Fixed locking for non-MVCC snapshots

.github/workflows/build.yml

@@ -8,8 +8,6 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - postgres: 17
-            os: ubuntu-22.04
           - postgres: 16
             os: ubuntu-22.04
           - postgres: 15
@@ -23,7 +21,7 @@ jobs:
           - postgres: 11
             os: ubuntu-20.04
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v3
       - uses: ankane/setup-postgres@v1
         with:
           postgres-version: ${{ matrix.postgres }}
@@ -45,7 +43,7 @@ jobs:
     runs-on: macos-latest
     if: ${{ !startsWith(github.ref_name, 'windows') }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v3
       - uses: ankane/setup-postgres@v1
         with:
           postgres-version: 14
@@ -67,7 +65,7 @@ jobs:
     runs-on: windows-latest
     if: ${{ !startsWith(github.ref_name, 'mac') }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v3
       - uses: ankane/setup-postgres@v1
         with:
           postgres-version: 14

CHANGELOG.md

@@ -1,16 +1,7 @@
-## 0.5.2 (unreleased)
+## 0.5.1 (unreleased)
 
-- Improved performance of HNSW
-- Added support for on-disk parallel index builds for HNSW
-- Reduced memory usage for HNSW index builds
-- Reduced WAL generation for HNSW index builds
-- Fixed error with logical replication
-- Fixed `invalid memory alloc request size` error with HNSW index build
-
-## 0.5.1 (2023-10-10)
-
-- Improved performance of HNSW index builds
-- Added check for MVCC-compliant snapshot for index scans
+- Improved performance of index scans for IVFFlat after updates and deletes
+- Fixed locking for index scans for HNSW
 
 ## 0.5.0 (2023-08-28)
 

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": "0.5.1",
+	"version": "0.5.0",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.5.1",
+			"version": "0.5.0",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.5.1
+EXTVERSION = 0.5.0
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.5.1
+EXTVERSION = 0.5.0
 
 OBJS = 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\vector.obj
 HEADERS = src\vector.h
@@ -56,7 +56,7 @@ install:
 	copy $(EXTENSION).control "$(SHAREDIR)\extension"
 	copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
 	mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
-	for %f in ($(HEADERS)) do copy %f "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
+	copy $(HEADERS) "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
 
 installcheck:
 	"$(BINDIR)\pg_regress" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)

README.md

@@ -10,7 +10,7 @@ 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
 
-[![Build Status](https://github.com/pgvector/pgvector/actions/workflows/build.yml/badge.svg)](https://github.com/pgvector/pgvector/actions)
+[![Build Status](https://github.com/pgvector/pgvector/workflows/build/badge.svg?branch=master)](https://github.com/pgvector/pgvector/actions)
 
 ## Installation
 
@@ -18,7 +18,7 @@ Compile and install the extension (supports Postgres 11+)
 
 ```sh
 cd /tmp
-git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
+git clone --branch v0.5.0 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -26,7 +26,7 @@ make install # may need sudo
 
 See the [installation notes](#installation-notes) if you run into issues
 
-You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), 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), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres)
 
 ## Getting Started
 
@@ -161,97 +161,8 @@ You can add an index to use approximate nearest neighbor search, which trades so
 
 Supported index types are:
 
-- [HNSW](#hnsw) - added in 0.5.0
 - [IVFFlat](#ivfflat)
-
-## HNSW
-
-An HNSW index creates a multilayer graph. It has better query performance than IVFFlat (in terms of speed-recall tradeoff), but has slower build times and uses more memory. Also, an index can be created without any data in the table since there isn’t a training step like IVFFlat.
-
-Add an index for each distance function you want to use.
-
-L2 distance
-
-```sql
-CREATE INDEX ON items USING hnsw (embedding vector_l2_ops);
-```
-
-Inner product
-
-```sql
-CREATE INDEX ON items USING hnsw (embedding vector_ip_ops);
-```
-
-Cosine distance
-
-```sql
-CREATE INDEX ON items USING hnsw (embedding vector_cosine_ops);
-```
-
-Vectors with up to 2,000 dimensions can be indexed.
-
-### Index Options
-
-Specify HNSW parameters
-
-- `m` - the max number of connections per layer (16 by default)
-- `ef_construction` - the size of the dynamic candidate list for constructing the graph (64 by default)
-
-```sql
-CREATE INDEX ON items USING hnsw (embedding vector_l2_ops) WITH (m = 16, ef_construction = 64);
-```
-
-A higher value of `ef_construction` provides better recall at the cost of index build time / insert speed.
-
-### Query Options
-
-Specify the size of the dynamic candidate list for search (40 by default)
-
-```sql
-SET hnsw.ef_search = 100;
-```
-
-A higher value provides better recall at the cost of speed.
-
-Use `SET LOCAL` inside a transaction to set it for a single query
-
-```sql
-BEGIN;
-SET LOCAL hnsw.ef_search = 100;
-SELECT ...
-COMMIT;
-```
-
-### Index Build Time
-
-Indexes build significantly faster when the graph fits into `maintenance_work_mem`
-
-```sql
-SET maintenance_work_mem = '8GB';
-```
-
-A notice is shown when the graph no longer fits
-
-```text
-NOTICE:  hnsw graph no longer fits into maintenance_work_mem after 100000 tuples
-DETAIL:  Building will take significantly more time.
-HINT:  Increase maintenance_work_mem to speed up builds.
-```
-
-Note: Do not set `maintenance_work_mem` so high that it exhausts the memory on the server
-
-### Indexing Progress
-
-Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING) with Postgres 12+
-
-```sql
-SELECT phase, round(100.0 * blocks_done / nullif(blocks_total, 0), 1) AS "%" FROM pg_stat_progress_create_index;
-```
-
-The phases for HNSW are:
-
-1. `initializing`
-2. `loading tuples`
+- [HNSW](#hnsw) - added in 0.5.0
 
 ## IVFFlat
 
@@ -304,32 +215,78 @@ SELECT ...
 COMMIT;
 ```
 
-### Index Build Time
+## HNSW
 
-Speed up index creation on large tables by increasing the number of parallel workers (2 by default)
+An HNSW index creates a multilayer graph. It has slower build times and uses more memory than IVFFlat, but has better query performance (in terms of speed-recall tradeoff). There’s no training step like IVFFlat, so the index can be created without any data in the table.
+
+Add an index for each distance function you want to use.
+
+L2 distance
 
 ```sql
-SET max_parallel_maintenance_workers = 7; -- plus leader
+CREATE INDEX ON items USING hnsw (embedding vector_l2_ops);
 ```
 
-For a large number of workers, you may also need to increase `max_parallel_workers` (8 by default)
+Inner product
 
-### Indexing Progress
+```sql
+CREATE INDEX ON items USING hnsw (embedding vector_ip_ops);
+```
+
+Cosine distance
+
+```sql
+CREATE INDEX ON items USING hnsw (embedding vector_cosine_ops);
+```
+
+Vectors with up to 2,000 dimensions can be indexed.
+
+### Index Options
+
+Specify HNSW parameters
+
+- `m` - the max number of connections per layer (16 by default)
+- `ef_construction` - the size of the dynamic candidate list for constructing the graph (64 by default)
+
+```sql
+CREATE INDEX ON items USING hnsw (embedding vector_l2_ops) WITH (m = 16, ef_construction = 64);
+```
+
+### Query Options
+
+Specify the size of the dynamic candidate list for search (40 by default)
+
+```sql
+SET hnsw.ef_search = 100;
+```
+
+A higher value provides better recall at the cost of speed.
+
+Use `SET LOCAL` inside a transaction to set it for a single query
+
+```sql
+BEGIN;
+SET LOCAL hnsw.ef_search = 100;
+SELECT ...
+COMMIT;
+```
+
+## Indexing Progress
 
 Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING) with Postgres 12+
 
 ```sql
-SELECT phase, round(100.0 * tuples_done / nullif(tuples_total, 0), 1) AS "%" FROM pg_stat_progress_create_index;
+SELECT phase, tuples_done, tuples_total FROM pg_stat_progress_create_index;
 ```
 
-The phases for IVFFlat are:
+The phases are:
 
 1. `initializing`
-2. `performing k-means`
-3. `assigning tuples`
+2. `performing k-means` - IVFFlat only
+3. `assigning tuples` - IVFFlat only
 4. `loading tuples`
 
-Note: `%` is only populated during the `loading tuples` phase
+Note: `tuples_done` and `tuples_total` are only populated during the `loading tuples` phase
 
 ## Filtering
 
@@ -348,7 +305,8 @@ CREATE INDEX ON items (category_id);
 Or a [partial index](https://www.postgresql.org/docs/current/indexes-partial.html) on the vector column for approximate search
 
 ```sql
-CREATE INDEX ON items USING hnsw (embedding vector_l2_ops) WHERE (category_id = 123);
+CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100)
+    WHERE (category_id = 123);
 ```
 
 Use [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) for approximate search on many different values of the `WHERE` columns
@@ -359,15 +317,13 @@ CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(cate
 
 ## Hybrid Search
 
-Use together with Postgres [full-text search](https://www.postgresql.org/docs/current/textsearch-intro.html) for hybrid search.
+Use together with Postgres [full-text search](https://www.postgresql.org/docs/current/textsearch-intro.html) for hybrid search ([Python example](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search.py)).
 
 ```sql
 SELECT id, content FROM items, plainto_tsquery('hello search') query
     WHERE textsearch @@ query ORDER BY ts_rank_cd(textsearch, query) DESC LIMIT 5;
 ```
 
-You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search_rrf.py) or a [cross-encoder](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search.py) to combine results.
-
 ## Performance
 
 Use `EXPLAIN ANALYZE` to debug performance.
@@ -404,21 +360,17 @@ Use pgvector from any language with a Postgres client. You can even generate and
 
 Language | Libraries / Examples
 --- | ---
-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)
+C# | [pgvector-dotnet](https://github.com/pgvector/pgvector-dotnet)
 Crystal | [pgvector-crystal](https://github.com/pgvector/pgvector-crystal)
 Dart | [pgvector-dart](https://github.com/pgvector/pgvector-dart)
 Elixir | [pgvector-elixir](https://github.com/pgvector/pgvector-elixir)
 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)
+Java, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
 Julia | [pgvector-julia](https://github.com/pgvector/pgvector-julia)
-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)
+Node.js | [pgvector-node](https://github.com/pgvector/pgvector-node)
 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)
@@ -426,7 +378,6 @@ R | [pgvector-r](https://github.com/pgvector/pgvector-r)
 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)
-Zig | [pgvector-zig](https://github.com/pgvector/pgvector-zig)
 
 ## Frequently Asked Questions
 
@@ -442,63 +393,6 @@ Yes, pgvector uses the write-ahead log (WAL), which allows for replication and p
 
 You’ll need to use [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction) at the moment.
 
-#### Can I store vectors with different dimensions in the same column?
-
-You can use `vector` as the type (instead of `vector(3)`).
-
-```sql
-CREATE TABLE embeddings (model_id bigint, item_id bigint, embedding vector, PRIMARY KEY (model_id, item_id));
-```
-
-However, you can only create indexes on rows with the same number of dimensions (using [expression](https://www.postgresql.org/docs/current/indexes-expressional.html) and [partial](https://www.postgresql.org/docs/current/indexes-partial.html) indexing):
-
-```sql
-CREATE INDEX ON embeddings USING hnsw ((embedding::vector(3)) vector_l2_ops) WHERE (model_id = 123);
-```
-
-and query with:
-
-```sql
-SELECT * FROM embeddings WHERE model_id = 123 ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
-```
-
-#### Can I store vectors with more precision?
-
-You can use the `double precision[]` or `numeric[]` type to store vectors with more precision.
-
-```sql
-CREATE TABLE items (id bigserial PRIMARY KEY, embedding double precision[]);
-
--- use {} instead of [] for Postgres arrays
-INSERT INTO items (embedding) VALUES ('{1,2,3}'), ('{4,5,6}');
-```
-
-Optionally, add a [check constraint](https://www.postgresql.org/docs/current/ddl-constraints.html) to ensure data can be converted to the `vector` type and has the expected dimensions.
-
-```sql
-ALTER TABLE items ADD CHECK (vector_dims(embedding::vector) = 3);
-```
-
-Use [expression indexing](https://www.postgresql.org/docs/current/indexes-expressional.html) to index (at a lower precision):
-
-```sql
-CREATE INDEX ON items USING hnsw ((embedding::vector(3)) vector_l2_ops);
-```
-
-and query with:
-
-```sql
-SELECT * FROM items ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
-```
-
-#### Do indexes need to fit into memory?
-
-No, but like other index types, you’ll likely see better performance if they do. You can get the size of an index with:
-
-```sql
-SELECT pg_size_pretty(pg_relation_size('index_name'));
-```
-
 ## Troubleshooting
 
 #### Why isn’t a query using an index?
@@ -512,8 +406,6 @@ SELECT ...
 COMMIT;
 ```
 
-Also, if the table is small, a table scan may be faster.
-
 #### Why isn’t a query using a parallel table scan?
 
 The planner doesn’t consider [out-of-line storage](https://www.postgresql.org/docs/current/storage-toast.html) in cost estimates, which can make a serial scan look cheaper. You can reduce the cost of a parallel scan for a query with:
@@ -582,7 +474,7 @@ sum(vector) → vector | sum | 0.5.0
 If your machine has multiple Postgres installations, specify the path to [pg_config](https://www.postgresql.org/docs/current/app-pgconfig.html) with:
 
 ```sh
-export PG_CONFIG=/Library/PostgreSQL/16/bin/pg_config
+export PG_CONFIG=/Applications/Postgres.app/Contents/Versions/latest/bin/pg_config
 ```
 
 Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
@@ -591,14 +483,6 @@ Then re-run the installation instructions (run `make clean` before `make` if nee
 sudo --preserve-env=PG_CONFIG make install
 ```
 
-A few common paths on Mac are:
-
-- EDB installer - `/Library/PostgreSQL/16/bin/pg_config`
-- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@16/bin/pg_config`
-- Homebrew (x86-64) - `/usr/local/opt/postgresql@16/bin/pg_config`
-
-Note: Replace `16` with your Postgres server version
-
 ### Missing Header
 
 If compilation fails with `fatal error: postgres.h: No such file or directory`, make sure Postgres development files are installed on the server.
@@ -606,14 +490,10 @@ If compilation fails with `fatal error: postgres.h: No such file or directory`,
 For Ubuntu and Debian, use:
 
 ```sh
-sudo apt install postgresql-server-dev-16
+sudo apt install postgresql-server-dev-15
 ```
 
-Note: Replace `16` 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.
+Note: Replace `15` with your Postgres server version
 
 ### Windows
 
@@ -628,8 +508,8 @@ Note: The exact path will vary depending on your Visual Studio version and editi
 Then use `nmake` to build:
 
 ```cmd
-set "PGROOT=C:\Program Files\PostgreSQL\16"
-git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
+set "PGROOT=C:\Program Files\PostgreSQL\15"
+git clone --branch v0.5.0 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -650,7 +530,7 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
+git clone --branch v0.5.0 https://github.com/pgvector/pgvector.git
 cd pgvector
 docker build --build-arg PG_MAJOR=15 -t myuser/pgvector .
 ```
@@ -678,22 +558,22 @@ pgxn install vector
 Debian and Ubuntu packages are available from the [PostgreSQL APT Repository](https://wiki.postgresql.org/wiki/Apt). Follow the [setup instructions](https://wiki.postgresql.org/wiki/Apt#Quickstart) and run:
 
 ```sh
-sudo apt install postgresql-16-pgvector
+sudo apt install postgresql-15-pgvector
 ```
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `15` with your Postgres server version
 
 ### Yum
 
 RPM packages are available from the [PostgreSQL Yum Repository](https://yum.postgresql.org/). Follow the [setup instructions](https://www.postgresql.org/download/linux/redhat/) for your distribution and run:
 
 ```sh
-sudo yum install pgvector_16
+sudo yum install pgvector_15
 # or
-sudo dnf install pgvector_16
+sudo dnf install pgvector_15
 ```
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `15` with your Postgres server version
 
 ### conda-forge
 
@@ -715,7 +595,7 @@ pgvector is available on [these providers](https://github.com/pgvector/pgvector/
 
 ## Upgrading
 
-[Install](#installation) the latest version (use the same method as the original installation). Then in each database you want to upgrade, run:
+Install the latest version. Then in each database you want to upgrade, run:
 
 ```sql
 ALTER EXTENSION vector UPDATE;
@@ -803,25 +683,7 @@ make prove_installcheck PROVE_TESTS=test/t/001_wal.pl  # TAP test
 To enable benchmarking:
 
