Compare commits

..

1 Commits

Author SHA1 Message Date
Andrew Kane
7c6694e0ef Added bound option 2022-02-13 01:41:06 -08:00
60 changed files with 539 additions and 1744 deletions

View File

@@ -1,6 +1,6 @@
root = true
[*.{c,h,pl,pm,sql}]
[*.{c,h,pl}]
indent_style = tab
indent_size = tab
tab_width = 4

View File

@@ -1,72 +1,38 @@
name: build
on: [push, pull_request]
jobs:
ubuntu:
build:
runs-on: ${{ matrix.os }}
if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest]
postgres: [14, 13, 12, 11, 10, 9.6]
include:
- postgres: 15
os: ubuntu-22.04
- postgres: 14
os: ubuntu-22.04
- postgres: 13
os: ubuntu-20.04
- postgres: 12
os: ubuntu-20.04
- postgres: 11
os: ubuntu-20.04
- os: macos-latest
postgres: 14
steps:
- uses: actions/checkout@v3
- uses: ankane/setup-postgres@v1
with:
postgres-version: ${{ matrix.postgres }}
dev-files: true
- run: make
- run: |
export PG_CONFIG=`which pg_config`
sudo --preserve-env=PG_CONFIG make install
- run: make installcheck
- if: ${{ failure() }}
run: cat regression.diffs
- run: |
sudo apt-get update
sudo apt-get install libipc-run-perl
- run: make prove_installcheck
mac:
runs-on: macos-latest
if: ${{ !startsWith(github.ref_name, 'windows') }}
steps:
- uses: actions/checkout@v3
- uses: ankane/setup-postgres@v1
with:
postgres-version: 14
- run: make
- run: make install
- run: make installcheck
- if: ${{ failure() }}
run: cat regression.diffs
- run: |
brew install cpanm
cpanm --notest IPC::Run
wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_5.tar.gz
tar xf REL_14_5.tar.gz
- run: make prove_installcheck PROVE_FLAGS="-I ./postgres-REL_14_5/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
windows:
runs-on: windows-latest
if: ${{ !startsWith(github.ref_name, 'mac') }}
steps:
- uses: actions/checkout@v3
- uses: ankane/setup-postgres@v1
with:
postgres-version: 14
- run: |
call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
nmake /NOLOGO /F Makefile.win && ^
nmake /NOLOGO /F Makefile.win install && ^
nmake /NOLOGO /F Makefile.win installcheck && ^
nmake /NOLOGO /F Makefile.win clean && ^
nmake /NOLOGO /F Makefile.win uninstall
shell: cmd
- uses: actions/checkout@v2
- uses: ankane/setup-postgres@v1
with:
postgres-version: ${{ matrix.postgres }}
- if: ${{ startsWith(matrix.os, 'ubuntu') }}
run: sudo apt-get update && sudo apt-get install postgresql-server-dev-${{ matrix.postgres }} libipc-run-perl
- run: make
- if: ${{ startsWith(matrix.os, 'ubuntu') }}
run: |
export PG_CONFIG=`which pg_config`
sudo --preserve-env=PG_CONFIG make install
- if: ${{ startsWith(matrix.os, 'macos') }}
run: make install
- run: make installcheck
- if: ${{ failure() }}
run: cat regression.diffs
- if: ${{ startsWith(matrix.os, 'ubuntu') }}
run: make prove_installcheck
- if: ${{ startsWith(matrix.os, 'macos') }}
run: |
brew install cpanm && cpanm IPC::Run
wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_1.tar.gz
tar xf REL_14_1.tar.gz
make prove_installcheck PROVE=prove PERL5LIB=postgres-REL_14_1/src/test/perl

7
.gitignore vendored
View File

@@ -1,14 +1,7 @@
/dist/
/log/
/results/
/tmp_check/
/sql/vector--?.?.?.sql
regression.*
*.o
*.so
*.bc
*.dll
*.dylib
*.obj
*.lib
*.exp

View File

@@ -1,50 +1,3 @@
## 0.4.2 (unreleased)
- Added notice when index created with little data
- Fixed installation error with Postgres 12.0-12.2
## 0.4.1 (2023-03-21)
- Improved performance of cosine distance
- Fixed index scan count
## 0.4.0 (2023-01-11)
If upgrading with Postgres < 13, see [this note](https://github.com/pgvector/pgvector#040).
- Changed text representation for vector elements to match `real`
- Changed storage for vector from `plain` to `extended`
- Increased max dimensions for vector from 1024 to 16000
- Increased max dimensions for index from 1024 to 2000
- Improved accuracy of text parsing for certain inputs
- Added `avg` aggregate for vector
- Added experimental support for Windows
- Dropped support for Postgres 10
## 0.3.2 (2022-11-22)
- Fixed `invalid memory alloc request size` error
## 0.3.1 (2022-11-02)
If upgrading from 0.2.7 or 0.3.0, [recreate](https://github.com/pgvector/pgvector#031) all `ivfflat` indexes after upgrading to ensure all data is indexed.
- Fixed issue with inserts silently corrupting `ivfflat` indexes (introduced in 0.2.7)
- Fixed segmentation fault with index creation when lists > 6500
## 0.3.0 (2022-10-15)
- Added support for Postgres 15
- Dropped support for Postgres 9.6
## 0.2.7 (2022-07-31)
- Fixed `unexpected data beyond EOF` error
## 0.2.6 (2022-05-22)
- Improved performance of index creation for Postgres < 12
## 0.2.5 (2022-02-11)
- Reduced memory usage during index creation

View File

@@ -1,11 +1,9 @@
ARG PG_MAJOR=15
FROM postgres:$PG_MAJOR
ARG PG_MAJOR
FROM postgres:14
COPY . /tmp/pgvector
RUN apt-get update && \
apt-get install -y --no-install-recommends build-essential postgresql-server-dev-$PG_MAJOR && \
apt-get install -y --no-install-recommends build-essential postgresql-server-dev-14 && \
cd /tmp/pgvector && \
make clean && \
make OPTFLAGS="" && \
@@ -13,6 +11,6 @@ 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-$PG_MAJOR && \
apt-get remove -y build-essential postgresql-server-dev-14 && \
apt-get autoremove -y && \
rm -rf /var/lib/apt/lists/*

View File

@@ -1,4 +1,4 @@
Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
Portions Copyright (c) 1996-2021, PostgreSQL Global Development Group
Portions Copyright (c) 1994, The Regents of the University of California

View File

@@ -2,7 +2,7 @@
"name": "vector",
"abstract": "Open-source vector similarity search for Postgres",
"description": "Supports L2 distance, inner product, and cosine distance",
"version": "0.4.1",
"version": "0.2.5",
"maintainer": [
"Andrew Kane <andrew@ankane.org>"
],
@@ -12,7 +12,7 @@
"prereqs": {
"runtime": {
"requires": {
"PostgreSQL": "11.0.0"
"PostgreSQL": "9.6.0"
}
}
},
@@ -20,7 +20,7 @@
"vector": {
"file": "sql/vector.sql",
"docfile": "README.md",
"version": "0.4.1",
"version": "0.2.5",
"abstract": "Open-source vector similarity search for Postgres"
}
},

View File

@@ -1,5 +1,5 @@
EXTENSION = vector
EXTVERSION = 0.4.1
EXTVERSION = 0.2.5
MODULE_big = vector
DATA = $(wildcard sql/*--*.sql)
@@ -7,14 +7,13 @@ OBJS = src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.
TESTS = $(wildcard test/sql/*.sql)
REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
REGRESS_OPTS = --inputdir=test --load-extension=vector
REGRESS_OPTS = --inputdir=test
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
@@ -41,14 +40,6 @@ PG_CONFIG ?= pg_config
PGXS := $(shell $(PG_CONFIG) --pgxs)
include $(PGXS)
# for Mac
ifeq ($(PROVE),)
PROVE = prove
endif
# for Postgres 15
PROVE_FLAGS += -I ./test/perl
prove_installcheck:
rm -rf $(CURDIR)/tmp_check
cd $(srcdir) && TESTDIR='$(CURDIR)' PATH="$(bindir):$$PATH" PGPORT='6$(DEF_PGPORT)' PG_REGRESS='$(top_builddir)/src/test/regress/pg_regress' $(PROVE) $(PG_PROVE_FLAGS) $(PROVE_FLAGS) $(if $(PROVE_TESTS),$(PROVE_TESTS),test/t/*.pl)
@@ -62,4 +53,4 @@ dist:
.PHONY: docker
docker:
docker build --pull --no-cache --platform linux/amd64 -t ankane/pgvector:latest .
docker build --pull --no-cache -t ankane/pgvector:latest .

View File

@@ -1,70 +0,0 @@
EXTENSION = vector
EXTVERSION = 0.4.1
OBJS = src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
REGRESS = btree cast copy functions input ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged
REGRESS_OPTS = --inputdir=test --load-extension=vector
# For /arch flags
# https://learn.microsoft.com/en-us/cpp/build/reference/arch-minimum-cpu-architecture
OPTFLAGS =
# For auto-vectorization:
# - MSVC (needs /O2 /fp:fast) - https://learn.microsoft.com/en-us/cpp/parallel/auto-parallelization-and-auto-vectorization?#auto-vectorizer
PG_CFLAGS = $(PG_CFLAGS) $(OPTFLAGS) /O2 /fp:fast
# Debug MSVC auto-vectorization
# https://learn.microsoft.com/en-us/cpp/error-messages/tool-errors/vectorizer-and-parallelizer-messages
# PG_CFLAGS = $(PG_CFLAGS) /Qvec-report:2
all: sql\$(EXTENSION)--$(EXTVERSION).sql
sql\$(EXTENSION)--$(EXTVERSION).sql: sql\$(EXTENSION).sql
copy sql\$(EXTENSION).sql $@
# TODO use pg_config
!ifndef PGROOT
!error PGROOT is not set
!endif
BINDIR = $(PGROOT)\bin
INCLUDEDIR = $(PGROOT)\include
INCLUDEDIR_SERVER = $(PGROOT)\include\server
LIBDIR = $(PGROOT)\lib
PKGLIBDIR = $(PGROOT)\lib
SHAREDIR = $(PGROOT)\share
CFLAGS = /nologo /I"$(INCLUDEDIR_SERVER)\port\win32_msvc" /I"$(INCLUDEDIR_SERVER)\port\win32" /I"$(INCLUDEDIR_SERVER)" /I"$(INCLUDEDIR)"
CFLAGS = $(CFLAGS) $(PG_CFLAGS)
SHLIB = $(EXTENSION).dll
LIBS = "$(LIBDIR)\postgres.lib"
.c.obj:
$(CC) $(CFLAGS) /c $< /Fo$@
$(SHLIB): $(OBJS)
$(CC) $(CFLAGS) $(OBJS) $(LIBS) /link /DLL /OUT:$(SHLIB)
all: $(SHLIB)
install:
copy $(SHLIB) "$(PKGLIBDIR)"
copy $(EXTENSION).control "$(SHAREDIR)\extension"
copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
installcheck:
"$(BINDIR)\pg_regress" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)
uninstall:
del /f "$(PKGLIBDIR)\$(SHLIB)"
del /f "$(SHAREDIR)\extension\$(EXTENSION).control"
del /f "$(SHAREDIR)\extension\vector--*.sql"
clean:
del /f $(SHLIB) $(EXTENSION).lib $(EXTENSION).exp
del /f $(OBJS)
del /f sql\$(EXTENSION)--$(EXTVERSION).sql
del /f /s /q results regression.diffs regression.out tmp_check tmp_check_iso log output_iso

