 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:pg18-trixie
+docker pull pgvector/pgvector:pg17-trixie
 ```
 
-This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `18` 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).
 
 Supported tags are:
 
-- `pg18-trixie`, `0.8.1-pg18-trixie`
-- `pg18-bookworm`, `0.8.1-pg18-bookworm`, `pg18`, `0.8.1-pg18`
 - `pg17-trixie`, `0.8.1-pg17-trixie`
 - `pg17-bookworm`, `0.8.1-pg17-bookworm`, `pg17`, `0.8.1-pg17`
 - `pg16-trixie`, `0.8.1-pg16-trixie`
@@ -1165,7 +1161,7 @@ You can also build the image manually:
 ```sh
 git clone --branch v0.8.1 https://github.com/pgvector/pgvector.git
 cd pgvector
-docker build --pull --build-arg PG_MAJOR=18 -t myuser/pgvector .
+docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector .
 ```
 
 If you increase `maintenance_work_mem`, make sure `--shm-size` is at least that size to avoid an error with parallel HNSW index builds.
@@ -1182,7 +1178,7 @@ With Homebrew Postgres, you can use:
 brew install pgvector
 ```
 
-Note: This only adds it to the `postgresql@18` and `postgresql@17` formulas
+Note: This only adds it to the `postgresql@17` and `postgresql@14` formulas
 
 ### PGXN
 
@@ -1197,22 +1193,22 @@ 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-18-pgvector
+sudo apt install postgresql-17-pgvector
 ```
 
-Note: Replace `18` 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_18
+sudo yum install pgvector_17
 # or
-sudo dnf install pgvector_18
+sudo dnf install pgvector_17
 ```
 
-Note: Replace `18` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### pkg
 
@@ -1271,6 +1267,7 @@ 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)
+- [Turbocharging Vector Databases using Modern SSDs](https://www.vldb.org/pvldb/volumes/18/paper/Turbocharging%20Vector%20Databases%20using%20Modern%20SSDs)
 
 ## History
 

src/hnsw.h

@@ -362,13 +362,6 @@ typedef union
 	ItemPointerData indextid;
 }			HnswUnvisited;
 
-typedef struct HnswReadStreamData
-{
-	HnswUnvisited *unvisited;
-	int			unvisitedLength;
-	int			visited;
-}			HnswReadStreamData;
-
 typedef struct HnswScanOpaqueData
 {
 	const		HnswTypeInfo *typeInfo;
@@ -424,13 +417,13 @@ 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, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples, bool maintenance);
+List	   *HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, 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, HnswSupport * support, int m, int efConstruction, bool existing, bool maintenance);
+void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing);
 HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, HnswQuery * q, Relation rel, HnswSupport * support, bool loadVec);
 void		HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building);
 void		HnswSetNeighborTuple(char *base, HnswNeighborTuple ntup, HnswElement e, int m);

src/hnswbuild.c

@@ -54,10 +54,6 @@
 #include "utils/datum.h"
 #include "utils/memutils.h"
 
-#if PG_VERSION_NUM >= 160000
-#include "varatt.h"
-#endif
-
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_progress.h"
 #else
@@ -135,6 +131,69 @@ HnswBuildAppendPage(Relation index, Buffer *buf, Page *page, ForkNumber forkNum)
 	HnswInitPage(*buf, *page);
 }
 