 ```sh
-make clean && PG_CFLAGS="-DIVFFLAT_BENCH" make && make install
-```
-
-To show memory usage:
-
-```sh
-make clean && PG_CFLAGS="-DHNSW_MEMORY -DIVFFLAT_MEMORY" make && make install
-```
-
-To enable assertions:
-
-```sh
-make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
-```
-
-To get k-means metrics:
-
-```sh
-make clean && PG_CFLAGS="-DIVFFLAT_KMEANS_DEBUG" make && make install
+make clean && PG_CFLAGS=-DIVFFLAT_BENCH make && make install
 ```
 
 Resources for contributors

sql/vector--0.5.0--0.5.1.sql

@@ -1,2 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.5.1'" to load this file. \quit

src/hnsw.c

@@ -14,7 +14,6 @@
 #endif
 
 int			hnsw_ef_search;
-bool		hnsw_enable_parallel_build;
 static relopt_kind hnsw_relopt_kind;
 
 /*
@@ -40,11 +39,6 @@ HnswInit(void)
 	DefineCustomIntVariable("hnsw.ef_search", "Sets the size of the dynamic candidate list for search",
 							"Valid range is 1..1000.", &hnsw_ef_search,
 							HNSW_DEFAULT_EF_SEARCH, HNSW_MIN_EF_SEARCH, HNSW_MAX_EF_SEARCH, PGC_USERSET, 0, NULL, NULL, NULL);
-
-	/* Behind a variable for now since can be slower than building in memory */
-	DefineCustomBoolVariable("hnsw.enable_parallel_build", "Enables or disables building indexes in parallel",
-							 NULL, &hnsw_enable_parallel_build,
-							 false, PGC_USERSET, 0, NULL, NULL, NULL);
 }
 
 /*
@@ -79,14 +73,12 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	int			m;
 	int			entryLevel;
 	Relation	index;
-	double		selectivity = 1;
-	ListCell   *lc;
 #if PG_VERSION_NUM < 120000
 	List	   *qinfos;
 #endif
 
-	/* Never use index without order or limit */
-	if (path->indexorderbys == NULL || root->limit_tuples < 0)
+	/* Never use index without order */
+	if (path->indexorderbys == NULL)
 	{
 		*indexStartupCost = DBL_MAX;
 		*indexTotalCost = DBL_MAX;
@@ -96,29 +88,6 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 		return;
 	}
 
-	/* Get the selectivity of non-index conditions */
-	foreach(lc, path->indexinfo->indrestrictinfo)
-	{
-		RestrictInfo *rinfo = lfirst(lc);
-
-		if (rinfo->norm_selec >= 0 && rinfo->norm_selec <= 1 && rinfo->norm_selec != (Selectivity) DEFAULT_INEQ_SEL)
-			selectivity *= rinfo->norm_selec;
-	}
-
-	/*
-	 * Do not use index if limit + offset > expected tuples unless
-	 * enable_seqscan = off
-	 */
-	if (root->limit_tuples > hnsw_ef_search * selectivity)
-	{
-		*indexStartupCost = 1.0e10 - 1;
-		*indexTotalCost = 1.0e10 - 1;
-		*indexSelectivity = 0;
-		*indexCorrelation = 0;
-		*indexPages = 0;
-		return;
-	}
-
 	MemSet(&costs, 0, sizeof(costs));
 
 	index = index_open(path->indexinfo->indexoid, NoLock);

src/hnsw.h

@@ -4,9 +4,7 @@
 #include "postgres.h"
 
 #include "access/generic_xlog.h"
-#include "access/parallel.h"
 #include "access/reloptions.h"
-#include "lib/ilist.h"
 #include "nodes/execnodes.h"
 #include "port.h"				/* for random() */
 #include "utils/sampling.h"
@@ -16,10 +14,6 @@
 #error "Requires PostgreSQL 11+"
 #endif
 
-#if PG_VERSION_NUM < 120000
-#include "access/relscan.h"
-#endif
-
 #define HNSW_MAX_DIM 2000
 
 /* Support functions */
@@ -63,9 +57,7 @@
 /* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
 #define PROGRESS_HNSW_PHASE_LOAD		2
 
-#define HNSW_MAX_SIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - sizeof(ItemIdData))
-
-#define HNSW_ELEMENT_TUPLE_SIZE(size)	MAXALIGN(offsetof(HnswElementTupleData, data) + (size))
+#define HNSW_ELEMENT_TUPLE_SIZE(_dim)	MAXALIGN(offsetof(HnswElementTupleData, vec) + VECTOR_SIZE(_dim))
 #define HNSW_NEIGHBOR_TUPLE_SIZE(level, m)	MAXALIGN(offsetof(HnswNeighborTupleData, indextids) + ((level) + 2) * (m) * sizeof(ItemPointerData))
 
 #define HnswPageGetOpaque(page)	((HnswPageOpaque) PageGetSpecialPointer(page))
@@ -73,10 +65,8 @@
 
 #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)
 #else
 #define RandomDouble() (((double) random()) / MAX_RANDOM_VALUE)
-#define SeedRandom(seed) srandom(seed)
 #endif
 
 #if PG_VERSION_NUM < 130000
@@ -94,28 +84,24 @@
 #define HnswGetMl(m) (1 / log(m))
 
 /* Ensure fits on page and in uint8 */
-#define HnswGetMaxLevel(m) Min(((BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - offsetof(HnswNeighborTupleData, indextids) - sizeof(ItemIdData)) / (sizeof(ItemPointerData)) / (m)) - 2, 255)
-
-#define HnswGetNeighbors(element, lc) (AssertMacro((element)->level >= (lc)), &(element)->neighbors[lc])
+#define HnswGetMaxLevel(m) Min(((BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - offsetof(HnswNeighborTupleData, indextids) - sizeof(ItemIdData)) / (sizeof(ItemPointerData)) / m) - 2, 255)
 
 /* Variables */
 extern int	hnsw_ef_search;
-extern bool hnsw_enable_parallel_build;
+
+typedef struct HnswNeighborArray HnswNeighborArray;
 
 typedef struct HnswElementData
 {
-	slist_node	next;
-	ItemPointerData heaptids[HNSW_HEAPTIDS];
-	uint8		heaptidsLength;
+	List	   *heaptids;
 	uint8		level;
 	uint8		deleted;
-	uint32		hash;
-	struct HnswNeighborArray *neighbors;
+	HnswNeighborArray *neighbors;
 	BlockNumber blkno;
 	OffsetNumber offno;
 	OffsetNumber neighborOffno;
 	BlockNumber neighborPage;
-	Datum		value;
+	Vector	   *vec;
 }			HnswElementData;
 
 typedef HnswElementData * HnswElement;
@@ -124,13 +110,11 @@ typedef struct HnswCandidate
 {
 	HnswElement element;
 	float		distance;
-	bool		closer;
 }			HnswCandidate;
 
 typedef struct HnswNeighborArray
 {
 	int			length;
-	bool		closerSet;
 	HnswCandidate *items;
 }			HnswNeighborArray;
 
@@ -148,59 +132,6 @@ typedef struct HnswOptions
 	int			efConstruction; /* size of dynamic candidate list */
 }			HnswOptions;
 
-typedef struct HnswGraph
-{
-	slist_head	elements;
-	HnswElement entryPoint;
-	long		memoryUsed;
-	long		memoryTotal;
-	bool		flushed;
-	double		indtuples;
-}			HnswGraph;
-
-typedef struct HnswSpool
-{
-	Relation	heap;
-	Relation	index;
-}			HnswSpool;
-
-typedef struct HnswShared
-{
-	/* Immutable state */
-	Oid			heaprelid;
-	Oid			indexrelid;
-	bool		isconcurrent;
-	int			scantuplesortstates;
-
-	/* Worker progress */
-	ConditionVariable workersdonecv;
-
-	/* Mutex for mutable state */
-	slock_t		mutex;
-
-	/* Mutable state */
-	int			nparticipantsdone;
-	double		reltuples;
-	HnswGraph	graphData;
-
-#if PG_VERSION_NUM < 120000
-	ParallelHeapScanDescData heapdesc;	/* must come last */
-#endif
-}			HnswShared;
-
-#if PG_VERSION_NUM >= 120000
-#define ParallelTableScanFromHnswShared(shared) \
-	(ParallelTableScanDesc) ((char *) (shared) + BUFFERALIGN(sizeof(HnswShared)))
-#endif
-
-typedef struct HnswLeader
-{
-	ParallelContext *pcxt;
-	int			nparticipanttuplesorts;
-	HnswShared *hnswshared;
-	Snapshot	snapshot;
-}			HnswLeader;
-
 typedef struct HnswBuildState
 {
 	/* Info */
@@ -224,19 +155,16 @@ typedef struct HnswBuildState
 	Oid			collation;
 
 	/* Variables */
-	HnswGraph	graphData;
-	HnswGraph  *graph;
+	List	   *elements;
+	HnswElement entryPoint;
 	double		ml;
 	int			maxLevel;
+	double		maxInMemoryElements;
+	bool		flushed;
 	Vector	   *normvec;
 
 	/* Memory */
-	MemoryContext graphCtx;
 	MemoryContext tmpCtx;
-
-	/* Parallel builds */
-	HnswLeader *hnswleader;
-	HnswShared *hnswshared;
 }			HnswBuildState;
 
 typedef struct HnswMetaPageData
@@ -272,7 +200,7 @@ typedef struct HnswElementTupleData
 	ItemPointerData heaptids[HNSW_HEAPTIDS];
 	ItemPointerData neighbortid;
 	uint16		unused2;
-	Vector		data;
+	Vector		vec;
 }			HnswElementTupleData;
 
 typedef HnswElementTupleData * HnswElementTuple;
@@ -318,7 +246,7 @@ typedef struct HnswVacuumState
 	Oid			collation;
 
 	/* Variables */
-	struct tidhash_hash *deleted;
+	HTAB	   *deleted;
 	BufferAccessStrategy bas;
 	HnswNeighborTuple ntup;
 	HnswElementData highestPoint;
@@ -330,30 +258,33 @@ typedef struct HnswVacuumState
 /* Methods */
 int			HnswGetM(Relation index);
 int			HnswGetEfConstruction(Relation index);
-FmgrInfo   *HnswOptionalProcInfo(Relation index, uint16 procnum);
+FmgrInfo   *HnswOptionalProcInfo(Relation rel, uint16 procnum);
 bool		HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
+void		HnswCommitBuffer(Buffer buf, GenericXLogState *state);
 Buffer		HnswNewBuffer(Relation index, ForkNumber forkNum);
 void		HnswInitPage(Buffer buf, Page page);
+void		HnswInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state);
 void		HnswInit(void);
 List	   *HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement);
 HnswElement HnswGetEntryPoint(Relation index);
 void		HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
 HnswElement HnswInitElement(ItemPointer tid, int m, double ml, int maxLevel);
+void		HnswFreeElement(HnswElement element);
 HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
 void		HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing);
+HnswElement HnswFindDuplicate(HnswElement e);
 HnswCandidate *HnswEntryCandidate(HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
-void		HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building);
+void		HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum);
 void		HnswSetNeighborTuple(HnswNeighborTuple ntup, HnswElement e, int m);
 void		HnswAddHeapTid(HnswElement element, ItemPointer heaptid);
 void		HnswInitNeighbors(HnswElement element, int m);
-bool		HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel, bool building);
-void		HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building);
+bool		HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel);
+void		HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting);
 void		HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec);
 void		HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec);
 void		HnswSetElementTuple(HnswElementTuple etup, HnswElement element);
 void		HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
 void		HnswLoadNeighbors(HnswElement element, Relation index, int m);
-PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);
 
 /* Index access methods */
 IndexBuildResult *hnswbuild(Relation heap, Relation index, IndexInfo *indexInfo);
@@ -371,31 +302,4 @@ void		hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys,
 bool		hnswgettuple(IndexScanDesc scan, ScanDirection dir);
 void		hnswendscan(IndexScanDesc scan);
 
-/* Hash tables */
-typedef struct TidHashEntry
-{
-	ItemPointerData tid;
-	char		status;
-}			TidHashEntry;
-
-#define SH_PREFIX tidhash
-#define SH_ELEMENT_TYPE TidHashEntry
-#define SH_KEY_TYPE ItemPointerData
-#define SH_SCOPE extern
-#define SH_DECLARE
-#include "lib/simplehash.h"
-
-typedef struct PointerHashEntry
-{
-	uintptr_t	ptr;
-	char		status;
-}			PointerHashEntry;
-
-#define SH_PREFIX pointerhash
-#define SH_ELEMENT_TYPE PointerHashEntry
-#define SH_KEY_TYPE uintptr_t
-#define SH_SCOPE extern
-#define SH_DECLARE
-#include "lib/simplehash.h"
-
 #endif

src/hnswbuild.c

@@ -2,16 +2,12 @@
 
 #include <math.h>
 
-#include "access/parallel.h"
-#include "access/xact.h"
 #include "catalog/index.h"
 #include "hnsw.h"
 #include "miscadmin.h"
 #include "lib/pairingheap.h"
 #include "nodes/pg_list.h"
 #include "storage/bufmgr.h"
-#include "tcop/tcopprot.h"
-#include "utils/datum.h"
 #include "utils/memutils.h"
 
 #if PG_VERSION_NUM >= 140000
@@ -39,27 +35,6 @@
 #define UpdateProgress(index, val) ((void)val)
 #endif
 
-#if PG_VERSION_NUM >= 140000
-#include "utils/backend_status.h"
-#include "utils/wait_event.h"
-#endif
-
-#if PG_VERSION_NUM >= 120000
-#include "access/table.h"
-#include "optimizer/optimizer.h"
-#else
-#include "access/heapam.h"
-#include "optimizer/planner.h"
-#include "pgstat.h"
-#endif
-
-#define PARALLEL_KEY_HNSW_SHARED		UINT64CONST(0xA000000000000001)
-#define PARALLEL_KEY_QUERY_TEXT			UINT64CONST(0xA000000000000002)
-
-#if PG_VERSION_NUM < 130000
-#define GENERATIONCHUNK_RAWSIZE (SIZEOF_SIZE_T + SIZEOF_VOID_P * 2)
-#endif
-
 /*
  * Create the metapage
  */
@@ -70,11 +45,11 @@ CreateMetaPage(HnswBuildState * buildstate)
 	ForkNumber	forkNum = buildstate->forkNum;
 	Buffer		buf;
 	Page		page;
+	GenericXLogState *state;
 	HnswMetaPage metap;
 
 	buf = HnswNewBuffer(index, forkNum);
-	page = BufferGetPage(buf);
-	HnswInitPage(buf, page);
+	HnswInitRegisterPage(index, &buf, &page, &state);
 
 	/* Set metapage data */
 	metap = HnswPageGetMeta(page);
@@ -90,15 +65,14 @@ CreateMetaPage(HnswBuildState * buildstate)
 	((PageHeader) page)->pd_lower =
 		((char *) metap + sizeof(HnswMetaPageData)) - (char *) page;
 
-	MarkBufferDirty(buf);
-	UnlockReleaseBuffer(buf);
+	HnswCommitBuffer(buf, state);
 }
 
 /*
  * Add a new page
  */
 static void
-HnswBuildAppendPage(Relation index, Buffer *buf, Page *page, ForkNumber forkNum)
+HnswBuildAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum)
 {
 	/* Add a new page */
 	Buffer		newbuf = HnswNewBuffer(index, forkNum);
@@ -108,6 +82,7 @@ HnswBuildAppendPage(Relation index, Buffer *buf, Page *page, ForkNumber forkNum)
 
 	/* Commit */
 	MarkBufferDirty(*buf);
+	GenericXLogFinish(*state);
 	UnlockReleaseBuffer(*buf);
 
 	/* Can take a while, so ensure we can interrupt */
@@ -118,7 +93,8 @@ HnswBuildAppendPage(Relation index, Buffer *buf, Page *page, ForkNumber forkNum)
 
 	/* Prepare new page */
 	*buf = newbuf;
-	*page = BufferGetPage(*buf);
+	*state = GenericXLogStart(index);
+	*page = GenericXLogRegisterBuffer(*state, *buf, GENERIC_XLOG_FULL_IMAGE);
 	HnswInitPage(*buf, *page);
 }
 
@@ -130,52 +106,46 @@ CreateElementPages(HnswBuildState * buildstate)
 {
 	Relation	index = buildstate->index;
 	ForkNumber	forkNum = buildstate->forkNum;
-	Size		etupAllocSize;
+	int			dimensions = buildstate->dimensions;
+	Size		etupSize;
 	Size		maxSize;
 	HnswElementTuple etup;
 	HnswNeighborTuple ntup;
 	BlockNumber insertPage;
 	Buffer		buf;
 	Page		page;
-	slist_iter	iter;
+	GenericXLogState *state;
+	ListCell   *lc;
 
 	/* Calculate sizes */
-	etupAllocSize = BLCKSZ;
-	maxSize = HNSW_MAX_SIZE;
+	maxSize = BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData));
+	etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
 
 	/* Allocate once */
-	etup = palloc0(etupAllocSize);
-	ntup = palloc0(BLCKSZ);
+	etup = palloc0(etupSize);
+	ntup = palloc0(maxSize);
 
 	/* Prepare first page */
 	buf = HnswNewBuffer(index, forkNum);
-	page = BufferGetPage(buf);
+	state = GenericXLogStart(index);
+	page = GenericXLogRegisterBuffer(state, buf, GENERIC_XLOG_FULL_IMAGE);
 	HnswInitPage(buf, page);
 
-	slist_foreach(iter, &buildstate->graph->elements)
+	foreach(lc, buildstate->elements)
 	{
-		HnswElement element = slist_container(HnswElementData, next, iter.cur);
-		Size		etupSize;
+		HnswElement element = lfirst(lc);
 		Size		ntupSize;
 		Size		combinedSize;
 
-		/* Zero memory for each element */
-		MemSet(etup, 0, etupAllocSize);
+		HnswSetElementTuple(etup, element);
 
 		/* Calculate sizes */
-		etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(DatumGetPointer(element->value)));
 		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, buildstate->m);
 		combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
 
-		/* Initial size check */
-		if (etupSize > etupAllocSize)
-			elog(ERROR, "index tuple too large");
-
-		HnswSetElementTuple(etup, element);
-
 		/* Keep element and neighbors on the same page if possible */
 		if (PageGetFreeSpace(page) < etupSize || (combinedSize <= maxSize && PageGetFreeSpace(page) < combinedSize))
-			HnswBuildAppendPage(index, &buf, &page, forkNum);
+			HnswBuildAppendPage(index, &buf, &page, &state, forkNum);
 
 		/* Calculate offsets */
 		element->blkno = BufferGetBlockNumber(buf);
@@ -199,7 +169,7 @@ CreateElementPages(HnswBuildState * buildstate)
 
 		/* Add new page if needed */
 		if (PageGetFreeSpace(page) < ntupSize)
-			HnswBuildAppendPage(index, &buf, &page, forkNum);
+			HnswBuildAppendPage(index, &buf, &page, &state, forkNum);
 
 		/* Add placeholder for neighbors */
 		if (PageAddItem(page, (Item) ntup, ntupSize, InvalidOffsetNumber, false, false) != element->neighborOffno)
@@ -210,9 +180,10 @@ CreateElementPages(HnswBuildState * buildstate)
 
 	/* Commit */
 	MarkBufferDirty(buf);
+	GenericXLogFinish(state);
 	UnlockReleaseBuffer(buf);
 
-	HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_ALWAYS, buildstate->graph->entryPoint, insertPage, forkNum, true);
+	HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_ALWAYS, buildstate->entryPoint, insertPage, forkNum);
 
 	pfree(etup);
 	pfree(ntup);
@@ -227,17 +198,18 @@ CreateNeighborPages(HnswBuildState * buildstate)
 	Relation	index = buildstate->index;
 	ForkNumber	forkNum = buildstate->forkNum;
 	int			m = buildstate->m;
-	slist_iter	iter;
+	ListCell   *lc;
 	HnswNeighborTuple ntup;
 
 	/* Allocate once */
 	ntup = palloc0(BLCKSZ);
 
-	slist_foreach(iter, &buildstate->graph->elements)
+	foreach(lc, buildstate->elements)
 	{
-		HnswElement e = slist_container(HnswElementData, next, iter.cur);
+		HnswElement e = lfirst(lc);
 		Buffer		buf;
 		Page		page;
+		GenericXLogState *state;
 		Size		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
 
 		/* Can take a while, so ensure we can interrupt */
@@ -246,7 +218,8 @@ CreateNeighborPages(HnswBuildState * buildstate)
 
 		buf = ReadBufferExtended(index, forkNum, e->neighborPage, RBM_NORMAL, NULL);
 		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-		page = BufferGetPage(buf);
+		state = GenericXLogStart(index);
+		page = GenericXLogRegisterBuffer(state, buf, 0);
 
 		HnswSetNeighborTuple(ntup, e, m);
 
@@ -255,29 +228,26 @@ CreateNeighborPages(HnswBuildState * buildstate)
 
 		/* Commit */
 		MarkBufferDirty(buf);
+		GenericXLogFinish(state);
 		UnlockReleaseBuffer(buf);
 	}
 
 	pfree(ntup);
 }
 
-#ifdef HNSW_MEMORY
 /*
- * Show memory usage
+ * Free elements
  */
 static void
-ShowMemoryUsage(HnswBuildState * buildstate)
+FreeElements(HnswBuildState * buildstate)
 {
-#if PG_VERSION_NUM >= 130000
-	elog(INFO, "graph memory: %zu MB, total memory: %zu MB",
-		 MemoryContextMemAllocated(buildstate->graphCtx, false) / (1024 * 1024),
-		 MemoryContextMemAllocated(CurrentMemoryContext, true) / (1024 * 1024));
-#else
-	MemoryContextStats(CurrentMemoryContext);
-	elog(INFO, "estimated memory: %zu MB", buildstate->memoryUsed / (1024 * 1024));
-#endif
+	ListCell   *lc;
+
+	foreach(lc, buildstate->elements)
+		HnswFreeElement(lfirst(lc));
+
+	list_free(buildstate->elements);
 }
-#endif
 
 /*
  * Flush pages
@@ -285,78 +255,25 @@ ShowMemoryUsage(HnswBuildState * buildstate)
 static void
 FlushPages(HnswBuildState * buildstate)
 {
-#ifdef HNSW_MEMORY
-	ShowMemoryUsage(buildstate);
-#endif
-
 	CreateMetaPage(buildstate);
 	CreateElementPages(buildstate);
 	CreateNeighborPages(buildstate);
 
-	buildstate->graph->flushed = true;
-	MemoryContextReset(buildstate->graphCtx);
+	buildstate->flushed = true;
+	FreeElements(buildstate);
 }
 
-#if PG_VERSION_NUM < 130000
 /*
- * Get the memory used by an element
- */
-static long
-HnswElementMemory(HnswElement e, int m)
-{
-	long		elementSize = sizeof(HnswElementData);
-
-	elementSize += sizeof(HnswNeighborArray) * (e->level + 1);
-	elementSize += sizeof(HnswCandidate) * (m * (e->level + 2));