370
README.md
View File

@@ -2,21 +2,22 @@
Open-source vector similarity search for Postgres
Supports
```sql
CREATE TABLE table (column vector(3));
CREATE INDEX ON table USING ivfflat (column vector_l2_ops);
SELECT * FROM table ORDER BY column <-> '[1,2,3]' LIMIT 5;
```
- exact and approximate nearest neighbor search
- L2 distance, inner product, and cosine distance
- any [language](#languages) with a Postgres client
Supports 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)
## Installation
Compile and install the extension (supports Postgres 11+)
Compile and install the extension (supports Postgres 9.6+)
```sh
cd /tmp
git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
git clone --branch v0.2.5 https://github.com/pgvector/pgvector.git
cd pgvector
make
make install # may need sudo
@@ -28,178 +29,81 @@ Then load it in databases where you want to use it
CREATE EXTENSION vector;
```
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)
You can also install it with [Docker](#docker), [Homebrew](#homebrew), or [PGXN](#pgxn)
## Getting Started
Create a vector column with 3 dimensions
Create a vector column with 3 dimensions (replace `table` and `column` with non-reserved names)
```sql
CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
CREATE TABLE table (column vector(3));
```
Insert vectors
Insert values
```sql
INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
INSERT INTO table VALUES ('[1,2,3]'), ('[4,5,6]');
```
Get the nearest neighbors by L2 distance
Get the nearest neighbor by L2 distance
```sql
SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
SELECT * FROM table ORDER BY column <-> '[3,1,2]' LIMIT 1;
```
Also supports inner product (`<#>`) and cosine distance (`<=>`)
Note: `<#>` returns the negative inner product since Postgres only supports `ASC` order index scans on operators
## Storing
Create a new table with a vector column
```sql
CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
```
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;
```
Note: Combine with `ORDER BY` and `LIMIT` to use an index
#### 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 (embedding <#> '[3,1,2]') * -1 AS inner_product FROM items;
```
For cosine similarity, use 1 - cosine distance
```sql
SELECT 1 - (embedding <=> '[3,1,2]') AS cosine_similarity FROM items;
```
#### Aggregates
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
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) - a good place to start is `lists / 10` for up to 1M rows and `sqrt(lists)` for over 1M rows
Add an index for each distance function you want to use.
Speed up queries with an approximate index. Add an index for each distance function you want to use.
L2 distance
```sql
CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100);
CREATE INDEX ON table USING ivfflat (column vector_l2_ops);
```
Inner product
```sql
CREATE INDEX ON items USING ivfflat (embedding vector_ip_ops) WITH (lists = 100);
CREATE INDEX ON table USING ivfflat (column vector_ip_ops);
```
Cosine distance
```sql
CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);
CREATE INDEX ON table USING ivfflat (column vector_cosine_ops);
```
Vectors with up to 2,000 dimensions can be indexed.
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.
### Index Options
Specify the number of inverted lists (100 by default)
```sql
CREATE INDEX ON table USING ivfflat (column opclass) WITH (lists = 100);
```
A [good place to start](https://github.com/facebookresearch/faiss/issues/112) is `4 * sqrt(rows)`
### Query Options
Specify the number of probes (1 by default)
```sql
SET ivfflat.probes = 10;
SET ivfflat.probes = 1;
```
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 wont use the index)
A higher value improves recall at the cost of speed.
Use `SET LOCAL` inside a transaction to set it for a single query
```sql
BEGIN;
SET LOCAL ivfflat.probes = 10;
SET LOCAL ivfflat.probes = 1;
SELECT ...
COMMIT;
```
@@ -215,9 +119,10 @@ SELECT phase, tuples_done, tuples_total FROM pg_stat_progress_create_index;
The phases are:
1. `initializing`
2. `performing k-means`
3. `sorting tuples`
4. `loading tuples`
2. `sampling table`
3. `performing k-means`
4. `sorting tuples`
5. `loading tuples`
Note: `tuples_done` and `tuples_total` are only populated during the `loading tuples` phase
@@ -226,97 +131,30 @@ Note: `tuples_done` and `tuples_total` are only populated during the `loading tu
Consider [partial indexes](https://www.postgresql.org/docs/current/indexes-partial.html) for queries with a `WHERE` clause
```sql
SELECT * FROM items WHERE category_id = 123 ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
CREATE INDEX ON table USING ivfflat (column opclass) WHERE (other_column = 123);
```
can be indexed with:
```sql
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`.
```sql
CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
```
To index many different values of `other_column`, consider [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) on `other_column`.
## 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);
CREATE INDEX ON table USING ivfflat (column opclass) WITH (lists = 1000);
```
## Languages
Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
Language | Libraries / Examples
--- | ---
C++ | [pgvector-cpp](https://github.com/pgvector/pgvector-cpp)
C# | [pgvector-dotnet](https://github.com/pgvector/pgvector-dotnet)
Crystal | [pgvector-crystal](https://github.com/pgvector/pgvector-crystal)
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, 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)
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)
## Frequently Asked Questions
#### How many vectors can be stored in a single table?
A non-partitioned table has a limit of 32 TB by default in Postgres. A partitioned table can have thousands of partitions of that size.
#### Is replication supported?
Yes, pgvector uses the write-ahead log (WAL), which allows for replication and point-in-time recovery.
#### What if I want to index vectors with more than 2,000 dimensions?
Two things you can try are:
1. use dimensionality reduction
2. compile Postgres with a larger block size (`./configure --with-blocksize=32`) and edit the limit in `src/ivfflat.h`
## Reference
### Vector Type
Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a float, and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 1024 dimensions.
### Vector Operators
@@ -332,53 +170,40 @@ Operator | Description
Function | Description
--- | ---
cosine_distance(vector, vector) → double precision | cosine distance
inner_product(vector, vector) → double precision | inner product
l2_distance(vector, vector) → double precision | Euclidean distance
vector_dims(vector) → integer | number of dimensions
vector_norm(vector) → double precision | Euclidean norm
cosine_distance(vector, vector) | cosine distance
inner_product(vector, vector) | inner product
l2_distance(vector, vector) | Euclidean distance
vector_dims(vector) | number of dimensions
vector_norm(vector) | Euclidean norm
### Aggregate Functions
## Libraries
Function | Description
--- | ---
avg(vector) → vector | arithmetic mean
Libraries that use pgvector:
## Installation Notes
- [pgvector-python](https://github.com/pgvector/pgvector-python) (Python)
- [Neighbor](https://github.com/ankane/neighbor) (Ruby)
- [pgvector-ruby](https://github.com/pgvector/pgvector-ruby) (Ruby)
- [pgvector-node](https://github.com/pgvector/pgvector-node) (Node.js)
- [pgvector-go](https://github.com/pgvector/pgvector-go) (Go)
- [pgvector-rust](https://github.com/pgvector/pgvector-rust) (Rust)
- [pgvector-cpp](https://github.com/pgvector/pgvector-cpp) (C++)
### Postgres Location
## Frequently Asked Questions
If your machine has multiple Postgres installations, specify the path to [pg_config](https://www.postgresql.org/docs/current/app-pgconfig.html) with:
#### How many vectors can be stored in a single table?
```sh
export PG_CONFIG=/Applications/Postgres.app/Contents/Versions/latest/bin/pg_config
```
A non-partitioned table has a limit of 32 TB by default in Postgres. A partitioned table can have thousands of partitions of that size.
Then re-run the installation instructions (run `make clean` before `make` if needed)
#### Is replication supported?
### Missing Header
Yes, pgvector uses the write-ahead log (WAL), which allows for replication and point-in-time recovery.
If compilation fails with `fatal error: postgres.h: No such file or directory`, make sure Postgres development files are installed on the server.
#### What if my data has more than 1024 dimensions?
For Ubuntu and Debian, use:
Two things you can try are:
```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
```
1. use dimensionality reduction
2. compile Postgres with a larger block size (`./configure --with-blocksize=32`) and edit the limit in `src/vector.h`
## Additional Installation Methods
@@ -390,22 +215,22 @@ Get the [Docker image](https://hub.docker.com/r/ankane/pgvector) with:
docker pull ankane/pgvector
```
This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (run it the same way).
This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres).
You can also build the image manually:
You can also build the image manually
```sh
git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
git clone --branch v0.2.5 https://github.com/pgvector/pgvector.git
cd pgvector
docker build -t pgvector .
```
### Homebrew
With Homebrew Postgres, you can use:
On Mac with Homebrew Postgres, you can use:
```sh
brew install pgvector
brew install pgvector/brew/pgvector
```
### PGXN
@@ -416,39 +241,14 @@ 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:
```sh
conda install -c conda-forge pgvector
```
This method is [community-maintained](https://github.com/conda-forge/pgvector-feedstock) by [@mmcauliffe](https://github.com/mmcauliffe)
## Hosted Postgres
pgvector is available on [these providers](https://github.com/pgvector/pgvector/issues/54).
To request a new extension on other providers:
Some Postgres providers only support specific extensions. To request a new extension:
- 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)
- Azure Database - vote or comment on [this page](https://feedback.azure.com/d365community/idea/7b423322-6189-ed11-a81b-000d3ae49307)
- Google Cloud SQL - follow the instructions on [this page](https://cloud.google.com/sql/docs/postgres/extensions#requesting-support-for-a-new-extension)
- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/app-framework-services/p/pgvector-extension-for-postgresql)
- Heroku Postgres - vote or comment on [this page](https://github.com/heroku/roadmap/issues/156)
- Azure Database for PostgreSQL - follow the instructions on [this page](https://docs.microsoft.com/en-us/azure/postgresql/concepts-extensions#next-steps)
## Upgrading
@@ -458,32 +258,6 @@ Install the latest version and run:
ALTER EXTENSION vector UPDATE;
```
## Upgrade Notes
### 0.4.0
If upgrading with Postgres < 13, remove this line from `sql/vector--0.3.2--0.4.0.sql`:
```sql
ALTER TYPE vector SET (STORAGE = extended);
```
Then run `make install` and `ALTER EXTENSION vector UPDATE;`.
### 0.3.1
If upgrading from 0.2.7 or 0.3.0, recreate all `ivfflat` indexes after upgrading to ensure all data is indexed.
```sql
-- Postgres 12+
REINDEX INDEX CONCURRENTLY index_name;
-- Postgres < 12
CREATE INDEX CONCURRENTLY temp_name ON table USING ivfflat (column opclass);
DROP INDEX CONCURRENTLY index_name;
ALTER INDEX temp_name RENAME TO index_name;
```
## Thanks
Thanks to:

View File

@@ -7,13 +7,13 @@ DROP CAST (double precision[] AS vector);
DROP CAST (numeric[] AS vector);
CREATE CAST (integer[] AS vector)
WITH FUNCTION array_to_vector(integer[], integer, boolean) AS ASSIGNMENT;
WITH FUNCTION array_to_vector(integer[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (real[] AS vector)
WITH FUNCTION array_to_vector(real[], integer, boolean) AS ASSIGNMENT;
WITH FUNCTION array_to_vector(real[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (double precision[] AS vector)
WITH FUNCTION array_to_vector(double precision[], integer, boolean) AS ASSIGNMENT;
WITH FUNCTION array_to_vector(double precision[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (numeric[] AS vector)
WITH FUNCTION array_to_vector(numeric[], integer, boolean) AS ASSIGNMENT;
WITH FUNCTION array_to_vector(numeric[], integer, boolean) AS ASSIGNMENT;

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.2.6'" to load this file. \quit

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.2.7'" to load this file. \quit

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.3.0'" to load this file. \quit

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.3.1'" to load this file. \quit

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.3.2'" to load this file. \quit

View File

@@ -1,23 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.0'" to load this file. \quit
-- remove this single line for Postgres < 13
ALTER TYPE vector SET (STORAGE = extended);
CREATE FUNCTION vector_accum(double precision[], vector) RETURNS double precision[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_avg(double precision[]) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_combine(double precision[], double precision[]) RETURNS double precision[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE AGGREGATE avg(vector) (
SFUNC = vector_accum,
STYPE = double precision[],
FINALFUNC = vector_avg,
COMBINEFUNC = vector_combine,
INITCOND = '{0}',
PARALLEL = SAFE
);

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.1'" to load this file. \quit

View File

@@ -25,8 +25,7 @@ CREATE TYPE vector (
OUTPUT = vector_out,
TYPMOD_IN = vector_typmod_in,
RECEIVE = vector_recv,
SEND = vector_send,
STORAGE = extended
SEND = vector_send
);
-- functions
@@ -84,26 +83,6 @@ CREATE FUNCTION vector_negative_inner_product(vector, vector) RETURNS float8
CREATE FUNCTION vector_spherical_distance(vector, vector) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_accum(double precision[], vector) RETURNS double precision[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_avg(double precision[]) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_combine(double precision[], double precision[]) RETURNS double precision[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- aggregates
CREATE AGGREGATE avg(vector) (
SFUNC = vector_accum,
STYPE = double precision[],
FINALFUNC = vector_avg,
COMBINEFUNC = vector_combine,
INITCOND = '{0}',
PARALLEL = SAFE
);
-- cast functions
CREATE FUNCTION vector(vector, integer, boolean) RETURNS vector

