Use maintenance_io_concurrency where needed [skip ci]

This reverts commit 33ca8a61e2.

.dockerignore

@@ -1,8 +1,15 @@
 /.git/
 /dist/
+/log/
 /results/
 /tmp_check/
 /sql/vector--?.?.?.sql
 regression.*
 *.o
 *.so
+*.bc
+*.dll
+*.dylib
+*.obj
+*.lib
+*.exp

.github/workflows/build.yml

@@ -13,13 +13,13 @@ jobs:
           - postgres: 17
             os: ubuntu-24.04
           - postgres: 16
-            os: ubuntu-22.04
+            os: ubuntu-24.04-arm
           - postgres: 15
             os: ubuntu-22.04
           - postgres: 14
-            os: ubuntu-20.04
+            os: ubuntu-22.04-arm
           - postgres: 13
-            os: ubuntu-20.04
+            os: ubuntu-22.04
     steps:
       - uses: actions/checkout@v4
       - uses: ankane/setup-postgres@v1
@@ -28,7 +28,7 @@ jobs:
           dev-files: true
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
+          PG_CFLAGS: ${{ 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
@@ -46,8 +46,8 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - postgres: 16
-            os: macos-14
+          - postgres: 17
+            os: macos-15
           - postgres: 14
             os: macos-13
     steps:
@@ -57,7 +57,7 @@ jobs:
           postgres-version: ${{ matrix.postgres }}
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter
+          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-unknown-warning-option
       - run: make install
       - run: make installcheck
       - if: ${{ failure() }}
@@ -70,12 +70,13 @@ jobs:
           tar xf $TAG.tar.gz
           mv postgres-$TAG postgres
         env:
-          TAG: ${{ matrix.postgres == 16 && 'REL_16_2' || 'REL_14_11' }}
+          TAG: ${{ matrix.postgres == 17 && 'REL_17_2' || 'REL_14_15' }}
       - run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl"
         env:
           PERL5LIB: /Users/runner/perl5/lib/perl5
-      - run: make clean && $(brew --prefix llvm@15)/bin/scan-build --status-bugs make
+      - run: make clean && $(brew --prefix llvm@$LLVM_VERSION)/bin/scan-build --status-bugs make
         env:
+          LLVM_VERSION: ${{ matrix.os == 'macos-15' && 18 || 15 }}
           PG_CFLAGS: -DUSE_ASSERT_CHECKING
   windows:
     runs-on: windows-latest
@@ -125,7 +126,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: ankane/setup-postgres-valgrind@v1
         with:
-          postgres-version: 16
+          postgres-version: 17
           check-ub: yes
       - run: make OPTFLAGS=""
       - run: sudo --preserve-env=PG_CONFIG make install

CHANGELOG.md

@@ -1,3 +1,7 @@
+## 0.8.1 (unreleased)
+
+- Improved performance of HNSW index scans for Postgres 17
+
 ## 0.8.0 (2024-10-30)
 
 - Added support for iterative index scans

LICENSE

@@ -1,4 +1,4 @@
-Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
+Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
 
 Portions Copyright (c) 1994, The Regents of the University of California
 

README.md

@@ -33,18 +33,10 @@ You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#p
 
 ### Windows
 
-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:
+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
-call "C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvars64.bat"
-```
-
-Note: The exact path will vary depending on your Visual Studio version and edition
-
-Then use `nmake` to build:
-
-```cmd
-set "PGROOT=C:\Program Files\PostgreSQL\16"
+set "PGROOT=C:\Program Files\PostgreSQL\17"
 cd %TEMP%
 git clone --branch v0.8.0 https://github.com/pgvector/pgvector.git
 cd pgvector
@@ -52,8 +44,6 @@ 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).
@@ -84,7 +74,7 @@ Get the nearest neighbors by L2 distance
 SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
-Also supports inner product (`<#>`), cosine distance (`<=>`), and L1 distance (`<+>`, added in 0.7.0)
+Also supports inner product (`<#>`), cosine distance (`<=>`), and L1 distance (`<+>`)
 
 Note: `<#>` returns the negative inner product since Postgres only supports `ASC` order index scans on operators
 
@@ -148,9 +138,9 @@ Supported distance functions are:
 - `<->` - L2 distance
 - `<#>` - (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)
+- `<+>` - L1 distance
+- `<~>` - Hamming distance (binary vectors)
+- `<%>` - Jaccard distance (binary vectors)
 