+static void
+WriteElement(char *base, HnswElement element, HnswElementTuple etup, HnswNeighborTuple ntup, HnswBuildState * buildstate, Size maxSize, Buffer *buf, Page *page, Relation index, ForkNumber forkNum)
+{
+	Size		etupSize;
+	Size		ntupSize;
+	Size		combinedSize;
+	Pointer		valuePtr;
+
+	/* Skip if already added */
+	if (BlockNumberIsValid(element->neighborPage))
+		return;
+
+	valuePtr = HnswPtrAccess(base, element->value);
+
+	/* Zero memory for each element */
+	MemSet(etup, 0, HNSW_TUPLE_ALLOC_SIZE);
+
+	/* Calculate sizes */
+	etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(valuePtr));
+	ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, buildstate->m);
+	combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
+
+	/* Initial size check */
+	if (etupSize > HNSW_TUPLE_ALLOC_SIZE)
+		ereport(ERROR,
+				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+				 errmsg("index tuple too large")));
+
+	HnswSetElementTuple(base, etup, element);
+
+	/* Keep element and neighbors on the same page if possible */
+	if (PageGetFreeSpace(*page) < etupSize || (combinedSize <= maxSize && PageGetFreeSpace(*page) < combinedSize))
+		HnswBuildAppendPage(index, buf, page, forkNum);
+
+	/* Calculate offsets */
+	element->blkno = BufferGetBlockNumber(*buf);
+	element->offno = OffsetNumberNext(PageGetMaxOffsetNumber(*page));
+	if (combinedSize <= maxSize)
+	{
+		element->neighborPage = element->blkno;
+		element->neighborOffno = OffsetNumberNext(element->offno);
+	}
+	else
+	{
+		element->neighborPage = element->blkno + 1;
+		element->neighborOffno = FirstOffsetNumber;
+	}
+
+	ItemPointerSet(&etup->neighbortid, element->neighborPage, element->neighborOffno);
+
+	/* Add element */
+	if (PageAddItem(*page, (Item) etup, etupSize, InvalidOffsetNumber, false, false) != element->offno)
+		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
+
+	/* Add new page if needed */
+	if (PageGetFreeSpace(*page) < ntupSize)
+		HnswBuildAppendPage(index, buf, page, forkNum);
+
+	/* Add placeholder for neighbors */
+	if (PageAddItem(*page, (Item) ntup, ntupSize, InvalidOffsetNumber, false, false) != element->neighborOffno)
+		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
+}
+
 /*
  * Create graph pages
  */
@@ -168,61 +227,24 @@ CreateGraphPages(HnswBuildState * buildstate)
 	while (!HnswPtrIsNull(base, iter))
 	{
 		HnswElement element = HnswPtrAccess(base, iter);
-		Size		etupSize;
-		Size		ntupSize;
-		Size		combinedSize;
-		Pointer		valuePtr = HnswPtrAccess(base, element->value);
 
 		/* Update iterator */
 		iter = element->next;
 
-		/* Zero memory for each element */
-		MemSet(etup, 0, HNSW_TUPLE_ALLOC_SIZE);
+		WriteElement(base, element, etup, ntup, buildstate, maxSize, &buf, &page, index, forkNum);
 
-		/* Calculate sizes */
-		etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(valuePtr));
-		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, buildstate->m);
-		combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
-
-		/* Initial size check */
-		if (etupSize > HNSW_TUPLE_ALLOC_SIZE)
-			ereport(ERROR,
-					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-					 errmsg("index tuple too large")));
-
-		HnswSetElementTuple(base, etup, element);
-
-		/* Keep element and neighbors on the same page if possible */
-		if (PageGetFreeSpace(page) < etupSize || (combinedSize <= maxSize && PageGetFreeSpace(page) < combinedSize))
-			HnswBuildAppendPage(index, &buf, &page, forkNum);
-
-		/* Calculate offsets */
-		element->blkno = BufferGetBlockNumber(buf);
-		element->offno = OffsetNumberNext(PageGetMaxOffsetNumber(page));
-		if (combinedSize <= maxSize)
+		for (int lc = element->level; lc >= 0; lc--)
 		{
-			element->neighborPage = element->blkno;
-			element->neighborOffno = OffsetNumberNext(element->offno);
+			HnswNeighborArray *neighbors = HnswGetNeighbors(base, element, lc);
+
+			for (int i = 0; i < neighbors->length; i++)
+			{
+				HnswCandidate *hc = &neighbors->items[i];
+				HnswElement hce = HnswPtrAccess(base, hc->element);
+
+				WriteElement(base, hce, etup, ntup, buildstate, maxSize, &buf, &page, index, forkNum);
+			}
 		}
-		else
-		{
-			element->neighborPage = element->blkno + 1;
-			element->neighborOffno = FirstOffsetNumber;
-		}
-
-		ItemPointerSet(&etup->neighbortid, element->neighborPage, element->neighborOffno);
-
-		/* Add element */
-		if (PageAddItem(page, (Item) etup, etupSize, InvalidOffsetNumber, false, false) != element->offno)
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-
-		/* Add new page if needed */
-		if (PageGetFreeSpace(page) < ntupSize)
-			HnswBuildAppendPage(index, &buf, &page, forkNum);
-
-		/* Add placeholder for neighbors */
-		if (PageAddItem(page, (Item) ntup, ntupSize, InvalidOffsetNumber, false, false) != element->neighborOffno)
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
 	}
 