View File

@@ -6,7 +6,6 @@
#include "ivfflat.h"
#include "miscadmin.h"
#include "storage/bufmgr.h"
#include "utils/memutils.h"
#if PG_VERSION_NUM >= 140000
#include "utils/backend_progress.h"
@@ -23,8 +22,13 @@
#define PROGRESS_CREATEIDX_TUPLES_DONE 0
#endif
#if PG_VERSION_NUM >= 110000
#include "catalog/pg_operator_d.h"
#include "catalog/pg_type_d.h"
#else
#include "catalog/pg_operator.h"
#include "catalog/pg_type.h"
#endif
#if PG_VERSION_NUM >= 130000
#define CALLBACK_ITEM_POINTER ItemPointer tid
@@ -39,16 +43,20 @@
#endif
/*
* Add sample
* Callback for sampling
*/
static void
AddSample(Datum *values, IvfflatBuildState * buildstate)
SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
bool *isnull, bool tupleIsAlive, void *state)
{
IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
VectorArray samples = buildstate->samples;
int targsamples = samples->maxlen;
Datum value = values[0];
/* Detoast once for all calls */
Datum value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
/* Skip nulls */
if (isnull[0])
return;
/*
* Normalize with KMEANS_NORM_PROC since spherical distance function
@@ -72,11 +80,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
if (buildstate->rowstoskip <= 0)
{
#if PG_VERSION_NUM >= 150000
int k = (int) (targsamples * sampler_random_fract(&buildstate->rstate.randstate));
#else
int k = (int) (targsamples * sampler_random_fract(buildstate->rstate.randstate));
#endif
Assert(k >= 0 && k < targsamples);
VectorArraySet(samples, k, DatumGetVector(value));
@@ -86,31 +90,6 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
}
}
/*
* Callback for sampling
*/
static void
SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
bool *isnull, bool tupleIsAlive, void *state)
{
IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
MemoryContext oldCtx;
/* Skip nulls */
if (isnull[0])
return;
/* Use memory context since detoast can allocate */
oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
/* Add sample */
AddSample(values, state);
/* Reset memory context */
MemoryContextSwitchTo(oldCtx);
MemoryContextReset(buildstate->tmpCtx);
}
/*
* Sample rows with same logic as ANALYZE
*/
@@ -120,9 +99,11 @@ SampleRows(IvfflatBuildState * buildstate)
int targsamples = buildstate->samples->maxlen;
BlockNumber totalblocks = RelationGetNumberOfBlocks(buildstate->heap);
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_SAMPLE);
buildstate->rowstoskip = -1;
BlockSampler_Init(&buildstate->bs, totalblocks, targsamples, RandomInt());
BlockSampler_Init(&buildstate->bs, totalblocks, targsamples, random());
reservoir_init_selection_state(&buildstate->rstate, targsamples);
while (BlockSampler_HasMore(&buildstate->bs))
@@ -131,29 +112,39 @@ SampleRows(IvfflatBuildState * buildstate)
#if PG_VERSION_NUM >= 120000
table_index_build_range_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
false, true, false, targblock, 1, SampleCallback, (void *) buildstate, NULL);
false, true, true, targblock, 1, SampleCallback, (void *) buildstate, NULL);
#elif PG_VERSION_NUM >= 110000
IndexBuildHeapRangeScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
true, true, targblock, 1, SampleCallback, (void *) buildstate, NULL);
#else
IndexBuildHeapRangeScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
false, true, targblock, 1, SampleCallback, (void *) buildstate, NULL);
true, true, targblock, 1, SampleCallback, (void *) buildstate);
#endif
}
}
/*
* Add tuple to sort
* Callback for table_index_build_scan
*/
static void
AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
bool *isnull, bool tupleIsAlive, void *state)
{
IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
double distance;
double minDistance = DBL_MAX;
int closestCenter = -1;
VectorArray centers = buildstate->centers;
TupleTableSlot *slot = buildstate->slot;
Datum value = values[0];
int i;
/* Detoast once for all calls */
Datum value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
#if PG_VERSION_NUM < 130000
ItemPointer tid = &hup->t_self;
#endif
if (isnull[0])
return;
/* Normalize if needed */
if (buildstate->normprocinfo != NULL)
@@ -176,18 +167,18 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
#ifdef IVFFLAT_KMEANS_DEBUG
buildstate->inertia += minDistance;
buildstate->listSums[closestCenter] += minDistance;
buildstate->listCounts[closestCenter]++;
#endif
/* Create a virtual tuple */
ExecClearTuple(slot);
slot->tts_values[0] = Int32GetDatum(closestCenter);
slot->tts_isnull[0] = false;
slot->tts_values[1] = PointerGetDatum(tid);
slot->tts_values[1] = Int32GetDatum(ItemPointerGetBlockNumberNoCheck(tid));
slot->tts_isnull[1] = false;
slot->tts_values[2] = value;
slot->tts_values[2] = Int32GetDatum(ItemPointerGetOffsetNumberNoCheck(tid));
slot->tts_isnull[2] = false;
slot->tts_values[3] = value;
slot->tts_isnull[3] = false;
ExecStoreVirtualTuple(slot);
/*
@@ -201,35 +192,6 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
buildstate->indtuples++;
}
/*
* Callback for table_index_build_scan
*/
static void
BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
bool *isnull, bool tupleIsAlive, void *state)
{
IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
MemoryContext oldCtx;
#if PG_VERSION_NUM < 130000
ItemPointer tid = &hup->t_self;
#endif
/* Skip nulls */
if (isnull[0])
return;
/* Use memory context since detoast can allocate */
oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
/* Add tuple to sort */
AddTupleToSort(index, tid, values, buildstate);
/* Reset memory context */
MemoryContextSwitchTo(oldCtx);
MemoryContextReset(buildstate->tmpCtx);
}
/*
* Get index tuple from sort state
*/
@@ -238,15 +200,23 @@ GetNextTuple(Tuplesortstate *sortstate, TupleDesc tupdesc, TupleTableSlot *slot,
{
Datum value;
bool isnull;
int tupblk;
int tupoff;
#if PG_VERSION_NUM >= 100000
if (tuplesort_gettupleslot(sortstate, true, false, slot, NULL))
#else
if (tuplesort_gettupleslot(sortstate, true, slot, NULL))
#endif
{
*list = DatumGetInt32(slot_getattr(slot, 1, &isnull));
value = slot_getattr(slot, 3, &isnull);
tupblk = DatumGetInt32(slot_getattr(slot, 2, &isnull));
tupoff = DatumGetInt32(slot_getattr(slot, 3, &isnull));
value = slot_getattr(slot, 4, &isnull);
/* Form the index tuple */
*itup = index_form_tuple(tupdesc, &value, &isnull);
(*itup)->t_tid = *((ItemPointer) DatumGetPointer(slot_getattr(slot, 2, &isnull)));
ItemPointerSet(&(*itup)->t_tid, tupblk, tupoff);
}
else
*list = -1;
@@ -263,8 +233,8 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
GenericXLogState *state;
int list;
IndexTuple itup = NULL; /* silence compiler warning */
BlockNumber startPage;
BlockNumber insertPage;
BlockNumber startPage = InvalidBlockNumber;
BlockNumber insertPage = InvalidBlockNumber;
Size itemsz;
int i;
int64 inserted = 0;
@@ -289,7 +259,7 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
CHECK_FOR_INTERRUPTS();
buf = IvfflatNewBuffer(index, forkNum);
IvfflatInitRegisterPage(index, &buf, &page, &state);
IvfflatInitPage(index, &buf, &page, &state);
startPage = BufferGetBlockNumber(buf);
@@ -338,9 +308,6 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
if (buildstate->dimensions < 0)
elog(ERROR, "column does not have dimensions");
if (buildstate->dimensions > IVFFLAT_MAX_DIM)
elog(ERROR, "column cannot have more than %d dimensions for ivfflat index", IVFFLAT_MAX_DIM);
buildstate->reltuples = 0;
buildstate->indtuples = 0;
@@ -358,13 +325,18 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
/* Create tuple description for sorting */
#if PG_VERSION_NUM >= 120000
buildstate->tupdesc = CreateTemplateTupleDesc(3);
buildstate->tupdesc = CreateTemplateTupleDesc(4);
#else
buildstate->tupdesc = CreateTemplateTupleDesc(3, false);
buildstate->tupdesc = CreateTemplateTupleDesc(4, false);
#endif
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 3, "vector", RelationGetDescr(index)->attrs[0].atttypid, -1, 0);
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 2, "blkno", INT4OID, -1, 0);
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 3, "offset", INT4OID, -1, 0);
#if PG_VERSION_NUM >= 110000
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 4, "vector", RelationGetDescr(index)->attrs[0].atttypid, -1, 0);
#else
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 4, "vector", RelationGetDescr(index)->attrs[0]->atttypid, -1, 0);
#endif
#if PG_VERSION_NUM >= 120000
buildstate->slot = MakeSingleTupleTableSlot(buildstate->tupdesc, &TTSOpsVirtual);
@@ -378,14 +350,8 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
/* Reuse for each tuple */
buildstate->normvec = InitVector(buildstate->dimensions);
buildstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
"Ivfflat build temporary context",
ALLOCSET_DEFAULT_SIZES);
#ifdef IVFFLAT_KMEANS_DEBUG
buildstate->inertia = 0;
buildstate->listSums = palloc0(sizeof(double) * buildstate->lists);
buildstate->listCounts = palloc0(sizeof(int) * buildstate->lists);
#endif
}
@@ -395,16 +361,9 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
static void
FreeBuildState(IvfflatBuildState * buildstate)
{
VectorArrayFree(buildstate->centers);
pfree(buildstate->centers);
pfree(buildstate->listInfo);
pfree(buildstate->normvec);
#ifdef IVFFLAT_KMEANS_DEBUG
pfree(buildstate->listSums);
pfree(buildstate->listCounts);
#endif
MemoryContextDelete(buildstate->tmpCtx);
}
/*
@@ -415,8 +374,6 @@ ComputeCenters(IvfflatBuildState * buildstate)
{
int numSamples;
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_KMEANS);
/* Target 50 samples per list, with at least 10000 samples */
/* The number of samples has a large effect on index build time */
numSamples = buildstate->lists * 50;
@@ -431,23 +388,14 @@ 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 */
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_KMEANS);
IvfflatBench("k-means", IvfflatKmeans(buildstate->index, buildstate->samples, buildstate->centers));
/* Free samples before we allocate more memory */
VectorArrayFree(buildstate->samples);
pfree(buildstate->samples);
}
/*
@@ -462,7 +410,7 @@ CreateMetaPage(Relation index, int dimensions, int lists, ForkNumber forkNum)
IvfflatMetaPage metap;
buf = IvfflatNewBuffer(index, forkNum);
IvfflatInitRegisterPage(index, &buf, &page, &state);
IvfflatInitPage(index, &buf, &page, &state);
/* Set metapage data */
metap = IvfflatPageGetMeta(page);
@@ -495,7 +443,7 @@ CreateListPages(Relation index, VectorArray centers, int dimensions,
list = palloc(itemsz);
buf = IvfflatNewBuffer(index, forkNum);
IvfflatInitRegisterPage(index, &buf, &page, &state);
IvfflatInitPage(index, &buf, &page, &state);
for (i = 0; i < lists; i++)
{
@@ -523,51 +471,6 @@ CreateListPages(Relation index, VectorArray centers, int dimensions,
pfree(list);
}
/*
* Print k-means metrics
*/
#ifdef IVFFLAT_KMEANS_DEBUG
static void
PrintKmeansMetrics(IvfflatBuildState * buildstate)
{
elog(INFO, "inertia: %.3e", buildstate->inertia);
/* Calculate Davies-Bouldin index */
if (buildstate->lists > 1)
{
double db = 0.0;
/* Calculate average distance */
for (int i = 0; i < buildstate->lists; i++)
{
if (buildstate->listCounts[i] > 0)
buildstate->listSums[i] /= buildstate->listCounts[i];
}
for (int i = 0; i < buildstate->lists; i++)
{
double max = 0.0;
double distance;
for (int j = 0; j < buildstate->lists; j++)
{
if (j == i)
continue;
distance = DatumGetFloat8(FunctionCall2Coll(buildstate->procinfo, buildstate->collation, PointerGetDatum(VectorArrayGet(buildstate->centers, i)), PointerGetDatum(VectorArrayGet(buildstate->centers, j))));
distance = (buildstate->listSums[i] + buildstate->listSums[j]) / distance;
if (distance > max)
max = distance;
}
db += max;
}
db /= buildstate->lists;
elog(INFO, "davies-bouldin: %.3f", db);
}
}
#endif
/*
* Create entry pages
*/
@@ -581,7 +484,11 @@ CreateEntryPages(IvfflatBuildState * buildstate, ForkNumber forkNum)
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_SORT);
#if PG_VERSION_NUM >= 110000
buildstate->sortstate = tuplesort_begin_heap(buildstate->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, maintenance_work_mem, NULL, false);
#else
buildstate->sortstate = tuplesort_begin_heap(buildstate->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, maintenance_work_mem, false);
#endif
/* Add tuples to sort */
if (buildstate->heap != NULL)
@@ -589,9 +496,12 @@ CreateEntryPages(IvfflatBuildState * buildstate, ForkNumber forkNum)
#if PG_VERSION_NUM >= 120000
buildstate->reltuples = table_index_build_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
true, true, BuildCallback, (void *) buildstate, NULL);
#else
#elif PG_VERSION_NUM >= 110000
buildstate->reltuples = IndexBuildHeapScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
true, BuildCallback, (void *) buildstate, NULL);
#else
buildstate->reltuples = IndexBuildHeapScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
true, BuildCallback, (void *) buildstate);
#endif
}
@@ -599,7 +509,7 @@ CreateEntryPages(IvfflatBuildState * buildstate, ForkNumber forkNum)
tuplesort_performsort(buildstate->sortstate);
#ifdef IVFFLAT_KMEANS_DEBUG
PrintKmeansMetrics(buildstate);
elog(INFO, "inertia: %.3e", buildstate->inertia);
#endif
/* Insert */