-	elementSize += VARSIZE_ANY(DatumGetPointer(e->value));
-	/* Each allocation has a chunk header */
-	elementSize += (e->level + 4) * GENERATIONCHUNK_RAWSIZE;
-	/* Add an extra 5% for alignment and other overhead */
-	return elementSize * 1.05;
-}
-#endif
-
-/*
- * Find duplicate element
+ * Insert tuple
  */
 static bool
-HnswFindDuplicateInMemory(HnswElement element)
-{
-	HnswNeighborArray *neighbors = HnswGetNeighbors(element, 0);
-
-	for (int i = 0; i < neighbors->length; i++)
-	{
-		HnswCandidate *neighbor = &neighbors->items[i];
-
-		/* Exit early since ordered by distance */
-		if (!datumIsEqual(element->value, neighbor->element->value, false, -1))
-			return false;
-
-		/* Check for space */
-		if (neighbor->element->heaptidsLength < HNSW_HEAPTIDS)
-		{
-			HnswAddHeapTid(neighbor->element, &element->heaptids[0]);
-			return true;
-		}
-	}
-
-	return false;
-}
-
-/*
- * Insert tuple into in-memory graph
- */
-static bool
-InsertTupleInMemory(Relation index, Datum *values, ItemPointer heaptid, HnswBuildState * buildstate)
+InsertTuple(Relation index, Datum *values, HnswElement element, HnswBuildState * buildstate, HnswElement * dup)
 {
 	FmgrInfo   *procinfo = buildstate->procinfo;
 	Oid			collation = buildstate->collation;
-	HnswGraph  *graph = buildstate->graph;
-	HnswElement entryPoint = graph->entryPoint;
+	HnswElement entryPoint = buildstate->entryPoint;
 	int			efConstruction = buildstate->efConstruction;
 	int			m = buildstate->m;
-	MemoryContext oldCtx;
-	HnswElement element;
 
 	/* Detoast once for all calls */
 	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
@@ -368,67 +285,35 @@ InsertTupleInMemory(Relation index, Datum *values, ItemPointer heaptid, HnswBuil
 			return false;
 	}
 
-	/* Allocate element in graph memory context */
-	oldCtx = MemoryContextSwitchTo(buildstate->graphCtx);
-	element = HnswInitElement(heaptid, buildstate->m, buildstate->ml, buildstate->maxLevel);
-	element->value = datumCopy(value, false, -1);
-	MemoryContextSwitchTo(oldCtx);
-
-	/* Update memory usage */
-#if PG_VERSION_NUM >= 130000
-	graph->memoryUsed = MemoryContextMemAllocated(buildstate->graphCtx, false);
-#else
-	graph->memoryUsed += HnswElementMemory(element, buildstate->m);
-#endif
+	/* Copy value to element so accessible outside of memory context */
+	memcpy(element->vec, DatumGetVector(value), VECTOR_SIZE(buildstate->dimensions));
 
 	/* Insert element in graph */
 	HnswInsertElement(element, entryPoint, NULL, procinfo, collation, m, efConstruction, false);
 
 	/* Look for duplicate */
-	if (HnswFindDuplicateInMemory(element))
+	*dup = HnswFindDuplicate(element);
+
+	/* Update neighbors if needed */
+	if (*dup == NULL)
 	{
-		/* No need to free element since memory unlikely to be reallocated */
-		return true;
-	}
+		for (int lc = element->level; lc >= 0; lc--)
+		{
+			int			lm = HnswGetLayerM(m, lc);
+			HnswNeighborArray *neighbors = &element->neighbors[lc];
 
-	/* Add element */
-	slist_push_head(&graph->elements, &element->next);
-
-	/* Update neighbors */
-	for (int lc = element->level; lc >= 0; lc--)
-	{
-		int			lm = HnswGetLayerM(m, lc);
-		HnswNeighborArray *neighbors = HnswGetNeighbors(element, lc);
-
-		for (int i = 0; i < neighbors->length; i++)
-			HnswUpdateConnection(element, &neighbors->items[i], lm, lc, NULL, NULL, procinfo, collation);
+			for (int i = 0; i < neighbors->length; i++)
+				HnswUpdateConnection(element, &neighbors->items[i], lm, lc, NULL, NULL, procinfo, collation);
+		}
 	}
 
 	/* Update entry point if needed */
-	if (entryPoint == NULL || element->level > entryPoint->level)
-		graph->entryPoint = element;
+	if (*dup == NULL && (entryPoint == NULL || element->level > entryPoint->level))
+		buildstate->entryPoint = element;
 
-	return true;
-}
+	UpdateProgress(PROGRESS_CREATEIDX_TUPLES_DONE, ++buildstate->indtuples);
 
-/*
- * Acquire a lock if needed
- */
-static inline void
-HnswLockAcquire(HnswShared * hnswshared)
-{
-	if (hnswshared)
-		SpinLockAcquire(&hnswshared->mutex);
-}
-
-/*
- * Release a lock if needed
- */
-static inline void
-HnswLockRelease(HnswShared * hnswshared)
-{
-	if (hnswshared)
-		SpinLockRelease(&hnswshared->mutex);
+	return *dup == NULL;
 }
 
 /*
@@ -439,9 +324,9 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
 			  bool *isnull, bool tupleIsAlive, void *state)
 {
 	HnswBuildState *buildstate = (HnswBuildState *) state;
-	HnswGraph  *graph = buildstate->graph;
-	HnswShared *hnswshared = buildstate->hnswshared;
 	MemoryContext oldCtx;
+	HnswElement element;
+	HnswElement dup = NULL;
 	bool		inserted;
 
 #if PG_VERSION_NUM < 130000
@@ -452,50 +337,69 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
 	if (isnull[0])
 		return;
 
-	/* Flush pages if needed */
-	if (!graph->flushed && graph->memoryUsed >= graph->memoryTotal)
+	if (buildstate->indtuples >= buildstate->maxInMemoryElements)
 	{
-		ereport(NOTICE,
-				(errmsg("hnsw graph no longer fits into maintenance_work_mem after " INT64_FORMAT " tuples", (int64) graph->indtuples),
-				 errdetail("Building will take significantly more time."),
-				 errhint("Increase maintenance_work_mem to speed up builds.")));
+		if (!buildstate->flushed)
+		{
+			ereport(NOTICE,
+					(errmsg("hnsw graph no longer fits into maintenance_work_mem after " INT64_FORMAT " tuples", (int64) buildstate->indtuples),
+					 errdetail("Building will take significantly more time."),
+					 errhint("Increase maintenance_work_mem to speed up builds.")));
 
-		FlushPages(buildstate);
+			FlushPages(buildstate);
+		}
+
+		oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
+
+		if (HnswInsertTuple(buildstate->index, values, isnull, tid, buildstate->heap))
+			UpdateProgress(PROGRESS_CREATEIDX_TUPLES_DONE, ++buildstate->indtuples);
+
+		/* Reset memory context */
+		MemoryContextSwitchTo(oldCtx);
+		MemoryContextReset(buildstate->tmpCtx);
+
+		return;
 	}
 
+	/* Allocate necessary memory outside of memory context */
+	element = HnswInitElement(tid, buildstate->m, buildstate->ml, buildstate->maxLevel);
+	element->vec = palloc(VECTOR_SIZE(buildstate->dimensions));
+
+	/* Use memory context since detoast can allocate */
 	oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
 
 	/* Insert tuple */
-	if (graph->flushed)
-		inserted = HnswInsertTuple(index, values, isnull, tid, buildstate->heap, true);
-	else
-		inserted = InsertTupleInMemory(index, values, tid, buildstate);
-
-	/* Update progress */
-	if (inserted)
-	{
-		HnswLockAcquire(hnswshared);
-		UpdateProgress(PROGRESS_CREATEIDX_TUPLES_DONE, ++graph->indtuples);
-		HnswLockRelease(hnswshared);
-	}
+	inserted = InsertTuple(index, values, element, buildstate, &dup);
 
 	/* Reset memory context */
 	MemoryContextSwitchTo(oldCtx);
 	MemoryContextReset(buildstate->tmpCtx);
+
+	/* Add outside memory context */
+	if (dup != NULL)
+		HnswAddHeapTid(dup, tid);
+
+	/* Add to buildstate or free */
+	if (inserted)
+		buildstate->elements = lappend(buildstate->elements, element);
+	else
+		HnswFreeElement(element);
 }
 
 /*
- * Initialize the graph
+ * Get the max number of elements that fit into maintenance_work_mem
  */
-static void
-InitGraph(HnswGraph * graph)
+static double
+HnswGetMaxInMemoryElements(int m, double ml, int dimensions)
 {
-	slist_init(&graph->elements);
-	graph->entryPoint = NULL;
-	graph->memoryUsed = 0;
-	graph->memoryTotal = maintenance_work_mem * 1024L;
-	graph->flushed = false;
-	graph->indtuples = 0;
+	Size		elementSize = sizeof(HnswElementData);
+	double		avgLevel = -log(0.5) * ml;
+
+	elementSize += sizeof(HnswNeighborArray) * (avgLevel + 1);
+	elementSize += sizeof(HnswCandidate) * (m * (avgLevel + 2));
+	elementSize += sizeof(ItemPointerData);
+	elementSize += VECTOR_SIZE(dimensions);
+	return (maintenance_work_mem * 1024L) / elementSize;
 }
 
 /*
@@ -531,26 +435,19 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 	buildstate->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
 	buildstate->collation = index->rd_indcollation[0];
 
-	InitGraph(&buildstate->graphData);
-	buildstate->graph = &buildstate->graphData;
+	buildstate->elements = NIL;
+	buildstate->entryPoint = NULL;
 	buildstate->ml = HnswGetMl(buildstate->m);
 	buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
+	buildstate->maxInMemoryElements = HnswGetMaxInMemoryElements(buildstate->m, buildstate->ml, buildstate->dimensions);
+	buildstate->flushed = false;
 
 	/* Reuse for each tuple */
 	buildstate->normvec = InitVector(buildstate->dimensions);
 
-	buildstate->graphCtx = GenerationContextCreate(CurrentMemoryContext,
-												   "Hnsw build graph context",
-#if PG_VERSION_NUM >= 150000
-												   1024 * 1024, 1024 * 1024,
-#endif
-												   1024 * 1024);
 	buildstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
 											   "Hnsw build temporary context",
 											   ALLOCSET_DEFAULT_SIZES);
-
-	buildstate->hnswleader = NULL;
-	buildstate->hnswshared = NULL;
 }
 
 /*
@@ -560,427 +457,26 @@ static void
 FreeBuildState(HnswBuildState * buildstate)
 {
 	pfree(buildstate->normvec);
-	MemoryContextDelete(buildstate->graphCtx);
 	MemoryContextDelete(buildstate->tmpCtx);
 }
 
-/*
- * Within leader, wait for end of heap scan
- */
-static double
-ParallelHeapScan(HnswBuildState * buildstate)
-{
-	HnswShared *hnswshared = buildstate->hnswleader->hnswshared;
-	int			nparticipanttuplesorts;
-	double		reltuples;
-
-	nparticipanttuplesorts = buildstate->hnswleader->nparticipanttuplesorts;
-	for (;;)
-	{
-		SpinLockAcquire(&hnswshared->mutex);
-		if (hnswshared->nparticipantsdone == nparticipanttuplesorts)
-		{
-			buildstate->graph = &hnswshared->graphData;
-			reltuples = hnswshared->reltuples;
-			SpinLockRelease(&hnswshared->mutex);
-			break;
-		}
-		SpinLockRelease(&hnswshared->mutex);
-
-		ConditionVariableSleep(&hnswshared->workersdonecv,
-							   WAIT_EVENT_PARALLEL_CREATE_INDEX_SCAN);
-	}
-
-	ConditionVariableCancelSleep();
-
-	return reltuples;
-}
-
-/*
- * Perform a worker's portion of a parallel insert
- */
-static void
-HnswParallelScanAndInsert(HnswSpool * hnswspool, HnswShared * hnswshared, bool progress)
-{
-	HnswBuildState buildstate;
-#if PG_VERSION_NUM >= 120000
-	TableScanDesc scan;
-#else
-	HeapScanDesc scan;
-#endif
-	double		reltuples;
-	IndexInfo  *indexInfo;
-
-	/* Join parallel scan */
-	indexInfo = BuildIndexInfo(hnswspool->index);
-	indexInfo->ii_Concurrent = hnswshared->isconcurrent;
-	InitBuildState(&buildstate, hnswspool->heap, hnswspool->index, indexInfo, MAIN_FORKNUM);
-	buildstate.graph = &hnswshared->graphData;
-	buildstate.hnswshared = hnswshared;
-#if PG_VERSION_NUM >= 120000
-	scan = table_beginscan_parallel(hnswspool->heap,
-									ParallelTableScanFromHnswShared(hnswshared));
-	reltuples = table_index_build_scan(hnswspool->heap, hnswspool->index, indexInfo,
-									   true, progress, BuildCallback,
-									   (void *) &buildstate, scan);
-#else
-	scan = heap_beginscan_parallel(hnswspool->heap, &hnswshared->heapdesc);
-	reltuples = IndexBuildHeapScan(hnswspool->heap, hnswspool->index, indexInfo,
-								   true, BuildCallback,
-								   (void *) &buildstate, scan);
-#endif
-
-	/* Record statistics */
-	SpinLockAcquire(&hnswshared->mutex);
-	hnswshared->nparticipantsdone++;
-	hnswshared->reltuples += reltuples;
-	SpinLockRelease(&hnswshared->mutex);
-
-	/* Log statistics */
-	if (progress)
-		ereport(DEBUG1, (errmsg("leader processed " INT64_FORMAT " tuples", (int64) reltuples)));
-	else
-		ereport(DEBUG1, (errmsg("worker processed " INT64_FORMAT " tuples", (int64) reltuples)));
-
-	/* Notify leader */
-	ConditionVariableSignal(&hnswshared->workersdonecv);
-
-	FreeBuildState(&buildstate);
-}
-
-/*
- * Perform work within a launched parallel process
- */
-void
-HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc)
-{
-	char	   *sharedquery;
-	HnswSpool  *hnswspool;
-	HnswShared *hnswshared;
-	Relation	heapRel;
-	Relation	indexRel;
-	LOCKMODE	heapLockmode;
-	LOCKMODE	indexLockmode;
-
-	/* Set debug_query_string for individual workers first */
-	sharedquery = shm_toc_lookup(toc, PARALLEL_KEY_QUERY_TEXT, true);
-	debug_query_string = sharedquery;
-
-	/* Report the query string from leader */
-	pgstat_report_activity(STATE_RUNNING, debug_query_string);
-
-	/* Look up shared state */
-	hnswshared = shm_toc_lookup(toc, PARALLEL_KEY_HNSW_SHARED, false);
-
-	/* Open relations using lock modes known to be obtained by index.c */
-	if (!hnswshared->isconcurrent)
-	{
-		heapLockmode = ShareLock;
-		indexLockmode = AccessExclusiveLock;
-	}
-	else
-	{
-		heapLockmode = ShareUpdateExclusiveLock;
-		indexLockmode = RowExclusiveLock;
-	}
-
-	/* Open relations within worker */
-#if PG_VERSION_NUM >= 120000
-	heapRel = table_open(hnswshared->heaprelid, heapLockmode);
-#else
-	heapRel = heap_open(hnswshared->heaprelid, heapLockmode);
-#endif
-	indexRel = index_open(hnswshared->indexrelid, indexLockmode);
-
-	/* Initialize worker's own spool */
-	hnswspool = (HnswSpool *) palloc0(sizeof(HnswSpool));
-	hnswspool->heap = heapRel;
-	hnswspool->index = indexRel;
-
-	/* Perform inserts */
-	HnswParallelScanAndInsert(hnswspool, hnswshared, false);
-
-	/* Close relations within worker */
-	index_close(indexRel, indexLockmode);
-#if PG_VERSION_NUM >= 120000
-	table_close(heapRel, heapLockmode);
-#else
-	heap_close(heapRel, heapLockmode);
-#endif
-}
-
-/*
- * End parallel build
- */
-static void
-HnswEndParallel(HnswLeader * hnswleader)
-{
-	/* Shutdown worker processes */
-	WaitForParallelWorkersToFinish(hnswleader->pcxt);
-
-	/* Free last reference to MVCC snapshot, if one was used */
-	if (IsMVCCSnapshot(hnswleader->snapshot))
-		UnregisterSnapshot(hnswleader->snapshot);
-	DestroyParallelContext(hnswleader->pcxt);
-	ExitParallelMode();
-}
-
-/*
- * Return size of shared memory required for parallel index build
- */
-static Size
-ParallelEstimateShared(Relation heap, Snapshot snapshot)
-{
-#if PG_VERSION_NUM >= 120000
-	return add_size(BUFFERALIGN(sizeof(HnswShared)), table_parallelscan_estimate(heap, snapshot));
-#else
-	if (!IsMVCCSnapshot(snapshot))
-	{
-		Assert(snapshot == SnapshotAny);
-		return sizeof(HnswShared);
-	}
-
-	return add_size(offsetof(HnswShared, heapdesc) +
-					offsetof(ParallelHeapScanDescData, phs_snapshot_data),
-					EstimateSnapshotSpace(snapshot));
-#endif
-}
-
-/*
- * Within leader, participate as a parallel worker
- */
-static void
-HnswLeaderParticipateAsWorker(HnswBuildState * buildstate)
-{
-	HnswLeader *hnswleader = buildstate->hnswleader;
-	HnswSpool  *leaderworker;
-
-	/* Allocate memory and initialize private spool */
-	leaderworker = (HnswSpool *) palloc0(sizeof(HnswSpool));
-	leaderworker->heap = buildstate->heap;
-	leaderworker->index = buildstate->index;
-
-	/* Perform work common to all participants */
-	HnswParallelScanAndInsert(leaderworker, hnswleader->hnswshared, true);
-}
-
-/*
- * Begin parallel build
- */
-static void
-HnswBeginParallel(HnswBuildState * buildstate, bool isconcurrent, int request)
-{
-	ParallelContext *pcxt;
-	int			scantuplesortstates;
-	Snapshot	snapshot;
-	Size		esthnswshared;
-	HnswShared *hnswshared;
-	HnswLeader *hnswleader = (HnswLeader *) palloc0(sizeof(HnswLeader));
-	bool		leaderparticipates = true;
-	int			querylen;
-
-#ifdef DISABLE_LEADER_PARTICIPATION
-	leaderparticipates = false;
-#endif
-
-	/* Enter parallel mode and create context */
-	EnterParallelMode();
-	Assert(request > 0);
-#if PG_VERSION_NUM >= 120000
-	pcxt = CreateParallelContext("vector", "HnswParallelBuildMain", request);
-#else
-	pcxt = CreateParallelContext("vector", "HnswParallelBuildMain", request, true);
-#endif
-
-	scantuplesortstates = leaderparticipates ? request + 1 : request;
-
-	/* Get snapshot for table scan */
-	if (!isconcurrent)
-		snapshot = SnapshotAny;
-	else
-		snapshot = RegisterSnapshot(GetTransactionSnapshot());
-
-	/* Estimate size of workspaces */
-	esthnswshared = ParallelEstimateShared(buildstate->heap, snapshot);
-	shm_toc_estimate_chunk(&pcxt->estimator, esthnswshared);
-	shm_toc_estimate_keys(&pcxt->estimator, 1);
-
-	/* Finally, estimate PARALLEL_KEY_QUERY_TEXT space */
-	if (debug_query_string)
-	{
-		querylen = strlen(debug_query_string);
-		shm_toc_estimate_chunk(&pcxt->estimator, querylen + 1);
-		shm_toc_estimate_keys(&pcxt->estimator, 1);
-	}
-	else
-		querylen = 0;			/* keep compiler quiet */
-
-	/* Everyone's had a chance to ask for space, so now create the DSM */
-	InitializeParallelDSM(pcxt);
-
-	/* If no DSM segment was available, back out (do serial build) */
-	if (pcxt->seg == NULL)
-	{
-		if (IsMVCCSnapshot(snapshot))
-			UnregisterSnapshot(snapshot);
-		DestroyParallelContext(pcxt);
-		ExitParallelMode();
-		return;
-	}
-
-	/* Store shared build state, for which we reserved space */
-	hnswshared = (HnswShared *) shm_toc_allocate(pcxt->toc, esthnswshared);
-	/* Initialize immutable state */
-	hnswshared->heaprelid = RelationGetRelid(buildstate->heap);
-	hnswshared->indexrelid = RelationGetRelid(buildstate->index);
-	hnswshared->isconcurrent = isconcurrent;
-	hnswshared->scantuplesortstates = scantuplesortstates;
-	ConditionVariableInit(&hnswshared->workersdonecv);
-	SpinLockInit(&hnswshared->mutex);
-	/* Initialize mutable state */
-	hnswshared->nparticipantsdone = 0;
-	hnswshared->reltuples = 0;
-	InitGraph(&hnswshared->graphData);
-	/* TODO Support in-memory builds */
-	hnswshared->graphData.memoryTotal = 0;
-	hnswshared->graphData.flushed = true;
-#if PG_VERSION_NUM >= 120000
-	table_parallelscan_initialize(buildstate->heap,
-								  ParallelTableScanFromHnswShared(hnswshared),
-								  snapshot);
-#else
-	heap_parallelscan_initialize(&hnswshared->heapdesc, buildstate->heap, snapshot);
-#endif
-
-	shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_SHARED, hnswshared);
-
-	/* Store query string for workers */
-	if (debug_query_string)
-	{
-		char	   *sharedquery;
-
-		sharedquery = (char *) shm_toc_allocate(pcxt->toc, querylen + 1);
-		memcpy(sharedquery, debug_query_string, querylen + 1);
-		shm_toc_insert(pcxt->toc, PARALLEL_KEY_QUERY_TEXT, sharedquery);
-	}
-
-	/* Launch workers, saving status for leader/caller */
-	LaunchParallelWorkers(pcxt);
-	hnswleader->pcxt = pcxt;
-	hnswleader->nparticipanttuplesorts = pcxt->nworkers_launched;
-	if (leaderparticipates)
-		hnswleader->nparticipanttuplesorts++;
-	hnswleader->hnswshared = hnswshared;
-	hnswleader->snapshot = snapshot;
-
-	/* If no workers were successfully launched, back out (do serial build) */
-	if (pcxt->nworkers_launched == 0)
-	{
-		HnswEndParallel(hnswleader);
-		return;
-	}
-
-	/* Log participants */
-	ereport(DEBUG1, (errmsg("using %d parallel workers", pcxt->nworkers_launched)));
-
-	/* Save leader state now that it's clear build will be parallel */
-	buildstate->hnswleader = hnswleader;
-
-	/* Join heap scan ourselves */
-	if (leaderparticipates)
-		HnswLeaderParticipateAsWorker(buildstate);
-
-	/* Wait for all launched workers */