 Get the nearest neighbors to a row
 
@@ -237,19 +227,19 @@ Cosine distance
 CREATE INDEX ON items USING hnsw (embedding vector_cosine_ops);
 ```
 
-L1 distance - added in 0.7.0
+L1 distance
 
 ```sql
 CREATE INDEX ON items USING hnsw (embedding vector_l1_ops);
 ```
 
-Hamming distance - added in 0.7.0
+Hamming distance
 
 ```sql
 CREATE INDEX ON items USING hnsw (embedding bit_hamming_ops);
 ```
 
-Jaccard distance - added in 0.7.0
+Jaccard distance
 
 ```sql
 CREATE INDEX ON items USING hnsw (embedding bit_jaccard_ops);
@@ -258,9 +248,9 @@ CREATE INDEX ON items USING hnsw (embedding bit_jaccard_ops);
 Supported types are:
 
 - `vector` - up to 2,000 dimensions
-- `halfvec` - up to 4,000 dimensions (added in 0.7.0)
-- `bit` - up to 64,000 dimensions (added in 0.7.0)
-- `sparsevec` - up to 1,000 non-zero elements (added in 0.7.0)
+- `halfvec` - up to 4,000 dimensions
+- `bit` - up to 64,000 dimensions
+- `sparsevec` - up to 1,000 non-zero elements
 
 ### Index Options
 
@@ -314,13 +304,15 @@ Note: Do not set `maintenance_work_mem` so high that it exhausts the memory on t
 
 Like other index types, it’s faster to create an index after loading your initial data
 
-Starting with 0.6.0, you can also speed up index creation by increasing the number of parallel workers (2 by default)
+You can also speed up index creation by increasing the number of parallel workers (2 by default)
 
 ```sql
 SET max_parallel_maintenance_workers = 7; -- plus leader
 ```
 
-For a large number of workers, you may also need to increase `max_parallel_workers` (8 by default)
+For a large number of workers, you may need to increase `max_parallel_workers` (8 by default)
+
+The [index options](#index-options) also have a significant impact on build time (use the defaults unless seeing low recall)
 
 ### Indexing Progress
 
@@ -367,7 +359,7 @@ Cosine distance
 CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);
 ```
 
-Hamming distance - added in 0.7.0
+Hamming distance
 
 ```sql
 CREATE INDEX ON items USING ivfflat (embedding bit_hamming_ops) WITH (lists = 100);
@@ -376,8 +368,8 @@ CREATE INDEX ON items USING ivfflat (embedding bit_hamming_ops) WITH (lists = 10
 Supported types are:
 
 - `vector` - up to 2,000 dimensions
-- `halfvec` - up to 4,000 dimensions (added in 0.7.0)
-- `bit` - up to 64,000 dimensions (added in 0.7.0)
+- `halfvec` - up to 4,000 dimensions
+- `bit` - up to 64,000 dimensions
 
 ### Query Options
 
@@ -477,8 +469,6 @@ CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(cate
 
 ## Iterative Index Scans
 
-*Added in 0.8.0*
-
 With approximate indexes, queries with filtering can return less results since filtering is applied *after* the index is scanned. Starting with 0.8.0, you can enable iterative index scans, which will automatically scan more of the index until enough results are found (or it reaches `hnsw.max_scan_tuples` or `ivfflat.max_probes`).
 
 Iterative scans can use strict or relaxed ordering.
@@ -502,9 +492,11 @@ With relaxed ordering, you can use a [materialized CTE](https://www.postgresql.o
 ```sql
 WITH relaxed_results AS MATERIALIZED (
     SELECT id, embedding <-> '[1,2,3]' AS distance FROM items WHERE category_id = 123 ORDER BY distance LIMIT 5
-) SELECT * FROM relaxed_results ORDER BY distance;
+) SELECT * FROM relaxed_results ORDER BY distance + 0;
 ```
 
+Note: `+ 0` is needed for Postgres 17+
+
 For queries that filter by distance, use a materialized CTE and place the distance filter outside of it for best performance (due to the [current behavior](https://www.postgresql.org/message-id/flat/CAOdR5yGUoMQ6j7M5hNUXrySzaqZVGf_Ne%2B8fwZMRKTFxU1nbJg%40mail.gmail.com) of the Postgres executor)
 
 ```sql
@@ -549,8 +541,6 @@ Note: If this is lower than `ivfflat.probes`, `ivfflat.probes` will be used
 
 ## Half-Precision Vectors
 
-*Added in 0.7.0*
-
 Use the `halfvec` type to store half-precision vectors
 