View File

@@ -13,6 +13,7 @@
#endif
int ivfflat_probes;
int ivfflat_bound;
static relopt_kind ivfflat_relopt_kind;
/*
@@ -32,6 +33,10 @@ _PG_init(void)
DefineCustomIntVariable("ivfflat.probes", "Sets the number of probes",
"Valid range is 1..lists.", &ivfflat_probes,
1, 1, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
DefineCustomIntVariable("ivfflat.bound", "Sets the max results from index (experimental)",
NULL, &ivfflat_bound,
0, 0, INT_MAX, PGC_USERSET, 0, NULL, NULL, NULL);
}
/*
@@ -45,6 +50,8 @@ ivfflatbuildphasename(int64 phasenum)
{
case PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE:
return "initializing";
case PROGRESS_IVFFLAT_PHASE_SAMPLE:
return "sampling table";
case PROGRESS_IVFFLAT_PHASE_KMEANS:
return "performing k-means";
case PROGRESS_IVFFLAT_PHASE_SORT:
@@ -64,7 +71,9 @@ static void
ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
Cost *indexStartupCost, Cost *indexTotalCost,
Selectivity *indexSelectivity, double *indexCorrelation
#if PG_VERSION_NUM >= 100000
,double *indexPages
#endif
)
{
GenericCosts costs;
@@ -82,7 +91,9 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
*indexTotalCost = DBL_MAX;
*indexSelectivity = 0;
*indexCorrelation = 0;
#if PG_VERSION_NUM >= 100000
*indexPages = 0;
#endif
return;
}
@@ -110,7 +121,9 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
*indexTotalCost = costs.indexTotalCost;
*indexSelectivity = costs.indexSelectivity;
*indexCorrelation = costs.indexCorrelation;
#if PG_VERSION_NUM >= 100000
*indexPages = costs.numIndexPages;
#endif
}
/*
@@ -156,7 +169,7 @@ ivfflatvalidate(Oid opclassoid)
*
* See https://www.postgresql.org/docs/current/index-api.html
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(ivfflathandler);
PG_FUNCTION_INFO_V1(ivfflathandler);
Datum
ivfflathandler(PG_FUNCTION_ARGS)
{
@@ -178,8 +191,12 @@ ivfflathandler(PG_FUNCTION_ARGS)
amroutine->amstorage = false;
amroutine->amclusterable = false;
amroutine->ampredlocks = false;
#if PG_VERSION_NUM >= 100000
amroutine->amcanparallel = false;
#endif
#if PG_VERSION_NUM >= 110000
amroutine->amcaninclude = false;
#endif
#if PG_VERSION_NUM >= 130000
amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
@@ -212,9 +229,11 @@ ivfflathandler(PG_FUNCTION_ARGS)
amroutine->amrestrpos = NULL;
/* Interface functions to support parallel index scans */
#if PG_VERSION_NUM >= 100000
amroutine->amestimateparallelscan = NULL;
amroutine->aminitparallelscan = NULL;
amroutine->amparallelrescan = NULL;
#endif
PG_RETURN_POINTER(amroutine);
}

View File

@@ -3,27 +3,20 @@
#include "postgres.h"
#if PG_VERSION_NUM < 110000
#error "Requires PostgreSQL 11+"
#endif
#include "access/generic_xlog.h"
#include "access/reloptions.h"
#include "nodes/execnodes.h"
#include "port.h" /* for strtof() and random() */
#include "utils/sampling.h"
#include "utils/tuplesort.h"
#include "vector.h"
#if PG_VERSION_NUM >= 150000
#include "common/pg_prng.h"
#endif
#ifdef IVFFLAT_BENCH
#include "portability/instr_time.h"
#endif
#define IVFFLAT_MAX_DIM 2000
#if PG_VERSION_NUM < 90600
#error "Requires PostgreSQL 9.6+"
#endif
/* Support functions */
#define IVFFLAT_DISTANCE_PROC 1
@@ -44,9 +37,10 @@
/* Build phases */
/* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
#define PROGRESS_IVFFLAT_PHASE_KMEANS 2
#define PROGRESS_IVFFLAT_PHASE_SORT 3
#define PROGRESS_IVFFLAT_PHASE_LOAD 4
#define PROGRESS_IVFFLAT_PHASE_SAMPLE 2
#define PROGRESS_IVFFLAT_PHASE_KMEANS 3
#define PROGRESS_IVFFLAT_PHASE_SORT 4
#define PROGRESS_IVFFLAT_PHASE_LOAD 5
#define IVFFLAT_LIST_SIZE(_dim) (offsetof(IvfflatListData, center) + VECTOR_SIZE(_dim))
@@ -68,26 +62,21 @@
#define IvfflatBench(name, code) (code)
#endif
#if PG_VERSION_NUM >= 150000
#define RandomDouble() pg_prng_double(&pg_global_prng_state)
#define RandomInt() pg_prng_uint32(&pg_global_prng_state)
#else
#define RandomDouble() (((double) random()) / MAX_RANDOM_VALUE)
#define RandomInt() random()
#if PG_VERSION_NUM < 100000
#define ItemPointerGetBlockNumberNoCheck ItemPointerGetBlockNumber
#define ItemPointerGetOffsetNumberNoCheck ItemPointerGetOffsetNumber
#endif
/* Variables */
extern int ivfflat_probes;
/* Exported functions */
PGDLLEXPORT void _PG_init(void);
extern int ivfflat_bound;
typedef struct VectorArrayData
{
int length;
int maxlen;
int dim;
Vector *items;
Vector items[FLEXIBLE_ARRAY_MEMBER];
} VectorArrayData;
typedef VectorArrayData * VectorArray;
@@ -134,8 +123,6 @@ typedef struct IvfflatBuildState
#ifdef IVFFLAT_KMEANS_DEBUG
double inertia;
double *listSums;
int *listCounts;
#endif
/* Sampling */
@@ -147,9 +134,6 @@ typedef struct IvfflatBuildState
Tuplesortstate *sortstate;
TupleDesc tupdesc;
TupleTableSlot *slot;
/* Memory */
MemoryContext tmpCtx;
} IvfflatBuildState;
typedef struct IvfflatMetaPageData
@@ -182,19 +166,10 @@ typedef IvfflatListData * IvfflatList;
typedef struct IvfflatScanList
{
pairingheap_node ph_node;
BlockNumber startPage;
double distance;
} IvfflatScanList;
typedef struct IvfflatScanItem
{
pairingheap_node ph_node;
BlockNumber searchPage;
double distance;
ItemPointerData tid;
} IvfflatScanItem;
typedef struct IvfflatScanOpaqueData
{
int probes;
@@ -212,28 +187,19 @@ typedef struct IvfflatScanOpaqueData
FmgrInfo *normprocinfo;
Oid collation;
/* Items */
int maxItems;
int itemCount;
pairingheap *itemQueue;
IvfflatScanItem *items;
IvfflatScanItem **sortedItems;
/* Lists */
pairingheap *listQueue;
IvfflatScanList lists[FLEXIBLE_ARRAY_MEMBER]; /* must come last */
} IvfflatScanOpaqueData;
typedef IvfflatScanOpaqueData * IvfflatScanOpaque;
#define VECTOR_ARRAY_SIZE(_length, _dim) (sizeof(VectorArrayData) + (_length) * VECTOR_SIZE(_dim))
#define VECTOR_ARRAY_OFFSET(_arr, _offset) ((char*) (_arr)->items + (_offset) * VECTOR_SIZE((_arr)->dim))
#define VECTOR_ARRAY_SIZE(_length, _dim) (offsetof(VectorArrayData, items) + _length * VECTOR_SIZE(_dim))
#define VECTOR_ARRAY_OFFSET(_arr, _offset) ((char*) _arr + offsetof(VectorArrayData, items) + (_offset) * VECTOR_SIZE(_arr->dim))
#define VectorArrayGet(_arr, _offset) ((Vector *) VECTOR_ARRAY_OFFSET(_arr, _offset))
#define VectorArraySet(_arr, _offset, _val) memcpy(VECTOR_ARRAY_OFFSET(_arr, _offset), _val, VECTOR_SIZE((_arr)->dim))
#define VectorArraySet(_arr, _offset, _val) (memcpy(VECTOR_ARRAY_OFFSET(_arr, _offset), _val, VECTOR_SIZE(_arr->dim)))
/* Methods */
void _PG_init(void);
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 rel, uint16 procnum);
@@ -243,8 +209,7 @@ void IvfflatUpdateList(Relation index, GenericXLogState *state, ListInfo listIn
void IvfflatCommitBuffer(Buffer buf, GenericXLogState *state);
void IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum);
Buffer IvfflatNewBuffer(Relation index, ForkNumber forkNum);
void IvfflatInitPage(Buffer buf, Page page);
void IvfflatInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state);
void IvfflatInitPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state);
/* Index access methods */
IndexBuildResult *ivfflatbuild(Relation heap, Relation index, IndexInfo *indexInfo);
@@ -253,7 +218,9 @@ bool ivfflatinsert(Relation index, Datum *values, bool *isnull, ItemPointer hea
#if PG_VERSION_NUM >= 140000
,bool indexUnchanged
#endif
#if PG_VERSION_NUM >= 100000
,IndexInfo *indexInfo
#endif
);
IndexBulkDeleteResult *ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats, IndexBulkDeleteCallback callback, void *callback_state);
IndexBulkDeleteResult *ivfflatvacuumcleanup(IndexVacuumInfo *info, IndexBulkDeleteResult *stats);

View File

@@ -4,7 +4,6 @@
#include "ivfflat.h"
#include "storage/bufmgr.h"
#include "utils/memutils.h"
/*
* Find the list that minimizes the distance function
@@ -54,15 +53,24 @@ FindInsertPage(Relation rel, Datum *values, BlockNumber *insertPage, ListInfo *
}
}
/*
* Prepare to insert an index tuple
*/
static void
LoadInsertPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, BlockNumber insertPage)
{
*buf = ReadBuffer(index, insertPage);
LockBuffer(*buf, BUFFER_LOCK_EXCLUSIVE);
*state = GenericXLogStart(index);
*page = GenericXLogRegisterBuffer(*state, *buf, 0);
}
/*
* Insert a tuple into the index
*/
static void
InsertTuple(Relation rel, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel)
InsertTuple(Relation rel, IndexTuple itup, Relation heapRel, Datum *values)
{
IndexTuple itup;
Datum value;
FmgrInfo *normprocinfo;
Buffer buf;
Page page;
GenericXLogState *state;
@@ -71,42 +79,19 @@ InsertTuple(Relation rel, Datum *values, bool *isnull, ItemPointer heap_tid, Rel
ListInfo listInfo;
BlockNumber originalInsertPage;
/* Detoast once for all calls */
value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
/* Normalize if needed */
normprocinfo = IvfflatOptionalProcInfo(rel, IVFFLAT_NORM_PROC);
if (normprocinfo != NULL)
{
if (!IvfflatNormValue(normprocinfo, rel->rd_indcollation[0], &value, NULL))
return;
}
/* Find the insert page - sets the page and list info */
FindInsertPage(rel, values, &insertPage, &listInfo);
Assert(BlockNumberIsValid(insertPage));
originalInsertPage = insertPage;
/* Form tuple */
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)));
LoadInsertPage(rel, &buf, &page, &state, insertPage);
/* Find a page to insert the item */
for (;;)
while (PageGetFreeSpace(page) < itemsz)
{
buf = ReadBuffer(rel, insertPage);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
state = GenericXLogStart(rel);
page = GenericXLogRegisterBuffer(state, buf, 0);
if (PageGetFreeSpace(page) >= itemsz)
break;
insertPage = IvfflatPageGetOpaque(page)->nextblkno;
if (BlockNumberIsValid(insertPage))
@@ -114,50 +99,15 @@ InsertTuple(Relation rel, Datum *values, bool *isnull, ItemPointer heap_tid, Rel
/* Move to next page */
GenericXLogAbort(state);
UnlockReleaseBuffer(buf);
LoadInsertPage(rel, &buf, &page, &state, insertPage);
}
else
{
Buffer metabuf;
Buffer newbuf;
Page newpage;
/*
* From ReadBufferExtended: Caller is responsible for ensuring
* that only one backend tries to extend a relation at the same
* time!
*/
metabuf = ReadBuffer(rel, IVFFLAT_METAPAGE_BLKNO);
LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
/* Add a new page */
newbuf = IvfflatNewBuffer(rel, MAIN_FORKNUM);
newpage = GenericXLogRegisterBuffer(state, newbuf, GENERIC_XLOG_FULL_IMAGE);
IvfflatAppendPage(rel, &buf, &page, &state, MAIN_FORKNUM);
/* Init new page */
IvfflatInitPage(newbuf, newpage);
/* Update insert page */
insertPage = BufferGetBlockNumber(newbuf);
/* Update previous buffer */
IvfflatPageGetOpaque(page)->nextblkno = insertPage;
/* Commit */
MarkBufferDirty(newbuf);
MarkBufferDirty(buf);
GenericXLogFinish(state);
/* Unlock extend relation lock as early as possible */
UnlockReleaseBuffer(metabuf);
/* Unlock previous buffer */
UnlockReleaseBuffer(buf);
/* Prepare new buffer */
state = GenericXLogStart(rel);
buf = newbuf;
page = GenericXLogRegisterBuffer(state, buf, 0);
break;
insertPage = BufferGetBlockNumber(buf);
}
}
@@ -181,31 +131,36 @@ ivfflatinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid,
#if PG_VERSION_NUM >= 140000
,bool indexUnchanged
#endif
#if PG_VERSION_NUM >= 100000
,IndexInfo *indexInfo
#endif
)
{
MemoryContext oldCtx;
MemoryContext insertCtx;
IndexTuple itup;
Datum value;
FmgrInfo *normprocinfo;
/* Skip nulls */
if (isnull[0])
return false;
/*
* Use memory context since detoast, IvfflatNormValue, and
* index_form_tuple can allocate
*/
insertCtx = AllocSetContextCreate(CurrentMemoryContext,
"Ivfflat insert temporary context",
ALLOCSET_DEFAULT_SIZES);
oldCtx = MemoryContextSwitchTo(insertCtx);
value = values[0];
/* Insert tuple */
InsertTuple(index, values, isnull, heap_tid, heap);
/* Normalize if needed */
normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
if (normprocinfo != NULL)
{
if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, NULL))
return false;
}
/* Delete memory context */
MemoryContextSwitchTo(oldCtx);
MemoryContextDelete(insertCtx);
itup = index_form_tuple(RelationGetDescr(index), &value, isnull);
itup->t_tid = *heap_tid;
InsertTuple(index, itup, heap, &value);
pfree(itup);
/* Clean up if we allocated a new value */
if (value != values[0])
pfree(DatumGetPointer(value));
return false;
}

