Test BufferAccessStrategy [skip ci]

.github/workflows/build.yml

@@ -17,7 +17,7 @@ jobs:
           - postgres: 12
             os: ubuntu-20.04
           - postgres: 11
-            os: ubuntu-18.04
+            os: ubuntu-20.04
     steps:
       - uses: actions/checkout@v3
       - uses: ankane/setup-postgres@v1

CHANGELOG.md

@@ -1,3 +1,7 @@
+## 0.4.2 (unreleased)
+
+- Added notice when index created with little data
+
 ## 0.4.1 (2023-03-21)
 
 - Improved performance of cosine distance

Dockerfile

@@ -1,9 +1,11 @@
-FROM postgres:15
+ARG PG_MAJOR=15
+FROM postgres:$PG_MAJOR
+ARG PG_MAJOR
 
 COPY . /tmp/pgvector
 
 RUN apt-get update && \
-		apt-get install -y --no-install-recommends build-essential postgresql-server-dev-15 && \
+		apt-get install -y --no-install-recommends build-essential postgresql-server-dev-$PG_MAJOR && \
 		cd /tmp/pgvector && \
 		make clean && \
 		make OPTFLAGS="" && \
@@ -11,6 +13,6 @@ RUN apt-get update && \
 		mkdir /usr/share/doc/pgvector && \
 		cp LICENSE README.md /usr/share/doc/pgvector && \
 		rm -r /tmp/pgvector && \
-		apt-get remove -y build-essential postgresql-server-dev-15 && \
+		apt-get remove -y build-essential postgresql-server-dev-$PG_MAJOR && \
 		apt-get autoremove -y && \
 		rm -rf /var/lib/apt/lists/*

Makefile

@@ -14,6 +14,7 @@ OPTFLAGS = -march=native
 # Mac ARM doesn't support -march=native
 ifeq ($(shell uname -s), Darwin)
 	ifeq ($(shell uname -p), arm)
+		# no difference with -march=armv8.5-a
 		OPTFLAGS =
 	endif
 endif

README.md

@@ -2,13 +2,7 @@
 
 Open-source vector similarity search for Postgres
 
-```sql
-CREATE TABLE items (embedding vector(3));
-CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops);
-SELECT * FROM items ORDER BY embedding <-> '[1,2,3]' LIMIT 5;
-```
-
-Supports L2 distance, inner product, and cosine distance
+Supports exact and approximate nearest neighbor search for L2 distance, inner product, and cosine distance
 
 [![Build Status](https://github.com/pgvector/pgvector/workflows/build/badge.svg?branch=master)](https://github.com/pgvector/pgvector/actions)
 
@@ -17,6 +11,7 @@ Supports L2 distance, inner product, and cosine distance
 Compile and install the extension (supports Postgres 11+)
 
 ```sh
+cd /tmp
 git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
@@ -29,41 +24,88 @@ Then load it in databases where you want to use it
 CREATE EXTENSION vector;
 ```
 
-You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), or [conda-forge](#conda-forge)
+See the [installation notes](#installation-notes) if you run into issues
+
+You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [Yum](#yum), or [conda-forge](#conda-forge)
 
 ## Getting Started
 
 Create a vector column with 3 dimensions
 
 ```sql
-CREATE TABLE items (embedding vector(3));
+CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
 ```
 
 Insert values
 
 ```sql
-INSERT INTO items VALUES ('[1,2,3]'), ('[4,5,6]');
+INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
 ```
 
-Get the nearest neighbor by L2 distance
+Get the nearest neighbors by L2 distance
 
 ```sql
-SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 1;
+SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
 Also supports inner product (`<#>`) and cosine distance (`<=>`)
 
 Note: `<#>` returns the negative inner product since Postgres only supports `ASC` order index scans on operators
 
-## Querying
+## Storing
 
-Use a `SELECT` clause to get the distance
+Create a new table with a vector column
 
 ```sql
-SELECT embedding <-> '[3,1,2]' AS distance FROM items;
+CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
 ```
 
-Use a `WHERE` clause to get rows within a certain distance
+Or add a vector column to an existing table
+
+```sql
+ALTER TABLE items ADD COLUMN embedding vector(3);
+```
+
+Insert vectors
+
+```sql
+INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
+```
+
+Upsert vectors
+
+```sql
+INSERT INTO items (id, embedding) VALUES (1, '[1,2,3]'), (2, '[4,5,6]')
+    ON CONFLICT (id) DO UPDATE SET embedding = EXCLUDED.embedding;
+```
+
+Update vectors
+
+```sql
+UPDATE items SET embedding = '[1,2,3]' WHERE id = 1;
+```
+
+Delete vectors
+
+```sql
+DELETE FROM items WHERE id = 1;
+```
+
+## Querying
+
+Get the nearest neighbors to a vector
+
+```sql
+SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
+```
+
+Get the nearest neighbors to a row
+
+```sql
+SELECT * FROM items WHERE id != 1 ORDER BY embedding <-> (SELECT embedding FROM items WHERE id = 1) LIMIT 5;
+```
+
+Get rows within a certain distance
 