 	insertPage = BufferGetBlockNumber(buf);
@@ -402,7 +424,7 @@ UpdateNeighborsInMemory(char *base, HnswSupport * support, HnswElement e, int m)
  * Update graph in memory
  */
 static void
-UpdateGraphInMemory(HnswSupport * support, HnswElement element, int m, 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;
@@ -461,10 +483,10 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false, true);
+	HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false);
 
 	/* Update graph in memory */
-	UpdateGraphInMemory(support, element, m, entryPoint, buildstate);
+	UpdateGraphInMemory(support, element, m, efConstruction, entryPoint, buildstate);
 
 	/* Release entry lock */
 	LWLockRelease(entryLock);
@@ -1058,7 +1080,7 @@ ComputeParallelWorkers(Relation heap, Relation index)
  * Build graph
  */
 static void
-BuildGraph(HnswBuildState * buildstate)
+BuildGraph(HnswBuildState * buildstate, ForkNumber forkNum)
 {
 	int			parallel_workers = 0;
 
@@ -1106,7 +1128,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 
 	InitBuildState(buildstate, heap, index, indexInfo, forkNum);
 
-	BuildGraph(buildstate);
+	BuildGraph(buildstate, forkNum);
 
 	if (RelationNeedsWAL(index) || forkNum == INIT_FORKNUM)
 		log_newpage_range(index, forkNum, 0, RelationGetNumberOfBlocksInFork(index, forkNum), true);

src/hnswinsert.c

@@ -9,10 +9,6 @@
 #include "utils/datum.h"
 #include "utils/memutils.h"
 
-#if PG_VERSION_NUM >= 160000
-#include "varatt.h"
-#endif
-
 /*
  * Get the insert page
  */
@@ -664,7 +660,7 @@ FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
  * Update graph on disk
  */
 static void
-UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, HnswElement entryPoint, bool building)
+UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
 {
 	BlockNumber newInsertPage = InvalidBlockNumber;
 
@@ -729,10 +725,10 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false, building);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false);
 
 	/* Update graph on disk */
-	UpdateGraphOnDisk(index, support, element, m, entryPoint, building);
+	UpdateGraphOnDisk(index, support, element, m, efConstruction, entryPoint, building);
 
 	/* Release lock */
 	UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);

src/hnswscan.c

@@ -37,11 +37,11 @@ GetScanItems(IndexScanDesc scan, Datum value)
 
 	for (int lc = entryPoint->level; lc >= 1; lc--)
 	{
-		w = HnswSearchLayer(base, q, ep, 1, lc, index, support, m, false, NULL, NULL, NULL, true, NULL, false);
+		w = HnswSearchLayer(base, q, ep, 1, lc, index, support, m, false, NULL, NULL, NULL, true, NULL);
 		ep = w;
 	}
 
-	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, &so->v, hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF ? &so->discarded : NULL, true, &so->tuples, false);
+	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, &so->v, hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF ? &so->discarded : NULL, true, &so->tuples);
 }
 
 /*
@@ -72,7 +72,7 @@ ResumeScanItems(IndexScanDesc scan)
 		ep = lappend(ep, sc);
 	}
 
-	return HnswSearchLayer(base, &so->q, ep, batch_size, 0, index, &so->support, so->m, false, NULL, &so->v, &so->discarded, false, &so->tuples, false);
+	return HnswSearchLayer(base, &so->q, ep, batch_size, 0, index, &so->support, so->m, false, NULL, &so->v, &so->discarded, false, &so->tuples);
 }
 