View File

@@ -16,7 +16,7 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
FmgrInfo *procinfo;
Oid collation;
int i;
int64 j;
int j;
double distance;
double sum;
double choice;
@@ -29,7 +29,7 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
collation = index->rd_indcollation[0];
/* Choose an initial center uniformly at random */
VectorArraySet(centers, 0, VectorArrayGet(samples, RandomInt() % samples->length));
VectorArraySet(centers, 0, VectorArrayGet(samples, random() % samples->length));
centers->length++;
for (j = 0; j < numSamples; j++)
@@ -66,7 +66,7 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
break;
/* Choose new center using weighted probability distribution. */
choice = sum * RandomDouble();
choice = sum * (((double) random()) / MAX_RANDOM_VALUE);
for (j = 0; j < numSamples - 1; j++)
{
choice -= weight[j];
@@ -145,7 +145,7 @@ QuickCenters(Relation index, VectorArray samples, VectorArray centers)
vec->dim = dimensions;
for (j = 0; j < dimensions; j++)
vec->x[j] = RandomDouble();
vec->x[j] = ((double) random()) / MAX_RANDOM_VALUE;
/* Normalize if needed (only needed for random centers) */
if (normprocinfo != NULL)
@@ -172,8 +172,8 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
Vector *vec;
Vector *newCenter;
int iteration;
int64 j;
int64 k;
int j;
int k;
int dimensions = centers->dim;
int numCenters = centers->maxlen;
int numSamples = samples->length;
@@ -217,10 +217,6 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
errmsg("memory required is %zu MB, maintenance_work_mem is %d MB",
totalSize / (1024 * 1024) + 1, maintenance_work_mem / 1024)));
/* Ensure indexing does not overflow */
if (numCenters * numCenters > INT_MAX)
elog(ERROR, "Indexing overflow detected. Please report a bug.");
/* Set support functions */
procinfo = index_getprocinfo(index, 1, IVFFLAT_KMEANS_DISTANCE_PROC);
normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_KMEANS_NORM_PROC);
@@ -233,7 +229,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
lowerBound = palloc_extended(lowerBoundSize, MCXT_ALLOC_HUGE);
upperBound = palloc(upperBoundSize);
s = palloc(sSize);
halfcdist = palloc_extended(halfcdistSize, MCXT_ALLOC_HUGE);
halfcdist = palloc(halfcdistSize);
newcdist = palloc(newcdistSize);
newCenters = VectorArrayInit(numCenters, dimensions);
@@ -253,6 +249,8 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
minDistance = DBL_MAX;
closestCenter = -1;
vec = VectorArrayGet(samples, j);
/* Find closest center */
for (k = 0; k < numCenters; k++)
{
@@ -405,7 +403,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
{
/* TODO Handle empty centers properly */
for (k = 0; k < dimensions; k++)
vec->x[k] = RandomDouble();
vec->x[k] = ((double) random()) / MAX_RANDOM_VALUE;
}
/* Normalize if needed */
@@ -443,7 +441,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
break;
}
VectorArrayFree(newCenters);
pfree(newCenters);
pfree(centerCounts);
pfree(closestCenters);
pfree(lowerBound);

View File

@@ -1,41 +1,30 @@
#include "postgres.h"
#include <float.h>
#include "access/relscan.h"
#include "ivfflat.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "storage/bufmgr.h"
#if PG_VERSION_NUM >= 110000
#include "catalog/pg_operator_d.h"
#include "catalog/pg_type_d.h"
#else
#include "catalog/pg_operator.h"
#include "catalog/pg_type.h"
#endif
/*
* Compare list distances
*/
static int
CompareLists(const pairingheap_node *a, const pairingheap_node *b, void *arg)
CompareLists(const void *a, const void *b)
{
if (((const IvfflatScanList *) a)->distance > ((const IvfflatScanList *) b)->distance)
double diff = (((IvfflatScanList *) a)->distance - ((IvfflatScanList *) b)->distance);
if (diff > 0)
return 1;
if (((const IvfflatScanList *) a)->distance < ((const IvfflatScanList *) b)->distance)
return -1;
return 0;
}
/*
* Compare item distances
*/
static int
CompareItems(const pairingheap_node *a, const pairingheap_node *b, void *arg)
{
if (((const IvfflatScanItem *) a)->distance > ((const IvfflatScanItem *) b)->distance)
return 1;
if (((const IvfflatScanItem *) a)->distance < ((const IvfflatScanItem *) b)->distance)
if (diff < 0)
return -1;
return 0;
@@ -56,8 +45,6 @@ GetScanLists(IndexScanDesc scan, Datum value)
int listCount = 0;
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
double distance;
IvfflatScanList *scanlist;
double maxDistance = DBL_MAX;
/* Search all list pages */
while (BlockNumberIsValid(nextblkno))
@@ -75,39 +62,22 @@ GetScanLists(IndexScanDesc scan, Datum value)
/* Use procinfo from the index instead of scan key for performance */
distance = DatumGetFloat8(FunctionCall2Coll(so->procinfo, so->collation, PointerGetDatum(&list->center), value));
if (listCount < so->probes)
{
scanlist = &so->lists[listCount];
scanlist->startPage = list->startPage;
scanlist->distance = distance;
listCount++;
/* Add to heap */
pairingheap_add(so->listQueue, &scanlist->ph_node);
/* Calculate max distance */
if (listCount == so->probes)
maxDistance = ((IvfflatScanList *) pairingheap_first(so->listQueue))->distance;
}
else if (distance < maxDistance)
{
/* Remove */
scanlist = (IvfflatScanList *) pairingheap_remove_first(so->listQueue);
/* Reuse */
scanlist->startPage = list->startPage;
scanlist->distance = distance;
pairingheap_add(so->listQueue, &scanlist->ph_node);
/* Update max distance */
maxDistance = ((IvfflatScanList *) pairingheap_first(so->listQueue))->distance;
}
so->lists[listCount].startPage = list->startPage;
so->lists[listCount].distance = distance;
listCount++;
}
nextblkno = IvfflatPageGetOpaque(cpage)->nextblkno;
UnlockReleaseBuffer(cbuf);
}
/* Sort by distance */
/* TODO Use heap for performance */
qsort(so->lists, listCount, sizeof(IvfflatScanList), CompareLists);
if (so->probes > listCount)
so->probes = listCount;
}
/*
@@ -125,11 +95,14 @@ GetScanItems(IndexScanDesc scan, Datum value)
OffsetNumber maxoffno;
Datum datum;
bool isnull;
TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
int i;
double distance;
IvfflatScanItem *scanitem;
double maxDistance = DBL_MAX;
TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
#if PG_VERSION_NUM >= 120000
TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
#else
TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc);
#endif
/*
* Reuse same set of shared buffers for scan
@@ -138,10 +111,14 @@ GetScanItems(IndexScanDesc scan, Datum value)
*/
BufferAccessStrategy bas = GetAccessStrategy(BAS_BULKREAD);
/* Set the max number of results */
if (ivfflat_bound > 0)
tuplesort_set_bound(so->sortstate, ivfflat_bound);
/* Search closest probes lists */
while (!pairingheap_is_empty(so->listQueue))
for (i = 0; i < so->probes; i++)
{
searchPage = ((IvfflatScanList *) pairingheap_remove_first(so->listQueue))->startPage;
searchPage = so->lists[i].startPage;
/* Search all entry pages for list */
while (BlockNumberIsValid(searchPage))
@@ -155,40 +132,25 @@ GetScanItems(IndexScanDesc scan, Datum value)
{
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offno));
datum = index_getattr(itup, 1, tupdesc, &isnull);
distance = DatumGetFloat8(FunctionCall2Coll(so->procinfo, so->collation, datum, value));
if (so->itemCount < so->maxItems)
{
scanitem = &so->items[so->itemCount];
scanitem->searchPage = searchPage;
scanitem->tid = itup->t_tid;
scanitem->distance = distance;
so->itemCount++;
/*
* Add virtual tuple
*
* Use procinfo from the index instead of scan key for
* performance
*/
ExecClearTuple(slot);
slot->tts_values[0] = FunctionCall2Coll(so->procinfo, so->collation, datum, value);
slot->tts_isnull[0] = false;
slot->tts_values[1] = Int32GetDatum((int) ItemPointerGetBlockNumberNoCheck(&itup->t_tid));
slot->tts_isnull[1] = false;
slot->tts_values[2] = Int32GetDatum((int) ItemPointerGetOffsetNumberNoCheck(&itup->t_tid));
slot->tts_isnull[2] = false;
slot->tts_values[3] = Int32GetDatum((int) searchPage);
slot->tts_isnull[3] = false;
ExecStoreVirtualTuple(slot);
/* Add to heap */
pairingheap_add(so->itemQueue, &scanitem->ph_node);
/* Calculate max distance */
if (so->itemCount == so->maxItems)
{
maxDistance = ((IvfflatScanItem *) pairingheap_first(so->itemQueue))->distance;
scanitem = &so->items[so->itemCount];
}
}
else if (distance < maxDistance)
{
/* Reuse */
scanitem->searchPage = searchPage;
scanitem->tid = itup->t_tid;
scanitem->distance = distance;
pairingheap_add(so->itemQueue, &scanitem->ph_node);
/* Remove */
scanitem = (IvfflatScanItem *) pairingheap_remove_first(so->itemQueue);
/* Update max distance */
maxDistance = ((IvfflatScanItem *) pairingheap_first(so->itemQueue))->distance;
}
tuplesort_puttupleslot(so->sortstate, slot);
}
searchPage = IvfflatPageGetOpaque(page)->nextblkno;
@@ -197,10 +159,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
}
}
for (i = 0; i < so->itemCount; i++)
so->sortedItems[i] = (IvfflatScanItem *) pairingheap_remove_first(so->itemQueue);
Assert(pairingheap_is_empty(so->itemQueue));
tuplesort_performsort(so->sortstate);
}
/*
@@ -212,31 +171,46 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
IndexScanDesc scan;
IvfflatScanOpaque so;
int lists;
int probes = ivfflat_probes;
AttrNumber attNums[] = {1};
Oid sortOperators[] = {Float8LessOperator};
Oid sortCollations[] = {InvalidOid};
bool nullsFirstFlags[] = {false};
scan = RelationGetIndexScan(index, nkeys, norderbys);
lists = IvfflatGetLists(scan->indexRelation);
if (probes > lists)
probes = lists;
so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + probes * sizeof(IvfflatScanList));
so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + lists * sizeof(IvfflatScanList));
so->buf = InvalidBuffer;
so->first = true;
so->probes = probes;
/* Set support functions */
so->procinfo = index_getprocinfo(index, 1, IVFFLAT_DISTANCE_PROC);
so->normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
so->collation = index->rd_indcollation[0];
so->listQueue = pairingheap_allocate(CompareLists, scan);
/* Create tuple description for sorting */
#if PG_VERSION_NUM >= 120000
so->tupdesc = CreateTemplateTupleDesc(4);
#else
so->tupdesc = CreateTemplateTupleDesc(4, false);
#endif
TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "blkno", INT4OID, -1, 0);
TupleDescInitEntry(so->tupdesc, (AttrNumber) 3, "offset", INT4OID, -1, 0);
TupleDescInitEntry(so->tupdesc, (AttrNumber) 4, "indexblkno", INT4OID, -1, 0);
so->maxItems = 10;
so->itemCount = 0;
so->itemQueue = pairingheap_allocate(CompareItems, scan);
so->items = palloc(sizeof(IvfflatScanItem) * (so->maxItems + 1));
so->sortedItems = palloc(sizeof(IvfflatScanItem *) * so->maxItems);
/* Prep sort */
#if PG_VERSION_NUM >= 110000
so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
#else
so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, false);
#endif
#if PG_VERSION_NUM >= 120000
so->slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsMinimalTuple);
#else
so->slot = MakeSingleTupleTableSlot(so->tupdesc);
#endif
scan->opaque = so;
@@ -251,10 +225,13 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
{
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
#if PG_VERSION_NUM >= 130000
if (!so->first)
tuplesort_reset(so->sortstate);
#endif
so->first = true;
pairingheap_reset(so->listQueue);
pairingheap_reset(so->itemQueue);
so->itemCount = 0;
so->probes = ivfflat_probes;
if (keys && scan->numberOfKeys > 0)
memmove(scan->keyData, keys, scan->numberOfKeys * sizeof(ScanKeyData));
@@ -281,9 +258,6 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
{
Datum value;
/* Count index scan for stats */
pgstat_count_index_scan(scan->indexRelation);
/* Safety check */
if (scan->orderByData == NULL)
elog(ERROR, "cannot scan ivfflat index without order");
@@ -294,10 +268,6 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
value = scan->orderByData->sk_argument;
/* Value should not be compressed or toasted */
Assert(!VARATT_IS_COMPRESSED(DatumGetPointer(value)));
Assert(!VARATT_IS_EXTENDED(DatumGetPointer(value)));
if (so->normprocinfo != NULL)
{
/* No items will match if normalization fails */
@@ -314,18 +284,20 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
pfree(DatumGetPointer(value));
}
if (so->itemCount > 0)
#if PG_VERSION_NUM >= 100000
if (tuplesort_gettupleslot(so->sortstate, true, false, so->slot, NULL))
#else
if (tuplesort_gettupleslot(so->sortstate, true, so->slot, NULL))
#endif
{
IvfflatScanItem *scanitem;
so->itemCount--;
scanitem = so->sortedItems[so->itemCount];
BlockNumber blkno = DatumGetInt32(slot_getattr(so->slot, 2, &so->isnull));
OffsetNumber offset = DatumGetInt32(slot_getattr(so->slot, 3, &so->isnull));
BlockNumber indexblkno = DatumGetInt32(slot_getattr(so->slot, 4, &so->isnull));
#if PG_VERSION_NUM >= 120000
scan->xs_heaptid = scanitem->tid;
ItemPointerSet(&scan->xs_heaptid, blkno, offset);
#else
scan->xs_ctup.t_self = scanitem->tid;
ItemPointerSet(&scan->xs_ctup.t_self, blkno, offset);
#endif
if (BufferIsValid(so->buf))
@@ -337,7 +309,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
*
* https://www.postgresql.org/docs/current/index-locking.html
*/
so->buf = ReadBuffer(scan->indexRelation, scanitem->searchPage);
so->buf = ReadBuffer(scan->indexRelation, indexblkno);
scan->xs_recheckorderby = false;
return true;
@@ -358,11 +330,7 @@ ivfflatendscan(IndexScanDesc scan)
if (BufferIsValid(so->buf))
ReleaseBuffer(so->buf);
pairingheap_free(so->listQueue);
pairingheap_free(so->itemQueue);
pfree(so->items);
pfree(so->sortedItems);
tuplesort_end(so->sortstate);
pfree(so);
scan->opaque = NULL;

