Compare commits

...

144 Commits

Author SHA1 Message Date
Andrew Kane
bb127fce8d Try again 2022-12-09 02:01:39 -08:00
Andrew Kane
7ce4e1f53e Skip IPC::Run 2022-12-09 01:57:10 -08:00
Andrew Kane
6e41c81412 Skip tests 2022-12-09 01:06:03 -08:00
Andrew Kane
98ec05145b Install IPC::Run 2022-12-09 00:57:28 -08:00
Andrew Kane
e302c99d4a Extract tar 2022-12-09 00:52:34 -08:00
Andrew Kane
d6ab29e772 Try again 2022-12-09 00:47:45 -08:00
Andrew Kane
6ad96fe8ca Use set 2022-12-08 23:05:11 -08:00
Andrew Kane
e52ef4f2b5 Added prove_installcheck for Windows 2022-12-08 22:57:46 -08:00
Andrew Kane
8c81213592 Improved installcheck on Windows 2022-12-08 22:33:19 -08:00
Andrew Kane
f77b044303 Simplified tests 2022-12-08 22:11:49 -08:00
Andrew Kane
fab7b50c3a Updated readme [skip ci] 2022-12-08 16:03:22 -08:00
Andrew Kane
9e2e726010 Improved macros [skip ci] 2022-12-08 15:42:22 -08:00
Andrew Kane
2621f9f947 Added experimental support for Windows (including auto-vectorization) - closes #49 2022-12-08 13:27:26 -08:00
Andrew Kane
4fbe55f97e Split CI into separate groups of steps 2022-12-08 00:26:48 -08:00
Andrew Kane
9c9489e888 Removed nested FLEXIBLE_ARRAY_MEMBER for Windows 2022-12-07 19:53:48 -08:00
Andrew Kane
d376011087 Added VectorArrayFree 2022-12-07 19:45:44 -08:00
Andrew Kane
5107e057ad Improved CI [skip ci] 2022-12-07 16:35:03 -08:00
Andrew Kane
1c1812aeac Updated indentation [skip ci] 2022-12-07 16:34:01 -08:00
Andrew Kane
da12d9d8b0 Fixed copy test on Windows [skip ci] 2022-12-07 16:10:35 -08:00
Andrew Kane
de26e1e78a Perform version check earlier [skip ci] 2022-12-07 15:52:35 -08:00
Andrew Kane
8691d2ad53 Include port.h to fix buggy strtof on some platforms 2022-12-07 15:47:39 -08:00
Andrew Kane
573200c188 Use pg_prng_uint32 for Postgres 15 2022-12-07 15:28:11 -08:00
Andrew Kane
cd7cab804f Added port.h for random() function on Windows 2022-12-07 15:17:32 -08:00
Andrew Kane
84a8bdd661 Fixed FLOAT_SHORTEST_DECIMAL_LEN for Postgres < 12 [skip ci] 2022-12-02 10:54:14 -08:00
Andrew Kane
7a13a6cd3c Updated changelog [skip ci] 2022-12-01 21:23:22 -08:00
Andrew Kane
35d5cb9f06 Fixed CI (behavior of extra_float_digits changed between Postgres 11 and 12) 2022-12-01 21:07:57 -08:00
Andrew Kane
bc6b9e4bba Added more tests for text output 2022-12-01 20:45:27 -08:00
Andrew Kane
d605948be4 Updated changelog [skip ci] 2022-12-01 20:28:17 -08:00
Andrew Kane
37a784d3f6 Fixed improved vector text representation for Postgres < 12 2022-12-01 15:41:12 -08:00
Andrew Kane
c74bb5495b Improved vector text representation - fixes #46 2022-12-01 14:57:47 -08:00
Andrew Kane
cc35af160e Use strtof like float4in to avoid a double-rounding problem 2022-12-01 14:41:09 -08:00
Andrew Kane
a7f712b5a4 Version bump to 0.3.2 [skip ci] 2022-11-22 13:46:33 -08:00
Andrew Kane
a532d82bda Use palloc_extended for halfcdist 2022-11-17 19:52:46 -08:00
Andrew Kane
ccec96ba54 Fixed invalid memory alloc request size error - fixes #43 2022-11-17 14:12:49 -08:00
Andrew Kane
b7c5849ba7 Switched to dev-files option on CI 2022-11-03 15:24:12 -07:00
Andrew Kane
135147893a Updated comment [skip ci] 2022-11-02 16:55:20 -07:00
Andrew Kane
00391953dd Updated comment [skip ci] 2022-11-02 16:54:27 -07:00
Andrew Kane
81bba1f37e Improved order of insert operations 2022-11-02 16:41:56 -07:00
Andrew Kane
2d8b7e5f81 Version bump to 0.3.1 [skip ci] 2022-11-02 15:24:29 -07:00
Andrew Kane
cebae5e5ff Fixed link [skip ci] 2022-11-02 15:21:44 -07:00
Andrew Kane
4c3332fc4d Added upgrade notes [skip ci] 2022-11-02 15:20:53 -07:00
Andrew Kane
386a3b5dd5 Improved test for inserts 2022-11-02 14:29:41 -07:00
Andrew Kane
fb819eb8b2 Added comment [skip ci] 2022-11-02 12:56:32 -07:00
Andrew Kane
3f8407460e Improved test 2022-11-02 12:50:50 -07:00
Andrew Kane
059e88dda5 Fixed randomness in WAL test 2022-11-02 12:41:54 -07:00
Andrew Kane
12f158d4bb Fixed warning 2022-11-02 12:27:09 -07:00
Andrew Kane