 /*
@@ -193,10 +193,6 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 
 		/* Count index scan for stats */
 		pgstat_count_index_scan(scan->indexRelation);
-#if PG_VERSION_NUM >= 180000
-		if (scan->instrument)
-			scan->instrument->nsearches++;
-#endif
 
 		/* Safety check */
 		if (scan->orderByData == NULL)

src/hnswutils.c

@@ -15,14 +15,6 @@
 #include "utils/memdebug.h"
 #include "utils/rel.h"
 
-#if PG_VERSION_NUM >= 160000
-#include "varatt.h"
-#endif
-
-#if PG_VERSION_NUM >= 170000
-#include "storage/read_stream.h"
-#endif
-
 #if PG_VERSION_NUM < 170000
 static inline uint64
 murmurhash64(uint64 data)
@@ -261,6 +253,7 @@ HnswInitElement(char *base, ItemPointer heaptid, int m, double ml, int maxLevel,
 	element->deleted = 0;
 	/* Start at one to make it easier to find issues */
 	element->version = 1;
+	element->neighborPage = InvalidBlockNumber;
 
 	HnswInitNeighbors(base, element, m, allocator);
 
@@ -533,12 +526,14 @@ HnswGetDistance(Datum a, Datum b, HnswSupport * support)
  * Load an element and optionally get its distance from q
  */
 static void
-HnswLoadElementImpl(Buffer buf, OffsetNumber offno, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance, HnswElement * element)
+HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance, HnswElement * element)
 {
+	Buffer		buf;
 	Page		page;
 	HnswElementTuple etup;
 
 	/* Read vector */
+	buf = ReadBuffer(index, blkno);
 	LockBuffer(buf, BUFFER_LOCK_SHARE);
 	page = BufferGetPage(buf);
 
@@ -559,7 +554,7 @@ HnswLoadElementImpl(Buffer buf, OffsetNumber offno, double *distance, HnswQuery
 	if (distance == NULL || maxDistance == NULL || *distance < *maxDistance)
 	{
 		if (*element == NULL)
-			*element = HnswInitElementFromBlock(BufferGetBlockNumber(buf), offno);
+			*element = HnswInitElementFromBlock(blkno, offno);
 
 		HnswLoadElementFromTuple(*element, etup, true, loadVec);
 	}
@@ -573,9 +568,7 @@ HnswLoadElementImpl(Buffer buf, OffsetNumber offno, double *distance, HnswQuery
 void
 HnswLoadElement(HnswElement element, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance)
 {
-	Buffer		buf = ReadBuffer(index, element->blkno);
-
-	HnswLoadElementImpl(buf, element->offno, distance, q, index, support, loadVec, maxDistance, &element);
+	HnswLoadElementImpl(element->blkno, element->offno, distance, q, index, support, loadVec, maxDistance, &element);
 }
 
 /*
@@ -815,31 +808,11 @@ HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *u
 	}
 }
 
-#if PG_VERSION_NUM >= 170000
-/*
- * Get next block number for read stream
- */
-static BlockNumber
-HnswReadStreamNextBlock(ReadStream *stream, void *callback_private_data, void *per_buffer_data)
-{
-	HnswReadStreamData *streamData = callback_private_data;
-	OffsetNumber *offno = per_buffer_data;
-	HnswUnvisited *uv;
-
-	if (streamData->visited == streamData->unvisitedLength)
-		return InvalidBlockNumber;
-
-	uv = &streamData->unvisited[streamData->visited++];
-	*offno = ItemPointerGetOffsetNumber(&uv->indextid);
-	return ItemPointerGetBlockNumber(&uv->indextid);
-}
-#endif
-
 /*
  * Algorithm 2 from paper
  */
 List *
-HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples, bool maintenance)
+HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples)
 {
 	List	   *w = NIL;
 	pairingheap *C = pairingheap_allocate(CompareNearestCandidates, NULL);
@@ -854,27 +827,6 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 	int			unvisitedLength;
 	bool		inMemory = index == NULL;
 