View File

@@ -10,25 +10,14 @@
VectorArray
VectorArrayInit(int maxlen, int dimensions)
{
VectorArray res = palloc(sizeof(VectorArrayData));
VectorArray res = palloc0(VECTOR_ARRAY_SIZE(maxlen, dimensions));
res->length = 0;
res->maxlen = maxlen;
res->dim = dimensions;
res->items = palloc_extended(maxlen * VECTOR_SIZE(dimensions), MCXT_ALLOC_ZERO | MCXT_ALLOC_HUGE);
return res;
}
/*
* Free a vector array
*/
void
VectorArrayFree(VectorArray arr)
{
pfree(arr->items);
pfree(arr);
}
/*
* Print vector array - useful for debugging
*/
@@ -86,7 +75,7 @@ IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * resul
if (norm > 0)
{
v = DatumGetVector(*value);
v = (Vector *) DatumGetPointer(*value);
if (result == NULL)
result = InitVector(v->dim);
@@ -118,22 +107,13 @@ IvfflatNewBuffer(Relation index, ForkNumber forkNum)
* Init page
*/
void
IvfflatInitPage(Buffer buf, Page page)
{
PageInit(page, BufferGetPageSize(buf), sizeof(IvfflatPageOpaqueData));
IvfflatPageGetOpaque(page)->nextblkno = InvalidBlockNumber;
IvfflatPageGetOpaque(page)->page_id = IVFFLAT_PAGE_ID;
}
/*
* Init and register page
*/
void
IvfflatInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state)
IvfflatInitPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state)
{
*state = GenericXLogStart(index);
*page = GenericXLogRegisterBuffer(*state, *buf, GENERIC_XLOG_FULL_IMAGE);
IvfflatInitPage(*buf, *page);
PageInit(*page, BufferGetPageSize(*buf), sizeof(IvfflatPageOpaqueData));
IvfflatPageGetOpaque(*page)->nextblkno = InvalidBlockNumber;
IvfflatPageGetOpaque(*page)->page_id = IVFFLAT_PAGE_ID;
}
/*
@@ -155,27 +135,17 @@ IvfflatCommitBuffer(Buffer buf, GenericXLogState *state)
void
IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum)
{
/* Get new buffer */
Buffer newbuf = IvfflatNewBuffer(index, forkNum);
Page newpage = GenericXLogRegisterBuffer(*state, newbuf, GENERIC_XLOG_FULL_IMAGE);
Buffer prevbuf = *buf;
/* Update the previous buffer */
IvfflatPageGetOpaque(*page)->nextblkno = BufferGetBlockNumber(newbuf);
/* Get new buffer */
*buf = IvfflatNewBuffer(index, forkNum);
/* Update and commit previous buffer */
IvfflatPageGetOpaque(*page)->nextblkno = BufferGetBlockNumber(*buf);
IvfflatCommitBuffer(prevbuf, *state);
/* Init new page */
IvfflatInitPage(newbuf, newpage);
/* Commit */
MarkBufferDirty(*buf);
MarkBufferDirty(newbuf);
GenericXLogFinish(*state);
/* Unlock */
UnlockReleaseBuffer(*buf);
*state = GenericXLogStart(index);
*page = GenericXLogRegisterBuffer(*state, newbuf, GENERIC_XLOG_FULL_IMAGE);
*buf = newbuf;
IvfflatInitPage(index, buf, page, state);
}
/*

View File

@@ -143,11 +143,6 @@ ivfflatvacuumcleanup(IndexVacuumInfo *info, IndexBulkDeleteResult *stats)
{
Relation rel = info->index;
if (info->analyze_only)
return stats;
/* stats is NULL if ambulkdelete not called */
/* OK to return NULL if index not changed */
if (stats == NULL)
return NULL;