 ```sql
@@ -559,8 +549,6 @@ CREATE TABLE items (id bigserial PRIMARY KEY, embedding halfvec(3));
 
 ## Half-Precision Indexing
 
-*Added in 0.7.0*
-
 Index vectors at half precision for smaller indexes
 
 ```sql
@@ -582,24 +570,16 @@ CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
 INSERT INTO items (embedding) VALUES ('000'), ('111');
 ```
 
-Get the nearest neighbors by Hamming distance (added in 0.7.0)
+Get the nearest neighbors by Hamming distance
 
 ```sql
 SELECT * FROM items ORDER BY embedding <~> '101' LIMIT 5;
 ```
 
-Or (before 0.7.0)
-
-```sql
-SELECT * FROM items ORDER BY bit_count(embedding # '101') LIMIT 5;
-```
-
 Also supports Jaccard distance (`<%>`)
 
 ## Binary Quantization
 
-*Added in 0.7.0*
-
 Use expression indexing for binary quantization
 
 ```sql
@@ -622,8 +602,6 @@ SELECT * FROM (
 
 ## Sparse Vectors
 
-*Added in 0.7.0*
-
 Use the `sparsevec` type to store sparse vectors
 
 ```sql
@@ -657,8 +635,6 @@ You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python
 
 ## Indexing Subvectors
 
-*Added in 0.7.0*
-
 Use expression indexing to index subvectors
 
 ```sql
@@ -772,8 +748,6 @@ SELECT query, calls, ROUND((total_plan_time + total_exec_time) / calls) AS avg_t
     FROM pg_stat_statements ORDER BY total_plan_time + total_exec_time DESC LIMIT 20;
 ```
 
-Note: Replace `total_plan_time + total_exec_time` with `total_time` for Postgres < 13
-
 Monitor recall by comparing results from approximate search with exact search.
 
 ```sql
@@ -801,13 +775,17 @@ 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)
 Crystal | [pgvector-crystal](https://github.com/pgvector/pgvector-crystal)
+D | [pgvector-d](https://github.com/pgvector/pgvector-d)
 Dart | [pgvector-dart](https://github.com/pgvector/pgvector-dart)
 Elixir | [pgvector-elixir](https://github.com/pgvector/pgvector-elixir)
+Erlang | [pgvector-erlang](https://github.com/pgvector/pgvector-erlang)
+Fortran | [pgvector-fortran](https://github.com/pgvector/pgvector-fortran)
+Gleam | [pgvector-gleam](https://github.com/pgvector/pgvector-gleam)
 Go | [pgvector-go](https://github.com/pgvector/pgvector-go)
 Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell)
 Java, Kotlin, Groovy, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
 JavaScript, TypeScript | [pgvector-node](https://github.com/pgvector/pgvector-node)
-Julia | [pgvector-julia](https://github.com/pgvector/pgvector-julia)
+Julia | [Pgvector.jl](https://github.com/pgvector/Pgvector.jl)
 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)
@@ -816,6 +794,7 @@ 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)
 R | [pgvector-r](https://github.com/pgvector/pgvector-r)
+Raku | [pgvector-raku](https://github.com/pgvector/pgvector-raku)
 Ruby | [pgvector-ruby](https://github.com/pgvector/pgvector-ruby), [Neighbor](https://github.com/ankane/neighbor)
 Rust | [pgvector-rust](https://github.com/pgvector/pgvector-rust)
 Swift | [pgvector-swift](https://github.com/pgvector/pgvector-swift)
@@ -837,7 +816,7 @@ You can use [half-precision indexing](#half-precision-indexing) to index up to 4
 
 #### Can I store vectors with different dimensions in the same column?
 
-You can use `vector` as the type (instead of `vector(3)`).
+You can use `vector` as the type (instead of `vector(n)`).
 