-	WaitForParallelWorkersToAttach(pcxt);
-}
-
-/*
- * Compute parallel workers
- */
-static int
-ComputeParallelWorkers(Relation heap, Relation index)
-{
-	int			parallel_workers;
-
-	/* Make sure it's safe to use parallel workers */
-	parallel_workers = plan_create_index_workers(RelationGetRelid(heap), RelationGetRelid(index));
-	if (parallel_workers == 0)
-		return 0;
-
-	/* Use parallel_workers storage parameter on table if set */
-	parallel_workers = RelationGetParallelWorkers(heap, -1);
-	if (parallel_workers != -1)
-		return Min(parallel_workers, max_parallel_maintenance_workers);
-
-	return max_parallel_maintenance_workers;
-}
-
 /*
  * Build graph
  */
 static void
 BuildGraph(HnswBuildState * buildstate, ForkNumber forkNum)
 {
-	int			parallel_workers = 0;
-
 	UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_HNSW_PHASE_LOAD);
 
-	/* Calculate parallel workers */
-	if (buildstate->heap != NULL && hnsw_enable_parallel_build)
-		parallel_workers = ComputeParallelWorkers(buildstate->heap, buildstate->index);
-
-	/* Attempt to launch parallel worker scan when required */
-	if (parallel_workers > 0)
-	{
-		/* TODO Support in-memory builds */
-		FlushPages(buildstate);
-		HnswBeginParallel(buildstate, buildstate->indexInfo->ii_Concurrent, parallel_workers);
-	}
-
-	/* Add tuples to graph */
-	if (buildstate->heap != NULL)
-	{
-		if (buildstate->hnswleader)
-			buildstate->reltuples = ParallelHeapScan(buildstate);
-		else
-		{
 #if PG_VERSION_NUM >= 120000
-			buildstate->reltuples = table_index_build_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
-														   true, true, BuildCallback, (void *) buildstate, NULL);
+	buildstate->reltuples = table_index_build_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
+												   true, true, BuildCallback, (void *) buildstate, NULL);
 #else
-			buildstate->reltuples = IndexBuildHeapScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
-													   true, BuildCallback, (void *) buildstate, NULL);
+	buildstate->reltuples = IndexBuildHeapScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
+											   true, BuildCallback, (void *) buildstate, NULL);
 #endif
-		}
-
-		buildstate->indtuples = buildstate->graph->indtuples;
-	}
-
-	/* Flush pages */
-	if (!buildstate->graph->flushed)
-		FlushPages(buildstate);
-
-	/* End parallel build */
-	if (buildstate->hnswleader)
-		HnswEndParallel(buildstate->hnswleader);
 }
 
-#if PG_VERSION_NUM < 110008
-void
-log_newpage_range(Relation rel, ForkNumber forkNum, BlockNumber startblk, BlockNumber endblk, bool page_std)
-{
-	for (BlockNumber blkno = startblk; blkno < endblk; blkno++)
-	{
-		Buffer		buf = ReadBufferExtended(rel, forkNum, blkno, RBM_NORMAL, NULL);
-
-		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-		MarkBufferDirty(buf);
-		log_newpage_buffer(buf, page_std);
-		UnlockReleaseBuffer(buf);
-	}
-}
-#endif
-
 /*
  * Build the index
  */
@@ -988,16 +484,13 @@ static void
 BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 		   HnswBuildState * buildstate, ForkNumber forkNum)
 {
-#ifdef HNSW_MEMORY
-	SeedRandom(42);
-#endif
-
 	InitBuildState(buildstate, heap, index, indexInfo, forkNum);
 
-	BuildGraph(buildstate, forkNum);
+	if (buildstate->heap != NULL)
+		BuildGraph(buildstate, forkNum);
 
-	if (RelationNeedsWAL(index))
-		log_newpage_range(index, forkNum, 0, RelationGetNumberOfBlocks(index), true);
+	if (!buildstate->flushed)
+		FlushPages(buildstate);
 
 	FreeBuildState(buildstate);
 }

src/hnswinsert.c

@@ -5,7 +5,6 @@
 #include "hnsw.h"
 #include "storage/bufmgr.h"
 #include "storage/lmgr.h"
-#include "utils/datum.h"
 #include "utils/memutils.h"
 
 /*
@@ -93,7 +92,7 @@ HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size
  * Add a new page
  */
 static void
-HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState *state, Page page, bool building)
+HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState *state, Page page)
 {
 	/* Add a new page */
 	LockRelationForExtension(index, ExclusiveLock);
@@ -101,11 +100,7 @@ HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState
 	UnlockRelationForExtension(index, ExclusiveLock);
 
 	/* Init new page */
-	if (building)
-		*npage = BufferGetPage(*nbuf);
-	else
-		*npage = GenericXLogRegisterBuffer(state, *nbuf, GENERIC_XLOG_FULL_IMAGE);
-
+	*npage = GenericXLogRegisterBuffer(state, *nbuf, GENERIC_XLOG_FULL_IMAGE);
 	HnswInitPage(*nbuf, *npage);
 
 	/* Update previous buffer */
@@ -116,7 +111,7 @@ HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState
  * Add to element and neighbor pages
  */
 static void
-WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPage, BlockNumber *updatedInsertPage, bool building)
+WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPage, BlockNumber *updatedInsertPage)
 {
 	Buffer		buf;
 	Page		page;
@@ -128,6 +123,7 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
 	Size		minCombinedSize;
 	HnswElementTuple etup;
 	BlockNumber currentPage = insertPage;
+	int			dimensions = e->vec->dim;
 	HnswNeighborTuple ntup;
 	Buffer		nbuf;
 	Page		npage;
@@ -136,10 +132,10 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
 	BlockNumber newInsertPage = InvalidBlockNumber;
 
 	/* Calculate sizes */
-	etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(DatumGetPointer(e->value)));
+	etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
 	ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
 	combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
-	maxSize = HNSW_MAX_SIZE;
+	maxSize = BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData));
 	minCombinedSize = etupSize + HNSW_NEIGHBOR_TUPLE_SIZE(0, m) + sizeof(ItemIdData);
 
 	/* Prepare element tuple */
@@ -156,16 +152,8 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
 		buf = ReadBuffer(index, currentPage);
 		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
 
-		if (building)
-		{
-			state = NULL;
-			page = BufferGetPage(buf);
-		}
-		else
-		{
-			state = GenericXLogStart(index);
-			page = GenericXLogRegisterBuffer(state, buf, 0);
-		}
+		state = GenericXLogStart(index);
+		page = GenericXLogRegisterBuffer(state, buf, 0);
 
 		/* Keep track of first page where element at level 0 can fit */
 		if (!BlockNumberIsValid(newInsertPage) && PageGetFreeSpace(page) >= minCombinedSize)
@@ -185,12 +173,7 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
 		if (HnswFreeOffset(index, buf, page, e, ntupSize, &nbuf, &npage, &freeOffno, &freeNeighborOffno, &newInsertPage))
 		{
 			if (nbuf != buf)
-			{
-				if (building)
-					npage = BufferGetPage(nbuf);
-				else
-					npage = GenericXLogRegisterBuffer(state, nbuf, 0);
-			}
+				npage = GenericXLogRegisterBuffer(state, nbuf, 0);
 
 			break;
 		}
@@ -199,7 +182,7 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
 		/* Skip if both tuples can fit on the same page */
 		if (combinedSize > maxSize && PageGetFreeSpace(page) >= etupSize && !BlockNumberIsValid(HnswPageGetOpaque(page)->nextblkno))
 		{
-			HnswInsertAppendPage(index, &nbuf, &npage, state, page, building);
+			HnswInsertAppendPage(index, &nbuf, &npage, state, page);
 			break;
 		}
 
@@ -208,8 +191,7 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
 		if (BlockNumberIsValid(currentPage))
 		{
 			/* Move to next page */
-			if (!building)
-				GenericXLogAbort(state);
+			GenericXLogAbort(state);
 			UnlockReleaseBuffer(buf);
 		}
 		else
@@ -217,33 +199,24 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
 			Buffer		newbuf;
 			Page		newpage;
 
-			HnswInsertAppendPage(index, &newbuf, &newpage, state, page, building);
+			HnswInsertAppendPage(index, &newbuf, &newpage, state, page);
 
 			/* Commit */
-			if (building)
-				MarkBufferDirty(buf);
-			else
-				GenericXLogFinish(state);
+			MarkBufferDirty(newbuf);
+			MarkBufferDirty(buf);
+			GenericXLogFinish(state);
 
 			/* Unlock previous buffer */
 			UnlockReleaseBuffer(buf);
 
 			/* Prepare new buffer */
+			state = GenericXLogStart(index);
 			buf = newbuf;
-			if (building)
-			{
-				state = NULL;
-				page = BufferGetPage(buf);
-			}
-			else
-			{
-				state = GenericXLogStart(index);
-				page = GenericXLogRegisterBuffer(state, buf, 0);
-			}
+			page = GenericXLogRegisterBuffer(state, buf, 0);
 
 			/* Create new page for neighbors if needed */
 			if (PageGetFreeSpace(page) < combinedSize)
-				HnswInsertAppendPage(index, &nbuf, &npage, state, page, building);
+				HnswInsertAppendPage(index, &nbuf, &npage, state, page);
 			else
 			{
 				nbuf = buf;
@@ -297,14 +270,10 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
 	}
 
 	/* Commit */
-	if (building)
-	{
-		MarkBufferDirty(buf);
-		if (nbuf != buf)
-			MarkBufferDirty(nbuf);
-	}
-	else
-		GenericXLogFinish(state);
+	MarkBufferDirty(buf);
+	if (nbuf != buf)
+		MarkBufferDirty(nbuf);
+	GenericXLogFinish(state);
 	UnlockReleaseBuffer(buf);
 	if (nbuf != buf)
 		UnlockReleaseBuffer(nbuf);
@@ -338,12 +307,12 @@ ConnectionExists(HnswElement e, HnswNeighborTuple ntup, int startIdx, int lm)
  * Update neighbors
  */
 void
-HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building)
+HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting)
 {
 	for (int lc = e->level; lc >= 0; lc--)
 	{
 		int			lm = HnswGetLayerM(m, lc);
-		HnswNeighborArray *neighbors = HnswGetNeighbors(e, lc);
+		HnswNeighborArray *neighbors = &e->neighbors[lc];
 
 		for (int i = 0; i < neighbors->length; i++)
 		{
@@ -379,16 +348,8 @@ HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswE
 			/* Register page */
 			buf = ReadBuffer(index, hc->element->neighborPage);
 			LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-			if (building)
-			{
-				state = NULL;
-				page = BufferGetPage(buf);
-			}
-			else
-			{
-				state = GenericXLogStart(index);
-				page = GenericXLogRegisterBuffer(state, buf, 0);
-			}
+			state = GenericXLogStart(index);
+			page = GenericXLogRegisterBuffer(state, buf, 0);
 
 			/* Get tuple */
 			itemid = PageGetItemId(page, offno);
@@ -430,12 +391,10 @@ HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswE
 					elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
 
 				/* Commit */
-				if (building)
-					MarkBufferDirty(buf);
-				else
-					GenericXLogFinish(state);
+				MarkBufferDirty(buf);
+				GenericXLogFinish(state);
 			}
-			else if (!building)
+			else
 				GenericXLogAbort(state);
 
 			UnlockReleaseBuffer(buf);
@@ -447,34 +406,23 @@ HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswE
  * Add a heap TID to an existing element
  */
 static bool
-HnswAddDuplicate(Relation index, HnswElement element, HnswElement dup, bool building)
+HnswAddDuplicate(Relation index, HnswElement element, HnswElement dup)
 {
 	Buffer		buf;
 	Page		page;
 	GenericXLogState *state;
-	ItemId		itemid;
+	Size		etupSize = HNSW_ELEMENT_TUPLE_SIZE(dup->vec->dim);
 	HnswElementTuple etup;
-	Size		etupSize;
 	int			i;
 
 	/* Read page */
 	buf = ReadBuffer(index, dup->blkno);
 	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-	if (building)
-	{
-		state = NULL;
-		page = BufferGetPage(buf);
-	}
-	else
-	{
-		state = GenericXLogStart(index);
-		page = GenericXLogRegisterBuffer(state, buf, 0);
-	}
+	state = GenericXLogStart(index);
+	page = GenericXLogRegisterBuffer(state, buf, 0);
 
 	/* Find space */
-	itemid = PageGetItemId(page, dup->offno);
-	etup = (HnswElementTuple) PageGetItem(page, itemid);
-	etupSize = ItemIdGetLength(itemid);
+	etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, dup->offno));
 	for (i = 0; i < HNSW_HEAPTIDS; i++)
 	{
 		if (!ItemPointerIsValid(&etup->heaptids[i]))
@@ -484,84 +432,61 @@ HnswAddDuplicate(Relation index, HnswElement element, HnswElement dup, bool buil
 	/* Either being deleted or we lost our chance to another backend */
 	if (i == 0 || i == HNSW_HEAPTIDS)
 	{
-		if (!building)
-			GenericXLogAbort(state);
+		GenericXLogAbort(state);
 		UnlockReleaseBuffer(buf);
 		return false;
 	}
 
 	/* Add heap TID */
-	etup->heaptids[i] = element->heaptids[0];
+	etup->heaptids[i] = *((ItemPointer) linitial(element->heaptids));
 
 	/* Overwrite tuple */
 	if (!PageIndexTupleOverwrite(page, dup->offno, (Item) etup, etupSize))
 		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
 
 	/* Commit */
-	if (building)
-		MarkBufferDirty(buf);
-	else
-		GenericXLogFinish(state);
+	MarkBufferDirty(buf);
+	GenericXLogFinish(state);
 	UnlockReleaseBuffer(buf);
 
 	return true;
 }
 
-/*
- * Find duplicate element
- */
-static bool
-HnswFindDuplicate(Relation index, HnswElement element, bool building)
-{
-	HnswNeighborArray *neighbors = HnswGetNeighbors(element, 0);
-
-	for (int i = 0; i < neighbors->length; i++)
-	{
-		HnswCandidate *neighbor = &neighbors->items[i];
-
-		/* Exit early since ordered by distance */
-		if (!datumIsEqual(element->value, neighbor->element->value, false, -1))
-			return false;
-
-		if (HnswAddDuplicate(index, element, neighbor->element, building))
-			return true;
-	}
-
-	return false;
-}
-
 /*
  * Write changes to disk
  */
 static void
-WriteElement(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
+WriteElement(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement dup, HnswElement entryPoint)
 {
 	BlockNumber newInsertPage = InvalidBlockNumber;
 
-	/* Look for duplicate */
-	if (HnswFindDuplicate(index, element, building))
-		return;
+	/* Try to add to existing page */
+	if (dup != NULL)
+	{
+		if (HnswAddDuplicate(index, element, dup))
+			return;
+	}
 
 	/* Write element and neighbor tuples */
-	WriteNewElementPages(index, element, m, GetInsertPage(index), &newInsertPage, building);
+	WriteNewElementPages(index, element, m, GetInsertPage(index), &newInsertPage);
 
 	/* Update insert page if needed */
 	if (BlockNumberIsValid(newInsertPage))
-		HnswUpdateMetaPage(index, 0, NULL, newInsertPage, MAIN_FORKNUM, building);
+		HnswUpdateMetaPage(index, 0, NULL, newInsertPage, MAIN_FORKNUM);
 
 	/* Update neighbors */
-	HnswUpdateNeighborPages(index, procinfo, collation, element, m, false, building);
+	HnswUpdateNeighborPages(index, procinfo, collation, element, m, false);
 
-	/* Update entry point if needed */
+	/* Update metapage if needed */
 	if (entryPoint == NULL || element->level > entryPoint->level)
-		HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_GREATER, element, InvalidBlockNumber, MAIN_FORKNUM, building);
+		HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_GREATER, element, InvalidBlockNumber, MAIN_FORKNUM);
 }
 
 /*
  * Insert a tuple into the index
  */
 bool
-HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel, bool building)
+HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel)
 {
 	Datum		value;
 	FmgrInfo   *normprocinfo;
@@ -571,6 +496,7 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
 	int			efConstruction = HnswGetEfConstruction(index);
 	FmgrInfo   *procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
 	Oid			collation = index->rd_indcollation[0];
+	HnswElement dup;
 	LOCKMODE	lockmode = ShareLock;
 
 	/* Detoast once for all calls */
@@ -596,7 +522,7 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
 
 	/* Create an element */
 	element = HnswInitElement(heap_tid, m, HnswGetMl(m), HnswGetMaxLevel(m));
-	element->value = value;
+	element->vec = DatumGetVector(value);
 
 	/* Prevent concurrent inserts when likely updating entry point */
 	if (entryPoint == NULL || element->level > entryPoint->level)
@@ -615,8 +541,11 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
 	/* Insert element in graph */
 	HnswInsertElement(element, entryPoint, index, procinfo, collation, m, efConstruction, false);
 
+	/* Look for duplicate */
+	dup = HnswFindDuplicate(element);
+
 	/* Write to disk */
-	WriteElement(index, procinfo, collation, element, m, efConstruction, entryPoint, building);
+	WriteElement(index, procinfo, collation, element, m, efConstruction, dup, entryPoint);
 
 	/* Release lock */
 	UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);
@@ -650,7 +579,7 @@ hnswinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid,
 	oldCtx = MemoryContextSwitchTo(insertCtx);
 
 	/* Insert tuple */
-	HnswInsertTuple(index, values, isnull, heap_tid, heap, false);
+	HnswInsertTuple(index, values, isnull, heap_tid, heap);
 
 	/* Delete memory context */
 	MemoryContextSwitchTo(oldCtx);

src/hnswscan.c

@@ -113,6 +113,12 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
 
 	scan->opaque = so;
 
+	/*
+	 * Get a shared lock. This allows vacuum to ensure no in-flight scans
+	 * before marking tuples as deleted.
+	 */
+	LockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
+
 	return scan;
 }
 
@@ -160,25 +166,11 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 		if (scan->orderByData == NULL)
 			elog(ERROR, "cannot scan hnsw index without order");
 
-		/* Requires MVCC-compliant snapshot as not able to maintain a pin */
-		/* https://www.postgresql.org/docs/current/index-locking.html */
-		if (!IsMVCCSnapshot(scan->xs_snapshot))
-			elog(ERROR, "non-MVCC snapshots are not supported with hnsw");
-
 		/* Get scan value */
 		value = GetScanValue(scan);
 
-		/*
-		 * Get a shared lock. This allows vacuum to ensure no in-flight scans
-		 * before marking tuples as deleted.
-		 */
-		LockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
-
 		so->w = GetScanItems(scan, value);
 
-		/* Release shared lock */
-		UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
-
 		so->first = false;
 	}
 
@@ -188,13 +180,15 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 		ItemPointer heaptid;
 
 		/* Move to next element if no valid heap TIDs */
-		if (hc->element->heaptidsLength == 0)
+		if (list_length(hc->element->heaptids) == 0)
 		{
 			so->w = list_delete_last(so->w);
 			continue;
 		}
 
-		heaptid = &hc->element->heaptids[--hc->element->heaptidsLength];
+		heaptid = llast(hc->element->heaptids);
+
+		hc->element->heaptids = list_delete_last(hc->element->heaptids);
 
 		MemoryContextSwitchTo(oldCtx);
 
@@ -204,6 +198,15 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 		scan->xs_ctup.t_self = *heaptid;
 #endif
 
+		/*
+		 * Typically, an index scan must maintain a pin on the index page
+		 * holding the item last returned by amgettuple. However, this is not
+		 * needed with the current vacuum strategy, which ensures scans do not
+		 * visit tuples in danger of being marked as deleted.
+		 *
+		 * https://www.postgresql.org/docs/current/index-locking.html
+		 */
+
 		scan->xs_recheckorderby = false;
 		return true;
 	}
@@ -220,6 +223,9 @@ hnswendscan(IndexScanDesc scan)
 {
 	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
 
+	/* Release shared lock */
+	UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
+
 	MemoryContextDelete(so->tmpCtx);
 
 	pfree(so);

src/hnswutils.c

@@ -4,92 +4,8 @@
 
 #include "hnsw.h"
 #include "storage/bufmgr.h"
-#include "utils/datum.h"
 #include "vector.h"
 