 ```sql
 SELECT * FROM items WHERE embedding <-> '[3,1,2]' < 5;
@@ -71,61 +113,89 @@ SELECT * FROM items WHERE embedding <-> '[3,1,2]' < 5;
 
 Note: Combine with `ORDER BY` and `LIMIT` to use an index
 
-Get the average of vectors
+#### Distances
+
+Get the distance
+
+```sql
+SELECT embedding <-> '[3,1,2]' AS distance FROM items;
+```
+
+For inner product, multiply by -1 (since `<#>` returns the negative inner product)
+
+```sql
+SELECT -1 * (embedding <#> '[3,1,2]') AS inner_product FROM items;
+```
+
+For cosine similarity, use 1 - cosine distance
+
+```sql
+SELECT 1 - (embedding <=> '[3,1,2]') AS cosine_similarity FROM items;
+```
+
+#### Averaging
+
+Average vectors
 
 ```sql
 SELECT AVG(embedding) FROM items;
 ```
 
+Average groups of vectors
+
+```sql
+SELECT category_id, AVG(embedding) FROM items GROUP BY category_id;
+```
+
 ## Indexing
 
-Speed up queries with an approximate index. Add an index for each distance function you want to use.
+By default, pgvector performs exact nearest neighbor search, which provides perfect recall.
+
+You can add an index to use approximate nearest neighbor search, which trades some recall for performance. Unlike typical indexes, you will see different results for queries after adding an approximate index.
+
+Three keys to achieving good recall are:
+
+1. Create the index *after* the table has some data
+2. Choose an appropriate number of lists (a good place to start is `rows / 1000` for up to 1M rows and `sqrt(rows)` for over 1M rows)
+3. When querying, specify an appropriate number of [probes](#query-options) (higher is better for recall, lower is better for speed)
+
+Add an index for each distance function you want to use.
 
 L2 distance
 
-```sql
-CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops);
-```
-
-Inner product
-
-```sql
-CREATE INDEX ON items USING ivfflat (embedding vector_ip_ops);
-```
-
-Cosine distance
-
-```sql
-CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops);
-```
-
-Indexes should be created after the table has some data for optimal clustering. Also, unlike typical indexes which only affect performance, you may see different results for queries after adding an approximate index. Vectors with up to 2,000 dimensions can be indexed.
-
-### Index Options
-
-Specify the number of inverted lists (100 by default)
-
 ```sql
 CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100);
 ```
 
-A [good place to start](https://github.com/facebookresearch/faiss/issues/112) is `4 * sqrt(rows)`
+Inner product
+
+```sql
+CREATE INDEX ON items USING ivfflat (embedding vector_ip_ops) WITH (lists = 100);
+```
+
+Cosine distance
+
+```sql
+CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);
+```
+
+Vectors with up to 2,000 dimensions can be indexed.
 
 ### Query Options
 
 Specify the number of probes (1 by default)
 
 ```sql
-SET ivfflat.probes = 1;
+SET ivfflat.probes = 10;
 ```
 
-A higher value improves recall at the cost of speed, and it can be set to the number of lists for exact nearest neighbor search (at which point the planner won’t use the index)
+A higher value provides better recall at the cost of speed, and it can be set to the number of lists for exact nearest neighbor search (at which point the planner won’t use the index)
 
 Use `SET LOCAL` inside a transaction to set it for a single query
 
 ```sql
 BEGIN;
-SET LOCAL ivfflat.probes = 1;
+SET LOCAL ivfflat.probes = 10;
 SELECT ...
 COMMIT;
 ```
@@ -158,7 +228,7 @@ SELECT * FROM items WHERE category_id = 123 ORDER BY embedding <-> '[3,1,2]' LIM
 can be indexed with:
 
 ```sql
-CREATE INDEX ON items USING ivfflat (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);
 ```
 
 To index many different values of `category_id`, consider [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) on `category_id`.
@@ -169,24 +239,34 @@ CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(cate
 
 ## Performance
 
+Use `EXPLAIN ANALYZE` to debug performance.
+
+```sql
+EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
+```
+
+### Exact Search
+
 To speed up queries without an index, increase `max_parallel_workers_per_gather`.
 
 ```sql
 SET max_parallel_workers_per_gather = 4;
 ```
 