 ```sql
 CREATE TABLE embeddings (model_id bigint, item_id bigint, embedding vector, PRIMARY KEY (model_id, item_id));
@@ -937,7 +916,7 @@ ALTER TABLE items ALTER COLUMN embedding SET STORAGE PLAIN;
 
 #### Why are there less results for a query after adding an HNSW index?
 
-Results are limited by the size of the dynamic candidate list (`hnsw.ef_search`). There may be even less results due to dead tuples or filtering conditions in the query. We recommend setting `hnsw.ef_search` to at least twice the `LIMIT` of the query. If you need more than 500 results, use an IVFFlat index instead.
+Results are limited by the size of the dynamic candidate list (`hnsw.ef_search`), which is 40 by default. There may be even less results due to dead tuples or filtering conditions in the query. Enabling [iterative index scans](#iterative-index-scans) can help address this.
 
 Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
 
@@ -949,7 +928,7 @@ The index was likely created with too little data for the number of lists. Drop
 DROP INDEX index_name;
 ```
 
-Results can also be limited by the number of probes (`ivfflat.probes`).
+Results can also be limited by the number of probes (`ivfflat.probes`). Enabling [iterative index scans](#iterative-index-scans) can address this.
 
 Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
 
@@ -1116,7 +1095,13 @@ Note: Replace `17` with your Postgres server version
 
 ### Missing SDK
 
-If compilation fails and the output includes `warning: no such sysroot directory` on Mac, reinstall Xcode Command Line Tools.
+If compilation fails and the output includes `warning: no such sysroot directory` on Mac, your Postgres installation points to a path that no longer exists.
+
+```sh
+pg_config --cppflags
+```
+
+Reinstall Postgres to fix this.
 
 ### Portability
 
@@ -1134,6 +1119,14 @@ make OPTFLAGS=""
 
 If compilation fails with `Cannot open include file: 'postgres.h': No such file or directory`, make sure `PGROOT` is correct.
 
+### Mismatched Architecture
+
+If compilation fails with `error C2196: case value '4' already used`, make sure you’re using the `x64 Native Tools Command Prompt`. Then run `nmake /F Makefile.win clean` and re-run the installation instructions.
+
+### Missing Symbol
+
+If linking fails with `unresolved external symbol float_to_shortest_decimal_bufn` with Postgres 17.0-17.2, upgrade to Postgres 17.3+.
+
 ### Permissions
 
 If installation fails with `Access is denied`, re-run the installation instructions as an administrator.
@@ -1158,6 +1151,12 @@ cd 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.
+
+```sh
+docker run --shm-size=1g ...
+```
+
 ### Homebrew
 
 With Homebrew Postgres, you can use:
@@ -1203,7 +1202,7 @@ Note: Replace `17` with your Postgres server version
 Install the FreeBSD package with:
 
 ```sh
-pkg install postgresql15-pgvector
+pkg install postgresql16-pgvector
 ```
 
 or the port with:
@@ -1245,36 +1244,6 @@ You can check the version in the current database with:
 SELECT extversion FROM pg_extension WHERE extname = 'vector';
 ```
 
-## Upgrade Notes
-
-### 0.6.0
-
-#### Postgres 12
-
-If upgrading with Postgres 12, remove this line from `sql/vector--0.5.1--0.6.0.sql`:
-
-```sql
-ALTER TYPE vector SET (STORAGE = external);
-```
-
-Then run `make install` and `ALTER EXTENSION vector UPDATE;`.
-
-#### Docker
-
-The Docker image is now published in the `pgvector` org, and there are tags for each supported version of Postgres (rather than a `latest` tag).
-
-```sh
-docker pull pgvector/pgvector:pg16
-# or
-docker pull pgvector/pgvector:0.6.0-pg16
-```
-
-Also, if you’ve increased `maintenance_work_mem`, make sure `--shm-size` is at least that size to avoid an error with parallel HNSW index builds.
-
-```sh
-docker run --shm-size=1g ...
-```
-
 ## Thanks
 
