Switched to TidStore for HNSW vacuuming for Postgres 17+

Co-authored-by: Bhagyesh Chaturvedi <bhagyeshc@google.com>

.github/workflows/build.yml

@@ -23,14 +23,14 @@ jobs:
           - postgres: 13
             os: ubuntu-22.04
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: ankane/setup-postgres@v1
         with:
           postgres-version: ${{ matrix.postgres }}
           dev-files: true
       - run: make
         env:
-          PG_CFLAGS: ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }} -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
+          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }}
       - run: |
           export PG_CONFIG=`which pg_config`
           sudo --preserve-env=PG_CONFIG make install
@@ -48,18 +48,18 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - postgres: 17
-            os: macos-15
+          - postgres: 18
+            os: macos-26
           - postgres: 14
-            os: macos-13
+            os: macos-15-intel
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: ankane/setup-postgres@v1
         with:
           postgres-version: ${{ matrix.postgres }}
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-unknown-warning-option
+          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-unknown-warning-option ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }}
       - run: make install
       - run: make installcheck
       - if: ${{ failure() }}
@@ -72,27 +72,35 @@ jobs:
           tar xf $TAG.tar.gz
           mv postgres-$TAG postgres
         env:
-          TAG: ${{ matrix.postgres == 17 && 'REL_17_2' || 'REL_14_15' }}
+          TAG: ${{ matrix.postgres == 18 && 'REL_18_2' || 'REL_14_21' }}
       - run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl"
         env:
           PERL5LIB: /Users/runner/perl5/lib/perl5
       - run: make clean && $(brew --prefix llvm@$LLVM_VERSION)/bin/scan-build --status-bugs make
         env:
-          LLVM_VERSION: ${{ matrix.os == 'macos-15' && 18 || 15 }}
+          LLVM_VERSION: ${{ matrix.os == 'macos-26' && 20 || 18 }}
           PG_CFLAGS: -DUSE_ASSERT_CHECKING
   windows:
-    runs-on: windows-latest
+    runs-on: ${{ matrix.os }}
     if: ${{ !startsWith(github.ref_name, 'mac') }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - postgres: 17
+            os: windows-2025
+          - postgres: 14
+            os: windows-2022
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: ankane/setup-postgres@v1
         with:
-          postgres-version: 14
+          postgres-version: ${{ matrix.postgres }}
       - run: |
-          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
+          call "C:\Program Files\Microsoft Visual Studio\${{ matrix.os == 'windows-2025' && 18 || 2022 }}\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
           nmake /NOLOGO /F Makefile.win && ^
           nmake /NOLOGO /F Makefile.win install && ^
-          nmake /NOLOGO /F Makefile.win installcheck && ^
+          nmake /NOLOGO /F Makefile.win installcheck ${{ matrix.postgres != 17 && 'PG_REGRESS=$(PGROOT)\bin\pg_regress' || '' }} && ^
           nmake /NOLOGO /F Makefile.win clean && ^
           nmake /NOLOGO /F Makefile.win uninstall
         shell: cmd
@@ -125,10 +133,10 @@ jobs:
     if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: ankane/setup-postgres-valgrind@v1
         with:
-          postgres-version: 17
+          postgres-version: 18
           check-ub: yes
       - run: make OPTFLAGS=""
       - run: sudo --preserve-env=PG_CONFIG make install

CHANGELOG.md

@@ -1,3 +1,14 @@
+## 0.8.3 (unreleased)
+
+- Fixed possible index corruption with HNSW vacuuming
+- Fixed performance regression with Hamming distance and Jaccard distance with Postgres 18
+
+## 0.8.2 (2026-02-25)
+
+- Fixed buffer overflow with parallel HNSW index build - [more info](https://github.com/pgvector/pgvector/issues/959)
+- Improved `install` target on Windows
+- Fixed `Index Searches` in `EXPLAIN` output for Postgres 18
+
 ## 0.8.1 (2025-09-04)
 
 - Added support for Postgres 18 rc1

Dockerfile

@@ -5,7 +5,7 @@ ARG DEBIAN_CODENAME=bookworm
 FROM postgres:$PG_MAJOR-$DEBIAN_CODENAME
 ARG PG_MAJOR
 
-ADD https://github.com/pgvector/pgvector.git#v0.8.1 /tmp/pgvector
+ADD https://github.com/pgvector/pgvector.git#v0.8.2 /tmp/pgvector
 
 RUN apt-get update && \
 		apt-mark hold locales && \

LICENSE

@@ -1,4 +1,4 @@
-Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
+Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
 
 Portions Copyright (c) 1994, The Regents of the University of California
 

META.json

@@ -2,7 +2,7 @@
 	"name": "vector",
 	"abstract": "Open-source vector similarity search for Postgres",
 	"description": "Supports L2 distance, inner product, and cosine distance",
-	"version": "0.8.1",
+	"version": "0.8.2",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.8.1",
+			"version": "0.8.2",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.8.1
+EXTVERSION = 0.8.2
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*--*.sql)
@@ -27,6 +27,11 @@ ifneq ($(filter ppc64%, $(shell uname -m)), )
 	OPTFLAGS =
 endif
 
+# RISC-V64 doesn't support -march=native
+ifeq ($(shell uname -m), riscv64)
+	OPTFLAGS =
+endif
+
 # For auto-vectorization:
 # - GCC (needs -ftree-vectorize OR -O3) - https://gcc.gnu.org/projects/tree-ssa/vectorization.html
 # - Clang (could use pragma instead) - https://llvm.org/docs/Vectorizers.html

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.8.1
+EXTVERSION = 0.8.2
 
 DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql
 OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
@@ -31,6 +31,9 @@ LIBDIR = $(PGROOT)\lib
 PKGLIBDIR = $(PGROOT)\lib
 SHAREDIR = $(PGROOT)\share
 
+# Use $(PGROOT)\bin\pg_regress for Postgres < 17
+PG_REGRESS = $(LIBDIR)\pgxs\src\test\regress\pg_regress
+
 CFLAGS = /nologo /I"$(INCLUDEDIR_SERVER)\port\win32_msvc" /I"$(INCLUDEDIR_SERVER)\port\win32" /I"$(INCLUDEDIR_SERVER)" /I"$(INCLUDEDIR)"
 
 CFLAGS = $(CFLAGS) $(PG_CFLAGS)
@@ -54,11 +57,11 @@ install: all
 	copy $(SHLIB) "$(PKGLIBDIR)"
 	copy $(EXTENSION).control "$(SHAREDIR)\extension"
 	copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
-	mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
+	if not exist "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)" mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
 	for %f in ($(HEADERS)) do copy %f "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
 
 installcheck:
-	"$(BINDIR)\pg_regress" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)
+	"$(PG_REGRESS)" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)
 
 uninstall:
 	del /f "$(PKGLIBDIR)\$(SHLIB)"

README.md

@@ -11,6 +11,8 @@ Store your vectors with the rest of your data. Supports:
 
 Plus [ACID](https://en.wikipedia.org/wiki/ACID) compliance, point-in-time recovery, JOINs, and all of the other [great features](https://www.postgresql.org/about/) of Postgres
 
+Have a lot of vectors? Use [quantization](#scaling) to scale
+
 [![Build Status](https://github.com/pgvector/pgvector/actions/workflows/build.yml/badge.svg)](https://github.com/pgvector/pgvector/actions)
 
 ## Installation
@@ -21,7 +23,7 @@ Compile and install the extension (supports Postgres 13+)
 
 ```sh
 cd /tmp
-git clone --branch v0.8.1 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.2 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -29,16 +31,16 @@ make install # may need sudo
 
 See the [installation notes](#installation-notes---linux-and-mac) if you run into issues
 
-You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), [pkg](#pkg), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
+You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), [pkg](#pkg), [APK](#apk), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
 
 ### 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 `x64 Native Tools Command Prompt for VS [version]` as administrator. Then use `nmake` to build:
 
 ```cmd
-set "PGROOT=C:\Program Files\PostgreSQL\17"
+set "PGROOT=C:\Program Files\PostgreSQL\18"
 cd %TEMP%
-git clone --branch v0.8.1 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.2 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -314,6 +316,8 @@ For a large number of workers, you may need to increase `max_parallel_workers` (
 
 The [index options](#index-options) also have a significant impact on build time (use the defaults unless seeing low recall)
 
+Use [binary quantization](#binary-quantization) for faster build times at scale
+
 ### Indexing Progress
 
 Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING)
@@ -443,13 +447,7 @@ Exact indexes work well for conditions that match a low percentage of rows. Othe
 CREATE INDEX ON items USING hnsw (embedding vector_l2_ops);
 ```
 
-With approximate indexes, filtering is applied *after* the index is scanned. If a condition matches 10% of rows, with HNSW and the default `hnsw.ef_search` of 40, only 4 rows will match on average. For more rows, increase `hnsw.ef_search`.
-
-```sql
-SET hnsw.ef_search = 200;
-```
-
-Starting with 0.8.0, you can enable [iterative index scans](#iterative-index-scans), which will automatically scan more of the index when needed.
+With approximate indexes, filtering is applied *after* the index is scanned. If a condition matches 10% of rows, with HNSW and the default `hnsw.ef_search` of 40, only 4 rows will match on average. For more rows, enable [iterative index scans](#iterative-index-scans), which will automatically scan more of the index when needed.
 