View File

@@ -13,20 +13,13 @@
#include "utils/numeric.h"
#if PG_VERSION_NUM >= 120000
#include "common/shortest_dec.h"
#include "utils/float.h"
#else
#include <float.h>
#endif
#if PG_VERSION_NUM < 130000
#define TYPALIGN_DOUBLE 'd'
#define TYPALIGN_INT 'i'
#endif
#define STATE_DIMS(x) (ARR_DIMS(x)[0] - 1)
#define CreateStateDatums(dim) palloc(sizeof(Datum) * (dim + 1))
PG_MODULE_MAGIC;
/*
@@ -86,30 +79,6 @@ CheckElement(float value)
errmsg("infinite value not allowed in vector")));
}
/*
* Check state array
*/
static float8 *
CheckStateArray(ArrayType *statearray, const char *caller)
{
if (ARR_NDIM(statearray) != 1 ||
ARR_DIMS(statearray)[0] < 1 ||
ARR_HASNULL(statearray) ||
ARR_ELEMTYPE(statearray) != FLOAT8OID)
elog(ERROR, "%s: expected state array", caller);
return (float8 *) ARR_DATA_PTR(statearray);
}
#if PG_VERSION_NUM < 120003
static pg_noinline void
float_overflow_error(void)
{
ereport(ERROR,
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
errmsg("value out of range: overflow")));
}
#endif
/*
* Print vector - useful for debugging
*/
@@ -137,14 +106,14 @@ PrintVector(char *msg, Vector * vector)
/*
* Convert textual representation to internal representation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_in);
PG_FUNCTION_INFO_V1(vector_in);
Datum
vector_in(PG_FUNCTION_ARGS)
{
char *str = PG_GETARG_CSTRING(0);
int32 typmod = PG_GETARG_INT32(2);
int i;
float x[VECTOR_MAX_DIM];
double x[VECTOR_MAX_DIM];
int dim = 0;
char *pt;
char *stringEnd;
@@ -167,8 +136,7 @@ vector_in(PG_FUNCTION_ARGS)
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("vector cannot have more than %d dimensions", VECTOR_MAX_DIM)));
/* Use strtof like float4in to avoid a double-rounding problem */
x[dim] = strtof(pt, &stringEnd);
x[dim] = strtod(pt, &stringEnd);
CheckElement(x[dim]);
dim++;
@@ -214,68 +182,35 @@ vector_in(PG_FUNCTION_ARGS)
/*
* Convert internal representation to textual representation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_out);
PG_FUNCTION_INFO_V1(vector_out);
Datum
vector_out(PG_FUNCTION_ARGS)
{
Vector *vector = PG_GETARG_VECTOR_P(0);
StringInfoData buf;
int dim = vector->dim;
char *buf;
char *ptr;
int i;
int n;
#if PG_VERSION_NUM < 120000
int ndig = FLT_DIG + extra_float_digits;
initStringInfo(&buf);
if (ndig < 1)
ndig = 1;
#define FLOAT_SHORTEST_DECIMAL_LEN (ndig + 10)
#endif
/*
* Need:
*
* dim * (FLOAT_SHORTEST_DECIMAL_LEN - 1) bytes for
* float_to_shortest_decimal_bufn
*
* dim - 1 bytes for separator
*
* 3 bytes for [, ], and \0
*/
buf = (char *) palloc(FLOAT_SHORTEST_DECIMAL_LEN * dim + 2);
ptr = buf;
*ptr = '[';
ptr++;
appendStringInfoChar(&buf, '[');
for (i = 0; i < dim; i++)
{
if (i > 0)
{
*ptr = ',';
ptr++;
}
appendStringInfoString(&buf, ",");
#if PG_VERSION_NUM >= 120000
n = float_to_shortest_decimal_bufn(vector->x[i], ptr);
#else
n = sprintf(ptr, "%.*g", ndig, vector->x[i]);
#endif
ptr += n;
appendStringInfoString(&buf, float8out_internal(vector->x[i]));
}
*ptr = ']';
ptr++;
*ptr = '\0';
appendStringInfoChar(&buf, ']');
PG_FREE_IF_COPY(vector, 0);
PG_RETURN_CSTRING(buf);
PG_RETURN_CSTRING(buf.data);
}
/*
* Convert type modifier
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_typmod_in);
PG_FUNCTION_INFO_V1(vector_typmod_in);
Datum
vector_typmod_in(PG_FUNCTION_ARGS)
{
@@ -306,7 +241,7 @@ vector_typmod_in(PG_FUNCTION_ARGS)
/*
* Convert external binary representation to internal representation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_recv);
PG_FUNCTION_INFO_V1(vector_recv);
Datum
vector_recv(PG_FUNCTION_ARGS)
{
@@ -338,7 +273,7 @@ vector_recv(PG_FUNCTION_ARGS)
/*
* Convert internal representation to the external binary representation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_send);
PG_FUNCTION_INFO_V1(vector_send);
Datum
vector_send(PG_FUNCTION_ARGS)
{
@@ -358,7 +293,7 @@ vector_send(PG_FUNCTION_ARGS)
/*
* Convert vector to vector
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector);
PG_FUNCTION_INFO_V1(vector);
Datum
vector(PG_FUNCTION_ARGS)
{
@@ -373,7 +308,7 @@ vector(PG_FUNCTION_ARGS)
/*
* Convert array to vector
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(array_to_vector);
PG_FUNCTION_INFO_V1(array_to_vector);
Datum
array_to_vector(PG_FUNCTION_ARGS)
{
@@ -416,7 +351,7 @@ array_to_vector(PG_FUNCTION_ARGS)
else if (ARR_ELEMTYPE(array) == FLOAT4OID)
result->x[i] = DatumGetFloat4(elemsp[i]);
else if (ARR_ELEMTYPE(array) == NUMERICOID)
result->x[i] = DatumGetFloat4(DirectFunctionCall1(numeric_float4, elemsp[i]));
result->x[i] = DatumGetFloat4(DirectFunctionCall1(numeric_float4, NumericGetDatum(elemsp[i])));
else
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
@@ -431,7 +366,7 @@ array_to_vector(PG_FUNCTION_ARGS)
/*
* Convert vector to float4[]
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_to_float4);
PG_FUNCTION_INFO_V1(vector_to_float4);
Datum
vector_to_float4(PG_FUNCTION_ARGS)
{
@@ -454,23 +389,20 @@ vector_to_float4(PG_FUNCTION_ARGS)
/*
* Get the L2 distance between vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(l2_distance);
PG_FUNCTION_INFO_V1(l2_distance);
Datum
l2_distance(PG_FUNCTION_ARGS)
{
Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0;
double diff;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
{
diff = ax[i] - bx[i];
diff = a->x[i] - b->x[i];
distance += diff * diff;
}
@@ -481,23 +413,20 @@ l2_distance(PG_FUNCTION_ARGS)
* Get the L2 squared distance between vectors
* This saves a sqrt calculation
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_l2_squared_distance);
PG_FUNCTION_INFO_V1(vector_l2_squared_distance);
Datum
vector_l2_squared_distance(PG_FUNCTION_ARGS)
{
Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0;
double diff;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
{
diff = ax[i] - bx[i];
diff = a->x[i] - b->x[i];
distance += diff * diff;
}
@@ -507,21 +436,18 @@ vector_l2_squared_distance(PG_FUNCTION_ARGS)
/*
* Get the inner product of two vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(inner_product);
PG_FUNCTION_INFO_V1(inner_product);
Datum
inner_product(PG_FUNCTION_ARGS)
{
Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
distance += ax[i] * bx[i];
distance += a->x[i] * b->x[i];
PG_RETURN_FLOAT8(distance);
}
@@ -529,21 +455,18 @@ inner_product(PG_FUNCTION_ARGS)
/*
* Get the negative inner product of two vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_negative_inner_product);
PG_FUNCTION_INFO_V1(vector_negative_inner_product);
Datum
vector_negative_inner_product(PG_FUNCTION_ARGS)
{
Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
distance += ax[i] * bx[i];
distance += a->x[i] * b->x[i];
PG_RETURN_FLOAT8(distance * -1);
}
@@ -551,30 +474,26 @@ vector_negative_inner_product(PG_FUNCTION_ARGS)
/*
* Get the cosine distance between two vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(cosine_distance);
PG_FUNCTION_INFO_V1(cosine_distance);
Datum
cosine_distance(PG_FUNCTION_ARGS)
{
Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0;
double norma = 0.0;
double normb = 0.0;
CheckDims(a, b);
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
{
distance += ax[i] * bx[i];
norma += ax[i] * ax[i];
normb += bx[i] * bx[i];
distance += a->x[i] * b->x[i];
norma += a->x[i] * a->x[i];
normb += b->x[i] * b->x[i];
}
/* Use sqrt(a * b) over sqrt(a) * sqrt(b) */
PG_RETURN_FLOAT8(1 - (distance / sqrt(norma * normb)));
PG_RETURN_FLOAT8(1 - (distance / (sqrt(norma) * sqrt(normb))));
}
/*
@@ -582,7 +501,7 @@ cosine_distance(PG_FUNCTION_ARGS)
* Currently uses angular distance since needs to satisfy triangle inequality
* Assumes inputs are unit vectors (skips norm)
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_spherical_distance);
PG_FUNCTION_INFO_V1(vector_spherical_distance);
Datum
vector_spherical_distance(PG_FUNCTION_ARGS)
{
@@ -592,7 +511,6 @@ 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];
@@ -608,7 +526,7 @@ vector_spherical_distance(PG_FUNCTION_ARGS)
/*
* Get the dimensions of a vector
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_dims);
PG_FUNCTION_INFO_V1(vector_dims);
Datum
vector_dims(PG_FUNCTION_ARGS)
{
@@ -620,17 +538,15 @@ vector_dims(PG_FUNCTION_ARGS)
/*
* Get the L2 norm of a vector
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_norm);
PG_FUNCTION_INFO_V1(vector_norm);
Datum
vector_norm(PG_FUNCTION_ARGS)
{
Vector *a = PG_GETARG_VECTOR_P(0);
float *ax = a->x;
double norm = 0.0;
/* Auto-vectorized */
for (int i = 0; i < a->dim; i++)
norm += ax[i] * ax[i];
norm += a->x[i] * a->x[i];
PG_RETURN_FLOAT8(sqrt(norm));
}
@@ -638,25 +554,20 @@ vector_norm(PG_FUNCTION_ARGS)
/*
* Add vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_add);
PG_FUNCTION_INFO_V1(vector_add);
Datum
vector_add(PG_FUNCTION_ARGS)
{
Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
Vector *result;
float *rx;
int i;
CheckDims(a, b);
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];
for (i = 0; i < a->dim; i++)
result->x[i] = a->x[i] + b->x[i];
PG_RETURN_POINTER(result);
}
@@ -664,25 +575,20 @@ vector_add(PG_FUNCTION_ARGS)
/*
* Subtract vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_sub);
PG_FUNCTION_INFO_V1(vector_sub);
Datum
vector_sub(PG_FUNCTION_ARGS)
{
Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
Vector *result;
float *rx;
int i;
CheckDims(a, b);
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];
for (i = 0; i < a->dim; i++)
result->x[i] = a->x[i] - b->x[i];
PG_RETURN_POINTER(result);
}
@@ -711,7 +617,7 @@ vector_cmp_internal(Vector * a, Vector * b)
/*
* Less than
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_lt);
PG_FUNCTION_INFO_V1(vector_lt);
Datum
vector_lt(PG_FUNCTION_ARGS)
{
@@ -724,7 +630,7 @@ vector_lt(PG_FUNCTION_ARGS)
/*
* Less than or equal
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_le);
PG_FUNCTION_INFO_V1(vector_le);
Datum
vector_le(PG_FUNCTION_ARGS)
{
@@ -737,7 +643,7 @@ vector_le(PG_FUNCTION_ARGS)
/*
* Equal
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_eq);
PG_FUNCTION_INFO_V1(vector_eq);
Datum
vector_eq(PG_FUNCTION_ARGS)
{
@@ -750,7 +656,7 @@ vector_eq(PG_FUNCTION_ARGS)
/*
* Not equal
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_ne);
PG_FUNCTION_INFO_V1(vector_ne);
Datum
vector_ne(PG_FUNCTION_ARGS)
{
@@ -763,7 +669,7 @@ vector_ne(PG_FUNCTION_ARGS)
/*
* Greater than or equal
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_ge);
PG_FUNCTION_INFO_V1(vector_ge);
Datum
vector_ge(PG_FUNCTION_ARGS)
{
@@ -776,7 +682,7 @@ vector_ge(PG_FUNCTION_ARGS)
/*
* Greater than
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_gt);
PG_FUNCTION_INFO_V1(vector_gt);
Datum
vector_gt(PG_FUNCTION_ARGS)
{
@@ -789,7 +695,7 @@ vector_gt(PG_FUNCTION_ARGS)
/*
* Compare vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_cmp);
PG_FUNCTION_INFO_V1(vector_cmp);
Datum
vector_cmp(PG_FUNCTION_ARGS)
{
@@ -798,167 +704,3 @@ vector_cmp(PG_FUNCTION_ARGS)
PG_RETURN_INT32(vector_cmp_internal(a, b));
}
/*
* Accumulate vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_accum);
Datum
vector_accum(PG_FUNCTION_ARGS)
{
ArrayType *statearray = PG_GETARG_ARRAYTYPE_P(0);
Vector *newval = PG_GETARG_VECTOR_P(1);
float8 *statevalues;
int16 dim;
bool newarr;
float8 n;
Datum *statedatums;
float *x = newval->x;
ArrayType *result;
/* Check array before using */
statevalues = CheckStateArray(statearray, "vector_accum");
dim = STATE_DIMS(statearray);
newarr = dim == 0;
if (newarr)
dim = newval->dim;
else
CheckExpectedDim(dim, newval->dim);
n = statevalues[0] + 1.0;
statedatums = CreateStateDatums(dim);
statedatums[0] = Float8GetDatumFast(n);
if (newarr)
{
for (int i = 0; i < dim; i++)
statedatums[i + 1] = Float8GetDatumFast((double) x[i]);
}
else
{
for (int i = 0; i < dim; i++)
{
double v = statevalues[i + 1] + x[i];
if (isinf(v))
float_overflow_error();
statedatums[i + 1] = Float8GetDatumFast(v);
}
}
/* Use float8 array like float4_accum */
result = construct_array(statedatums, dim + 1,
FLOAT8OID,
sizeof(float8), FLOAT8PASSBYVAL, TYPALIGN_DOUBLE);
pfree(statedatums);
PG_RETURN_ARRAYTYPE_P(result);
}
/*
* Combine vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_combine);
Datum
vector_combine(PG_FUNCTION_ARGS)
{
ArrayType *statearray1 = PG_GETARG_ARRAYTYPE_P(0);
ArrayType *statearray2 = PG_GETARG_ARRAYTYPE_P(1);
float8 *statevalues1;
float8 *statevalues2;
float8 n;
float8 n1;
float8 n2;
int16 dim;
Datum *statedatums;
ArrayType *result;
/* Check arrays before using */
statevalues1 = CheckStateArray(statearray1, "vector_combine");
statevalues2 = CheckStateArray(statearray2, "vector_combine");
n1 = statevalues1[0];
n2 = statevalues2[0];
if (n1 == 0.0)
{
n = n2;
dim = STATE_DIMS(statearray2);
statedatums = CreateStateDatums(dim);
for (int i = 1; i <= dim; i++)
statedatums[i] = Float8GetDatumFast(statevalues2[i]);
}
else if (n2 == 0.0)
{
n = n1;
dim = STATE_DIMS(statearray1);
statedatums = CreateStateDatums(dim);
for (int i = 1; i <= dim; i++)
statedatums[i] = Float8GetDatumFast(statevalues1[i]);
}
else
{
n = n1 + n2;
dim = STATE_DIMS(statearray1);
CheckExpectedDim(dim, STATE_DIMS(statearray2));
statedatums = CreateStateDatums(dim);
for (int i = 1; i <= dim; i++)
{
double v = statevalues1[i] + statevalues2[i];
if (isinf(v))
float_overflow_error();
statedatums[i] = Float8GetDatumFast(v);
}
}
statedatums[0] = Float8GetDatumFast(n);
result = construct_array(statedatums, dim + 1,
FLOAT8OID,
sizeof(float8), FLOAT8PASSBYVAL, TYPALIGN_DOUBLE);
pfree(statedatums);
PG_RETURN_ARRAYTYPE_P(result);
}
/*
* Average vectors
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_avg);
Datum
vector_avg(PG_FUNCTION_ARGS)
{
ArrayType *statearray = PG_GETARG_ARRAYTYPE_P(0);
float8 *statevalues;
float8 n;
uint16 dim;
Vector *result;
float v;
/* Check array before using */
statevalues = CheckStateArray(statearray, "vector_avg");
n = statevalues[0];
/* SQL defines AVG of no values to be NULL */
if (n == 0.0)
PG_RETURN_NULL();
/* Create vector */
dim = STATE_DIMS(statearray);
result = InitVector(dim);
for (int i = 0; i < dim; i++)
{
v = statevalues[i + 1] / n;
CheckElement(v);
result->x[i] = v;
}
PG_RETURN_POINTER(result);
}

View File

@@ -3,11 +3,7 @@
#include "postgres.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
#define VECTOR_MAX_DIM 16000
#define VECTOR_MAX_DIM 1024
#define VECTOR_SIZE(_dim) (offsetof(Vector, x) + sizeof(float)*(_dim))
#define DatumGetVector(x) ((Vector *) PG_DETOAST_DATUM(x))

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT ARRAY[1,2,3]::vector;
array
---------
@@ -22,12 +24,6 @@ SELECT ARRAY[1,2,3]::float8[]::vector;
[1,2,3]
(1 row)
SELECT ARRAY[1,2,3]::numeric[]::vector;
array
---------
[1,2,3]
(1 row)
SELECT '{NULL}'::real[]::vector;
ERROR: array must not containing NULLs
SELECT '{NaN}'::real[]::vector;
@@ -44,8 +40,8 @@ SELECT '[1,2,3]'::vector::real[];
{1,2,3}
(1 row)
SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
ERROR: vector cannot have more than 16000 dimensions
SELECT array_agg(n)::vector FROM generate_series(1, 1025) n;
ERROR: vector cannot have more than 1024 dimensions
-- ensure no error
SELECT ARRAY[1,2,3] = ARRAY[1,2,3];
?column?

View File

@@ -1,8 +1,10 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val vector(3));
\copy t TO 'results/data.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/data.bin' WITH (FORMAT binary)
\copy t TO '/tmp/data.bin' WITH (FORMAT binary)
\copy t2 FROM '/tmp/data.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;
val
---------

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT '[1,2,3]'::vector + '[4,5,6]';
?column?
----------
@@ -22,28 +24,10 @@ SELECT round(vector_norm('[1,1]')::numeric, 5);
1.41421
(1 row)
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
SELECT round(l2_distance('[1,2]', '[0,0]')::numeric, 5);
round
---------
2.23607
(1 row)
SELECT l2_distance('[1,2]', '[3]');
@@ -56,10 +40,10 @@ SELECT inner_product('[1,2]', '[3,4]');
SELECT inner_product('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT cosine_distance('[1,2]', '[2,4]');
cosine_distance
-----------------
0
SELECT round(cosine_distance('[1,2]', '[2,4]')::numeric, 5);
round
---------
0.00000
(1 row)
SELECT cosine_distance('[1,2]', '[0,0]');
@@ -68,37 +52,5 @@ SELECT cosine_distance('[1,2]', '[0,0]');
NaN
(1 row)
SELECT cosine_distance('[1,1]', '[1,1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,1]', '[-1,-1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
avg
-----------
[2,3.5,5]
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
avg
-----------
[2,3.5,5]
(1 row)
SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
avg
-----
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
ERROR: expected 2 dimensions, not 1

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT '[1,2,3]'::vector;
vector
---------
@@ -10,12 +12,6 @@ SELECT '[-1,2,3]'::vector;
[-1,2,3]
(1 row)
SELECT '[1.23456]'::vector;
vector
-----------
[1.23456]
(1 row)
SELECT '[hello,1]'::vector;
ERROR: invalid input syntax for type vector: "hello"
LINE 1: SELECT '[hello,1]'::vector;

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
CREATE INDEX ON t USING ivfflat (val) WITH (lists = 0);

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
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);

View File

@@ -1,8 +0,0 @@
use PostgreSQL::Test::Cluster;
sub get_new_node
{
return PostgreSQL::Test::Cluster->new(@_);
}
1;

View File

@@ -1,3 +0,0 @@
use PostgreSQL::Test::Utils;
1;

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));

View File

@@ -1,15 +1,17 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT ARRAY[1,2,3]::vector;
SELECT ARRAY[1.0,2.0,3.0]::vector;
SELECT ARRAY[1,2,3]::float4[]::vector;
SELECT ARRAY[1,2,3]::float8[]::vector;
SELECT ARRAY[1,2,3]::numeric[]::vector;
SELECT '{NULL}'::real[]::vector;
SELECT '{NaN}'::real[]::vector;
SELECT '{Infinity}'::real[]::vector;
SELECT '{-Infinity}'::real[]::vector;
SELECT '{}'::real[]::vector;
SELECT '[1,2,3]'::vector::real[];
SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
SELECT array_agg(n)::vector FROM generate_series(1, 1025) n;
-- ensure no error
SELECT ARRAY[1,2,3] = ARRAY[1,2,3];