 Thanks to:

src/hnsw.c

@@ -259,6 +259,11 @@ hnswhandler(PG_FUNCTION_ARGS)
 	amroutine->amoptsprocnum = 0;
 	amroutine->amcanorder = false;
 	amroutine->amcanorderbyop = true;
+#if PG_VERSION_NUM >= 180000
+	amroutine->amcanhash = false;
+	amroutine->amconsistentequality = false;
+	amroutine->amconsistentordering = false;
+#endif
 	amroutine->amcanbackward = false;	/* can change direction mid-scan */
 	amroutine->amcanunique = false;
 	amroutine->amcanmulticol = false;
@@ -291,6 +296,9 @@ hnswhandler(PG_FUNCTION_ARGS)
 	amroutine->amvacuumcleanup = hnswvacuumcleanup;
 	amroutine->amcanreturn = NULL;
 	amroutine->amcostestimate = hnswcostestimate;
+#if PG_VERSION_NUM >= 180000
+	amroutine->amgettreeheight = NULL;
+#endif
 	amroutine->amoptions = hnswoptions;
 	amroutine->amproperty = NULL;	/* TODO AMPROP_DISTANCE_ORDERABLE */
 	amroutine->ambuildphasename = hnswbuildphasename;
@@ -311,5 +319,10 @@ hnswhandler(PG_FUNCTION_ARGS)
 	amroutine->aminitparallelscan = NULL;
 	amroutine->amparallelrescan = NULL;
 
+#if PG_VERSION_NUM >= 180000
+	amroutine->amtranslatestrategy = NULL;
+	amroutine->amtranslatecmptype = NULL;
+#endif
+
 	PG_RETURN_POINTER(amroutine);
 }

src/hnsw.h

@@ -126,7 +126,7 @@ typedef struct HnswNeighborArray HnswNeighborArray;
 
 #define HnswPtrDeclare(type, relptrtype, ptrtype) \
 	relptr_declare(type, relptrtype); \
-	typedef union { type *ptr; relptrtype relptr; } ptrtype;
+	typedef union { type *ptr; relptrtype relptr; } ptrtype
 
 /* Pointers that can be absolute or relative */
 /* Use char for DatumPtr so works with Pointer */
@@ -362,6 +362,13 @@ typedef union
 	ItemPointerData indextid;
 }			HnswUnvisited;
 
+typedef struct HnswReadStreamData
+{
+	HnswUnvisited *unvisited;
+	int			unvisitedLength;
+	int			visited;
+}			HnswReadStreamData;
+
 typedef struct HnswScanOpaqueData
 {
 	const		HnswTypeInfo *typeInfo;
@@ -417,13 +424,13 @@ 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);
+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);
 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);
+void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing, bool maintenance);
 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

@@ -457,7 +457,7 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false);
+	HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false, true);
 
 	/* Update graph in memory */
 	UpdateGraphInMemory(support, element, m, efConstruction, entryPoint, buildstate);

src/hnswinsert.c

@@ -725,7 +725,7 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false, building);
 
 	/* Update graph on disk */
 	UpdateGraphOnDisk(index, support, element, m, efConstruction, entryPoint, building);

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);
+		w = HnswSearchLayer(base, q, ep, 1, lc, index, support, m, false, NULL, NULL, NULL, true, NULL, false);
 		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);
+	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);
 }
 
 /*
@@ -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);
+	return HnswSearchLayer(base, &so->q, ep, batch_size, 0, index, &so->support, so->m, false, NULL, &so->v, &so->discarded, false, &so->tuples, false);
 }
 
 /*

src/hnswutils.c

@@ -15,6 +15,10 @@
 #include "utils/memdebug.h"
 #include "utils/rel.h"
 
+#if PG_VERSION_NUM >= 170000
+#include "storage/read_stream.h"
+#endif
+
 #if PG_VERSION_NUM < 170000
 static inline uint64
 murmurhash64(uint64 data)
@@ -525,14 +529,12 @@ HnswGetDistance(Datum a, Datum b, HnswSupport * support)
  * Load an element and optionally get its distance from q
  */
 static void
-HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance, HnswElement * element)
+HnswLoadElementImpl(Buffer buf, 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);
 
@@ -553,7 +555,7 @@ HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, Hns
 	if (distance == NULL || maxDistance == NULL || *distance < *maxDistance)
 	{
 		if (*element == NULL)
-			*element = HnswInitElementFromBlock(blkno, offno);
+			*element = HnswInitElementFromBlock(BufferGetBlockNumber(buf), offno);
 
 		HnswLoadElementFromTuple(*element, etup, true, loadVec);
 	}
@@ -567,7 +569,9 @@ HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, Hns
 void
 HnswLoadElement(HnswElement element, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance)
 {
-	HnswLoadElementImpl(element->blkno, element->offno, distance, q, index, support, loadVec, maxDistance, &element);
+	Buffer		buf = ReadBuffer(index, element->blkno);
+
+	HnswLoadElementImpl(buf, element->offno, distance, q, index, support, loadVec, maxDistance, &element);
 }
 