 ```sql
 SET hnsw.iterative_scan = strict_order;
@@ -673,6 +671,10 @@ SHOW shared_buffers;
 
 Be sure to restart Postgres for changes to take effect.
 
+### Storing
+
+Use the `halfvec` type instead of `vector` for a smaller working set.
+
 ### Loading
 
 Use `COPY` for bulk loading data ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/loading/example.py)).
@@ -687,6 +689,8 @@ Add any indexes *after* loading the initial data for best performance.
 
 See index build time for [HNSW](#index-build-time) and [IVFFlat](#index-build-time-1).
 
+Use [binary quantization](#binary-quantization) for smaller indexes and faster build times at scale.
+
 In production environments, create indexes concurrently to avoid blocking writes.
 
 ```sql
@@ -695,10 +699,10 @@ CREATE INDEX CONCURRENTLY ...
 
 ### Querying
 
-Use `EXPLAIN ANALYZE` to debug performance.
+Use `EXPLAIN (ANALYZE, BUFFERS)` to debug performance.
 
 ```sql
-EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
+EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
 #### Exact Search
@@ -717,6 +721,8 @@ SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
 
 #### Approximate Search
 
+Use [binary quantization](#binary-quantization) with re-ranking to keep indexes in-memory at scale.
+
 To speed up queries with an IVFFlat index, increase the number of inverted lists (at the expense of recall).
 
 ```sql
@@ -732,21 +738,20 @@ REINDEX INDEX CONCURRENTLY index_name;
 VACUUM table_name;
 ```
 
+## Scaling
+
+For a smaller working set:
+
+1. Use the `halfvec` type instead of `vector` for tables
+2. Use [binary quantization](#binary-quantization) for indexes (with re-ranking for search)
+
+Scale vertically by increasing memory, CPU, and storage on a single instance. Use existing tools to [tune parameters](#tuning) and [monitor performance](#monitoring).
+
+Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus), [PgDog](https://github.com/pgdogdev/pgdog), or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus/example.py)).
+
 ## Monitoring
 
-Monitor performance with [pg_stat_statements](https://www.postgresql.org/docs/current/pgstatstatements.html) (be sure to add it to `shared_preload_libraries`).
-
-```sql
-CREATE EXTENSION pg_stat_statements;
-```
-
-Get the most time-consuming queries with:
-
-```sql
-SELECT query, calls, ROUND((total_plan_time + total_exec_time) / calls) AS avg_time_ms,
-    ROUND((total_plan_time + total_exec_time) / 60000) AS total_time_min
-    FROM pg_stat_statements ORDER BY total_plan_time + total_exec_time DESC LIMIT 20;
-```
+Use existing tools like [pg_stat_statements](https://www.postgresql.org/docs/current/pgstatstatements.html) or [PgHero](https://github.com/ankane/pghero) to monitor performance.
 
 Monitor recall by comparing results from approximate search with exact search.
 
@@ -757,23 +762,18 @@ SELECT ...
 COMMIT;
 ```
 
-## Scaling
-
-Scale pgvector the same way you scale Postgres.
-
-Scale vertically by increasing memory, CPU, and storage on a single instance. Use existing tools to [tune parameters](#tuning) and [monitor performance](#monitoring).
-
-Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus) or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus/example.py)).
-
 ## Languages
 
 Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
 
 Language | Libraries / Examples
 --- | ---
+Ada | [pgvector-ada](https://github.com/pgvector/pgvector-ada)
+Algol | [pgvector-algol](https://github.com/pgvector/pgvector-algol)
 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)
+COBOL | [pgvector-cobol](https://github.com/pgvector/pgvector-cobol)
 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)
@@ -790,14 +790,18 @@ 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)
+Prolog | [pgvector-prolog](https://github.com/pgvector/pgvector-prolog)
 Python | [pgvector-python](https://github.com/pgvector/pgvector-python)
 R | [pgvector-r](https://github.com/pgvector/pgvector-r)
+Racket | [pgvector-racket](https://github.com/pgvector/pgvector-racket)
 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)
+Tcl | [pgvector-tcl](https://github.com/pgvector/pgvector-tcl)
 Zig | [pgvector-zig](https://github.com/pgvector/pgvector-zig)
 
 ## Frequently Asked Questions
@@ -812,7 +816,7 @@ 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 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).
+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).
 
 #### Can I store vectors with different dimensions in the same column?
 
@@ -871,6 +875,8 @@ No, but like other index types, you’ll likely see better performance if they d
 SELECT pg_size_pretty(pg_relation_size('index_name'));
 ```
 
+Use [half-precision indexing](#half-precision-indexing) or [binary quantization](#binary-quantization) for smaller indexes.
+
 ## Troubleshooting
 
 #### Why isn’t a query using an index?
@@ -1064,7 +1070,7 @@ 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/17/bin/pg_config
+export PG_CONFIG=/Library/PostgreSQL/18/bin/pg_config
 ```
 
 Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
@@ -1075,11 +1081,11 @@ sudo --preserve-env=PG_CONFIG make install
 
 A few common paths on Mac are:
 
-- 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`
+- 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`
 
-Note: Replace `17` with your Postgres server version
+Note: Replace `18` with your Postgres server version
 
 ### Missing Header
 
@@ -1088,10 +1094,10 @@ 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-17
+sudo apt install postgresql-server-dev-18
 ```
 
-Note: Replace `17` with your Postgres server version
+Note: Replace `18` with your Postgres server version
 
 ### Missing SDK
 
@@ -1138,30 +1144,32 @@ 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:pg17-trixie
+docker pull pgvector/pgvector:pg18-trixie
 ```
 
-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).
+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).
 
 Supported tags are:
 
-- `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`
-- `pg16-bookworm`, `0.8.1-pg16-bookworm`, `pg16`, `0.8.1-pg16`
-- `pg15-trixie`, `0.8.1-pg15-trixie`
-- `pg15-bookworm`, `0.8.1-pg15-bookworm`, `pg15`, `0.8.1-pg15`
-- `pg14-trixie`, `0.8.1-pg14-trixie`
-- `pg14-bookworm`, `0.8.1-pg14-bookworm`, `pg14`, `0.8.1-pg14`
-- `pg13-trixie`, `0.8.1-pg13-trixie`
-- `pg13-bookworm`, `0.8.1-pg13-bookworm`, `pg13`, `0.8.1-pg13`
+- `pg18-trixie`, `0.8.2-pg18-trixie`
+- `pg18-bookworm`, `0.8.2-pg18-bookworm`, `pg18`, `0.8.2-pg18`
+- `pg17-trixie`, `0.8.2-pg17-trixie`
+- `pg17-bookworm`, `0.8.2-pg17-bookworm`, `pg17`, `0.8.2-pg17`
+- `pg16-trixie`, `0.8.2-pg16-trixie`
+- `pg16-bookworm`, `0.8.2-pg16-bookworm`, `pg16`, `0.8.2-pg16`
+- `pg15-trixie`, `0.8.2-pg15-trixie`
+- `pg15-bookworm`, `0.8.2-pg15-bookworm`, `pg15`, `0.8.2-pg15`
+- `pg14-trixie`, `0.8.2-pg14-trixie`
+- `pg14-bookworm`, `0.8.2-pg14-bookworm`, `pg14`, `0.8.2-pg14`
+- `pg13-trixie`, `0.8.2-pg13-trixie`
+- `pg13-bookworm`, `0.8.2-pg13-bookworm`, `pg13`, `0.8.2-pg13`
 
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.8.1 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.2 https://github.com/pgvector/pgvector.git
 cd pgvector
-docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector .
+docker build --pull --build-arg PG_MAJOR=18 -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.
@@ -1178,7 +1186,7 @@ With Homebrew Postgres, you can use:
 brew install pgvector
 ```
 
-Note: This only adds it to the `postgresql@17` and `postgresql@14` formulas
+Note: This only adds it to the `postgresql@18` and `postgresql@17` formulas
 
 ### PGXN
 
@@ -1193,22 +1201,22 @@ 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-17-pgvector
+sudo apt install postgresql-18-pgvector
 ```
 
-Note: Replace `17` with your Postgres server version
+Note: Replace `18` 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_17
+sudo yum install pgvector_18
 # or
-sudo dnf install pgvector_17
+sudo dnf install pgvector_18
 ```
 
-Note: Replace `17` with your Postgres server version
+Note: Replace `18` with your Postgres server version
 
 ### pkg
 
@@ -1225,6 +1233,14 @@ cd /usr/ports/databases/pgvector
 make install
 ```
 
+### APK
+
+Install the Alpine package with:
+
+```sh
+apk add postgresql-pgvector
+```
+
 ### conda-forge
 
 With Conda Postgres, install from [conda-forge](https://anaconda.org/conda-forge/pgvector) with:
@@ -1313,7 +1329,7 @@ make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
 To enable benchmarking:
 