-#if PG_VERSION_NUM >= 130000
-#include "common/hashfn.h"
-#else
-#include "utils/hashutils.h"
-#endif
-
-#if PG_VERSION_NUM < 170000
-static inline uint64
-murmurhash64(uint64 data)
-{
-	uint64		h = data;
-
-	h ^= h >> 33;
-	h *= 0xff51afd7ed558ccd;
-	h ^= h >> 33;
-	h *= 0xc4ceb9fe1a85ec53;
-	h ^= h >> 33;
-
-	return h;
-}
-#endif
-
-/* TID hash table */
-static uint32
-hash_tid(ItemPointerData tid)
-{
-	union
-	{
-		uint64		i;
-		ItemPointerData tid;
-	}			x;
-
-	/* Initialize unused bytes */
-	x.i = 0;
-	x.tid = tid;
-
-	return murmurhash64(x.i);
-}
-
-#define SH_PREFIX		tidhash
-#define SH_ELEMENT_TYPE	TidHashEntry
-#define SH_KEY_TYPE		ItemPointerData
-#define	SH_KEY			tid
-#define SH_HASH_KEY(tb, key)	hash_tid(key)
-#define SH_EQUAL(tb, a, b)		ItemPointerEquals(&a, &b)
-#define	SH_SCOPE		extern
-#define SH_DEFINE
-#include "lib/simplehash.h"
-
-/* Needed to include simplehash.h twice */
-#if PG_VERSION_NUM < 120000
-#undef SH_EQUAL
-#define sh_log2 pointerhash_sh_log2
-#define sh_pow2 pointerhash_sh_pow2
-#endif
-
-/* Pointer hash table */
-static uint32
-hash_pointer(uintptr_t ptr)
-{
-#if SIZEOF_VOID_P == 8
-	return murmurhash64((uint64) ptr);
-#else
-	return murmurhash32((uint32) ptr);
-#endif
-}
-
-#define SH_PREFIX		pointerhash
-#define SH_ELEMENT_TYPE	PointerHashEntry
-#define SH_KEY_TYPE		uintptr_t
-#define	SH_KEY			ptr
-#define SH_HASH_KEY(tb, key)	hash_pointer(key)
-#define SH_EQUAL(tb, a, b)		(a == b)
-#define	SH_SCOPE		extern
-#define SH_DEFINE
-#include "lib/simplehash.h"
-
-typedef union
-{
-	pointerhash_hash *pointers;
-	tidhash_hash *tids;
-}			visited_hash;
-
 /*
  * Get the max number of connections in an upper layer for each element in the index
  */
@@ -122,12 +38,12 @@ HnswGetEfConstruction(Relation index)
  * Get proc
  */
 FmgrInfo *
-HnswOptionalProcInfo(Relation index, uint16 procnum)
+HnswOptionalProcInfo(Relation rel, uint16 procnum)
 {
-	if (!OidIsValid(index_getprocid(index, 1, procnum)))
+	if (!OidIsValid(index_getprocid(rel, 1, procnum)))
 		return NULL;
 
-	return index_getprocinfo(index, 1, procnum);
+	return index_getprocinfo(rel, 1, procnum);
 }
 
 /*
@@ -184,6 +100,28 @@ HnswInitPage(Buffer buf, Page page)
 	HnswPageGetOpaque(page)->page_id = HNSW_PAGE_ID;
 }
 
+/*
+ * Init and register page
+ */
+void
+HnswInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state)
+{
+	*state = GenericXLogStart(index);
+	*page = GenericXLogRegisterBuffer(*state, *buf, GENERIC_XLOG_FULL_IMAGE);
+	HnswInitPage(*buf, *page);
+}
+
+/*
+ * Commit buffer
+ */
+void
+HnswCommitBuffer(Buffer buf, GenericXLogState *state)
+{
+	MarkBufferDirty(buf);
+	GenericXLogFinish(state);
+	UnlockReleaseBuffer(buf);
+}
+
 /*
  * Allocate neighbors
  */
@@ -202,7 +140,6 @@ HnswInitNeighbors(HnswElement element, int m)
 		a = &element->neighbors[lc];
 		a->length = 0;
 		a->items = palloc(sizeof(HnswCandidate) * lm);
-		a->closerSet = false;
 	}
 }
 
@@ -220,7 +157,7 @@ HnswInitElement(ItemPointer heaptid, int m, double ml, int maxLevel)
 	if (level > maxLevel)
 		level = maxLevel;
 
-	element->heaptidsLength = 0;
+	element->heaptids = NIL;
 	HnswAddHeapTid(element, heaptid);
 
 	element->level = level;
@@ -228,18 +165,33 @@ HnswInitElement(ItemPointer heaptid, int m, double ml, int maxLevel)
 
 	HnswInitNeighbors(element, m);
 
-	element->value = PointerGetDatum(NULL);
-
 	return element;
 }
 
+/*
+ * Free an element
+ */
+void
+HnswFreeElement(HnswElement element)
+{
+	list_free_deep(element->heaptids);
+	for (int lc = 0; lc <= element->level; lc++)
+		pfree(element->neighbors[lc].items);
+	pfree(element->neighbors);
+	pfree(element->vec);
+	pfree(element);
+}
+
 /*
  * Add a heap TID to an element
  */
 void
 HnswAddHeapTid(HnswElement element, ItemPointer heaptid)
 {
-	element->heaptids[element->heaptidsLength++] = *heaptid;
+	ItemPointer copy = palloc(sizeof(ItemPointerData));
+
+	ItemPointerCopy(heaptid, copy);
+	element->heaptids = lappend(element->heaptids, copy);
 }
 
 /*
@@ -253,7 +205,7 @@ HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno)
 	element->blkno = blkno;
 	element->offno = offno;
 	element->neighbors = NULL;
-	element->value = PointerGetDatum(NULL);
+	element->vec = NULL;
 	return element;
 }
 
@@ -331,7 +283,7 @@ HnswUpdateMetaPageInfo(Page page, int updateEntry, HnswElement entryPoint, Block
  * Update the metapage
  */
 void
-HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building)
+HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum)
 {
 	Buffer		buf;
 	Page		page;
@@ -339,24 +291,12 @@ HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, Bloc
 
 	buf = ReadBufferExtended(index, forkNum, HNSW_METAPAGE_BLKNO, RBM_NORMAL, NULL);
 	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-	if (building)
-	{
-		state = NULL;
-		page = BufferGetPage(buf);
-	}
-	else
-	{
-		state = GenericXLogStart(index);
-		page = GenericXLogRegisterBuffer(state, buf, 0);
-	}
+	state = GenericXLogStart(index);
+	page = GenericXLogRegisterBuffer(state, buf, 0);
 
 	HnswUpdateMetaPageInfo(page, updateEntry, entryPoint, insertPage);
 
-	if (building)
-		MarkBufferDirty(buf);
-	else
-		GenericXLogFinish(state);
-	UnlockReleaseBuffer(buf);
+	HnswCommitBuffer(buf, state);
 }
 
 /*
@@ -370,12 +310,12 @@ HnswSetElementTuple(HnswElementTuple etup, HnswElement element)
 	etup->deleted = 0;
 	for (int i = 0; i < HNSW_HEAPTIDS; i++)
 	{
-		if (i < element->heaptidsLength)
-			etup->heaptids[i] = element->heaptids[i];
+		if (i < list_length(element->heaptids))
+			etup->heaptids[i] = *((ItemPointer) list_nth(element->heaptids, i));
 		else
 			ItemPointerSetInvalid(&etup->heaptids[i]);
 	}
-	memcpy(&etup->data, DatumGetPointer(element->value), VARSIZE_ANY(DatumGetPointer(element->value)));
+	memcpy(&etup->vec, element->vec, VECTOR_SIZE(element->vec->dim));
 }
 
 /*
@@ -390,7 +330,7 @@ HnswSetNeighborTuple(HnswNeighborTuple ntup, HnswElement e, int m)
 
 	for (int lc = e->level; lc >= 0; lc--)
 	{
-		HnswNeighborArray *neighbors = HnswGetNeighbors(e, lc);
+		HnswNeighborArray *neighbors = &e->neighbors[lc];
 		int			lm = HnswGetLayerM(m, lc);
 
 		for (int i = 0; i < lm; i++)
@@ -448,7 +388,7 @@ LoadNeighborsFromPage(HnswElement element, Relation index, Page page, int m)
 		if (level < 0)
 			level = 0;
 
-		neighbors = HnswGetNeighbors(element, level);
+		neighbors = &element->neighbors[level];
 		hc = &neighbors->items[neighbors->length++];
 		hc->element = e;
 	}
@@ -482,7 +422,7 @@ HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHe
 	element->deleted = etup->deleted;
 	element->neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
 	element->neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
-	element->heaptidsLength = 0;
+	element->heaptids = NIL;
 
 	if (loadHeaptids)
 	{
@@ -497,7 +437,10 @@ HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHe
 	}
 
 	if (loadVec)
-		element->value = datumCopy(PointerGetDatum(&etup->data), false, -1);
+	{
+		element->vec = palloc(VECTOR_SIZE(etup->vec.dim));
+		memcpy(element->vec, &etup->vec, VECTOR_SIZE(etup->vec.dim));
+	}
 }
 
 /*
@@ -524,7 +467,7 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
 
 	/* Calculate distance */
 	if (distance != NULL)
-		*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->data)));
+		*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->vec)));
 
 	UnlockReleaseBuffer(buf);
 }
@@ -535,7 +478,7 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
 static float
 GetCandidateDistance(HnswCandidate * hc, Datum q, FmgrInfo *procinfo, Oid collation)
 {
-	return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q, hc->element->value));
+	return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q, PointerGetDatum(hc->element->vec)));
 }
 
 /*
@@ -600,22 +543,16 @@ CreatePairingHeapNode(HnswCandidate * c)
  * Add to visited
  */
 static inline void
-AddToVisited(visited_hash v, HnswCandidate * hc, Relation index, bool *found)
+AddToVisited(HTAB *v, HnswCandidate * hc, Relation index, bool *found)
 {
 	if (index == NULL)
-	{
-#if PG_VERSION_NUM >= 130000
-		pointerhash_insert_hash(v.pointers, (uintptr_t) hc->element, hc->element->hash, found);
-#else
-		pointerhash_insert(v.pointers, (uintptr_t) hc->element, found);
-#endif
-	}
+		hash_search(v, &hc->element, HASH_ENTER, found);
 	else
 	{
 		ItemPointerData indextid;
 
 		ItemPointerSet(&indextid, hc->element->blkno, hc->element->offno);
-		tidhash_insert(v.tids, indextid, found);
+		hash_search(v, &indextid, HASH_ENTER, found);
 	}
 }
 
@@ -625,26 +562,36 @@ AddToVisited(visited_hash v, HnswCandidate * hc, Relation index, bool *found)
 List *
 HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement)
 {
+	ListCell   *lc2;
+
 	List	   *w = NIL;
 	pairingheap *C = pairingheap_allocate(CompareNearestCandidates, NULL);
 	pairingheap *W = pairingheap_allocate(CompareFurthestCandidates, NULL);
 	int			wlen = 0;
-	visited_hash v;
-	ListCell   *lc2;
+	HASHCTL		hash_ctl;
+	HTAB	   *v;
 
 	/* Create hash table */
 	if (index == NULL)
-		v.pointers = pointerhash_create(CurrentMemoryContext, ef * m * 2, NULL);
+	{
+		hash_ctl.keysize = sizeof(HnswElement *);
+		hash_ctl.entrysize = sizeof(HnswElement *);
+	}
 	else
-		v.tids = tidhash_create(CurrentMemoryContext, ef * m * 2, NULL);
+	{
+		hash_ctl.keysize = sizeof(ItemPointerData);
+		hash_ctl.entrysize = sizeof(ItemPointerData);
+	}
+
+	hash_ctl.hcxt = CurrentMemoryContext;
+	v = hash_create("hnsw visited", 256, &hash_ctl, HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
 
 	/* Add entry points to v, C, and W */
 	foreach(lc2, ep)
 	{
 		HnswCandidate *hc = (HnswCandidate *) lfirst(lc2);
-		bool		found;
 
-		AddToVisited(v, hc, index, &found);
+		AddToVisited(v, hc, index, NULL);
 
 		pairingheap_add(C, &(CreatePairingHeapNode(hc)->ph_node));
 		pairingheap_add(W, &(CreatePairingHeapNode(hc)->ph_node));
@@ -654,7 +601,7 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
 		 * would be ideal to do this for inserts as well, but this could
 		 * affect insert performance.
 		 */
-		if (skipElement == NULL || hc->element->heaptidsLength != 0)
+		if (skipElement == NULL || list_length(hc->element->heaptids) != 0)
 			wlen++;
 	}
 
@@ -671,7 +618,7 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
 			HnswLoadNeighbors(c->element, index, m);
 
 		/* Get the neighborhood at layer lc */
-		neighborhood = HnswGetNeighbors(c->element, lc);
+		neighborhood = &c->element->neighbors[lc];
 
 		for (int i = 0; i < neighborhood->length; i++)
 		{
@@ -713,7 +660,7 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
 					 * vacuuming. It would be ideal to do this for inserts as
 					 * well, but this could affect insert performance.
 					 */
-					if (skipElement == NULL || e->element->heaptidsLength != 0)
+					if (skipElement == NULL || list_length(e->element->heaptids) != 0)
 					{
 						wlen++;
 
@@ -737,34 +684,6 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
 	return w;
 }
 
-/*
- * Compare candidate distances
- */
-static int
-#if PG_VERSION_NUM >= 130000
-CompareCandidateDistances(const ListCell *a, const ListCell *b)
-#else
-CompareCandidateDistances(const void *a, const void *b)
-#endif
-{
-	HnswCandidate *hca = lfirst((ListCell *) a);
-	HnswCandidate *hcb = lfirst((ListCell *) b);
-
-	if (hca->distance < hcb->distance)
-		return 1;
-
-	if (hca->distance > hcb->distance)
-		return -1;
-
-	if (hca->element < hcb->element)
-		return 1;
-
-	if (hca->element > hcb->element)
-		return -1;
-
-	return 0;
-}
-
 /*
  * Calculate the distance between elements
  */
@@ -774,27 +693,27 @@ HnswGetDistance(HnswElement a, HnswElement b, int lc, FmgrInfo *procinfo, Oid co
 	/* Look for cached distance */
 	if (a->neighbors != NULL)
 	{
-		HnswNeighborArray *neighbors = HnswGetNeighbors(a, lc);
+		Assert(a->level >= lc);
 
-		for (int i = 0; i < neighbors->length; i++)
+		for (int i = 0; i < a->neighbors[lc].length; i++)
 		{
-			if (neighbors->items[i].element == b)
-				return neighbors->items[i].distance;
+			if (a->neighbors[lc].items[i].element == b)
+				return a->neighbors[lc].items[i].distance;
 		}
 	}
 
 	if (b->neighbors != NULL)
 	{
-		HnswNeighborArray *neighbors = HnswGetNeighbors(b, lc);
+		Assert(b->level >= lc);
 
-		for (int i = 0; i < neighbors->length; i++)
+		for (int i = 0; i < b->neighbors[lc].length; i++)
 		{
-			if (neighbors->items[i].element == a)
-				return neighbors->items[i].distance;
+			if (b->neighbors[lc].items[i].element == a)
+				return b->neighbors[lc].items[i].distance;
 		}
 	}
 
-	return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, a->value, b->value));
+	return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(a->vec), PointerGetDatum(b->vec)));
 }
 
 /*
@@ -821,78 +740,33 @@ CheckElementCloser(HnswCandidate * e, List *r, int lc, FmgrInfo *procinfo, Oid c
  * Algorithm 4 from paper
  */
 static List *
-SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswElement e2, HnswCandidate * newCandidate, HnswCandidate * *pruned, bool sortCandidates)
+SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswCandidate * *pruned)
 {
 	List	   *r = NIL;
 	List	   *w = list_copy(c);
 	pairingheap *wd;
-	HnswNeighborArray *neighbors = HnswGetNeighbors(e2, lc);
-	bool		mustCalculate = !neighbors->closerSet;
-	List	   *added = NIL;
-	bool		removedAny = false;
 
 	if (list_length(w) <= m)
 		return w;
 
 	wd = pairingheap_allocate(CompareNearestCandidates, NULL);
 
-	/* Ensure order of candidates is deterministic for closer caching */
-	if (sortCandidates)
-		list_sort(w, CompareCandidateDistances);
-
 	while (list_length(w) > 0 && list_length(r) < m)
 	{
 		/* Assumes w is already ordered desc */
 		HnswCandidate *e = llast(w);
+		bool		closer;
 
 		w = list_delete_last(w);
 
-		/* Use previous state of r and wd to skip work when possible */
-		if (mustCalculate)
-			e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
-		else if (list_length(added) > 0)
-		{
-			/*
-			 * If the current candidate was closer, we only need to compare it
-			 * with the other candidates that we have added.
-			 */
-			if (e->closer)
-			{
-				e->closer = CheckElementCloser(e, added, lc, procinfo, collation);
+		closer = CheckElementCloser(e, r, lc, procinfo, collation);
 
-				if (!e->closer)
-					removedAny = true;
-			}
-			else
-			{
-				/*
-				 * If we have removed any candidates from closer, a candidate
-				 * that was not closer earlier might now be.
-				 */
-				if (removedAny)
-				{
-					e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
-					if (e->closer)
-						added = lappend(added, e);
-				}
-			}
-		}
-		else if (e == newCandidate)
-		{
-			e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
-			if (e->closer)
-				added = lappend(added, e);
-		}
-
-		if (e->closer)
+		if (closer)
 			r = lappend(r, e);
 		else
 			pairingheap_add(wd, &(CreatePairingHeapNode(e)->ph_node));
 	}
 
-	/* Cached value can only be used in future if sorted deterministically */
-	neighbors->closerSet = sortCandidates;
-
 	/* Keep pruned connections */
 	while (!pairingheap_is_empty(wd) && list_length(r) < m)
 		r = lappend(r, ((HnswPairingHeapNode *) pairingheap_remove_first(wd))->inner);
@@ -909,6 +783,30 @@ SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswE
 	return r;
 }
 
+/*
+ * Find duplicate element
+ */
+HnswElement
+HnswFindDuplicate(HnswElement e)
+{
+	HnswNeighborArray *neighbors = &e->neighbors[0];
+
+	for (int i = 0; i < neighbors->length; i++)
+	{
+		HnswCandidate *neighbor = &neighbors->items[i];
+
+		/* Exit early since ordered by distance */
+		if (vector_cmp_internal(e->vec, neighbor->element->vec) != 0)
+			break;
+
+		/* Check for space */
+		if (list_length(neighbor->element->heaptids) < HNSW_HEAPTIDS)
+			return neighbor->element;
+	}
+
+	return NULL;
+}
+
 /*
  * Add connections
  */
@@ -916,19 +814,41 @@ static void
 AddConnections(HnswElement element, List *neighbors, int m, int lc)
 {
 	ListCell   *lc2;
-	HnswNeighborArray *a = HnswGetNeighbors(element, lc);
+	HnswNeighborArray *a = &element->neighbors[lc];
 
 	foreach(lc2, neighbors)
 		a->items[a->length++] = *((HnswCandidate *) lfirst(lc2));
 }
 
+/*
+ * Compare candidate distances
+ */
+static int
+#if PG_VERSION_NUM >= 130000
+CompareCandidateDistances(const ListCell *a, const ListCell *b)
+#else
+CompareCandidateDistances(const void *a, const void *b)
+#endif
+{
+	HnswCandidate *hca = lfirst((ListCell *) a);
+	HnswCandidate *hcb = lfirst((ListCell *) b);
+
+	if (hca->distance < hcb->distance)
+		return 1;
+
+	if (hca->distance > hcb->distance)
+		return -1;
+
+	return 0;
+}
+
 /*
  * Update connections
  */
 void
 HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation)
 {
-	HnswNeighborArray *currentNeighbors = HnswGetNeighbors(hc->element, lc);
+	HnswNeighborArray *currentNeighbors = &hc->element->neighbors[lc];
 
 	HnswCandidate hc2;
 
@@ -951,19 +871,19 @@ HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int
 		/* Load elements on insert */
 		if (index != NULL)
 		{
-			Datum		q = hc->element->value;
+			Datum		q = PointerGetDatum(hc->element->vec);
 
 			for (int i = 0; i < currentNeighbors->length; i++)
 			{
 				HnswCandidate *hc3 = &currentNeighbors->items[i];
 
-				if (DatumGetPointer(hc3->element->value) == NULL)
+				if (hc3->element->vec == NULL)
 					HnswLoadElement(hc3->element, &hc3->distance, &q, index, procinfo, collation, true);
 				else
 					hc3->distance = GetCandidateDistance(hc3, q, procinfo, collation);
 
 				/* Prune element if being deleted */
-				if (hc3->element->heaptidsLength == 0)
+				if (list_length(hc3->element->heaptids) == 0)
 				{
 					pruned = &currentNeighbors->items[i];
 					break;
@@ -975,12 +895,13 @@ HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int
 		{
 			List	   *c = NIL;
 
-			/* Add candidates */
+			/* Add and sort candidates */
 			for (int i = 0; i < currentNeighbors->length; i++)
 				c = lappend(c, &currentNeighbors->items[i]);
 			c = lappend(c, &hc2);
+			list_sort(c, CompareCandidateDistances);
 
-			SelectNeighbors(c, m, lc, procinfo, collation, hc->element, &hc2, &pruned, true);
+			SelectNeighbors(c, m, lc, procinfo, collation, &pruned);
 
 			/* Should not happen */
 			if (pruned == NULL)
@@ -1021,7 +942,7 @@ RemoveElements(List *w, HnswElement skipElement)
 		if (skipElement != NULL && hc->element->blkno == skipElement->blkno && hc->element->offno == skipElement->offno)
 			continue;
 
-		if (hc->element->heaptidsLength != 0)
+		if (list_length(hc->element->heaptids) != 0)
 			w2 = lappend(w2, hc);
 	}
 
@@ -1038,15 +959,9 @@ HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, F
 	List	   *w;
 	int			level = element->level;
 	int			entryLevel;
-	Datum		q = element->value;
+	Datum		q = PointerGetDatum(element->vec);
 	HnswElement skipElement = existing ? element : NULL;
 
-#if PG_VERSION_NUM >= 130000
-	/* Precompute hash */
-	if (index == NULL)
-		element->hash = hash_pointer((uintptr_t) element);