 /*
@@ -807,11 +811,31 @@ 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)
+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	   *w = NIL;
 	pairingheap *C = pairingheap_allocate(CompareNearestCandidates, NULL);
@@ -826,6 +850,18 @@ 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 = maintenance ? READ_STREAM_MAINTENANCE : READ_STREAM_DEFAULT;
+
+		stream = read_stream_begin_relation(flags, NULL, index, MAIN_FORKNUM, HnswReadStreamNextBlock, &streamData, sizeof(OffsetNumber));
+	}
+#endif
+
 	if (v == NULL)
 	{
 		v = &vh;
@@ -888,8 +924,18 @@ 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_reset(stream);
+
+			streamData.unvisited = unvisited;
+			streamData.unvisitedLength = unvisitedLength;
+			streamData.visited = 0;
+#endif
+		}
+
 		/* OK to count elements instead of tuples */
 		if (tuples != NULL)
 			(*tuples) += unvisitedLength;
@@ -910,13 +956,24 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 			}
 			else
 			{
+				Buffer		buf;
+				OffsetNumber offno;
+
+#if PG_VERSION_NUM >= 170000
+				void	   *offnoPtr;
+
+				buf = read_stream_next_buffer(stream, &offnoPtr);
+				offno = *((OffsetNumber *) offnoPtr);
+#else
 				ItemPointer indextid = &unvisited[i].indextid;
-				BlockNumber blkno = ItemPointerGetBlockNumber(indextid);
-				OffsetNumber offno = ItemPointerGetOffsetNumber(indextid);
+
+				buf = ReadBuffer(index, ItemPointerGetBlockNumber(indextid));
+				offno = ItemPointerGetOffsetNumber(indextid);
+#endif
 
 				/* Avoid any allocations if not adding */
 				eElement = NULL;
-				HnswLoadElementImpl(blkno, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
+				HnswLoadElementImpl(buf, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
 
 				if (eElement == NULL)
 					continue;
@@ -972,6 +1029,11 @@ 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;
 }
 
@@ -1267,7 +1329,7 @@ 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)
+HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing, bool maintenance)
 {
 	List	   *ep;
 	List	   *w;
@@ -1294,7 +1356,7 @@ 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);
+		w = HnswSearchLayer(base, &q, ep, 1, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL, maintenance);
 		ep = w;
 	}
 
@@ -1313,7 +1375,7 @@ 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);
+		w = HnswSearchLayer(base, &q, ep, efConstruction, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL, maintenance);
 
 		/* Convert search candidates to candidates */
 		foreach(lc2, w)
@@ -1393,7 +1455,7 @@ hnsw_halfvec_support(PG_FUNCTION_ARGS)
 	};
 
 	PG_RETURN_POINTER(&typeInfo);
-};
+}
 
 FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnsw_bit_support);
 Datum
@@ -1406,7 +1468,7 @@ hnsw_bit_support(PG_FUNCTION_ARGS)
 	};
 
 	PG_RETURN_POINTER(&typeInfo);
-};
+}
 
 FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnsw_sparsevec_support);
 Datum
@@ -1419,4 +1481,4 @@ hnsw_sparsevec_support(PG_FUNCTION_ARGS)
 	};
 
 	PG_RETURN_POINTER(&typeInfo);
-};
+}

src/hnswvacuum.c

@@ -9,6 +9,10 @@
 #include "storage/lmgr.h"
 #include "utils/memutils.h"
 
+#if PG_VERSION_NUM >= 180000
+#define vacuum_delay_point() vacuum_delay_point(false)
+#endif
+
 /*
  * Check if deleted list contains an index TID
  */
@@ -204,7 +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);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, true, true);
 
 	/* Zero memory for each element */
 	MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);

src/ivfbuild.c

@@ -360,7 +360,7 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 	buildstate->sortdesc = CreateTemplateTupleDesc(3);
 	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
 	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
-	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 3, "vector", buildstate->tupdesc->attrs[0].atttypid, -1, 0);
+	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 3, "vector", TupleDescAttr(buildstate->tupdesc, 0)->atttypid, -1, 0);
 
 	buildstate->slot = MakeSingleTupleTableSlot(buildstate->sortdesc, &TTSOpsVirtual);
 