 ```sh
-make clean && PG_CFLAGS="-DIVFFLAT_BENCH" make && make install
+make clean && PG_CFLAGS="-DHNSW_BENCH -DIVFFLAT_BENCH" make && make install
 ```
 
 To show memory usage:

sql/vector--0.8.1--0.8.2.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.2'" to load this file. \quit

src/bitutils.c

@@ -31,10 +31,12 @@
 #define BIT_TARGET_CLONES
 #endif
 
-/* Use built-ins when possible for inlining */
-#if defined(HAVE__BUILTIN_POPCOUNT) && defined(HAVE_LONG_INT_64)
+/* Use built-ins when possible for Postgres < 19 for inlining */
+#if PG_VERSION_NUM >= 190000
+#define popcount64(x) pg_popcount64(x)
+#elif defined(HAVE__BUILTIN_POPCOUNT) && (defined(HAVE_LONG_INT_64) || SIZEOF_LONG == 8)
 #define popcount64(x) __builtin_popcountl(x)
-#elif defined(HAVE__BUILTIN_POPCOUNT) && defined(HAVE_LONG_LONG_INT_64)
+#elif defined(HAVE__BUILTIN_POPCOUNT) && (defined(HAVE_LONG_LONG_INT_64) || SIZEOF_LONG_LONG == 8)
 #define popcount64(x) __builtin_popcountll(x)
 #elif !defined(_MSC_VER)
 /* Fails to resolve with MSVC */
@@ -169,7 +171,7 @@ BitJaccardDistanceAvx512Popcount(uint32 bytes, unsigned char *ax, unsigned char
 #endif
 
 TARGET_XSAVE static bool
-SupportsAvx512Popcount()
+SupportsAvx512Popcount(void)
 {
 	unsigned int exx[4] = {0, 0, 0, 0};
 

src/bitvec.c

@@ -2,6 +2,7 @@
 
 #include "bitutils.h"
 #include "bitvec.h"
+#include "fmgr.h"
 #include "utils/varbit.h"
 #include "vector.h"
 

src/halfutils.c

@@ -1,5 +1,7 @@
 #include "postgres.h"
 
+#include <math.h>
+
 #include "halfutils.h"
 #include "halfvec.h"
 

src/halfvec.c

@@ -13,12 +13,20 @@
 #include "port.h"				/* for strtof() */
 #include "sparsevec.h"
 #include "utils/array.h"
-#include "utils/builtins.h"
 #include "utils/float.h"
+#include "utils/fmgrprotos.h"
 #include "utils/lsyscache.h"
-#include "utils/numeric.h"
+#include "utils/varbit.h"
 #include "vector.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
+#if PG_VERSION_NUM >= 170000
+#include "parser/scansup.h"
+#endif
+
 #define STATE_DIMS(x) (ARR_DIMS(x)[0] - 1)
 #define CreateStateDatums(dim) palloc(sizeof(Datum) * (dim + 1))
 
@@ -129,9 +137,9 @@ InitHalfVector(int dim)
 	return result;
 }
 
-/*
- * Check for whitespace, since array_isspace() is static
- */
+#if PG_VERSION_NUM >= 170000
+#define halfvec_isspace(ch) scanner_isspace(ch)
+#else
 static inline bool
 halfvec_isspace(char ch)
 {
@@ -144,6 +152,7 @@ halfvec_isspace(char ch)
 		return true;
 	return false;
 }
+#endif
 
 /*
  * Check state array

src/hnsw.c

@@ -1,18 +1,25 @@
 #include "postgres.h"
 
 #include <float.h>
+#include <limits.h>
 #include <math.h>
 
 #include "access/amapi.h"
+#include "access/genam.h"
 #include "access/reloptions.h"
 #include "commands/progress.h"
 #include "commands/vacuum.h"
+#include "fmgr.h"
 #include "hnsw.h"
 #include "miscadmin.h"
+#include "nodes/pg_list.h"
+#include "storage/lwlock.h"
 #include "utils/float.h"
 #include "utils/guc.h"
+#include "utils/relcache.h"
 #include "utils/selfuncs.h"
 #include "utils/spccache.h"
+#include "vector.h"
 
 #if PG_VERSION_NUM < 150000
 #define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
@@ -260,6 +267,68 @@ FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnswhandler);
 Datum
 hnswhandler(PG_FUNCTION_ARGS)
 {
+#if PG_VERSION_NUM >= 190000
+	static const IndexAmRoutine amroutine = {
+		.type = T_IndexAmRoutine,
+		.amstrategies = 0,
+		.amsupport = 3,
+		.amoptsprocnum = 0,
+		.amcanorder = false,
+		.amcanorderbyop = true,
+		.amcanhash = false,
+		.amconsistentequality = false,
+		.amconsistentordering = false,
+		.amcanbackward = false,
+		.amcanunique = false,
+		.amcanmulticol = false,
+		.amoptionalkey = true,
+		.amsearcharray = false,
+		.amsearchnulls = false,
+		.amstorage = false,
+		.amclusterable = false,
+		.ampredlocks = false,
+		.amcanparallel = false,
+		.amcanbuildparallel = true,
+		.amcaninclude = false,
+#if PG_VERSION_NUM >= 170000
+		.amusemaintenanceworkmem = true,
+#else
+		.amusemaintenanceworkmem = false,
+#endif
+		.amsummarizing = false,
+		.amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL,
+		.amkeytype = InvalidOid,
+
+		.ambuild = hnswbuild,
+		.ambuildempty = hnswbuildempty,
+		.aminsert = hnswinsert,
+		.aminsertcleanup = NULL,
+		.ambulkdelete = hnswbulkdelete,
+		.amvacuumcleanup = hnswvacuumcleanup,
+		.amcanreturn = NULL,
+		.amcostestimate = hnswcostestimate,
+		.amgettreeheight = NULL,
+		.amoptions = hnswoptions,
+		.amproperty = NULL,
+		.ambuildphasename = hnswbuildphasename,
+		.amvalidate = hnswvalidate,
+		.amadjustmembers = NULL,
+		.ambeginscan = hnswbeginscan,
+		.amrescan = hnswrescan,
+		.amgettuple = hnswgettuple,
+		.amgetbitmap = NULL,
+		.amendscan = hnswendscan,
+		.ammarkpos = NULL,
+		.amrestrpos = NULL,
+		.amestimateparallelscan = NULL,
+		.aminitparallelscan = NULL,
+		.amparallelrescan = NULL,
+		.amtranslatestrategy = NULL,
+		.amtranslatecmptype = NULL,
+	};
+
+	PG_RETURN_POINTER(&amroutine);
+#else
 	IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
 
 	amroutine->amstrategies = 0;
@@ -286,7 +355,11 @@ hnswhandler(PG_FUNCTION_ARGS)
 	amroutine->amcanbuildparallel = true;
 #endif
 	amroutine->amcaninclude = false;
-	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
+#if PG_VERSION_NUM >= 170000
+	amroutine->amusemaintenanceworkmem = true;
+#else
+	amroutine->amusemaintenanceworkmem = false;
+#endif
 #if PG_VERSION_NUM >= 160000
 	amroutine->amsummarizing = false;
 #endif
@@ -333,4 +406,5 @@ hnswhandler(PG_FUNCTION_ARGS)
 #endif
 
 	PG_RETURN_POINTER(amroutine);
+#endif
 }

src/hnsw.h

@@ -3,15 +3,35 @@
 
 #include "postgres.h"
 
+#include <math.h>
+
 #include "access/genam.h"
 #include "access/parallel.h"
 #include "lib/pairingheap.h"
 #include "nodes/execnodes.h"
 #include "port.h"				/* for random() */
+#include "storage/bufpage.h"
+#include "storage/condition_variable.h"
+#include "storage/lwlock.h"
+#include "storage/s_lock.h"
 #include "utils/relptr.h"
 #include "utils/sampling.h"
 #include "vector.h"
 
+#ifdef HNSW_BENCH
+#include "portability/instr_time.h"
+#endif
+
+#if PG_VERSION_NUM >= 190000
+typedef Pointer Item;
+#endif
+
+#if PG_VERSION_NUM >= 170000
+#define HnswTidStore TidStore
+#else
+#define HnswTidStore tidhash_hash
+#endif
+
 #define HNSW_MAX_DIM 2000
 #define HNSW_MAX_NNZ 1000
 
@@ -68,6 +88,21 @@
 #define HnswPageGetOpaque(page)	((HnswPageOpaque) PageGetSpecialPointer(page))
 #define HnswPageGetMeta(page)	((HnswMetaPageData *) PageGetContents(page))
 
+#ifdef HNSW_BENCH
+#define HnswBench(name, code) \
+	do { \
+		instr_time	start; \
+		instr_time	duration; \
+		INSTR_TIME_SET_CURRENT(start); \
+		(code); \
+		INSTR_TIME_SET_CURRENT(duration); \
+		INSTR_TIME_SUBTRACT(duration, start); \
+		elog(INFO, "%s: %.3f ms", name, INSTR_TIME_GET_MILLISEC(duration)); \
+	} while (0)
+#else
+#define HnswBench(name, code) (code)
+#endif
+
 #if PG_VERSION_NUM >= 150000
 #define RandomDouble() pg_prng_double(&pg_global_prng_state)
 #define SeedRandom(seed) pg_prng_seed(&pg_global_prng_state, seed)
@@ -398,10 +433,11 @@ typedef struct HnswVacuumState
 	HnswSupport support;
 
 	/* Variables */
-	struct tidhash_hash *deleted;
+	struct HnswTidStore *deleted;
 	BufferAccessStrategy bas;
 	HnswNeighborTuple ntup;
 	HnswElementData highestPoint;
+	HnswElementData fallbackPoint;
 
 	/* Memory */
 	MemoryContext tmpCtx;
@@ -424,7 +460,7 @@ void	   *HnswAlloc(HnswAllocator * allocator, Size size);
 HnswElement HnswInitElement(char *base, ItemPointer tid, int m, double ml, int maxLevel, HnswAllocator * alloc);
 HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
 void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing);
-HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, HnswQuery * q, Relation rel, HnswSupport * support, bool loadVec);
+HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement entryPoint, HnswQuery * q, Relation index, 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);
 void		HnswAddHeapTid(HnswElement element, ItemPointer heaptid);

src/hnswbuild.c

@@ -36,11 +36,14 @@
  */
 #include "postgres.h"
 
-#include <math.h>
+#include <limits.h>
 
+#include "access/genam.h"
 #include "access/parallel.h"
+#include "access/relscan.h"
 #include "access/table.h"
 #include "access/tableam.h"
+#include "access/tupdesc.h"
 #include "access/xact.h"
 #include "access/xloginsert.h"
 #include "catalog/index.h"
@@ -48,11 +51,19 @@
 #include "commands/progress.h"
 #include "hnsw.h"
 #include "miscadmin.h"
+#include "nodes/execnodes.h"
 #include "optimizer/optimizer.h"
 #include "storage/bufmgr.h"
+#include "storage/condition_variable.h"
 #include "tcop/tcopprot.h"
 #include "utils/datum.h"
 #include "utils/memutils.h"
+#include "utils/rel.h"
+#include "utils/snapmgr.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_progress.h"
@@ -69,6 +80,8 @@
 #define PARALLEL_KEY_HNSW_AREA			UINT64CONST(0xA000000000000002)
 #define PARALLEL_KEY_QUERY_TEXT			UINT64CONST(0xA000000000000003)
 
+#define HNSW_MAX_GRAPH_MEMORY (SIZE_MAX / 2)
+
 /*
  * Create the metapage
  */
@@ -398,7 +411,7 @@ UpdateNeighborsInMemory(char *base, HnswSupport * support, HnswElement e, int m)
  * Update graph in memory
  */
 static void
-UpdateGraphInMemory(HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, HnswBuildState * buildstate)
+UpdateGraphInMemory(HnswSupport * support, HnswElement element, int m, HnswElement entryPoint, HnswBuildState * buildstate)
 {
 	HnswGraph  *graph = buildstate->graph;
 	char	   *base = buildstate->hnswarea;
@@ -460,7 +473,7 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false);
 
 	/* Update graph in memory */
-	UpdateGraphInMemory(support, element, m, efConstruction, entryPoint, buildstate);
+	UpdateGraphInMemory(support, element, m, entryPoint, buildstate);
 
 	/* Release entry lock */
 	LWLockRelease(entryLock);
@@ -481,6 +494,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	LWLock	   *flushLock = &graph->flushLock;
 	char	   *base = buildstate->hnswarea;
 	Datum		value;
+	Size		memoryMargin;
 
 	/* Form index value */
 	if (!HnswFormIndexValue(&value, values, isnull, buildstate->typeInfo, support))
@@ -489,6 +503,9 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	/* Get datum size */
 	valueSize = VARSIZE_ANY(DatumGetPointer(value));
 
+	/* In a parallel build, add a margin so allocations never fail */
+	memoryMargin = base == NULL ? 0 : 1024 * 1024;
+
 	/* Ensure graph not flushed when inserting */
 	LWLockAcquire(flushLock, LW_SHARED);
 
@@ -510,7 +527,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	 * Check that we have enough memory available for the new element now that
 	 * we have the allocator lock, and flush pages if needed.
 	 */
-	if (graph->memoryUsed >= graph->memoryTotal)
+	if (graph->memoryUsed + memoryMargin >= graph->memoryTotal)
 	{
 		LWLockRelease(&graph->allocatorLock);
 
@@ -545,7 +562,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 
 	/* Copy the datum */
 	memcpy(valuePtr, DatumGetPointer(value), valueSize);
-	HnswPtrStore(base, element->value, valuePtr);
+	HnswPtrStore(base, element->value, (char *) valuePtr);
 
 	/* Create a lock for the element */
 	LWLockInitialize(&element->lock, hnsw_lock_tranche_id);
@@ -603,7 +620,7 @@ InitGraph(HnswGraph * graph, char *base, Size memoryTotal)
 	HnswPtrStore(base, graph->head, (HnswElement) NULL);
 	HnswPtrStore(base, graph->entryPoint, (HnswElement) NULL);
 	graph->memoryUsed = 0;
-	graph->memoryTotal = memoryTotal;
+	graph->memoryTotal = Min(memoryTotal, HNSW_MAX_GRAPH_MEMORY);
 	graph->flushed = false;
 	graph->indtuples = 0;
 	SpinLockInit(&graph->lock);
@@ -644,9 +661,17 @@ static void *
 HnswSharedMemoryAlloc(Size size, void *state)
 {
 	HnswBuildState *buildstate = (HnswBuildState *) state;
-	void	   *chunk = buildstate->hnswarea + buildstate->graph->memoryUsed;
+	Size		alignedSize = MAXALIGN(size);
+	void	   *chunk;
 
-	buildstate->graph->memoryUsed += MAXALIGN(size);
+	if (alignedSize > 1024 * 1024)
+		elog(ERROR, "hnsw allocation too large");
+
+	if (buildstate->graph->memoryUsed + alignedSize > buildstate->graph->memoryTotal)
+		elog(ERROR, "hnsw allocator out of memory");
+
+	chunk = buildstate->hnswarea + buildstate->graph->memoryUsed;
+	buildstate->graph->memoryUsed += alignedSize;
 	return chunk;
 }
 
@@ -778,7 +803,11 @@ HnswParallelScanAndInsert(Relation heapRel, Relation indexRel, HnswShared * hnsw
 	buildstate.hnswarea = hnswarea;
 	InitAllocator(&buildstate.allocator, &HnswSharedMemoryAlloc, &buildstate);
 	scan = table_beginscan_parallel(heapRel,
-									ParallelTableScanFromHnswShared(hnswshared));
+									ParallelTableScanFromHnswShared(hnswshared)
+#if PG_VERSION_NUM >= 190000
+									,SO_NONE
+#endif
+		);
 	reltuples = table_index_build_scan(heapRel, indexRel, indexInfo,
 									   true, progress, BuildCallback,
 									   (void *) &buildstate, scan);
@@ -932,6 +961,8 @@ HnswBeginParallel(HnswBuildState * buildstate, bool isconcurrent, int request)
 	if (esthnswarea > estother)
 		esthnswarea -= estother;
 
+	esthnswarea = Min(esthnswarea, HNSW_MAX_GRAPH_MEMORY);
+
 	shm_toc_estimate_chunk(&pcxt->estimator, esthnswarea);
 	shm_toc_estimate_keys(&pcxt->estimator, 2);
 
@@ -974,8 +1005,7 @@ HnswBeginParallel(HnswBuildState * buildstate, bool isconcurrent, int request)
 								  snapshot);
 
 	hnswarea = (char *) shm_toc_allocate(pcxt->toc, esthnswarea);
-	/* Report less than allocated so never fails */
-	InitGraph(&hnswshared->graphData, hnswarea, esthnswarea - 1024 * 1024);
+	InitGraph(&hnswshared->graphData, hnswarea, esthnswarea);
 
 	/*
 	 * Avoid base address for relptr for Postgres < 14.5
@@ -1054,7 +1084,7 @@ ComputeParallelWorkers(Relation heap, Relation index)
  * Build graph
  */
 static void
-BuildGraph(HnswBuildState * buildstate, ForkNumber forkNum)
+BuildGraph(HnswBuildState * buildstate)
 {
 	int			parallel_workers = 0;
 
@@ -1102,7 +1132,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 
 	InitBuildState(buildstate, heap, index, indexInfo, forkNum);
 
-	BuildGraph(buildstate, forkNum);
+	BuildGraph(buildstate);
 
 	if (RelationNeedsWAL(index) || forkNum == INIT_FORKNUM)
 		log_newpage_range(index, forkNum, 0, RelationGetNumberOfBlocksInFork(index, forkNum), true);

src/hnswinsert.c

@@ -1,13 +1,19 @@
 #include "postgres.h"
 
-#include <math.h>
-
+#include "access/genam.h"
 #include "access/generic_xlog.h"
 #include "hnsw.h"
+#include "nodes/execnodes.h"
 #include "storage/bufmgr.h"
 #include "storage/lmgr.h"
+#include "storage/lwlock.h"
 #include "utils/datum.h"
 #include "utils/memutils.h"
+#include "utils/rel.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 /*
  * Get the insert page
@@ -660,7 +666,7 @@ FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
  * Update graph on disk
  */
 static void
-UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
+UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, HnswElement entryPoint, bool building)
 {
 	BlockNumber newInsertPage = InvalidBlockNumber;
 
@@ -708,7 +714,7 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
 
 	/* Create an element */
 	element = HnswInitElement(base, heaptid, m, HnswGetMl(m), HnswGetMaxLevel(m), NULL);
-	HnswPtrStore(base, element->value, DatumGetPointer(value));
+	HnswPtrStore(base, element->value, (char *) DatumGetPointer(value));
 
 	/* Prevent concurrent inserts when likely updating entry point */
 	if (entryPoint == NULL || element->level > entryPoint->level)
@@ -728,7 +734,7 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
 	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false);
 
 	/* Update graph on disk */