-#endif
-
 	/* No neighbors if no entry point */
 	if (entryPoint == NULL)
 		return;
@@ -1085,12 +1000,7 @@ HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, F
 		else
 			lw = w;
 
-		/*
-		 * Candidates are sorted, but not deterministically. Could set
-		 * sortCandidates to true for in-memory builds to enable closer
-		 * caching, but there does not seem to be a difference in performance.
-		 */
-		neighbors = SelectNeighbors(lw, lm, lc, procinfo, collation, element, NULL, NULL, false);
+		neighbors = SelectNeighbors(lw, lm, lc, procinfo, collation, NULL);
 
 		AddConnections(element, neighbors, lm, lc);
 

src/hnswvacuum.c

@@ -12,9 +12,12 @@
  * Check if deleted list contains an index TID
  */
 static bool
-DeletedContains(tidhash_hash * deleted, ItemPointer indextid)
+DeletedContains(HTAB *deleted, ItemPointer indextid)
 {
-	return tidhash_lookup(deleted, *indextid) != NULL;
+	bool		found;
+
+	hash_search(deleted, indextid, HASH_FIND, &found);
+	return found;
 }
 
 /*
@@ -59,8 +62,7 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 		/* Iterate over nodes */
 		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
 		{
-			ItemId		itemid = PageGetItemId(page, offno);
-			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, itemid);
+			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
 			int			idx = 0;
 			bool		itemUpdated = false;
 
@@ -91,7 +93,7 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 
 				if (itemUpdated)
 				{
-					Size		etupSize = ItemIdGetLength(itemid);
+					Size		etupSize = HNSW_ELEMENT_TUPLE_SIZE(etup->vec.dim);
 
 					/* Mark rest as invalid */
 					for (int i = idx; i < HNSW_HEAPTIDS; i++)
@@ -107,13 +109,11 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 			if (!ItemPointerIsValid(&etup->heaptids[0]))
 			{
 				ItemPointerData ip;
-				bool		found;
 
 				/* Add to deleted list */
 				ItemPointerSet(&ip, blkno, offno);
 
-				tidhash_insert(vacuumstate->deleted, ip, &found);
-				Assert(!found);
+				(void) hash_search(vacuumstate->deleted, &ip, HASH_ENTER, NULL);
 			}
 			else if (etup->level > highestLevel && !(entryPoint != NULL && blkno == entryPoint->blkno && offno == entryPoint->offno))
 			{
@@ -128,7 +128,10 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 		blkno = HnswPageGetOpaque(page)->nextblkno;
 
 		if (updated)
+		{
+			MarkBufferDirty(buf);
 			GenericXLogFinish(state);
+		}
 		else
 			GenericXLogAbort(state);
 
@@ -206,7 +209,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 
 	/* Init fields */
 	HnswInitNeighbors(element, m);
-	element->heaptidsLength = 0;
+	element->heaptids = NIL;
 
 	/* Add element to graph, skipping itself */
 	HnswInsertElement(element, entryPoint, index, procinfo, collation, m, efConstruction, true);
@@ -226,11 +229,12 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
 
 	/* Commit */
+	MarkBufferDirty(buf);
 	GenericXLogFinish(state);
 	UnlockReleaseBuffer(buf);
 
 	/* Update neighbors */
-	HnswUpdateNeighborPages(index, procinfo, collation, element, m, true, false);
+	HnswUpdateNeighborPages(index, procinfo, collation, element, m, true);
 }
 
 /*
@@ -286,7 +290,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 			 * point is outdated and empty, the entry point will be empty
 			 * until an element is repaired.
 			 */
-			HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_ALWAYS, highestPoint, InvalidBlockNumber, MAIN_FORKNUM, false);
+			HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_ALWAYS, highestPoint, InvalidBlockNumber, MAIN_FORKNUM);
 		}
 		else
 		{
@@ -419,7 +423,7 @@ RepairGraph(HnswVacuumState * vacuumstate)
 			 * was replaced and highest point was outdated.
 			 */
 			if (entryPoint == NULL || element->level > entryPoint->level)
-				HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_GREATER, element, InvalidBlockNumber, MAIN_FORKNUM, false);
+				HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_GREATER, element, InvalidBlockNumber, MAIN_FORKNUM);
 
 			/* Release lock */
 			UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);
@@ -477,8 +481,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 		/* Update element and neighbors together */
 		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
 		{
-			ItemId		itemid = PageGetItemId(page, offno);
-			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, itemid);
+			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
 			HnswNeighborTuple ntup;
 			Size		etupSize;
 			Size		ntupSize;
@@ -506,7 +509,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 				continue;
 
 			/* Calculate sizes */
-			etupSize = ItemIdGetLength(itemid);
+			etupSize = HNSW_ELEMENT_TUPLE_SIZE(etup->vec.dim);
 			ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(etup->level, vacuumstate->m);
 
 			/* Get neighbor page */
@@ -529,7 +532,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 
 			/* Overwrite element */
 			etup->deleted = 1;
-			MemSet(&etup->data, 0, VARSIZE_ANY(&etup->data));
+			MemSet(&etup->vec.x, 0, etup->vec.dim * sizeof(float));
 
 			/* Overwrite neighbors */
 			for (int i = 0; i < ntup->count; i++)
@@ -544,6 +547,9 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 				elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
 
 			/* Commit */
+			MarkBufferDirty(buf);
+			if (nbuf != buf)
+				MarkBufferDirty(nbuf);
 			GenericXLogFinish(state);
 			if (nbuf != buf)
 				UnlockReleaseBuffer(nbuf);
@@ -564,7 +570,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 	}
 
 	/* Update insert page last, after everything has been marked as deleted */
-	HnswUpdateMetaPage(index, 0, NULL, insertPage, MAIN_FORKNUM, false);
+	HnswUpdateMetaPage(index, 0, NULL, insertPage, MAIN_FORKNUM);
 }
 
 /*
@@ -574,6 +580,7 @@ static void
 InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkDeleteResult *stats, IndexBulkDeleteCallback callback, void *callback_state)
 {
 	Relation	index = info->index;
+	HASHCTL		hash_ctl;
 
 	if (stats == NULL)
 		stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
@@ -595,7 +602,10 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
 	HnswGetMetaPageInfo(index, &vacuumstate->m, NULL);
 
 	/* Create hash table */
-	vacuumstate->deleted = tidhash_create(CurrentMemoryContext, 256, NULL);
+	hash_ctl.keysize = sizeof(ItemPointerData);
+	hash_ctl.entrysize = sizeof(ItemPointerData);
+	hash_ctl.hcxt = CurrentMemoryContext;
+	vacuumstate->deleted = hash_create("hnswbulkdelete indextids", 256, &hash_ctl, HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
 }
 
 /*
@@ -604,7 +614,7 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
 static void
 FreeVacuumState(HnswVacuumState * vacuumstate)
 {
-	tidhash_destroy(vacuumstate->deleted);
+	hash_destroy(vacuumstate->deleted);
 	FreeAccessStrategy(vacuumstate->bas);
 	pfree(vacuumstate->ntup);
 	MemoryContextDelete(vacuumstate->tmpCtx);

src/ivfbuild.c

@@ -506,30 +506,29 @@ CreateListPages(Relation index, VectorArray centers, int dimensions,
 	Buffer		buf;
 	Page		page;
 	GenericXLogState *state;
-	Size		listSize;
+	OffsetNumber offno;
+	Size		itemsz;
 	IvfflatList list;
 
-	listSize = MAXALIGN(IVFFLAT_LIST_SIZE(dimensions));
-	list = palloc(listSize);
+	itemsz = MAXALIGN(IVFFLAT_LIST_SIZE(dimensions));
+	list = palloc(itemsz);
 
 	buf = IvfflatNewBuffer(index, forkNum);
 	IvfflatInitRegisterPage(index, &buf, &page, &state);
 
 	for (int i = 0; i < lists; i++)
 	{
-		OffsetNumber offno;
-
 		/* Load list */
 		list->startPage = InvalidBlockNumber;
 		list->insertPage = InvalidBlockNumber;
 		memcpy(&list->center, VectorArrayGet(centers, i), VECTOR_SIZE(dimensions));
 
 		/* Ensure free space */
-		if (PageGetFreeSpace(page) < listSize)
+		if (PageGetFreeSpace(page) < itemsz)
 			IvfflatAppendPage(index, &buf, &page, &state, forkNum);
 
 		/* Add the item */
-		offno = PageAddItem(page, (Item) list, listSize, InvalidOffsetNumber, false, false);
+		offno = PageAddItem(page, (Item) list, itemsz, InvalidOffsetNumber, false, false);
 		if (offno == InvalidOffsetNumber)
 			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
 
@@ -543,10 +542,10 @@ CreateListPages(Relation index, VectorArray centers, int dimensions,
 	pfree(list);
 }
 
-#ifdef IVFFLAT_KMEANS_DEBUG
 /*
  * Print k-means metrics
  */
+#ifdef IVFFLAT_KMEANS_DEBUG
 static void
 PrintKmeansMetrics(IvfflatBuildState * buildstate)
 {

src/ivfflat.c

@@ -72,14 +72,12 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	double		ratio;
 	double		spc_seq_page_cost;
 	Relation	index;
-	double		selectivity = 1;
-	ListCell   *lc;
 #if PG_VERSION_NUM < 120000
 	List	   *qinfos;
 #endif
 
-	/* Never use index without order or limit */
-	if (path->indexorderbys == NULL || root->limit_tuples < 0)
+	/* Never use index without order */
+	if (path->indexorderbys == NULL)
 	{
 		*indexStartupCost = DBL_MAX;
 		*indexTotalCost = DBL_MAX;
@@ -107,29 +105,6 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	 */
 	costs.numIndexTuples = path->indexinfo->tuples * ratio;
 
-	/* Get the selectivity of non-index conditions */
-	foreach(lc, path->indexinfo->indrestrictinfo)
-	{
-		RestrictInfo *rinfo = lfirst(lc);
-
-		if (rinfo->norm_selec >= 0 && rinfo->norm_selec <= 1 && rinfo->norm_selec != (Selectivity) DEFAULT_INEQ_SEL)
-			selectivity *= rinfo->norm_selec;
-	}
-
-	/*
-	 * Do not use index if limit + offset > expected tuples unless
-	 * enable_seqscan = off
-	 */
-	if (root->limit_tuples > costs.numIndexTuples * selectivity)
-	{
-		*indexStartupCost = 1.0e10 - 1;
-		*indexTotalCost = 1.0e10 - 1;
-		*indexSelectivity = 0;
-		*indexCorrelation = 0;
-		*indexPages = 0;
-		return;
-	}
-
 #if PG_VERSION_NUM >= 120000
 	genericcostestimate(root, path, loop_count, &costs);
 #else

src/ivfflat.h

@@ -40,6 +40,9 @@
 #define IVFFLAT_METAPAGE_BLKNO	0
 #define IVFFLAT_HEAD_BLKNO		1	/* first list page */
 
+/* Must correspond to page numbers since page lock is used */
+#define IVFFLAT_SCAN_LOCK		0
+
 /* IVFFlat parameters */
 #define IVFFLAT_DEFAULT_LISTS	100
 #define IVFFLAT_MIN_LISTS		1
@@ -246,6 +249,9 @@ typedef struct IvfflatScanOpaqueData
 	int			probes;
 	int			dimensions;
 	bool		first;
+	bool		hasLock;
+	Buffer		buf;
+	ItemPointerData heaptid;
 
 	/* Sorting */
 	Tuplesortstate *sortstate;
@@ -275,7 +281,7 @@ VectorArray VectorArrayInit(int maxlen, int dimensions);
 void		VectorArrayFree(VectorArray arr);
 void		PrintVectorArray(char *msg, VectorArray arr);
 void		IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
-FmgrInfo   *IvfflatOptionalProcInfo(Relation index, uint16 procnum);
+FmgrInfo   *IvfflatOptionalProcInfo(Relation rel, uint16 procnum);
 bool		IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
 int			IvfflatGetLists(Relation index);
 void		IvfflatGetMetaPageInfo(Relation index, int *lists, int *dimensions);

src/ivfinsert.c

@@ -11,37 +11,36 @@
  * Find the list that minimizes the distance function
  */
 static void
-FindInsertPage(Relation index, Datum *values, BlockNumber *insertPage, ListInfo * listInfo)
+FindInsertPage(Relation rel, Datum *values, BlockNumber *insertPage, ListInfo * listInfo)
 {
+	Buffer		cbuf;
+	Page		cpage;
+	IvfflatList list;
+	double		distance;
 	double		minDistance = DBL_MAX;
 	BlockNumber nextblkno = IVFFLAT_HEAD_BLKNO;
 	FmgrInfo   *procinfo;
 	Oid			collation;
+	OffsetNumber offno;
+	OffsetNumber maxoffno;
 
 	/* Avoid compiler warning */
 	listInfo->blkno = nextblkno;
 	listInfo->offno = FirstOffsetNumber;
 
-	procinfo = index_getprocinfo(index, 1, IVFFLAT_DISTANCE_PROC);
-	collation = index->rd_indcollation[0];
+	procinfo = index_getprocinfo(rel, 1, IVFFLAT_DISTANCE_PROC);
+	collation = rel->rd_indcollation[0];
 
 	/* Search all list pages */
 	while (BlockNumberIsValid(nextblkno))
 	{
-		Buffer		cbuf;
-		Page		cpage;
-		OffsetNumber maxoffno;
-
-		cbuf = ReadBuffer(index, nextblkno);
+		cbuf = ReadBuffer(rel, nextblkno);
 		LockBuffer(cbuf, BUFFER_LOCK_SHARE);
 		cpage = BufferGetPage(cbuf);
 		maxoffno = PageGetMaxOffsetNumber(cpage);
 
-		for (OffsetNumber offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
+		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
 		{
-			IvfflatList list;
-			double		distance;
-
 			list = (IvfflatList) PageGetItem(cpage, PageGetItemId(cpage, offno));
 			distance = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, values[0], PointerGetDatum(&list->center)));
 
@@ -64,7 +63,7 @@ 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 rel, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel)
 {
 	IndexTuple	itup;
 	Datum		value;
@@ -81,33 +80,33 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 	value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
 
 	/* Normalize if needed */
-	normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
+	normprocinfo = IvfflatOptionalProcInfo(rel, IVFFLAT_NORM_PROC);
 	if (normprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, NULL))
+		if (!IvfflatNormValue(normprocinfo, rel->rd_indcollation[0], &value, NULL))
 			return;
 	}
 
 	/* Find the insert page - sets the page and list info */
-	FindInsertPage(index, values, &insertPage, &listInfo);
+	FindInsertPage(rel, values, &insertPage, &listInfo);
 	Assert(BlockNumberIsValid(insertPage));
 	originalInsertPage = insertPage;
 
 	/* Form tuple */
-	itup = index_form_tuple(RelationGetDescr(index), &value, isnull);
+	itup = index_form_tuple(RelationGetDescr(rel), &value, isnull);
 	itup->t_tid = *heap_tid;
 
 	/* Get tuple size */
 	itemsz = MAXALIGN(IndexTupleSize(itup));
-	Assert(itemsz <= BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(IvfflatPageOpaqueData)) - sizeof(ItemIdData));
+	Assert(itemsz <= BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(IvfflatPageOpaqueData)));
 
 	/* Find a page to insert the item */
 	for (;;)
 	{
-		buf = ReadBuffer(index, insertPage);
+		buf = ReadBuffer(rel, insertPage);
 		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
 
-		state = GenericXLogStart(index);
+		state = GenericXLogStart(rel);
 		page = GenericXLogRegisterBuffer(state, buf, 0);
 
 		if (PageGetFreeSpace(page) >= itemsz)
@@ -127,9 +126,9 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 			Page		newpage;
 
 			/* Add a new page */
-			LockRelationForExtension(index, ExclusiveLock);
-			newbuf = IvfflatNewBuffer(index, MAIN_FORKNUM);
-			UnlockRelationForExtension(index, ExclusiveLock);
+			LockRelationForExtension(rel, ExclusiveLock);
+			newbuf = IvfflatNewBuffer(rel, MAIN_FORKNUM);
+			UnlockRelationForExtension(rel, ExclusiveLock);
 
 			/* Init new page */
 			newpage = GenericXLogRegisterBuffer(state, newbuf, GENERIC_XLOG_FULL_IMAGE);
@@ -142,13 +141,15 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 			IvfflatPageGetOpaque(page)->nextblkno = insertPage;
 
 			/* Commit */
+			MarkBufferDirty(newbuf);
+			MarkBufferDirty(buf);
 			GenericXLogFinish(state);
 
 			/* Unlock previous buffer */
 			UnlockReleaseBuffer(buf);
 
 			/* Prepare new buffer */
-			state = GenericXLogStart(index);
+			state = GenericXLogStart(rel);
 			buf = newbuf;
 			page = GenericXLogRegisterBuffer(state, buf, 0);
 			break;
@@ -157,13 +158,13 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 
 	/* Add to next offset */
 	if (PageAddItem(page, (Item) itup, itemsz, InvalidOffsetNumber, false, false) == InvalidOffsetNumber)
-		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
+		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(rel));
 
 	IvfflatCommitBuffer(buf, state);
 
 	/* Update the insert page */
 	if (insertPage != originalInsertPage)
-		IvfflatUpdateList(index, listInfo, insertPage, originalInsertPage, InvalidBlockNumber, MAIN_FORKNUM);
+		IvfflatUpdateList(rel, listInfo, insertPage, originalInsertPage, InvalidBlockNumber, MAIN_FORKNUM);
 }
 
 /*

src/ivfkmeans.c

@@ -6,10 +6,6 @@
 #include "ivfflat.h"
 #include "miscadmin.h"
 
-#ifdef IVFFLAT_MEMORY
-#include "utils/memutils.h"
-#endif
-
 /*
  * Initialize with kmeans++
  *
@@ -21,6 +17,10 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
 	FmgrInfo   *procinfo;
 	Oid			collation;
 	int64		j;
+	double		distance;
+	double		sum;
+	double		choice;
+	Vector	   *vec;
 	float	   *weight = palloc(samples->length * sizeof(float));
 	int			numCenters = centers->maxlen;
 	int			numSamples = samples->length;
@@ -33,21 +33,17 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
 	centers->length++;
 
 	for (j = 0; j < numSamples; j++)
-		weight[j] = FLT_MAX;
+		weight[j] = DBL_MAX;
 
 	for (int i = 0; i < numCenters; i++)
 	{
-		double		sum;
-		double		choice;
-
 		CHECK_FOR_INTERRUPTS();
 
 		sum = 0.0;
 
 		for (j = 0; j < numSamples; j++)
 		{
-			Vector	   *vec = VectorArrayGet(samples, j);
-			double		distance;
+			vec = VectorArrayGet(samples, j);
 
 			/* Only need to compute distance for new center */
 			/* TODO Use triangle inequality to reduce distance calculations */
@@ -116,6 +112,7 @@ CompareVectors(const void *a, const void *b)
 static void
 QuickCenters(Relation index, VectorArray samples, VectorArray centers)
 {
+	Vector	   *vec;
 	int			dimensions = centers->dim;
 	Oid			collation = index->rd_indcollation[0];
 	FmgrInfo   *normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_KMEANS_NORM_PROC);
@@ -126,7 +123,7 @@ QuickCenters(Relation index, VectorArray samples, VectorArray centers)
 		qsort(samples->items, samples->length, VECTOR_SIZE(samples->dim), CompareVectors);
 		for (int i = 0; i < samples->length; i++)
 		{
-			Vector	   *vec = VectorArrayGet(samples, i);
+			vec = VectorArrayGet(samples, i);
 
 			if (i == 0 || CompareVectors(vec, VectorArrayGet(samples, i - 1)) != 0)
 			{
@@ -139,7 +136,7 @@ QuickCenters(Relation index, VectorArray samples, VectorArray centers)
 	/* Fill remaining with random data */
 	while (centers->length < centers->maxlen)
 	{
-		Vector	   *vec = VectorArrayGet(centers, centers->length);
+		vec = VectorArrayGet(centers, centers->length);
 
 		SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
 		vec->dim = dimensions;
@@ -155,23 +152,6 @@ QuickCenters(Relation index, VectorArray samples, VectorArray centers)
 	}
 }
 
-#ifdef IVFFLAT_MEMORY
-/*
- * Show memory usage
- */
-static void
-ShowMemoryUsage(Size estimatedSize)
-{
-#if PG_VERSION_NUM >= 130000
-	elog(INFO, "total memory: %zu MB",
-		 MemoryContextMemAllocated(CurrentMemoryContext, true) / (1024 * 1024));
-#else
-	MemoryContextStats(CurrentMemoryContext);
-#endif
-	elog(INFO, "estimated memory: %zu MB", estimatedSize / (1024 * 1024));
-}
-#endif
-
 /*
  * Use Elkan for performance. This requires distance function to satisfy triangle inequality.
  *
@@ -188,6 +168,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 	Oid			collation;
 	Vector	   *vec;
 	Vector	   *newCenter;
+	int			iteration;
 	int64		j;
 	int64		k;
 	int			dimensions = centers->dim;
@@ -201,6 +182,14 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 	float	   *s;
 	float	   *halfcdist;