@@ -1023,6 +1023,10 @@ ivfflatbuild(Relation heap, Relation index, IndexInfo *indexInfo)
 	IndexBuildResult *result;
 	IvfflatBuildState buildstate;
 
+#ifdef IVFFLAT_BENCH
+	SeedRandom(42);
+#endif
+
 	BuildIndex(heap, index, indexInfo, &buildstate, MAIN_FORKNUM);
 
 	result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult));

src/ivfflat.c

@@ -186,6 +186,11 @@ ivfflathandler(PG_FUNCTION_ARGS)
 	amroutine->amoptsprocnum = 0;
 	amroutine->amcanorder = false;
 	amroutine->amcanorderbyop = true;
+#if PG_VERSION_NUM >= 180000
+	amroutine->amcanhash = false;
+	amroutine->amconsistentequality = false;
+	amroutine->amconsistentordering = false;
+#endif
 	amroutine->amcanbackward = false;	/* can change direction mid-scan */
 	amroutine->amcanunique = false;
 	amroutine->amcanmulticol = false;
@@ -218,6 +223,9 @@ ivfflathandler(PG_FUNCTION_ARGS)
 	amroutine->amvacuumcleanup = ivfflatvacuumcleanup;
 	amroutine->amcanreturn = NULL;	/* tuple not included in heapsort */
 	amroutine->amcostestimate = ivfflatcostestimate;
+#if PG_VERSION_NUM >= 180000
+	amroutine->amgettreeheight = NULL;
+#endif
 	amroutine->amoptions = ivfflatoptions;
 	amroutine->amproperty = NULL;	/* TODO AMPROP_DISTANCE_ORDERABLE */
 	amroutine->ambuildphasename = ivfflatbuildphasename;
@@ -238,5 +246,10 @@ ivfflathandler(PG_FUNCTION_ARGS)
 	amroutine->aminitparallelscan = NULL;
 	amroutine->amparallelrescan = NULL;
 
+#if PG_VERSION_NUM >= 180000
+	amroutine->amtranslatestrategy = NULL;
+	amroutine->amtranslatecmptype = NULL;
+#endif
+
 	PG_RETURN_POINTER(amroutine);
 }

src/ivfflat.h

@@ -73,9 +73,11 @@
 #if PG_VERSION_NUM >= 150000
 #define RandomDouble() pg_prng_double(&pg_global_prng_state)
 #define RandomInt() pg_prng_uint32(&pg_global_prng_state)
+#define SeedRandom(seed) pg_prng_seed(&pg_global_prng_state, seed)
 #else
 #define RandomDouble() (((double) random()) / MAX_RANDOM_VALUE)
 #define RandomInt() random()
+#define SeedRandom(seed) srandom(seed)
 #endif
 
 /* Variables */

src/ivfutils.c

@@ -259,8 +259,8 @@ VectorUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
 	vec->dim = dimensions;
 
-	for (int k = 0; k < dimensions; k++)
-		vec->x[k] = x[k];
+	for (int i = 0; i < dimensions; i++)
+		vec->x[i] = x[i];
 }
 
 static void
@@ -271,8 +271,8 @@ HalfvecUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, HALFVEC_SIZE(dimensions));
 	vec->dim = dimensions;
 
-	for (int k = 0; k < dimensions; k++)
-		vec->x[k] = Float4ToHalfUnchecked(x[k]);
+	for (int i = 0; i < dimensions; i++)
+		vec->x[i] = Float4ToHalfUnchecked(x[i]);
 }
 
 static void
@@ -284,29 +284,33 @@ BitUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, VARBITTOTALLEN(dimensions));
 	VARBITLEN(vec) = dimensions;
 
-	for (uint32 k = 0; k < VARBITBYTES(vec); k++)
-		nx[k] = 0;
+	for (uint32 i = 0; i < VARBITBYTES(vec); i++)
+		nx[i] = 0;
 
-	for (int k = 0; k < dimensions; k++)
-		nx[k / 8] |= (x[k] > 0.5 ? 1 : 0) << (7 - (k % 8));
+	for (int i = 0; i < dimensions; i++)
+		nx[i / 8] |= (x[i] > 0.5 ? 1 : 0) << (7 - (i % 8));
 }
 
 static void
 VectorSumCenter(Pointer v, float *x)
 {
 	Vector	   *vec = (Vector *) v;
+	int			dim = vec->dim;
 