-	UpdateGraphOnDisk(index, support, element, m, efConstruction, entryPoint, building);
+	UpdateGraphOnDisk(index, support, element, m, entryPoint, building);
 
 	/* Release lock */
 	UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);

src/hnswscan.c

@@ -1,12 +1,23 @@
 #include "postgres.h"
 
+#include <limits.h>
+
+#include "access/genam.h"
 #include "access/relscan.h"
 #include "hnsw.h"
+#include "lib/pairingheap.h"
+#include "miscadmin.h"
+#include "nodes/pg_list.h"
 #include "pgstat.h"
-#include "storage/bufmgr.h"
 #include "storage/lmgr.h"
 #include "utils/float.h"
 #include "utils/memutils.h"
+#include "utils/relcache.h"
+#include "utils/snapmgr.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 /*
  * Algorithm 5 from paper
@@ -142,7 +153,7 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
 	/* Calculate max memory */
 	/* Add 256 extra bytes to fill last block when close */
 	maxMemory = (double) work_mem * hnsw_scan_mem_multiplier * 1024.0 + 256;
-	so->maxMemory = Min(maxMemory, (double) SIZE_MAX);
+	so->maxMemory = Min(maxMemory, (double) (SIZE_MAX / 2));
 
 	scan->opaque = so;
 
@@ -193,6 +204,10 @@ 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

@@ -2,18 +2,24 @@
 
 #include <math.h>
 
+#include "access/genam.h"
 #include "access/generic_xlog.h"
-#include "catalog/pg_type.h"
-#include "catalog/pg_type_d.h"
 #include "common/hashfn.h"
 #include "fmgr.h"
 #include "hnsw.h"
 #include "lib/pairingheap.h"
+#include "nodes/pg_list.h"
+#include "port/atomics.h"
 #include "sparsevec.h"
 #include "storage/bufmgr.h"
 #include "utils/datum.h"
 #include "utils/memdebug.h"
 #include "utils/rel.h"
+#include "vector.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 #if PG_VERSION_NUM < 170000
 static inline uint64
@@ -256,7 +262,7 @@ HnswInitElement(char *base, ItemPointer heaptid, int m, double ml, int maxLevel,
 
 	HnswInitNeighbors(base, element, m, allocator);
 
-	HnswPtrStore(base, element->value, (Pointer) NULL);
+	HnswPtrStore(base, element->value, (char *) NULL);
 
 	return element;
 }
@@ -282,7 +288,7 @@ HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno)
 	element->blkno = blkno;
 	element->offno = offno;
 	HnswPtrStore(base, element->neighbors, (HnswNeighborArrayPtr *) NULL);
-	HnswPtrStore(base, element->value, (Pointer) NULL);
+	HnswPtrStore(base, element->value, (char *) NULL);
 	return element;
 }
 
@@ -508,7 +514,7 @@ HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHe
 		char	   *base = NULL;
 		Datum		value = datumCopy(PointerGetDatum(&etup->data), false, -1);
 
-		HnswPtrStore(base, element->value, DatumGetPointer(value));
+		HnswPtrStore(base, element->value, (char *) DatumGetPointer(value));
 	}
 }
 
@@ -540,6 +546,9 @@ HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, Hns
 
 	Assert(HnswIsElementTuple(etup));
 
+	if (unlikely(etup->deleted))
+		elog(ERROR, "cannot load deleted element");
+
 	/* Calculate distance */
 	if (distance != NULL)
 	{
@@ -922,7 +931,7 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 					continue;
 			}
 
-			if (eElement == NULL || !(eDistance < f->distance || alwaysAdd))
+			if (!(eDistance < f->distance || alwaysAdd))
 			{
 				if (discarded != NULL)
 				{

src/hnswvacuum.c

@@ -1,13 +1,24 @@
 #include "postgres.h"
 
-#include <math.h>
-
+#include "access/genam.h"
 #include "access/generic_xlog.h"
 #include "commands/vacuum.h"
 #include "hnsw.h"
+#include "nodes/pg_list.h"
 #include "storage/bufmgr.h"
 #include "storage/lmgr.h"
 #include "utils/memutils.h"
+#include "utils/rel.h"
+
+#if PG_VERSION_NUM >= 170000
+#include "access/tidstore.h"
+#include "miscadmin.h"
+#include "postmaster/autovacuum.h"
+#endif
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 #if PG_VERSION_NUM >= 180000
 #define vacuum_delay_point() vacuum_delay_point(false)
@@ -17,9 +28,13 @@
  * Check if deleted list contains an index TID
  */
 static bool
-DeletedContains(tidhash_hash * deleted, ItemPointer indextid)
+DeletedContains(HnswTidStore * deleted, ItemPointer indextid)
 {
+#if PG_VERSION_NUM >= 170000
+	return TidStoreIsMember(deleted, indextid);
+#else
 	return tidhash_lookup(deleted, *indextid) != NULL;
+#endif
 }
 
 /*
@@ -32,17 +47,20 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 {
 	BlockNumber blkno = HNSW_HEAD_BLKNO;
 	HnswElement highestPoint = &vacuumstate->highestPoint;
+	HnswElement fallbackPoint = &vacuumstate->fallbackPoint;
 	Relation	index = vacuumstate->index;
 	BufferAccessStrategy bas = vacuumstate->bas;
-	HnswElement entryPoint = HnswGetEntryPoint(vacuumstate->index);
 	IndexBulkDeleteResult *stats = vacuumstate->stats;
 
-	/* Store separately since highestPoint.level is uint8 */
+	/* Store separately since HnswElement level is uint8 */
 	int			highestLevel = -1;
+	int			fallbackLevel = -1;
 
-	/* Initialize highest point */
+	/* Initialize highest point and fallback point */
 	highestPoint->blkno = InvalidBlockNumber;
 	highestPoint->offno = InvalidOffsetNumber;
+	fallbackPoint->blkno = InvalidBlockNumber;
+	fallbackPoint->offno = InvalidOffsetNumber;
 
 	while (BlockNumberIsValid(blkno))
 	{
@@ -52,6 +70,10 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 		OffsetNumber offno;
 		OffsetNumber maxoffno;
 		bool		updated = false;
+#if PG_VERSION_NUM >= 170000
+		OffsetNumber deletedoffs[MaxOffsetNumber];
+		int			ndeletedoffs = 0;
+#endif
 
 		vacuum_delay_point();
 
@@ -105,6 +127,9 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 
 			if (!ItemPointerIsValid(&etup->heaptids[0]))
 			{
+#if PG_VERSION_NUM >= 170000
+				deletedoffs[ndeletedoffs++] = offno;
+#else
 				ItemPointerData ip;
 				bool		found;
 
@@ -113,17 +138,39 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 
 				tidhash_insert(vacuumstate->deleted, ip, &found);
 				Assert(!found);
+#endif
 			}
-			else if (etup->level > highestLevel && !(entryPoint != NULL && blkno == entryPoint->blkno && offno == entryPoint->offno))
+			else if (etup->level > highestLevel)
 			{
-				/* Keep track of highest non-entry point */
+				if (BlockNumberIsValid(highestPoint->blkno))
+				{
+					/* Current highest point becomes fallback */
+					fallbackPoint->blkno = highestPoint->blkno;
+					fallbackPoint->offno = highestPoint->offno;
+					fallbackPoint->level = highestPoint->level;
+					fallbackLevel = highestLevel;
+				}
+
+				/* Keep track of highest point */
 				highestPoint->blkno = blkno;
 				highestPoint->offno = offno;
 				highestPoint->level = etup->level;
 				highestLevel = etup->level;
 			}
+			else if (etup->level > fallbackLevel)
+			{
+				/* Keep track of second highest point */
+				fallbackPoint->blkno = blkno;
+				fallbackPoint->offno = offno;
+				fallbackPoint->level = etup->level;
+				fallbackLevel = etup->level;
+			}
 		}
 
+#if PG_VERSION_NUM >= 170000
+		TidStoreSetBlockOffsets(vacuumstate->deleted, blkno, deletedoffs, ndeletedoffs);
+#endif
+
 		blkno = HnswPageGetOpaque(page)->nextblkno;
 
 		if (updated)
@@ -133,6 +180,10 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 
 		UnlockReleaseBuffer(buf);
 	}
+
+#ifdef HNSW_MEMORY
+	elog(INFO, "memory: %zu KB", MemoryContextMemAllocated(CurrentMemoryContext, true) / 1024);
+#endif
 }
 
 /*
@@ -174,7 +225,12 @@ NeedsUpdated(HnswVacuumState * vacuumstate, HnswElement element)
 	/* Also update if layer 0 is not full */
 	/* This could indicate too many candidates being deleted during insert */
 	if (!needsUpdated)
+	{
+		/* Keep clang-tidy happy */
+		Assert(ntup->count > 0);
+
 		needsUpdated = !ItemPointerIsValid(&ntup->indextids[ntup->count - 1]);
+	}
 
 	UnlockReleaseBuffer(buf);
 
@@ -259,12 +315,27 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 		/* Get a shared lock */
 		LockPage(index, HNSW_UPDATE_LOCK, ShareLock);
 
-		/* Load element */
-		HnswLoadElement(highestPoint, NULL, NULL, index, support, true, NULL);