-#if PG_VERSION_NUM >= 170000
-	HnswReadStreamData streamData;
-	ReadStream *stream = NULL;
-
-	if (!inMemory)
-	{
-		int			flags = READ_STREAM_DEFAULT;
-
-		if (maintenance)
-		{
-			flags |= READ_STREAM_MAINTENANCE;
-		}
-
-#if PG_VERSION_NUM >= 180000
-		flags |= READ_STREAM_USE_BATCHING;
-#endif
-
-		stream = read_stream_begin_relation(flags, NULL, index, MAIN_FORKNUM, HnswReadStreamNextBlock, &streamData, sizeof(OffsetNumber));
-	}
-#endif
-
 	if (v == NULL)
 	{
 		v = &vh;
@@ -937,23 +889,13 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 		if (inMemory)
 			HnswLoadUnvisitedFromMemory(base, cElement, unvisited, &unvisitedLength, v, lc, localNeighborhood, neighborhoodSize);
 		else
-		{
 			HnswLoadUnvisitedFromDisk(cElement, unvisited, &unvisitedLength, v, index, m, lm, lc);
 
-#if PG_VERSION_NUM >= 170000
-			read_stream_resume(stream);
-
-			streamData.unvisited = unvisited;
-			streamData.unvisitedLength = unvisitedLength;
-			streamData.visited = 0;
-#endif
-		}
-
 		/* OK to count elements instead of tuples */
 		if (tuples != NULL)
 			(*tuples) += unvisitedLength;
 
-		for (int i = 0;; i++)
+		for (int i = 0; i < unvisitedLength; i++)
 		{
 			HnswElement eElement;
 			HnswSearchCandidate *e;
@@ -964,40 +906,18 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 
 			if (inMemory)
 			{
-				if (i == unvisitedLength)
-					break;
-
 				eElement = unvisited[i].element;
 				eDistance = GetElementDistance(base, eElement, q, support);
 			}
 			else
 			{
-				Buffer		buf;
-				OffsetNumber offno;
-
-#if PG_VERSION_NUM >= 170000
-				void	   *offnoPtr;
-
-				buf = read_stream_next_buffer(stream, &offnoPtr);
-
-				if (!BufferIsValid(buf))
-					break;
-
-				offno = *((OffsetNumber *) offnoPtr);
-#else
-				ItemPointer indextid;
-
-				if (i == unvisitedLength)
-					break;
-
-				indextid = &unvisited[i].indextid;
-				buf = ReadBuffer(index, ItemPointerGetBlockNumber(indextid));
-				offno = ItemPointerGetOffsetNumber(indextid);
-#endif
+				ItemPointer indextid = &unvisited[i].indextid;
+				BlockNumber blkno = ItemPointerGetBlockNumber(indextid);
+				OffsetNumber offno = ItemPointerGetOffsetNumber(indextid);
 
 				/* Avoid any allocations if not adding */
 				eElement = NULL;
-				HnswLoadElementImpl(buf, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
+				HnswLoadElementImpl(blkno, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
 
 				if (eElement == NULL)
 					continue;
@@ -1053,11 +973,6 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 		w = lappend(w, sc);
 	}
 
-#if PG_VERSION_NUM >= 170000
-	if (!inMemory)
-		read_stream_end(stream);
-#endif
-
 	return w;
 }
 
@@ -1353,7 +1268,7 @@ PrecomputeHash(char *base, HnswElement element)
  * Algorithm 1 from paper
  */
 void
-HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing, bool maintenance)
+HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing)
 {
 	List	   *ep;
 	List	   *w;
@@ -1380,7 +1295,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
 	/* 1st phase: greedy search to insert level */
 	for (int lc = entryLevel; lc >= level + 1; lc--)
 	{
-		w = HnswSearchLayer(base, &q, ep, 1, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL, maintenance);
+		w = HnswSearchLayer(base, &q, ep, 1, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL);
 		ep = w;
 	}
 
@@ -1399,7 +1314,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
 		List	   *lw = NIL;
 		ListCell   *lc2;
 
-		w = HnswSearchLayer(base, &q, ep, efConstruction, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL, maintenance);
+		w = HnswSearchLayer(base, &q, ep, efConstruction, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL);
 
 		/* Convert search candidates to candidates */
 		foreach(lc2, w)

src/hnswvacuum.c

@@ -9,10 +9,6 @@
 #include "storage/lmgr.h"
 #include "utils/memutils.h"
 