 	float	   *newcdist;
+	int			changes;
+	double		minDistance;
+	int			closestCenter;
+	double		distance;
+	bool		rj;
+	bool		rjreset;
+	double		dxcx;
+	double		dxc;
 
 	/* Calculate allocation sizes */
 	Size		samplesSize = VECTOR_ARRAY_SIZE(samples->maxlen, samples->dim);
@@ -252,24 +241,20 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 		vec->dim = dimensions;
 	}
 
-#ifdef IVFFLAT_MEMORY
-	ShowMemoryUsage(totalSize);
-#endif
-
 	/* Pick initial centers */
 	InitCenters(index, samples, centers, lowerBound);
 
 	/* Assign each x to its closest initial center c(x) = argmin d(x,c) */
 	for (j = 0; j < numSamples; j++)
 	{
-		float		minDistance = FLT_MAX;
-		int			closestCenter = 0;
+		minDistance = DBL_MAX;
+		closestCenter = 0;
 
 		/* Find closest center */
 		for (k = 0; k < numCenters; k++)
 		{
 			/* TODO Use Lemma 1 in k-means++ initialization */
-			float		distance = lowerBound[j * numCenters + k];
+			distance = lowerBound[j * numCenters + k];
 
 			if (distance < minDistance)
 			{
@@ -283,14 +268,13 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 	}
 
 	/* Give 500 iterations to converge */
-	for (int iteration = 0; iteration < 500; iteration++)
+	for (iteration = 0; iteration < 500; iteration++)
 	{
-		int			changes = 0;
-		bool		rjreset;
-
 		/* Can take a while, so ensure we can interrupt */
 		CHECK_FOR_INTERRUPTS();
 
+		changes = 0;
+
 		/* Step 1: For all centers, compute distance */
 		for (j = 0; j < numCenters; j++)
 		{
@@ -298,8 +282,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 			for (k = j + 1; k < numCenters; k++)
 			{
-				float		distance = 0.5 * DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(vec), PointerGetDatum(VectorArrayGet(centers, k))));
-
+				distance = 0.5 * DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(vec), PointerGetDatum(VectorArrayGet(centers, k))));
 				halfcdist[j * numCenters + k] = distance;
 				halfcdist[k * numCenters + j] = distance;
 			}
@@ -308,12 +291,10 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 		/* For all centers c, compute s(c) */
 		for (j = 0; j < numCenters; j++)
 		{
-			float		minDistance = FLT_MAX;
+			minDistance = DBL_MAX;
 
 			for (k = 0; k < numCenters; k++)
 			{
-				float		distance;
-
 				if (j == k)
 					continue;
 
@@ -329,8 +310,6 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 		for (j = 0; j < numSamples; j++)
 		{
-			bool		rj;
-
 			/* Step 2: Identify all points x such that u(x) <= s(c(x)) */
 			if (upperBound[j] <= s[closestCenters[j]])
 				continue;
@@ -339,8 +318,6 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 			for (k = 0; k < numCenters; k++)
 			{
-				float		dxcx;
-
 				/* Step 3: For all remaining points x and centers c */
 				if (k == closestCenters[j])
 					continue;
@@ -370,7 +347,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 				/* Step 3b */
 				if (dxcx > lowerBound[j * numCenters + k] || dxcx > halfcdist[closestCenters[j] * numCenters + k])
 				{
-					float		dxc = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(vec), PointerGetDatum(VectorArrayGet(centers, k))));
+					dxc = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(vec), PointerGetDatum(VectorArrayGet(centers, k))));
 
 					/* d(x,c) calculated */
 					lowerBound[j * numCenters + k] = dxc;
@@ -384,6 +361,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 						changes++;
 					}
+
 				}
 			}
 		}
@@ -400,8 +378,6 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 		for (j = 0; j < numSamples; j++)
 		{
-			int			closestCenter;
-
 			vec = VectorArrayGet(samples, j);
 			closestCenter = closestCenters[j];
 
@@ -450,7 +426,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 		{
 			for (k = 0; k < numCenters; k++)
 			{
-				float		distance = lowerBound[j * numCenters + k] - newcdist[k];
+				distance = lowerBound[j * numCenters + k] - newcdist[k];
 
 				if (distance < 0)
 					distance = 0;
@@ -466,7 +442,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 		/* Step 7 */
 		for (j = 0; j < numCenters; j++)
-			VectorArraySet(centers, j, VectorArrayGet(newCenters, j));
+			memcpy(VectorArrayGet(centers, j), VectorArrayGet(newCenters, j), VECTOR_SIZE(dimensions));
 
 		if (changes == 0 && iteration != 0)
 			break;
@@ -489,6 +465,9 @@ static void
 CheckCenters(Relation index, VectorArray centers)
 {
 	FmgrInfo   *normprocinfo;
+	Oid			collation;
+	Vector	   *vec;
+	double		norm;
 
 	if (centers->length != centers->maxlen)
 		elog(ERROR, "Not enough centers. Please report a bug.");
@@ -496,7 +475,7 @@ CheckCenters(Relation index, VectorArray centers)
 	/* Ensure no NaN or infinite values */
 	for (int i = 0; i < centers->length; i++)
 	{
-		Vector	   *vec = VectorArrayGet(centers, i);
+		vec = VectorArrayGet(centers, i);
 
 		for (int j = 0; j < vec->dim; j++)
 		{
@@ -522,12 +501,11 @@ CheckCenters(Relation index, VectorArray centers)
 	normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
 	if (normprocinfo != NULL)
 	{
-		Oid			collation = index->rd_indcollation[0];
+		collation = index->rd_indcollation[0];
 
 		for (int i = 0; i < centers->length; i++)
 		{
-			double		norm = DatumGetFloat8(FunctionCall1Coll(normprocinfo, collation, PointerGetDatum(VectorArrayGet(centers, i))));
-
+			norm = DatumGetFloat8(FunctionCall1Coll(normprocinfo, collation, PointerGetDatum(VectorArrayGet(centers, i))));
 			if (norm == 0)
 				elog(ERROR, "Zero norm detected. Please report a bug.");
 		}

src/ivfscan.c

@@ -9,6 +9,7 @@
 #include "miscadmin.h"
 #include "pgstat.h"
 #include "storage/bufmgr.h"
+#include "storage/lmgr.h"
 
 /*
  * Compare list distances
@@ -143,6 +144,10 @@ GetScanItems(IndexScanDesc scan, Datum value)
 				bool		isnull;
 				ItemId		itemid = PageGetItemId(page, offno);
 
+				/* Skip dead tuples */
+				if (scan->ignore_killed_tuples && ItemIdIsDead(itemid))
+					continue;
+
 				itup = (IndexTuple) PageGetItem(page, itemid);
 				datum = index_getattr(itup, 1, tupdesc, &isnull);
 
@@ -157,6 +162,8 @@ GetScanItems(IndexScanDesc scan, Datum value)
 				slot->tts_isnull[0] = false;
 				slot->tts_values[1] = PointerGetDatum(&itup->t_tid);
 				slot->tts_isnull[1] = false;
+				slot->tts_values[2] = Int32GetDatum((int) searchPage);
+				slot->tts_isnull[2] = false;
 				ExecStoreVirtualTuple(slot);
 
 				tuplesort_puttupleslot(so->sortstate, slot);
@@ -181,6 +188,55 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	tuplesort_performsort(so->sortstate);
 }
 
+/*
+ * Mark prior tuple as dead
+ */
+static void
+MarkPriorTupleDead(IndexScanDesc scan)
+{
+	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
+	Buffer		buf = so->buf;
+	Page		page;
+	OffsetNumber maxoffno;
+
+	/* Safety check */
+	if (!BufferIsValid(so->buf) || !ItemPointerIsValid(&so->heaptid))
+		return;
+
+	/* Only a shared locked is needed for ItemIdMarkDead */
+	LockBuffer(buf, BUFFER_LOCK_SHARE);
+	page = BufferGetPage(buf);
+	maxoffno = PageGetMaxOffsetNumber(page);
+
+	for (OffsetNumber offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
+	{
+		ItemId		itemid = PageGetItemId(page, offno);
+		IndexTuple	itup = (IndexTuple) PageGetItem(page, itemid);
+
+		/*
+		 * Find tuple. Since buffer has been pinned, tuple cannot have been
+		 * vacuumed (and heap TID reused).
+		 */
+		if (ItemPointerEquals(&itup->t_tid, &so->heaptid))
+		{
+			/*
+			 * Make sure tuple has not already been marked dead to avoid extra
+			 * WAL if wal_log_hints or data checksums enabled
+			 */
+			if (!ItemIdIsDead(itemid))
+			{
+				ItemIdMarkDead(itemid);
+				MarkBufferDirtyHint(buf, true);
+			}
+
+			break;
+		}
+	}
+
+	/* Unlock buffer */
+	LockBuffer(buf, BUFFER_LOCK_UNLOCK);
+}
+
 /*
  * Prepare for an index scan
  */
@@ -206,7 +262,10 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 		probes = lists;
 
 	so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + probes * sizeof(IvfflatScanList));
+	so->buf = InvalidBuffer;
 	so->first = true;
+	so->hasLock = false;
+	ItemPointerSetInvalid(&so->heaptid);
 	so->probes = probes;
 	so->dimensions = dimensions;
 
@@ -217,12 +276,13 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 
 	/* Create tuple description for sorting */
 #if PG_VERSION_NUM >= 120000
-	so->tupdesc = CreateTemplateTupleDesc(2);
+	so->tupdesc = CreateTemplateTupleDesc(3);
 #else
-	so->tupdesc = CreateTemplateTupleDesc(2, false);
+	so->tupdesc = CreateTemplateTupleDesc(3, false);
 #endif
 	TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
 	TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "heaptid", TIDOID, -1, 0);
+	TupleDescInitEntry(so->tupdesc, (AttrNumber) 3, "indexblkno", INT4OID, -1, 0);
 
 	/* Prep sort */
 	so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
@@ -254,6 +314,7 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
 #endif
 
 	so->first = true;
+	ItemPointerSetInvalid(&so->heaptid);
 	pairingheap_reset(so->listQueue);
 
 	if (keys && scan->numberOfKeys > 0)
@@ -288,10 +349,12 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		if (scan->orderByData == NULL)
 			elog(ERROR, "cannot scan ivfflat index without order");
 
-		/* Requires MVCC-compliant snapshot as not able to pin during sorting */
-		/* https://www.postgresql.org/docs/current/index-locking.html */
-		if (!IsMVCCSnapshot(scan->xs_snapshot))
-			elog(ERROR, "non-MVCC snapshots are not supported with ivfflat");
+		/* Get a shared lock for non-MVCC snapshots */
+		if (!so->hasLock && !IsMVCCSnapshot(scan->xs_snapshot))
+		{
+			so->hasLock = true;
+			LockPage(scan->indexRelation, IVFFLAT_SCAN_LOCK, ShareLock);
+		}
 
 		if (scan->orderByData->sk_flags & SK_ISNULL)
 			value = PointerGetDatum(InitVector(so->dimensions));
@@ -316,10 +379,17 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		if (value != scan->orderByData->sk_argument)
 			pfree(DatumGetPointer(value));
 	}
+	else
+	{
+		/* Mark prior tuple as dead */
+		if (scan->kill_prior_tuple)
+			MarkPriorTupleDead(scan);
+	}
 
 	if (tuplesort_gettupleslot(so->sortstate, true, false, so->slot, NULL))
 	{
 		ItemPointer heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->slot, 2, &so->isnull));
+		BlockNumber indexblkno = DatumGetInt32(slot_getattr(so->slot, 3, &so->isnull));
 
 #if PG_VERSION_NUM >= 120000
 		scan->xs_heaptid = *heaptid;
@@ -327,6 +397,21 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		scan->xs_ctup.t_self = *heaptid;
 #endif
 
+		/* Keep track of info needed to mark tuple as dead */
+		so->heaptid = *heaptid;
+
+		/* Unpin buffer */
+		if (BufferIsValid(so->buf))
+			ReleaseBuffer(so->buf);
+
+		/*
+		 * An index scan must maintain a pin on the index page holding the
+		 * item last returned by amgettuple
+		 *
+		 * https://www.postgresql.org/docs/current/index-locking.html
+		 */
+		so->buf = ReadBuffer(scan->indexRelation, indexblkno);
+
 		scan->xs_recheckorderby = false;
 		return true;
 	}
@@ -342,6 +427,14 @@ ivfflatendscan(IndexScanDesc scan)
 {
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
 
+	/* Release pin */
+	if (BufferIsValid(so->buf))
+		ReleaseBuffer(so->buf);
+
+	/* Release lock */
+	if (so->hasLock)
+		UnlockPage(scan->indexRelation, IVFFLAT_SCAN_LOCK, ShareLock);
+
 	pairingheap_free(so->listQueue);
 	tuplesort_end(so->sortstate);
 

src/ivfutils.c

@@ -57,12 +57,12 @@ IvfflatGetLists(Relation index)
  * Get proc
  */
 FmgrInfo *
-IvfflatOptionalProcInfo(Relation index, uint16 procnum)
+IvfflatOptionalProcInfo(Relation rel, uint16 procnum)
 {
-	if (!OidIsValid(index_getprocid(index, 1, procnum)))
+	if (!OidIsValid(index_getprocid(rel, 1, procnum)))
 		return NULL;
 
-	return index_getprocinfo(index, 1, procnum);
+	return index_getprocinfo(rel, 1, procnum);
 }
 
 /*
@@ -136,6 +136,7 @@ IvfflatInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogStat
 void
 IvfflatCommitBuffer(Buffer buf, GenericXLogState *state)
 {
+	MarkBufferDirty(buf);
 	GenericXLogFinish(state);
 	UnlockReleaseBuffer(buf);
 }
@@ -159,6 +160,8 @@ IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **st
 	IvfflatInitPage(newbuf, newpage);
 
 	/* Commit */
+	MarkBufferDirty(*buf);
+	MarkBufferDirty(newbuf);
 	GenericXLogFinish(*state);
 
 	/* Unlock */

src/ivfvacuum.c

@@ -3,6 +3,7 @@
 #include "commands/vacuum.h"
 #include "ivfflat.h"
 #include "storage/bufmgr.h"
+#include "storage/lmgr.h"
 
 /*
  * Bulk delete tuples from the index
@@ -65,14 +66,10 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 
 				vacuum_delay_point();
 
-				buf = ReadBufferExtended(index, MAIN_FORKNUM, searchPage, RBM_NORMAL, bas);
+				/* Ensure no in-flight index scans for non-MVCC snapshots */
+				LockPage(index, IVFFLAT_SCAN_LOCK, ExclusiveLock);
 
-				/*
-				 * ambulkdelete cannot delete entries from pages that are
-				 * pinned by other backends
-				 *
-				 * https://www.postgresql.org/docs/current/index-locking.html
-				 */
+				buf = ReadBufferExtended(index, MAIN_FORKNUM, searchPage, RBM_NORMAL, bas);
 				LockBufferForCleanup(buf);
 
 				state = GenericXLogStart(index);
@@ -107,12 +104,15 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 				{
 					/* Delete tuples */
 					PageIndexMultiDelete(page, deletable, ndeletable);
+					MarkBufferDirty(buf);
 					GenericXLogFinish(state);
 				}
 				else
 					GenericXLogAbort(state);
 
 				UnlockReleaseBuffer(buf);
+
+				UnlockPage(index, IVFFLAT_SCAN_LOCK, ExclusiveLock);
 			}
 
 			/*

src/vector.c

@@ -89,7 +89,7 @@ CheckDim(int dim)
 }
 
 /*
- * Ensure finite element
+ * Ensure finite elements
  */
 static inline void
 CheckElement(float value)
@@ -177,15 +177,14 @@ PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_in);
 Datum
 vector_in(PG_FUNCTION_ARGS)
 {
-	char	   *lit = PG_GETARG_CSTRING(0);
+	char	   *str = PG_GETARG_CSTRING(0);
 	int32		typmod = PG_GETARG_INT32(2);
 	float		x[VECTOR_MAX_DIM];
 	int			dim = 0;
 	char	   *pt;
 	char	   *stringEnd;
 	Vector	   *result;
-	char	   *litcopy = pstrdup(lit);
-	char	   *str = litcopy;
+	char	   *lit = pstrdup(str);
 
 	while (vector_isspace(*str))
 		str++;
@@ -269,7 +268,7 @@ vector_in(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("vector must have at least 1 dimension")));
 
-	pfree(litcopy);
+	pfree(lit);
 
 	CheckExpectedDim(typmod, dim);
 
@@ -438,18 +437,17 @@ vector_send(PG_FUNCTION_ARGS)
 
 /*
  * Convert vector to vector
- * This is needed to check the type modifier
  */
 PGDLLEXPORT PG_FUNCTION_INFO_V1(vector);
 Datum
 vector(PG_FUNCTION_ARGS)
 {
-	Vector	   *vec = PG_GETARG_VECTOR_P(0);
+	Vector	   *arg = PG_GETARG_VECTOR_P(0);
 	int32		typmod = PG_GETARG_INT32(1);
 
-	CheckExpectedDim(typmod, vec->dim);
+	CheckExpectedDim(typmod, arg->dim);
 
-	PG_RETURN_POINTER(vec);
+	PG_RETURN_POINTER(arg);
 }
 
 /*
@@ -466,6 +464,7 @@ array_to_vector(PG_FUNCTION_ARGS)
 	bool		typbyval;
 	char		typalign;
 	Datum	   *elemsp;
+	bool	   *nullsp;
 	int			nelemsp;
 
 	if (ARR_NDIM(array) > 1)
@@ -479,7 +478,7 @@ array_to_vector(PG_FUNCTION_ARGS)
 				 errmsg("array must not contain nulls")));
 
 	get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
-	deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, NULL, &nelemsp);
+	deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, &nullsp, &nelemsp);
 
 	CheckDim(nelemsp);
 	CheckExpectedDim(typmod, nelemsp);
@@ -513,12 +512,6 @@ array_to_vector(PG_FUNCTION_ARGS)
 				 errmsg("unsupported array type")));
 	}
 
-	/*
-	 * Free allocation from deconstruct_array. Do not free individual elements
-	 * when pass-by-reference since they point to original array.
-	 */
-	pfree(elemsp);
-
 	/* Check elements */
 	for (int i = 0; i < result->dim; i++)
 		CheckElement(result->x[i]);
@@ -725,7 +718,7 @@ vector_spherical_distance(PG_FUNCTION_ARGS)
 }
 
 /*
- * Get the L1 distance between two vectors
+ * Get the L1 distance between vectors
  */
 PGDLLEXPORT PG_FUNCTION_INFO_V1(l1_distance);
 Datum
@@ -904,8 +897,8 @@ PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_lt);
 Datum
 vector_lt(PG_FUNCTION_ARGS)
 {
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
+	Vector	   *a = (Vector *) PG_GETARG_VECTOR_P(0);
+	Vector	   *b = (Vector *) PG_GETARG_VECTOR_P(1);
 
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) < 0);
 }
@@ -917,8 +910,8 @@ PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_le);
 Datum
 vector_le(PG_FUNCTION_ARGS)
 {
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
+	Vector	   *a = (Vector *) PG_GETARG_VECTOR_P(0);
+	Vector	   *b = (Vector *) PG_GETARG_VECTOR_P(1);
 
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) <= 0);
 }
@@ -930,8 +923,8 @@ PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_eq);
 Datum
 vector_eq(PG_FUNCTION_ARGS)
 {
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
+	Vector	   *a = (Vector *) PG_GETARG_VECTOR_P(0);
+	Vector	   *b = (Vector *) PG_GETARG_VECTOR_P(1);
 
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) == 0);
 }
@@ -943,8 +936,8 @@ PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_ne);
 Datum
 vector_ne(PG_FUNCTION_ARGS)
 {
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
+	Vector	   *a = (Vector *) PG_GETARG_VECTOR_P(0);
+	Vector	   *b = (Vector *) PG_GETARG_VECTOR_P(1);
 
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) != 0);
 }
@@ -956,8 +949,8 @@ PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_ge);
 Datum
 vector_ge(PG_FUNCTION_ARGS)
 {
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
+	Vector	   *a = (Vector *) PG_GETARG_VECTOR_P(0);
+	Vector	   *b = (Vector *) PG_GETARG_VECTOR_P(1);
 
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) >= 0);
 }
@@ -969,8 +962,8 @@ PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_gt);
 Datum
 vector_gt(PG_FUNCTION_ARGS)
 {
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
+	Vector	   *a = (Vector *) PG_GETARG_VECTOR_P(0);
+	Vector	   *b = (Vector *) PG_GETARG_VECTOR_P(1);
 
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) > 0);
 }
@@ -982,8 +975,8 @@ PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_cmp);
 Datum
 vector_cmp(PG_FUNCTION_ARGS)
 {
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
+	Vector	   *a = (Vector *) PG_GETARG_VECTOR_P(0);
+	Vector	   *b = (Vector *) PG_GETARG_VECTOR_P(1);
 
 	PG_RETURN_INT32(vector_cmp_internal(a, b));
 }

test/expected/hnsw_cosine.out

@@ -3,7 +3,7 @@ CREATE TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING hnsw (val vector_cosine_ops);
 INSERT INTO t (val) VALUES ('[1,2,4]');
-SELECT * FROM t ORDER BY val <=> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <=> '[3,3,3]';
    val   
 ---------
  [1,1,1]
@@ -11,13 +11,13 @@ SELECT * FROM t ORDER BY val <=> '[3,3,3]' LIMIT 5;
  [1,2,4]
 (3 rows)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]' LIMIT 5) t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
  count 
 -------
      3
 (1 row)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector) LIMIT 5) t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