+		/* Get latest entry point */
+		entryPoint = HnswGetEntryPoint(index);
 
-		/* Repair if needed */
-		if (NeedsUpdated(vacuumstate, highestPoint))
-			RepairGraphElement(vacuumstate, highestPoint, HnswGetEntryPoint(index));
+		/* Use fallback point if highest point is entry point */
+		if (entryPoint != NULL && entryPoint->blkno == highestPoint->blkno && entryPoint->offno == highestPoint->offno)
+		{
+			highestPoint = &vacuumstate->fallbackPoint;
+
+			if (!BlockNumberIsValid(highestPoint->blkno))
+				highestPoint = NULL;
+		}
+
+		if (highestPoint != NULL)
+		{
+			/* Load element */
+			HnswLoadElement(highestPoint, NULL, NULL, index, support, true, NULL);
+
+			/* Repair if needed */
+			if (NeedsUpdated(vacuumstate, highestPoint))
+				RepairGraphElement(vacuumstate, highestPoint, entryPoint);
+		}
 
 		/* Release lock */
 		UnlockPage(index, HNSW_UPDATE_LOCK, ShareLock);
@@ -431,6 +502,99 @@ RepairGraph(HnswVacuumState * vacuumstate)
 		/* Reset memory context */
 		MemoryContextSwitchTo(oldCtx);
 		MemoryContextReset(vacuumstate->tmpCtx);
+
+#ifdef HNSW_VACUUM_PROGRESS
+		if (!BlockNumberIsValid(blkno) || ((blkno - HNSW_HEAD_BLKNO) % 1000 == 0 && blkno != HNSW_HEAD_BLKNO))
+		{
+			BlockNumber totalBlocks = RelationGetNumberOfBlocks(index);
+			BlockNumber currentBlocks = BlockNumberIsValid(blkno) ? blkno : totalBlocks;
+
+			elog(INFO, "hnsw vacuum progress: %.1f%%", 100.0 * currentBlocks / totalBlocks);
+		}
+#endif
+	}
+}
+
+/*
+ * Confirm graph was repaired
+ */
+static void
+ConfirmRepaired(HnswVacuumState * vacuumstate)
+{
+	BlockNumber blkno = HNSW_HEAD_BLKNO;
+	Relation	index = vacuumstate->index;
+	BufferAccessStrategy bas = vacuumstate->bas;
+
+	while (BlockNumberIsValid(blkno))
+	{
+		Buffer		buf;
+		Page		page;
+		OffsetNumber offno;
+		OffsetNumber maxoffno;
+
+		vacuum_delay_point();
+
+		buf = ReadBufferExtended(index, MAIN_FORKNUM, blkno, RBM_NORMAL, bas);
+		LockBuffer(buf, BUFFER_LOCK_SHARE);
+		page = BufferGetPage(buf);
+		maxoffno = PageGetMaxOffsetNumber(page);
+
+		/* Iterate over nodes */
+		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
+		{
+			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
+			HnswNeighborTuple ntup;
+			Buffer		nbuf;
+			Page		npage;
+			BlockNumber neighborPage;
+			OffsetNumber neighborOffno;
+
+			/* Skip neighbor tuples */
+			if (!HnswIsElementTuple(etup))
+				continue;
+
+			/* Skip if being deleted */
+			if (!ItemPointerIsValid(&etup->heaptids[0]))
+				continue;
+
+			/* Get neighbor page */
+			neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
+			neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
+
+			if (neighborPage == blkno)
+			{
+				nbuf = buf;
+				npage = page;
+			}
+			else
+			{
+				nbuf = ReadBufferExtended(index, MAIN_FORKNUM, neighborPage, RBM_NORMAL, bas);
+				LockBuffer(nbuf, BUFFER_LOCK_SHARE);
+				npage = BufferGetPage(nbuf);
+			}
+
+			ntup = (HnswNeighborTuple) PageGetItem(npage, PageGetItemId(npage, neighborOffno));
+
+			/* Check neighbors */
+			for (int i = 0; i < ntup->count; i++)
+			{
+				ItemPointer indextid = &ntup->indextids[i];
+
+				if (!ItemPointerIsValid(indextid))
+					continue;
+
+				/* Check if in deleted list */
+				if (DeletedContains(vacuumstate->deleted, indextid))
+					elog(ERROR, "hnsw graph not repaired");
+			}
+
+			if (nbuf != buf)
+				UnlockReleaseBuffer(nbuf);
+		}
+
+		blkno = HnswPageGetOpaque(page)->nextblkno;
+
+		UnlockReleaseBuffer(buf);
 	}
 }
 
@@ -524,8 +688,9 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 			ntup = (HnswNeighborTuple) PageGetItem(npage, PageGetItemId(npage, neighborOffno));
 
 			/* Overwrite element */
+			/* Use memset instead of MemSet to keep clang-tidy happy */
 			etup->deleted = 1;
-			MemSet(&etup->data, 0, VARSIZE_ANY(&etup->data));
+			memset(&etup->data, 0, VARSIZE_ANY(&etup->data));
 
 			/* Overwrite neighbors */
 			for (int i = 0; i < ntup->count; i++)
@@ -596,7 +761,11 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
 	HnswGetMetaPageInfo(index, &vacuumstate->m, NULL);
 
 	/* Create hash table */
+#if PG_VERSION_NUM >= 170000
+	vacuumstate->deleted = TidStoreCreateLocal((AmAutoVacuumWorkerProcess() && autovacuum_work_mem != -1) ? autovacuum_work_mem : maintenance_work_mem, true);
+#else
 	vacuumstate->deleted = tidhash_create(CurrentMemoryContext, 256, NULL);
+#endif
 }
 
 /*
@@ -605,7 +774,11 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
 static void
 FreeVacuumState(HnswVacuumState * vacuumstate)
 {
+#if PG_VERSION_NUM >= 170000
+	TidStoreDestroy(vacuumstate->deleted);
+#else
 	tidhash_destroy(vacuumstate->deleted);
+#endif
 	FreeAccessStrategy(vacuumstate->bas);
 	pfree(vacuumstate->ntup);
 	MemoryContextDelete(vacuumstate->tmpCtx);
@@ -623,13 +796,16 @@ hnswbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 	InitVacuumState(&vacuumstate, info, stats, callback, callback_state);
 
 	/* Pass 1: Remove heap TIDs */
-	RemoveHeapTids(&vacuumstate);
+	HnswBench("RemoveHeapTids", RemoveHeapTids(&vacuumstate));
 
 	/* Pass 2: Repair graph */
-	RepairGraph(&vacuumstate);
+	HnswBench("RepairGraph", RepairGraph(&vacuumstate));
 
-	/* Pass 3: Mark as deleted */
-	MarkDeleted(&vacuumstate);
+	/* Pass 3: Confirm repaired */
+	HnswBench("ConfirmRepaired", ConfirmRepaired(&vacuumstate));
+
+	/* Pass 4: Mark as deleted */
+	HnswBench("MarkDeleted", MarkDeleted(&vacuumstate));
 
 	FreeVacuumState(&vacuumstate);
 

src/ivfbuild.c

@@ -2,23 +2,37 @@
 
 #include <float.h>
 
+#include "access/genam.h"
+#include "access/generic_xlog.h"
+#include "access/itup.h"
+#include "access/relscan.h"
 #include "access/table.h"
 #include "access/tableam.h"
+#include "access/tupdesc.h"
 #include "access/parallel.h"
 #include "access/xact.h"
-#include "bitvec.h"
+#include "access/xloginsert.h"
 #include "catalog/index.h"
 #include "catalog/pg_operator_d.h"
 #include "catalog/pg_type_d.h"
 #include "commands/progress.h"
-#include "halfvec.h"
+#include "fmgr.h"
 #include "ivfflat.h"
 #include "miscadmin.h"
+#include "nodes/execnodes.h"
 #include "optimizer/optimizer.h"
 #include "storage/bufmgr.h"
+#include "storage/condition_variable.h"
 #include "tcop/tcopprot.h"
 #include "utils/memutils.h"
-#include "vector.h"
+#include "utils/rel.h"
+#include "utils/sampling.h"
+#include "utils/snapmgr.h"
+#include "utils/tuplesort.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_progress.h"
@@ -49,15 +63,13 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
 
 	/*
-	 * Normalize with KMEANS_NORM_PROC since spherical distance function
-	 * expects unit vectors
+	 * Check with KMEANS_NORM_PROC that the value can be normalized since
+	 * spherical distance function expects unit vectors
 	 */
 	if (buildstate->kmeansnormprocinfo != NULL)
 	{
 		if (!IvfflatCheckNorm(buildstate->kmeansnormprocinfo, buildstate->collation, value))
 			return;
-
-		value = IvfflatNormValue(buildstate->typeInfo, buildstate->collation, value);
 	}
 
 	if (samples->length < targsamples)
@@ -68,7 +80,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 	else
 	{
 		if (buildstate->rowstoskip < 0)
-			buildstate->rowstoskip = reservoir_get_next_S(&buildstate->rstate, samples->length, targsamples);
+			buildstate->rowstoskip = reservoir_get_next_S(&buildstate->rstate, buildstate->samplerows, targsamples);
 
 		if (buildstate->rowstoskip <= 0)
 		{
@@ -84,6 +96,9 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 
 		buildstate->rowstoskip -= 1;
 	}
+
+	/* Increment after reservoir_get_next_S */
+	buildstate->samplerows += 1;
 }
 
 /*
@@ -120,6 +135,7 @@ SampleRows(IvfflatBuildState * buildstate)
 	int			targsamples = buildstate->samples->maxlen;
 	BlockNumber totalblocks = RelationGetNumberOfBlocks(buildstate->heap);
 
+	buildstate->samplerows = 0;