-#if PG_VERSION_NUM >= 160000
-#include "varatt.h"
-#endif
-
 #if PG_VERSION_NUM >= 180000
 #define vacuum_delay_point() vacuum_delay_point(false)
 #endif
@@ -212,7 +208,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 	element->heaptidsLength = 0;
 
 	/* Find neighbors for element, skipping itself */
-	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, true, true);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, true);
 
 	/* Zero memory for each element */
 	MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);

src/ivfbuild.c

@@ -20,10 +20,6 @@
 #include "utils/memutils.h"
 #include "vector.h"
 
-#if PG_VERSION_NUM >= 160000
-#include "varatt.h"
-#endif
-
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_progress.h"
 #else
@@ -142,7 +138,7 @@ SampleRows(IvfflatBuildState * buildstate)
  * Add tuple to sort
  */
 static void
-AddTupleToSort(ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
+AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
 {
 	double		distance;
 	double		minDistance = DBL_MAX;
@@ -219,7 +215,7 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
 	oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
 
 	/* Add tuple to sort */
-	AddTupleToSort(tid, values, buildstate);
+	AddTupleToSort(index, tid, values, buildstate);
 
 	/* Reset memory context */
 	MemoryContextSwitchTo(oldCtx);
@@ -474,8 +470,8 @@ CreateMetaPage(Relation index, int dimensions, int lists, ForkNumber forkNum)
  * Create list pages
  */
 static void
-CreateListPages(Relation index, VectorArray centers, int lists,
-				ForkNumber forkNum, ListInfo * *listInfo)
+CreateListPages(Relation index, VectorArray centers, int dimensions,
+				int lists, ForkNumber forkNum, ListInfo * *listInfo)
 {
 	Buffer		buf;
 	Page		page;
@@ -1008,7 +1004,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 
 	/* Create pages */
 	CreateMetaPage(index, buildstate->dimensions, buildstate->lists, forkNum);
-	CreateListPages(index, buildstate->centers, buildstate->lists, forkNum, &buildstate->listInfo);
+	CreateListPages(index, buildstate->centers, buildstate->dimensions, buildstate->lists, forkNum, &buildstate->listInfo);
 	CreateEntryPages(buildstate, forkNum);
 
 	/* Write WAL for initialization fork since GenericXLog functions do not */

src/ivfflat.h

@@ -13,10 +13,6 @@
 #include "utils/tuplesort.h"
 #include "vector.h"
 
-#if PG_VERSION_NUM >= 160000
-#include "varatt.h"
-#endif
-
 #if PG_VERSION_NUM >= 150000
 #include "common/pg_prng.h"
 #endif

src/ivfinsert.c

@@ -65,7 +65,7 @@ FindInsertPage(Relation index, Datum *values, BlockNumber *insertPage, ListInfo
  * Insert a tuple into the index
  */
 static void
-InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid)
+InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel)
 {
 	const		IvfflatTypeInfo *typeInfo = IvfflatGetTypeInfo(index);
 	IndexTuple	itup;
@@ -204,7 +204,7 @@ ivfflatinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid,
 	oldCtx = MemoryContextSwitchTo(insertCtx);
 
 	/* Insert tuple */
-	InsertTuple(index, values, isnull, heap_tid);
+	InsertTuple(index, values, isnull, heap_tid, heap);
 
 	/* Delete memory context */
 	MemoryContextSwitchTo(oldCtx);

src/ivfkmeans.c

@@ -13,10 +13,6 @@
 #include "utils/memutils.h"
 #include "vector.h"
 
-#if PG_VERSION_NUM >= 160000
-#include "varatt.h"
-#endif
-
 /*
  * Initialize with kmeans++
  *

src/ivfscan.c

@@ -355,10 +355,6 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 
 		/* Count index scan for stats */
 		pgstat_count_index_scan(scan->indexRelation);
-#if PG_VERSION_NUM >= 180000
-		if (scan->instrument)
-			scan->instrument->nsearches++;
-#endif
 
 		/* Safety check */
 		if (scan->orderByData == NULL)