-	for (int k = 0; k < vec->dim; k++)
-		x[k] += vec->x[k];
+	/* Auto-vectorized */
+	for (int i = 0; i < dim; i++)
+		x[i] += vec->x[i];
 }
 
 static void
 HalfvecSumCenter(Pointer v, float *x)
 {
 	HalfVector *vec = (HalfVector *) v;
+	int			dim = vec->dim;
 
-	for (int k = 0; k < vec->dim; k++)
-		x[k] += HalfToFloat4(vec->x[k]);
+	/* Auto-vectorized on aarch64 */
+	for (int i = 0; i < dim; i++)
+		x[i] += HalfToFloat4(vec->x[i]);
 }
 
 static void
@@ -314,8 +318,8 @@ BitSumCenter(Pointer v, float *x)
 {
 	VarBit	   *vec = (VarBit *) v;
 
-	for (int k = 0; k < VARBITLEN(vec); k++)
-		x[k] += (float) (((VARBITS(vec)[k / 8]) >> (7 - (k % 8))) & 0x01);
+	for (int i = 0; i < VARBITLEN(vec); i++)
+		x[i] += (float) (((VARBITS(vec)[i / 8]) >> (7 - (i % 8))) & 0x01);
 }
 
 /*
@@ -355,7 +359,7 @@ ivfflat_halfvec_support(PG_FUNCTION_ARGS)
 	};
 
 	PG_RETURN_POINTER(&typeInfo);
-};
+}
 
 FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_bit_support);
 Datum
@@ -370,4 +374,4 @@ ivfflat_bit_support(PG_FUNCTION_ARGS)
 	};
 
 	PG_RETURN_POINTER(&typeInfo);
-};
+}

src/ivfvacuum.c

@@ -5,6 +5,10 @@
 #include "ivfflat.h"
 #include "storage/bufmgr.h"
 
+#if PG_VERSION_NUM >= 180000
+#define vacuum_delay_point() vacuum_delay_point(false)
+#endif
+
 /*
  * Bulk delete tuples from the index
  */

src/vector.c

@@ -916,15 +916,19 @@ vector_concat(PG_FUNCTION_ARGS)
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 	Vector	   *result;
 	int			dim = a->dim + b->dim;
+	int			dim_a = a->dim;
+	int			dim_b = b->dim;
 
 	CheckDim(dim);
 	result = InitVector(dim);
 
-	for (int i = 0; i < a->dim; i++)
+	/* Auto-vectorized */
+	for (int i = 0; i < dim_a; i++)
 		result->x[i] = a->x[i];
 
-	for (int i = 0; i < b->dim; i++)
-		result->x[i + a->dim] = b->x[i];
+	/* Auto-vectorized */
+	for (int i = 0; i < dim_b; i++)
+		result->x[i + dim_a] = b->x[i];
 
 	PG_RETURN_POINTER(result);
 }

test/expected/hnsw_vector.out

@@ -123,6 +123,12 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
  [0,0,0]
 (3 rows)
 
+TRUNCATE t;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
+ val 
+-----
+(0 rows)
+
 RESET hnsw.iterative_scan;
 RESET hnsw.ef_search;
 DROP TABLE t;

test/expected/ivfflat_vector.out

@@ -110,6 +110,15 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
  [1,1,1]
 (2 rows)
 
+TRUNCATE t;
+NOTICE:  ivfflat index created with little data
+DETAIL:  This will cause low recall.
+HINT:  Drop the index until the table has more data.
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
+ val 
+-----
+(0 rows)
+
 RESET ivfflat.iterative_scan;
 RESET ivfflat.max_probes;
 DROP TABLE t;

test/sql/hnsw_vector.sql

@@ -70,6 +70,9 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 SET hnsw.iterative_scan = relaxed_order;
 SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 
+TRUNCATE t;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
+
 RESET hnsw.iterative_scan;
 RESET hnsw.ef_search;
 DROP TABLE t;

test/sql/ivfflat_vector.sql

@@ -59,6 +59,9 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 SET ivfflat.max_probes = 2;
 SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 
+TRUNCATE t;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
+
 RESET ivfflat.iterative_scan;
 RESET ivfflat.max_probes;
 DROP TABLE t;