+If vectors are normalized to length 1 (like [OpenAI embeddings](https://platform.openai.com/docs/guides/embeddings/which-distance-function-should-i-use)), use inner product for best performance.
+
+```sql
+SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
+```
+
+### Approximate Search
+
 To speed up queries with an index, increase the number of inverted lists (at the expense of recall).
 
 ```sql
 CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 1000);
 ```
 
-Use `EXPLAIN ANALYZE` to debug performance.
-
-```sql
-EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 1;
-```
-
 ## Languages
 
 Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
@@ -201,6 +281,7 @@ Java, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
 Julia | [pgvector-julia](https://github.com/pgvector/pgvector-julia)
 Lua | [pgvector-lua](https://github.com/pgvector/pgvector-lua)
 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)
 R | [pgvector-r](https://github.com/pgvector/pgvector-r)
@@ -256,6 +337,42 @@ Function | Description
 --- | ---
 avg(vector) → vector | arithmetic mean
 
+## Installation Notes
+
+### Postgres Location
+
+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=/Applications/Postgres.app/Contents/Versions/latest/bin/pg_config
+```
+
+Then re-run the installation instructions (run `make clean` before `make` if needed)
+
+### 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.
+
+For Ubuntu and Debian, use:
+
+```sh
+sudo apt-get install postgresql-server-dev-15
+```
+
+Note: Replace `15` with your Postgres server version
+
+### Windows
+
+Support for Windows is currently experimental. Use `nmake` to build:
+
+```cmd
+set "PGROOT=C:\Program Files\PostgreSQL\15"
+git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
+cd pgvector
+nmake /F Makefile.win
+nmake /F Makefile.win install
+```
+
 ## Additional Installation Methods
 
 ### Docker
@@ -281,7 +398,7 @@ docker build -t pgvector .
 With Homebrew Postgres, you can use:
 
 ```sh
-brew install pgvector/brew/pgvector
+brew install pgvector
 ```
 