 	buildstate->rowstoskip = -1;
 
 	BlockSampler_Init(&buildstate->bs, totalblocks, targsamples, RandomInt());
@@ -129,8 +145,24 @@ SampleRows(IvfflatBuildState * buildstate)
 	{
 		BlockNumber targblock = BlockSampler_Next(&buildstate->bs);
 
+		/* Set anyvisible to false like table_index_build_scan */
 		table_index_build_range_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
-									 false, true, false, targblock, 1, SampleCallback, (void *) buildstate, NULL);
+									 false, false, false, targblock, 1, SampleCallback, (void *) buildstate, NULL);
+	}
+
+	/* Normalize if needed */
+	if (buildstate->kmeansnormprocinfo != NULL)
+	{
+		VectorArray samples = buildstate->samples;
+
+		for (int i = 0; i < samples->length; i++)
+		{
+			Datum		value = PointerGetDatum(VectorArrayGet(samples, i));
+			Datum		normValue = IvfflatNormValue(buildstate->typeInfo, buildstate->collation, value);
+
+			VectorArraySet(samples, i, DatumGetPointer(normValue));
+			pfree(DatumGetPointer(normValue));
+		}
 	}
 }
 
@@ -138,7 +170,7 @@ SampleRows(IvfflatBuildState * buildstate)
  * Add tuple to sort
  */
 static void
-AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
+AddTupleToSort(ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
 {
 	double		distance;
 	double		minDistance = DBL_MAX;
@@ -215,7 +247,7 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
 	oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
 
 	/* Add tuple to sort */
-	AddTupleToSort(index, tid, values, buildstate);
+	AddTupleToSort(tid, values, buildstate);
 
 	/* Reset memory context */
 	MemoryContextSwitchTo(oldCtx);
@@ -361,9 +393,13 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
 	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
 	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 3, "vector", TupleDescAttr(buildstate->tupdesc, 0)->atttypid, -1, 0);
+#if PG_VERSION_NUM >= 190000
+	TupleDescFinalize(buildstate->sortdesc);
+#endif
 
 	buildstate->slot = MakeSingleTupleTableSlot(buildstate->sortdesc, &TTSOpsVirtual);
 
+	/* TODO Ensure within maintenance_work_mem */
 	buildstate->centers = VectorArrayInit(buildstate->lists, buildstate->dimensions, buildstate->typeInfo->itemSize(buildstate->dimensions));
 	buildstate->listInfo = palloc(sizeof(ListInfo) * buildstate->lists);
 
@@ -422,7 +458,7 @@ ComputeCenters(IvfflatBuildState * buildstate)
 	buildstate->samples = VectorArrayInit(numSamples, buildstate->dimensions, buildstate->centers->itemsize);
 	if (buildstate->heap != NULL)
 	{
-		SampleRows(buildstate);
+		IvfflatBench("sample rows", SampleRows(buildstate));
 
 		if (buildstate->samples->length < buildstate->lists)
 		{
@@ -470,8 +506,8 @@ CreateMetaPage(Relation index, int dimensions, int lists, ForkNumber forkNum)
  * Create list pages
  */
 static void
-CreateListPages(Relation index, VectorArray centers, int dimensions,
-				int lists, ForkNumber forkNum, ListInfo * *listInfo)
+CreateListPages(Relation index, VectorArray centers, int lists,
+				ForkNumber forkNum, ListInfo * *listInfo)
 {
 	Buffer		buf;
 	Page		page;
@@ -637,7 +673,11 @@ IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, S
 	ivfspool->sortstate = InitBuildSortState(buildstate.sortdesc, sortmem, coordinate);
 	buildstate.sortstate = ivfspool->sortstate;
 	scan = table_beginscan_parallel(ivfspool->heap,
-									ParallelTableScanFromIvfflatShared(ivfshared));
+									ParallelTableScanFromIvfflatShared(ivfshared)
+#if PG_VERSION_NUM >= 190000
+									,SO_NONE
+#endif
+		);
 	reltuples = table_index_build_scan(ivfspool->heap, ivfspool->index, indexInfo,
 									   true, progress, BuildCallback,
 									   (void *) &buildstate, scan);
@@ -1004,7 +1044,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 
 	/* Create pages */
 	CreateMetaPage(index, buildstate->dimensions, buildstate->lists, forkNum);
-	CreateListPages(index, buildstate->centers, buildstate->dimensions, buildstate->lists, forkNum, &buildstate->listInfo);
+	CreateListPages(index, buildstate->centers, buildstate->lists, forkNum, &buildstate->listInfo);
 	CreateEntryPages(buildstate, forkNum);
 
 	/* Write WAL for initialization fork since GenericXLog functions do not */

src/ivfflat.c

@@ -3,14 +3,19 @@
 #include <float.h>
 
 #include "access/amapi.h"
+#include "access/genam.h"
 #include "access/reloptions.h"
 #include "commands/progress.h"
 #include "commands/vacuum.h"
+#include "fmgr.h"
 #include "ivfflat.h"
+#include "nodes/pg_list.h"
 #include "utils/float.h"
 #include "utils/guc.h"
+#include "utils/relcache.h"
 #include "utils/selfuncs.h"
 #include "utils/spccache.h"
+#include "vector.h"
 
 #if PG_VERSION_NUM < 150000
 #define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
@@ -179,6 +184,64 @@ FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflathandler);
 Datum
 ivfflathandler(PG_FUNCTION_ARGS)
 {
+#if PG_VERSION_NUM >= 190000
+	static const IndexAmRoutine amroutine = {
+		.type = T_IndexAmRoutine,
+		.amstrategies = 0,
+		.amsupport = 5,
+		.amoptsprocnum = 0,
+		.amcanorder = false,
+		.amcanorderbyop = true,
+		.amcanhash = false,
+		.amconsistentequality = false,
+		.amconsistentordering = false,
+		.amcanbackward = false,
+		.amcanunique = false,
+		.amcanmulticol = false,
+		.amoptionalkey = true,
+		.amsearcharray = false,
+		.amsearchnulls = false,
+		.amstorage = false,
+		.amclusterable = false,
+		.ampredlocks = false,
+		.amcanparallel = false,
+		.amcanbuildparallel = true,
+		.amcaninclude = false,
+		.amusemaintenanceworkmem = false,
+		.amsummarizing = false,
+		.amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL,
+		.amkeytype = InvalidOid,
+
+		.ambuild = ivfflatbuild,
+		.ambuildempty = ivfflatbuildempty,
+		.aminsert = ivfflatinsert,
+		.aminsertcleanup = NULL,
+		.ambulkdelete = ivfflatbulkdelete,
+		.amvacuumcleanup = ivfflatvacuumcleanup,
+		.amcanreturn = NULL,
+		.amcostestimate = ivfflatcostestimate,
+		.amgettreeheight = NULL,
+		.amoptions = ivfflatoptions,
+		.amproperty = NULL,
+		.ambuildphasename = ivfflatbuildphasename,
+		.amvalidate = ivfflatvalidate,
+		.amadjustmembers = NULL,
+		.ambeginscan = ivfflatbeginscan,
+		.amrescan = ivfflatrescan,
+		.amgettuple = ivfflatgettuple,
+		.amgetbitmap = NULL,
+		.amendscan = ivfflatendscan,
+		.ammarkpos = NULL,
+		.amrestrpos = NULL,
+		.amestimateparallelscan = NULL,
+		.aminitparallelscan = NULL,
+		.amparallelrescan = NULL,
+		.amtranslatestrategy = NULL,
+		.amtranslatecmptype = NULL,
+	};
+
+	PG_RETURN_POINTER(&amroutine);
+#else
 	IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
 
 	amroutine->amstrategies = 0;
@@ -252,4 +315,5 @@ ivfflathandler(PG_FUNCTION_ARGS)
 #endif
 
 	PG_RETURN_POINTER(amroutine);
+#endif
 }

src/ivfflat.h

@@ -9,10 +9,15 @@
 #include "lib/pairingheap.h"
 #include "nodes/execnodes.h"
 #include "port.h"				/* for random() */
+#include "storage/condition_variable.h"
 #include "utils/sampling.h"
 #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
@@ -21,6 +26,10 @@
 #include "portability/instr_time.h"
 #endif
 
+#if PG_VERSION_NUM >= 190000
+typedef Pointer Item;
+#endif
+
 #define IVFFLAT_MAX_DIM 2000
 
 /* Support functions */
@@ -205,7 +214,8 @@ typedef struct IvfflatBuildState
 	/* Sampling */
 	BlockSamplerData bs;
 	ReservoirStateData rstate;
-	int			rowstoskip;
+	double		samplerows;
+	double		rowstoskip;
 
 	/* Sorting */
 	Tuplesortstate *sortstate;

src/ivfinsert.c

@@ -2,11 +2,16 @@
 
 #include <float.h>
 
+#include "access/genam.h"
 #include "access/generic_xlog.h"
+#include "access/itup.h"
+#include "fmgr.h"
 #include "ivfflat.h"
+#include "nodes/execnodes.h"
 #include "storage/bufmgr.h"
 #include "storage/lmgr.h"
 #include "utils/memutils.h"
+#include "utils/rel.h"
 
 /*
  * Find the list that minimizes the distance function