  count 
 -------
      3

test/expected/hnsw_ip.out

@@ -3,7 +3,7 @@ CREATE TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING hnsw (val vector_ip_ops);
 INSERT INTO t (val) VALUES ('[1,2,4]');
-SELECT * FROM t ORDER BY val <#> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <#> '[3,3,3]';
    val   
 ---------
  [1,2,4]
@@ -12,7 +12,7 @@ SELECT * FROM t ORDER BY val <#> '[3,3,3]' LIMIT 5;
  [0,0,0]
 (4 rows)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector) LIMIT 5) t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector)) t2;
  count 
 -------
      4

test/expected/hnsw_l2.out

@@ -3,7 +3,7 @@ CREATE TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING hnsw (val vector_l2_ops);
 INSERT INTO t (val) VALUES ('[1,2,4]');
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
    val   
 ---------
  [1,2,3]
@@ -12,7 +12,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
  [0,0,0]
 (4 rows)
 
-SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector) LIMIT 5;
+SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
    val   
 ---------
  [0,0,0]
@@ -28,7 +28,7 @@ SELECT COUNT(*) FROM t;
 (1 row)
 
 TRUNCATE t;
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
  val 
 -----
 (0 rows)

test/expected/hnsw_unlogged.out

@@ -2,7 +2,7 @@ SET enable_seqscan = off;
 CREATE UNLOGGED TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING hnsw (val vector_l2_ops);
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
    val   
 ---------
  [1,2,3]

test/expected/ivfflat_cosine.out

@@ -3,7 +3,7 @@ CREATE TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING ivfflat (val vector_cosine_ops) WITH (lists = 1);
 INSERT INTO t (val) VALUES ('[1,2,4]');
-SELECT * FROM t ORDER BY val <=> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <=> '[3,3,3]';
    val   
 ---------
  [1,1,1]
@@ -11,13 +11,13 @@ SELECT * FROM t ORDER BY val <=> '[3,3,3]' LIMIT 5;
  [1,2,4]
 (3 rows)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]' LIMIT 5) t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
  count 
 -------
      3
 (1 row)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector) LIMIT 5) t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
  count 
 -------
      3

test/expected/ivfflat_ip.out

@@ -3,7 +3,7 @@ CREATE TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING ivfflat (val vector_ip_ops) WITH (lists = 1);
 INSERT INTO t (val) VALUES ('[1,2,4]');
-SELECT * FROM t ORDER BY val <#> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <#> '[3,3,3]';
    val   
 ---------
  [1,2,4]
@@ -12,7 +12,7 @@ SELECT * FROM t ORDER BY val <#> '[3,3,3]' LIMIT 5;
  [0,0,0]
 (4 rows)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector) LIMIT 5) t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector)) t2;
  count 
 -------
      4

test/expected/ivfflat_l2.out

@@ -3,7 +3,7 @@ CREATE TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
 INSERT INTO t (val) VALUES ('[1,2,4]');
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
    val   
 ---------
  [1,2,3]
@@ -12,7 +12,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
  [0,0,0]
 (4 rows)
 
-SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector) LIMIT 5;
+SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
    val   
 ---------
  [0,0,0]
@@ -31,7 +31,7 @@ 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]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
  val 
 -----
 (0 rows)

test/expected/ivfflat_unlogged.out

@@ -2,7 +2,7 @@ SET enable_seqscan = off;
 CREATE UNLOGGED TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
    val   
 ---------
  [1,2,3]

test/sql/hnsw_cosine.sql

@@ -6,8 +6,8 @@ CREATE INDEX ON t USING hnsw (val vector_cosine_ops);
 
 INSERT INTO t (val) VALUES ('[1,2,4]');
 
-SELECT * FROM t ORDER BY val <=> '[3,3,3]' LIMIT 5;
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]' LIMIT 5) t2;
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector) LIMIT 5) t2;
+SELECT * FROM t ORDER BY val <=> '[3,3,3]';
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
 
 DROP TABLE t;

test/sql/hnsw_ip.sql

@@ -6,7 +6,7 @@ CREATE INDEX ON t USING hnsw (val vector_ip_ops);
 
 INSERT INTO t (val) VALUES ('[1,2,4]');
 
-SELECT * FROM t ORDER BY val <#> '[3,3,3]' LIMIT 5;
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector) LIMIT 5) t2;
+SELECT * FROM t ORDER BY val <#> '[3,3,3]';
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector)) t2;
 
 DROP TABLE t;

test/sql/hnsw_l2.sql

@@ -6,11 +6,11 @@ CREATE INDEX ON t USING hnsw (val vector_l2_ops);
 
 INSERT INTO t (val) VALUES ('[1,2,4]');
 
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
-SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector) LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
+SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
 SELECT COUNT(*) FROM t;
 
 TRUNCATE t;
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 
 DROP TABLE t;

test/sql/hnsw_unlogged.sql

@@ -4,6 +4,6 @@ CREATE UNLOGGED TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING hnsw (val vector_l2_ops);
 
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 
 DROP TABLE t;

test/sql/ivfflat_cosine.sql

@@ -6,8 +6,8 @@ CREATE INDEX ON t USING ivfflat (val vector_cosine_ops) WITH (lists = 1);
 
 INSERT INTO t (val) VALUES ('[1,2,4]');
 
-SELECT * FROM t ORDER BY val <=> '[3,3,3]' LIMIT 5;
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]' LIMIT 5) t2;
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector) LIMIT 5) t2;
+SELECT * FROM t ORDER BY val <=> '[3,3,3]';
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
 
 DROP TABLE t;

test/sql/ivfflat_ip.sql

@@ -6,7 +6,7 @@ CREATE INDEX ON t USING ivfflat (val vector_ip_ops) WITH (lists = 1);
 
 INSERT INTO t (val) VALUES ('[1,2,4]');
 
-SELECT * FROM t ORDER BY val <#> '[3,3,3]' LIMIT 5;
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector) LIMIT 5) t2;
+SELECT * FROM t ORDER BY val <#> '[3,3,3]';
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector)) t2;
 
 DROP TABLE t;

test/sql/ivfflat_l2.sql

@@ -6,11 +6,11 @@ CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
 
 INSERT INTO t (val) VALUES ('[1,2,4]');
 
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
-SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector) LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
+SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
 SELECT COUNT(*) FROM t;
 
 TRUNCATE t;
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 
 DROP TABLE t;

test/sql/ivfflat_unlogged.sql

@@ -4,6 +4,6 @@ CREATE UNLOGGED TABLE t (val vector(3));
 INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
 CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
 
-SELECT * FROM t ORDER BY val <-> '[3,3,3]' LIMIT 5;
+SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 
 DROP TABLE t;

test/t/007_ivfflat_inserts.pl

@@ -49,7 +49,7 @@ is(idx_scan(), 0);
 $count = $node->safe_psql("postgres", qq(
 	SET enable_seqscan = off;
 	SET ivfflat.probes = 100;
-	SELECT COUNT(*) FROM (SELECT v FROM tst ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 20000) t;
+	SELECT COUNT(*) FROM (SELECT v FROM tst ORDER BY v <-> (SELECT v FROM tst LIMIT 1)) t;
 ));
 is($count, $expected);
 is(idx_scan(), 1);

test/t/012_hnsw_build_recall.pl

@@ -83,32 +83,11 @@ for my $i (0 .. $#operators)
 		push(@expected, $res);
 	}
 
-	# Build index serially
-	$node->safe_psql("postgres", qq(
-		SET max_parallel_maintenance_workers = 0;
-		CREATE INDEX idx ON tst USING hnsw (v $opclass);
-	));
+	# Add index
+	$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v $opclass);");
 
-	# Test approximate results
 	my $min = $operator eq "<#>" ? 0.80 : 0.99;
 	test_recall($min, $operator);
-
-	$node->safe_psql("postgres", "DROP INDEX idx;");
-
-	# Build index in parallel
-	my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
-		SET client_min_messages = DEBUG;
-		SET min_parallel_table_scan_size = 1;
-		SET hnsw.enable_parallel_build = on;
-		CREATE INDEX idx ON tst USING hnsw (v $opclass);
-	));
-	is($ret, 0, $stderr);
-	like($stderr, qr/using \d+ parallel workers/);
-
-	# Test approximate results
-	test_recall($min, $operator);
-
-	$node->safe_psql("postgres", "DROP INDEX idx;");
 }
 
 done_testing();

test/t/014_hnsw_inserts.pl

@@ -42,7 +42,7 @@ for my $i (1 .. 20)
 
 	my $count = $node->safe_psql("postgres", qq(
 		SET enable_seqscan = off;
-		SELECT COUNT(*) FROM (SELECT v FROM tst ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 20) t;
+		SELECT COUNT(*) FROM (SELECT v FROM tst ORDER BY v <-> (SELECT v FROM tst LIMIT 1)) t;
 	));
 	is($count, 10);
 
@@ -63,7 +63,7 @@ $node->pgbench(
 my $count = $node->safe_psql("postgres", qq(
 	SET enable_seqscan = off;
 	SET hnsw.ef_search = 1000;
-	SELECT COUNT(*) FROM (SELECT v FROM tst ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 1000) t;
+	SELECT COUNT(*) FROM (SELECT v FROM tst ORDER BY v <-> (SELECT v FROM tst LIMIT 1)) t;
 ));
 # Elements may lose all incoming connections with the HNSW algorithm
 # Vacuuming can fix this if one of the elements neighbors is deleted

test/t/015_hnsw_duplicates.pl

@@ -26,7 +26,7 @@ sub test_duplicates
 	my $res = $node->safe_psql("postgres", qq(
 		SET enable_seqscan = off;
 		SET hnsw.ef_search = 1;
-		SELECT COUNT(*) FROM (SELECT * FROM tst ORDER BY v <-> '[1,1,1]' LIMIT 20) t;
+		SELECT COUNT(*) FROM (SELECT * FROM tst ORDER BY v <-> '[1,1,1]') t;
 	));
 	is($res, 10);
 }

test/t/018_hnsw_filtering.pl

@@ -1,111 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-my $dim = 3;
-my $nc = 50;
-my $limit = 20;
-
-my $array_sql = join(",", ('random()') x $dim);
-
-# Initialize node
-my $node = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table and index
-$node->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int4, t text);");
-$node->safe_psql("postgres",
-	"INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc, 'test ' || i FROM generate_series(1, 10000) i;"
-);
-$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
-$node->safe_psql("postgres", "ANALYZE tst;");
-
-# Generate query
-my @r = ();
-for (1 .. $dim)
-{
-	push(@r, rand());
-}
-my $query = "[" . join(",", @r) . "]";
-my $c = int(rand() * $nc);
-
-# Test attribute filtering
-my $explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Seq Scan/);
-
-# Test attribute filtering with few rows removed
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c != $c ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using idx/);
-
-# Test attribute filtering with few rows removed comparison
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c >= 1 ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using idx/);
-
-# Test attribute filtering with many rows removed comparison
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c < 1 ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Seq Scan/);
-
-# Test attribute filtering with few rows removed like
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE t LIKE '%%test%%' ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using idx/);
-
-# Test attribute filtering with many rows removed like
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE t LIKE '%%other%%' ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Seq Scan/);
-
-# Test distance filtering
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using idx/);
-
-# Test distance filtering greater than distance
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' > 1 ORDER BY v <-> '$query' LIMIT $limit;
-));
-# TODO Do not use index
-like($explain, qr/Index Scan using idx/);
-
-# Test distance filtering without order
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1;
-));
-like($explain, qr/Seq Scan/);
-
-# Test distance filtering without limit
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query';
-));
-like($explain, qr/Seq Scan/);
-
-# Test attribute index
-$node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);");
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Bitmap Index Scan on attribute_idx/);
-
-# Test partial index
-$node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING hnsw (v vector_l2_ops) WHERE (c = $c);");
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using partial_idx/);
-
-done_testing();

test/t/018_ivfflat_deletes.pl

@@ -0,0 +1,43 @@
+use strict;
+use warnings;
+use PostgresNode;
+use TestLib;
+use Test::More;
+
+my $dim = 3;
+
+my $array_sql = join(",", ('random()') x $dim);
+
+# Initialize node
+my $node = get_new_node('node');
+$node->init;
+$node->start;
+
+# Create table and index
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i serial, v vector($dim));");
+$node->safe_psql("postgres",
+	"INSERT INTO tst (v) SELECT ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
+);
+$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v vector_l2_ops);");
+
+# Delete data
+$node->safe_psql("postgres", "DELETE FROM tst WHERE i % 100 != 0;");
+
+my $exp = $node->safe_psql("postgres", qq(
+	SET enable_indexscan = off;
+	SELECT i FROM tst ORDER BY v <-> '[0,0,0]';
+));
+
+# Run twice to make sure correct tuples marked as dead
+for (1 .. 2)
+{
+	my $res = $node->safe_psql("postgres", qq(
+		SET enable_seqscan = off;
+		SET ivfflat.probes = 100;
+		SELECT i FROM tst ORDER BY v <-> '[0,0,0]';
+	));
+	is($res, $exp);
+}
+
+done_testing();

test/t/019_ivfflat_filtering.pl

@@ -1,99 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-my $dim = 3;
-my $nc = 50;
-my $limit = 20;
-
-my $array_sql = join(",", ('random()') x $dim);
-
-# Initialize node
-my $node = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table and index
-$node->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int4);");
-$node->safe_psql("postgres",
-	"INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc FROM generate_series(1, 10000) i;"
-);
-$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING ivfflat (v vector_l2_ops) WITH (lists = 100);");
-$node->safe_psql("postgres", "ANALYZE tst;");
-
-# Generate query
-my @r = ();
-for (1 .. $dim)
-{
-	push(@r, rand());
-}
-my $query = "[" . join(",", @r) . "]";
-my $c = int(rand() * $nc);
-
-# Test attribute filtering
-my $explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Seq Scan/);
-
-# Test attribute filtering with few rows removed
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c != $c ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using idx/);
-
-# Test attribute filtering with few rows removed comparison
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c >= 1 ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using idx/);
-
-# Test attribute filtering with many rows removed comparison
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c < 1 ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Seq Scan/);
-
-# Test distance filtering
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using idx/);
-
-# Test distance filtering greater than distance
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' > 1 ORDER BY v <-> '$query' LIMIT $limit;
-));
-# TODO Do not use index
-like($explain, qr/Index Scan using idx/);
-
-# Test distance filtering without order
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1;
-));
-like($explain, qr/Seq Scan/);
-
-# Test distance filtering without limit
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query';
-));
-like($explain, qr/Seq Scan/);
-
-# Test attribute index
-$node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);");
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Bitmap Index Scan on attribute_idx/);
-
-# Test partial index
-$node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING ivfflat (v vector_l2_ops) WITH (lists = 5) WHERE (c = $c);");
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
-));
-like($explain, qr/Index Scan using partial_idx/);
-
-done_testing();

test/t/020_ivfflat_limit.pl

@@ -1,64 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-# Initialize node
-my $node = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table and index
-$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;"
-);
-$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v vector_l2_ops) WITH (lists = 10);");
-
-# Test limit
-my $explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 100;
-));
-like($explain, qr/Index Scan/);
-
-# Test limit with probes
-$explain = $node->safe_psql("postgres", qq(
-	SET ivfflat.probes = 2;
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 200;
-));
-like($explain, qr/Index Scan/);
-
-# Test limit + offset
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 90 OFFSET 10;
-));
-like($explain, qr/Index Scan/);
-
-# Test limit > expected tuples
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 101;
-));
-like($explain, qr/Seq Scan/);
-
-# Test limit > expected tuples with probes
-$explain = $node->safe_psql("postgres", qq(
-	SET ivfflat.probes = 2;
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 201;
-));
-like($explain, qr/Seq Scan/);
-
-# Test limit + offset > expected tuples
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 91 OFFSET 10;
-));
-like($explain, qr/Seq Scan/);
-
-# Test no limit
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]';
-));
-like($explain, qr/Seq Scan/);
-
-done_testing();

test/t/021_hnsw_limit.pl

@@ -1,62 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-# Initialize node
-my $node = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table and index
-$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;"
-);
-$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops);");
-
-# Test limit
-my $explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 40;
-));
-like($explain, qr/Index Scan/);
-
-# Test limit with CTE
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE WITH cte AS (SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 40) SELECT * FROM cte;
-));
-like($explain, qr/Index Scan/);
-
-# Test limit + offset
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 30 OFFSET 10;
-));
-like($explain, qr/Index Scan/);
-
-# Test limit > ef_search
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 41;
-));
-like($explain, qr/Seq Scan/);
-
-# Test limit > ef_search with CTE
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE WITH cte AS (SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 41) SELECT * FROM cte;
-));
-like($explain, qr/Seq Scan/);
-
-# Test limit + offset > ef_search
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]' LIMIT 31 OFFSET 10;
-));
-like($explain, qr/Seq Scan/);
-
-# Test no limit
-$explain = $node->safe_psql("postgres", qq(
-	EXPLAIN ANALYZE SELECT * FROM tst ORDER BY v <-> '[1,2,3]';
-));
-like($explain, qr/Seq Scan/);
-
-done_testing();

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.5.1'
+default_version = '0.5.0'
 module_pathname = '$libdir/vector'
 relocatable = true