 ### PGXN
@@ -292,6 +409,18 @@ Install from the [PostgreSQL Extension Network](https://pgxn.org/dist/vector) wi
 pgxn install vector
 ```
 
+### 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_15
+# or
+sudo dnf install pgvector_15
+```
+
+Note: Replace `15` with your Postgres server version
+
 ### conda-forge
 
 With Conda Postgres, install from [conda-forge](https://anaconda.org/conda-forge/pgvector) with:
@@ -310,8 +439,9 @@ To request a new extension on other providers:
 
 - Amazon RDS - follow the instructions on [this page](https://aws.amazon.com/rds/postgresql/faqs/)
 - Google Cloud SQL - vote or comment on [this page](https://issuetracker.google.com/issues/265172065)
-- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/app-framework-services/p/pgvector-extension-for-postgresql)
 - Azure Database - vote or comment on [this page](https://feedback.azure.com/d365community/idea/7b423322-6189-ed11-a81b-000d3ae49307)
+- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/app-framework-services/p/pgvector-extension-for-postgresql)
+- Render - vote or comment on [this page](https://feedback.render.com/features/p/add-pgvector-extension-to-postgresql)
 
 ## Upgrading
 

src/ivfbuild.c

@@ -431,8 +431,18 @@ ComputeCenters(IvfflatBuildState * buildstate)
 	/* TODO Ensure within maintenance_work_mem */
 	buildstate->samples = VectorArrayInit(numSamples, buildstate->dimensions);
 	if (buildstate->heap != NULL)
+	{
 		SampleRows(buildstate);
 
+		if (buildstate->samples->length < buildstate->lists)
+		{
+			ereport(NOTICE,
+					(errmsg("ivfflat index created with little data"),
+					 errdetail("this will cause poor recall"),
+					 errhint("drop the index until the table has more data")));
+		}
+	}
+
 	/* Calculate centers */
 	IvfflatBench("k-means", IvfflatKmeans(buildstate->index, buildstate->samples, buildstate->centers));
 

src/ivfscan.c

@@ -111,6 +111,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	Datum		datum;
 	bool		isnull;
 	TupleDesc	tupdesc = RelationGetDescr(scan->indexRelation);
+	double		tuples = 0;
 
 #if PG_VERSION_NUM >= 120000
 	TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
@@ -123,7 +124,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	 *
 	 * See postgres/src/backend/storage/buffer/README for description
 	 */
-	BufferAccessStrategy bas = GetAccessStrategy(BAS_BULKREAD);
+	BufferAccessStrategy bas = GetAccessStrategy(BAS_NORMAL);
 
 	/* Search closest probes lists */
 	while (!pairingheap_is_empty(so->listQueue))
@@ -159,6 +160,8 @@ GetScanItems(IndexScanDesc scan, Datum value)
 				ExecStoreVirtualTuple(slot);
 
 				tuplesort_puttupleslot(so->sortstate, slot);
+
+				tuples++;
 			}
 
 			searchPage = IvfflatPageGetOpaque(page)->nextblkno;
@@ -167,6 +170,13 @@ GetScanItems(IndexScanDesc scan, Datum value)
 		}
 	}
 
+	/* TODO Scan more lists */
+	if (tuples < 100)
+		ereport(DEBUG1,
+				(errmsg("index scan found few tuples"),
+				 errdetail("index may have been created without data or lists is too high"),
+				 errhint("recreate the index and possibly decrease lists")));
+
 	tuplesort_performsort(so->sortstate);
 }
 

src/vector.c

@@ -467,6 +467,7 @@ l2_distance(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 	{
 		diff = ax[i] - bx[i];
@@ -493,6 +494,7 @@ vector_l2_squared_distance(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 	{
 		diff = ax[i] - bx[i];
@@ -517,6 +519,7 @@ inner_product(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		distance += ax[i] * bx[i];
 
@@ -538,6 +541,7 @@ vector_negative_inner_product(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		distance += ax[i] * bx[i];
 
@@ -561,6 +565,7 @@ cosine_distance(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 	{
 		distance += ax[i] * bx[i];
@@ -587,6 +592,7 @@ vector_spherical_distance(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		distance += a->x[i] * b->x[i];
 
@@ -622,6 +628,7 @@ vector_norm(PG_FUNCTION_ARGS)
 	float	   *ax = a->x;
 	double		norm = 0.0;
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		norm += ax[i] * ax[i];
 
@@ -646,6 +653,8 @@ vector_add(PG_FUNCTION_ARGS)
 
 	result = InitVector(a->dim);
 	rx = result->x;
+
+	/* Auto-vectorized */
 	for (int i = 0, imax = a->dim; i < imax; i++)
 		rx[i] = ax[i] + bx[i];
 
@@ -670,6 +679,8 @@ vector_sub(PG_FUNCTION_ARGS)
 
 	result = InitVector(a->dim);
 	rx = result->x;
+
+	/* Auto-vectorized */
 	for (int i = 0, imax = a->dim; i < imax; i++)
 		rx[i] = ax[i] - bx[i];
 

test/expected/functions.out

@@ -22,10 +22,28 @@ SELECT round(vector_norm('[1,1]')::numeric, 5);
  1.41421
 (1 row)
 
-SELECT round(l2_distance('[1,2]', '[0,0]')::numeric, 5);
-  round  
----------
- 2.23607
+SELECT vector_norm('[3,4]');
+ vector_norm 
+-------------
+           5
+(1 row)
+
+SELECT vector_norm('[0,1]');
+ vector_norm 
+-------------
+           1
+(1 row)
+
+SELECT l2_distance('[0,0]', '[3,4]');
+ l2_distance 
+-------------
+           5
+(1 row)
+
+SELECT l2_distance('[0,0]', '[0,1]');
+ l2_distance 
+-------------
+           1
 (1 row)
 
 SELECT l2_distance('[1,2]', '[3]');
@@ -38,10 +56,10 @@ SELECT inner_product('[1,2]', '[3,4]');
 
 SELECT inner_product('[1,2]', '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT round(cosine_distance('[1,2]', '[2,4]')::numeric, 5);
-  round  
----------
- 0.00000
+SELECT cosine_distance('[1,2]', '[2,4]');
+ cosine_distance 
+-----------------
+               0
 (1 row)
 
 SELECT cosine_distance('[1,2]', '[0,0]');

test/sql/functions.sql

@@ -2,15 +2,19 @@ SELECT '[1,2,3]'::vector + '[4,5,6]';
 SELECT '[1,2,3]'::vector - '[4,5,6]';
 
 SELECT vector_dims('[1,2,3]');
-SELECT round(vector_norm('[1,1]')::numeric, 5);
 
-SELECT round(l2_distance('[1,2]', '[0,0]')::numeric, 5);
+SELECT round(vector_norm('[1,1]')::numeric, 5);
+SELECT vector_norm('[3,4]');
+SELECT vector_norm('[0,1]');
+
+SELECT l2_distance('[0,0]', '[3,4]');
+SELECT l2_distance('[0,0]', '[0,1]');
 SELECT l2_distance('[1,2]', '[3]');
 
 SELECT inner_product('[1,2]', '[3,4]');
 SELECT inner_product('[1,2]', '[3]');
 
-SELECT round(cosine_distance('[1,2]', '[2,4]')::numeric, 5);
+SELECT cosine_distance('[1,2]', '[2,4]');
 SELECT cosine_distance('[1,2]', '[0,0]');
 SELECT cosine_distance('[1,1]', '[1,1]');
 SELECT cosine_distance('[1,1]', '[-1,-1]');