@@ -65,7 +70,7 @@ 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, Relation heapRel)
+InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid)
 {
 	const		IvfflatTypeInfo *typeInfo = IvfflatGetTypeInfo(index);
 	IndexTuple	itup;
@@ -204,7 +209,7 @@ ivfflatinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid,
 	oldCtx = MemoryContextSwitchTo(insertCtx);
 
 	/* Insert tuple */
-	InsertTuple(index, values, isnull, heap_tid, heap);
+	InsertTuple(index, values, isnull, heap_tid);
 
 	/* Delete memory context */
 	MemoryContextSwitchTo(oldCtx);

src/ivfkmeans.c

@@ -1,17 +1,19 @@
 #include "postgres.h"
 
 #include <float.h>
+#include <limits.h>
 #include <math.h>
 
-#include "bitvec.h"
-#include "halfutils.h"
-#include "halfvec.h"
+#include "access/genam.h"
+#include "fmgr.h"
 #include "ivfflat.h"
 #include "miscadmin.h"
-#include "utils/builtins.h"
-#include "utils/datum.h"
 #include "utils/memutils.h"
-#include "vector.h"
+#include "utils/relcache.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 /*
  * Initialize with kmeans++

src/ivfscan.c

@@ -2,15 +2,26 @@
 
 #include <float.h>
 
+#include "access/genam.h"
+#include "access/itup.h"
 #include "access/relscan.h"
+#include "access/tupdesc.h"
 #include "catalog/pg_operator_d.h"
 #include "catalog/pg_type_d.h"
+#include "fmgr.h"
 #include "lib/pairingheap.h"
 #include "ivfflat.h"
 #include "miscadmin.h"
 #include "pgstat.h"
 #include "storage/bufmgr.h"
 #include "utils/memutils.h"
+#include "utils/rel.h"
+#include "utils/snapmgr.h"
+#include "utils/tuplesort.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 #define GetScanList(ptr) pairingheap_container(IvfflatScanList, ph_node, ptr)
 #define GetScanListConst(ptr) pairingheap_const_container(IvfflatScanList, ph_node, ptr)
@@ -287,6 +298,9 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	so->tupdesc = CreateTemplateTupleDesc(2);
 	TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
 	TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "heaptid", TIDOID, -1, 0);
+#if PG_VERSION_NUM >= 190000
+	TupleDescFinalize(so->tupdesc);
+#endif
 
 	/* Prep sort */
 	so->sortstate = InitScanSortState(so->tupdesc);
@@ -355,6 +369,10 @@ 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)

src/ivfutils.c

@@ -1,13 +1,19 @@
 #include "postgres.h"
 
+#include "access/genam.h"
 #include "access/generic_xlog.h"
-#include "bitvec.h"
-#include "catalog/pg_type.h"
 #include "fmgr.h"
 #include "halfutils.h"
 #include "halfvec.h"
 #include "ivfflat.h"
 #include "storage/bufmgr.h"
+#include "utils/relcache.h"
+#include "utils/varbit.h"
+#include "vector.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
 
 /*
  * Allocate a vector array

src/ivfvacuum.c

@@ -1,9 +1,12 @@
 #include "postgres.h"
 
+#include "access/genam.h"
 #include "access/generic_xlog.h"
+#include "access/itup.h"
 #include "commands/vacuum.h"
 #include "ivfflat.h"
 #include "storage/bufmgr.h"
+#include "utils/relcache.h"
 
 #if PG_VERSION_NUM >= 180000
 #define vacuum_delay_point() vacuum_delay_point(false)

src/sparsevec.c

@@ -5,18 +5,27 @@
 
 #include "catalog/pg_type.h"
 #include "common/shortest_dec.h"
-#include "common/string.h"
 #include "fmgr.h"
 #include "halfutils.h"
 #include "halfvec.h"
+#include "lib/stringinfo.h"
 #include "libpq/pqformat.h"
 #include "sparsevec.h"
 #include "utils/array.h"
 #include "utils/builtins.h"
 #include "utils/float.h"
+#include "utils/fmgrprotos.h"
 #include "utils/lsyscache.h"
 #include "vector.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
+#if PG_VERSION_NUM >= 170000
+#include "parser/scansup.h"
+#endif
+
 typedef struct SparseInputElement
 {
 	int32		index;
@@ -150,9 +159,9 @@ InitSparseVector(int dim, int nnz)
 	return result;
 }
 
-/*
- * Check for whitespace, since array_isspace() is static
- */
+#if PG_VERSION_NUM >= 170000
+#define sparsevec_isspace(ch) scanner_isspace(ch)
+#else
 static inline bool
 sparsevec_isspace(char ch)
 {
@@ -165,6 +174,7 @@ sparsevec_isspace(char ch)
 		return true;
 	return false;
 }
+#endif
 
 /*
  * Compare indices
@@ -172,10 +182,10 @@ sparsevec_isspace(char ch)
 static int
 CompareIndices(const void *a, const void *b)
 {
-	if (((SparseInputElement *) a)->index < ((SparseInputElement *) b)->index)
+	if (((const SparseInputElement *) a)->index < ((const SparseInputElement *) b)->index)
 		return -1;
 
-	if (((SparseInputElement *) a)->index > ((SparseInputElement *) b)->index)
+	if (((const SparseInputElement *) a)->index > ((const SparseInputElement *) b)->index)
 		return 1;
 
 	return 0;

src/vector.c

@@ -16,16 +16,20 @@
 #include "port.h"				/* for strtof() */
 #include "sparsevec.h"
 #include "utils/array.h"
-#include "utils/builtins.h"
 #include "utils/float.h"
+#include "utils/fmgrprotos.h"
 #include "utils/lsyscache.h"
-#include "utils/numeric.h"
+#include "utils/varbit.h"
 #include "vector.h"
 
 #if PG_VERSION_NUM >= 160000
 #include "varatt.h"
 #endif
 
+#if PG_VERSION_NUM >= 170000
+#include "parser/scansup.h"
+#endif
+
 #define STATE_DIMS(x) (ARR_DIMS(x)[0] - 1)
 #define CreateStateDatums(dim) palloc(sizeof(Datum) * (dim + 1))
 
@@ -36,7 +40,7 @@
 #endif
 
 #if PG_VERSION_NUM >= 180000
-PG_MODULE_MAGIC_EXT(.name = "vector",.version = "0.8.1");
+PG_MODULE_MAGIC_EXT(.name = "vector", .version = "0.8.2");
 #else
 PG_MODULE_MAGIC;
 #endif
@@ -129,9 +133,9 @@ InitVector(int dim)
 	return result;
 }
 
-/*
- * Check for whitespace, since array_isspace() is static
- */
+#if PG_VERSION_NUM >= 170000
+#define vector_isspace(ch) scanner_isspace(ch)
+#else
 static inline bool
 vector_isspace(char ch)
 {
@@ -144,6 +148,7 @@ vector_isspace(char ch)
 		return true;
 	return false;
 }
+#endif
 
 /*
  * Check state array

test/t/005_ivfflat_query_recall.pl

@@ -16,29 +16,19 @@ $node->safe_psql("postgres",
 	"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 100000) i;"
 );
 
-# Check each index type
-my @operators = ("<->", "<#>", "<=>");
-my @opclasses = ("vector_l2_ops", "vector_ip_ops", "vector_cosine_ops");
+# Add index
+$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v vector_l2_ops);");
 
-for my $i (0 .. $#operators)
+# Test 100% recall
+for (1 .. 20)
 {
-	my $operator = $operators[$i];
-	my $opclass = $opclasses[$i];
-
-	# Add index
-	$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v $opclass);");
-
-	# Test 100% recall
-	for (1 .. 20)
-	{
-		my $id = int(rand() * 100000);
-		my $query = $node->safe_psql("postgres", "SELECT v FROM tst WHERE i = $id;");
-		my $res = $node->safe_psql("postgres", qq(
-			SET enable_seqscan = off;
-			SELECT v FROM tst ORDER BY v <-> '$query' LIMIT 1;
-		));
-		is($res, $query);
-	}
+	my $id = int(rand() * 100000);
+	my $query = $node->safe_psql("postgres", "SELECT v FROM tst WHERE i = $id;");
+	my $res = $node->safe_psql("postgres", qq(
+		SET enable_seqscan = off;
+		SELECT v FROM tst ORDER BY v <-> '$query' LIMIT 1;
+	));
+	is($res, $query);
 }
 
 done_testing();

test/t/045_hnsw_low_memory_build.pl

@@ -0,0 +1,29 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+# Initialize node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (v vector(3));");
+$node->safe_psql("postgres",
+	"INSERT INTO tst SELECT ARRAY[random(), random(), random()] FROM generate_series(1, 1000) i;"
+);
+
+my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
+	SET client_min_messages = DEBUG;
+	SET maintenance_work_mem = '3073kB';
+	ALTER TABLE tst SET (parallel_workers = 1);
+	CREATE INDEX ON tst USING hnsw (v vector_l2_ops);
+));
+is($ret, 0, $stderr);
+like($stderr, qr/using \d+ parallel workers/);
+like($stderr, qr/hnsw graph no longer fits into maintenance_work_mem after 0 tuples/);
+
+done_testing();

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.8.1'
+default_version = '0.8.2'
 module_pathname = '$libdir/vector'
 relocatable = true