View File

@@ -1,10 +1,13 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val vector(3));
\copy t TO 'results/data.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/data.bin' WITH (FORMAT binary)
\copy t TO '/tmp/data.bin' WITH (FORMAT binary)
\copy t2 FROM '/tmp/data.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;

View File

@@ -1,26 +1,18 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
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 vector_norm('[3,4]');
SELECT vector_norm('[0,1]');
SELECT l2_distance('[0,0]', '[3,4]');
SELECT l2_distance('[0,0]', '[0,1]');
SELECT round(l2_distance('[1,2]', '[0,0]')::numeric, 5);
SELECT l2_distance('[1,2]', '[3]');
SELECT inner_product('[1,2]', '[3,4]');
SELECT inner_product('[1,2]', '[3]');
SELECT cosine_distance('[1,2]', '[2,4]');
SELECT round(cosine_distance('[1,2]', '[2,4]')::numeric, 5);
SELECT cosine_distance('[1,2]', '[0,0]');
SELECT cosine_distance('[1,1]', '[1,1]');
SELECT cosine_distance('[1,1]', '[-1,-1]');
SELECT cosine_distance('[1,2]', '[3]');
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;

View File

@@ -1,6 +1,8 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT '[1,2,3]'::vector;
SELECT '[-1,2,3]'::vector;
SELECT '[1.23456]'::vector;
SELECT '[hello,1]'::vector;
SELECT '[NaN,1]'::vector;
SELECT '[Infinity,1]'::vector;

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));

View File

@@ -1,3 +1,5 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SET enable_seqscan = off;
CREATE UNLOGGED TABLE t (val vector(3));

View File

@@ -7,8 +7,6 @@ use PostgresNode;
use TestLib;
use Test::More tests => 31;
my $dim = 32;
my $node_primary;
my $node_replica;
@@ -20,20 +18,25 @@ sub test_index_replay
# Wait for replica to catch up
my $applname = $node_replica->name;
my $caughtup_query;
my $server_version_num = $node_primary->safe_psql("postgres", "SHOW server_version_num");
my $caughtup_query = "SELECT pg_current_wal_lsn() <= replay_lsn FROM pg_stat_replication WHERE application_name = '$applname';";
if ($server_version_num >= 100000) {
$caughtup_query = "SELECT pg_current_wal_lsn() <= replay_lsn FROM pg_stat_replication WHERE application_name = '$applname';";
} else {
# TODO figure out why replay location doesn't work
$caughtup_query = "SELECT pg_current_xlog_location() <= write_location FROM pg_stat_replication WHERE application_name = '$applname';";
}
$node_primary->poll_query_until('postgres', $caughtup_query)
or die "Timed out while waiting for replica 1 to catch up";
my @r = ();
for (1 .. $dim) {
push(@r, rand());
}
my $sql = join(",", @r);
my $r1 = rand();
my $r2 = rand();
my $r3 = rand();
my $queries = qq(
SET enable_seqscan = off;
SELECT * FROM tst ORDER BY v <-> '[$sql]' LIMIT 10;
SELECT * FROM tst ORDER BY v <-> '[$r1,$r2,$r3]' LIMIT 10;
);
# Run test queries and compare their result
@@ -44,21 +47,9 @@ sub test_index_replay
return;
}
# Use ARRAY[random(), random(), random(), ...] over
# SELECT array_agg(random()) FROM generate_series(1, $dim)
# to generate different values for each row
my $array_sql = join(",", ('random()') x $dim);
# Initialize primary node
$node_primary = get_new_node('primary');
$node_primary->init(allows_streaming => 1);
if ($dim > 32) {
# TODO use wal_keep_segments for Postgres < 13
$node_primary->append_conf('postgresql.conf', qq(wal_keep_size = 1GB));
}
if ($dim > 1500) {
$node_primary->append_conf('postgresql.conf', qq(maintenance_work_mem = 128MB));
}
$node_primary->start;
my $backup_name = 'my_backup';
@@ -73,9 +64,9 @@ $node_replica->start;
# Create ivfflat index on primary
$node_primary->safe_psql("postgres", "CREATE EXTENSION vector;");
$node_primary->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
$node_primary->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));");
$node_primary->safe_psql("postgres",
"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
"INSERT INTO tst SELECT i % 10, ARRAY[random(), random(), random()] FROM generate_series(1, 100000) i;"
);
$node_primary->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v);");
@@ -91,7 +82,7 @@ for my $i (1 .. 10)
test_index_replay("vacuum $i");
my ($start, $end) = (100001 + ($i - 1) * 10000, 100000 + $i * 10000);
$node_primary->safe_psql("postgres",
"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series($start, $end) i;"
"INSERT INTO tst SELECT i % 10, ARRAY[random(), random(), random()] FROM generate_series($start, $end) i;"
);
test_index_replay("insert $i");
}

View File

@@ -4,15 +4,6 @@ use PostgresNode;
use TestLib;
use Test::More tests => 1;
my $dim = 3;
my @r = ();
for (1 .. $dim) {
my $v = int(rand(1000)) + 1;
push(@r, "i % $v");
}
my $array_sql = join(", ", @r);
# Initialize node
my $node = get_new_node('node');
$node->init;
@@ -20,9 +11,9 @@ $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));");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
"INSERT INTO tst SELECT i % 10, ARRAY[i % 1000, i % 333, i % 55] FROM generate_series(1, 100000) i;"
);
$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v);");
@@ -33,7 +24,7 @@ my $size = $node->safe_psql("postgres", "SELECT pg_total_relation_size('tst_v_id
$node->safe_psql("postgres", "DELETE FROM tst;");
$node->safe_psql("postgres", "VACUUM tst;");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
"INSERT INTO tst SELECT i % 10, ARRAY[i % 1000, i % 333, i % 55] FROM generate_series(1, 100000) i;"
);
# Check size

View File

@@ -2,16 +2,15 @@ use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More tests => 9;
use Test::More tests => 2;
my $node;
my @queries = ();
my @expected;
my $limit = 20;
my @expected = ();
sub test_recall
{
my ($probes, $min, $operator) = @_;
my ($probes, $min) = @_;
my $correct = 0;
my $total = 0;
@@ -19,7 +18,7 @@ sub test_recall
my $actual = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET ivfflat.probes = $probes;
SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
SELECT i FROM tst ORDER BY v <-> '$queries[$i]' LIMIT 10;
));
my @actual_ids = split("\n", $actual);
my %actual_set = map { $_ => 1 } @actual_ids;
@@ -34,7 +33,7 @@ sub test_recall
}
}
cmp_ok($correct / $total, ">=", $min, $operator);
cmp_ok($correct / $total, ">=", $min);
}
# Initialize node
@@ -57,32 +56,17 @@ for (1..20) {
push(@queries, "[$r1,$r2,$r3]");
}
# Check each index type
my @operators = ("<->", "<#>", "<=>");
foreach (@operators) {
my $operator = $_;
# Get exact results
@expected = ();
foreach (@queries) {
my $res = $node->safe_psql("postgres", "SELECT i FROM tst ORDER BY v $operator '$_' LIMIT $limit;");
push(@expected, $res);
}
# Add index
my $opclass;
if ($operator == "<->") {
$opclass = "vector_l2_ops";
} elsif ($operator == "<#>") {
$opclass = "vector_ip_ops";
} else {
$opclass = "vector_cosine_ops";
}
$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v $opclass);");
# Test approximate results
test_recall(1, 0.75, $operator);
test_recall(10, 0.95, $operator);
test_recall(100, 1.0, $operator);
# Get exact results
foreach (@queries) {
my $res = $node->safe_psql("postgres", "SELECT i FROM tst ORDER BY v <-> '$_' LIMIT 10;");
push(@expected, $res);
}
# Add index
$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v);");
# Test approximate results
test_recall(1, 0.8);
# Test probes
test_recall(100, 1.0);

View File

@@ -1,45 +0,0 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More tests => 60;
# Initialize node
my $node = get_new_node('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4 primary key, v vector(3));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 100000) i;"
);
# Check each index type
my @operators = ("<->", "<#>", "<=>");
foreach (@operators) {
my $operator = $_;
# Add index
my $opclass;
if ($operator == "<->") {
$opclass = "vector_l2_ops";
} elsif ($operator == "<#>") {
$opclass = "vector_ip_ops";
} else {
$opclass = "vector_cosine_ops";
}
$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v $opclass);");
# Test 100% recall
for (1..20) {
my $i = int(rand() * 100000);
my $query = $node->safe_psql("postgres", "SELECT v FROM tst WHERE i = $i;");
my $res = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SELECT v FROM tst ORDER BY v <-> '$query' LIMIT 1;
));
is($res, $query);
}
}

View File

@@ -1,31 +0,0 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More tests => 3;
# Initialize node
my $node = get_new_node('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (v vector(3));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT ARRAY[random(), random(), random()] FROM generate_series(1, 100000) i;"
);
$node->safe_psql("postgres", "CREATE INDEX lists50 ON tst USING ivfflat (v) WITH (lists = 50);");
$node->safe_psql("postgres", "CREATE INDEX lists100 ON tst USING ivfflat (v) WITH (lists = 100);");
# Test prefers more lists
my $res = $node->safe_psql("postgres", "EXPLAIN SELECT v FROM tst ORDER BY v <-> '[0.5,0.5,0.5]' LIMIT 10;");
like($res, qr/lists100/);
unlike($res, qr/lists50/);
# Test errors with too much memory
my ($ret, $stdout, $stderr) = $node->psql("postgres",
"CREATE INDEX lists10000 ON tst USING ivfflat (v) WITH (lists = 10000);"
);
like($stderr, qr/memory required is/);

View File

@@ -1,55 +0,0 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More tests => 7;
my $dim = 768;
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;");
$node->safe_psql("postgres", "CREATE TABLE tst (v vector($dim));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
);
$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v);");
$node->pgbench(
"--no-vacuum --client=5 --transactions=100",
0,
[qr{actually processed}],
[qr{^$}],
"concurrent INSERTs",
{
"007_inserts" => "INSERT INTO tst SELECT ARRAY[$array_sql] FROM generate_series(1, 10) i;"
}
);
sub idx_scan
{
# Stats do not update instantaneously
# https://www.postgresql.org/docs/current/monitoring-stats.html#MONITORING-STATS-VIEWS
sleep(1);
$node->safe_psql("postgres", "SELECT idx_scan FROM pg_stat_user_indexes WHERE indexrelid = 'tst_v_idx'::regclass;");
}
my $expected = 10000 + 5 * 100 * 10;
my $count = $node->safe_psql("postgres", "SELECT COUNT(*) FROM tst;");
is($count, $expected);
is(idx_scan(), 0);
$count = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET ivfflat.probes = 100;
SELECT COUNT(*) FROM (SELECT v FROM tst ORDER BY v <-> (SELECT v FROM tst LIMIT 1)) t;
));
is($count, $expected);
is(idx_scan(), 1);

View File

@@ -1,35 +0,0 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More tests => 5;
# Initialize node
my $node = get_new_node('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (r1 real, r2 real, r3 real, v vector(3));");
$node->safe_psql("postgres", qq(
INSERT INTO tst SELECT r1, r2, r3, ARRAY[r1, r2, r3] FROM (
SELECT random() + 1.01 AS r1, random() + 2.01 AS r2, random() + 3.01 AS r3 FROM generate_series(1, 1000000) t
) i;
));
# Test avg
my $avg = $node->safe_psql("postgres", "SELECT AVG(v) FROM tst;");
like($avg, qr/\[1\.5/);
like($avg, qr/,2\.5/);
like($avg, qr/,3\.5/);
# Test matches real
my $r1 = $node->safe_psql("postgres", "SELECT AVG(r1)::float4 FROM tst;");
my $r2 = $node->safe_psql("postgres", "SELECT AVG(r2)::float4 FROM tst;");
my $r3 = $node->safe_psql("postgres", "SELECT AVG(r3)::float4 FROM tst;");
is($avg, "[$r1,$r2,$r3]");
# Test explain
my $explain = $node->safe_psql("postgres", "EXPLAIN SELECT AVG(v) FROM tst;");
like($explain, qr/Partial Aggregate/);

View File

@@ -1,32 +0,0 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More tests => 1;
my $dim = 1024;
# Initialize node
my $node = get_new_node('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (v1 vector(1024), v2 vector(1024), v3 vector(1024));");
# Test insert succeeds
$node->safe_psql("postgres",
"INSERT INTO tst SELECT array_agg(n), array_agg(n), array_agg(n) FROM generate_series(1, $dim) n"
);
# Change storage to PLAIN
$node->safe_psql("postgres", "ALTER TABLE tst ALTER COLUMN v1 SET STORAGE PLAIN");
$node->safe_psql("postgres", "ALTER TABLE tst ALTER COLUMN v2 SET STORAGE PLAIN");
$node->safe_psql("postgres", "ALTER TABLE tst ALTER COLUMN v3 SET STORAGE PLAIN");
# Test insert fails
my ($ret, $stdout, $stderr) = $node->psql("postgres",
"INSERT INTO tst SELECT array_agg(n), array_agg(n), array_agg(n) FROM generate_series(1, $dim) n"
);
like($stderr, qr/row is too big/);

View File

@@ -1,4 +1,4 @@
comment = 'vector data type and ivfflat access method'
default_version = '0.4.1'
default_version = '0.2.5'
module_pathname = '$libdir/vector'
relocatable = true