2a3c0a769a Ensure only one backend tries to extend relation at the same time - fixes #41 2022-11-02 12:23:49 -07:00
Andrew Kane
349d844fa1 Fixed issue with uninitialized pages - fixes #41 2022-11-02 11:01:47 -07:00
Andrew Kane
a57ca81d8f Switched to runtime check 2022-10-30 12:07:03 -07:00
Andrew Kane
dc4a79a7e7 Added check for indexing overflow 2022-10-30 12:01:24 -07:00
Andrew Kane
1652a130c5 Removed unused assignments 2022-10-30 11:55:17 -07:00
Andrew Kane
a6743bd206 Added .bc to .gitignore 2022-10-30 11:54:58 -07:00
Andrew Kane
e25627074f Removed unused code [skip ci] 2022-10-30 11:49:06 -07:00
Andrew Kane
b3cad93f3a Fixed segmentation fault with index creation when lists > 6500 - fixes #42 2022-10-30 11:14:58 -07:00
Andrew Kane
379a76098e Fixed version in META.json [skip ci] 2022-10-15 14:44:44 -07:00
Andrew Kane
5af768752f Version bump to 0.3.0 [skip ci] 2022-10-15 14:38:43 -07:00
Andrew Kane
f03bcc4fd8 Updated Dockerfile to Postgres 15 [skip ci] 2022-10-14 16:01:25 -07:00
Andrew Kane
74fc4b9aaa Added .pm to .editorconfig [skip ci] 2022-10-14 14:05:20 -07:00
Andrew Kane
6870ba96ba Removed Postgres 9.6 from CI 2022-10-14 13:47:07 -07:00
Andrew Kane
2575cfdaa8 Fixed CI 2022-10-14 13:45:16 -07:00
Andrew Kane
9ff5ceacca Dropped support for Postgres 9.6 [skip ci] 2022-10-14 13:44:13 -07:00
Andrew Kane
9a58ee609a Added Postgres 15 to CI 2022-10-14 13:42:04 -07:00
Andrew Kane
66a31bef36 Simplified define [skip ci] 2022-10-10 15:01:16 -07:00
Andrew Kane
163ecdf063 Use define for RandomDouble 2022-10-10 14:59:10 -07:00
Andrew Kane
c9c6b96eed Added support for Postgres 15 [skip ci] 2022-10-05 16:11:37 -07:00
Andrew Kane
48e9a6830b Added link to pgvector-elixir [skip ci] 2022-08-03 23:12:25 -07:00
Andrew Kane
f22b47ca46 Added link to pgvector-php - #30 [skip ci] 2022-08-03 13:50:34 -07:00
Andrew Kane
2a057c055c Updated example [skip ci] 2022-08-01 00:06:45 -07:00
Andrew Kane
b89b58b893 Updated examples to be runnable - fixes #24 [skip ci] 2022-07-31 22:36:46 -07:00
Andrew Kane
2e77650e1d Updated example [skip ci] 2022-07-31 22:34:03 -07:00
Andrew Kane
0a68b040fb DRY code 2022-07-31 21:56:44 -07:00
Andrew Kane
434dc904c1 Version bump to 0.2.7 [skip ci] 2022-07-31 21:42:15 -07:00
Andrew Kane
92a671eeb1 Revert "Commit pages with tuple"
This reverts commit 94eeb83abe.
2022-07-31 21:36:46 -07:00
Andrew Kane
94eeb83abe Commit pages with tuple 2022-07-31 21:29:29 -07:00
Andrew Kane
6866939090 Fixed indentation [skip ci] 2022-07-31 21:21:06 -07:00
Andrew Kane
ba2fb6ad8c Fixed order in IvfflatAppendPage 2022-07-31 21:03:57 -07:00
Andrew Kane
51fe8aff02 Fixed unexpected data beyond EOF error - fixes #29 2022-07-31 20:48:15 -07:00
Andrew Kane
df3dc151d5 Updated readme [skip ci] 2022-07-27 17:38:13 -07:00
Andrew Kane
6687430589 Fixed CI 2022-07-10 16:30:45 -04:00
Andrew Kane
f128a455d1 Made dimensions configurable for WAL test - #29 2022-07-10 16:01:05 -04:00
Andrew Kane
56cb5f3503 Version bump to 0.2.6 [skip ci] 2022-05-22 18:21:15 -07:00
Andrew Kane
ecbf46938f Moved metrics to separate function [skip ci] 2022-02-15 19:13:23 -08:00
Andrew Kane
4ca264ba02 Added Davies-Bouldin index [skip ci] 2022-02-15 19:08:47 -08:00
Andrew Kane
f5458414b8 Made recall test less flaky 2022-02-15 18:17:05 -08:00
Andrew Kane
e64ed39acb Disabled scan progress for sampling 2022-02-15 18:14:59 -08:00
Andrew Kane
0d025be9d3 Improved performance of index creation for Postgres < 12 2022-02-15 18:05:33 -08:00
Andrew Kane
fed60dce78 Added tests for lists 2022-02-15 17:04:20 -08:00
Andrew Kane
a37f5eea4a Improved compare method [skip ci] 2022-02-15 11:50:35 -08:00
Andrew Kane
4bdb27e85a Fixed style [skip ci] 2022-02-14 23:15:27 -08:00
Andrew Kane
38f869e0bd Added test for 100% recall 2022-02-14 23:13:36 -08:00
Andrew Kane
ad8acc00d4 Fixed spacing [skip ci] 2022-02-14 21:58:20 -08:00
Andrew Kane
01926a418e Print operator name for failed recall test [skip ci] 2022-02-14 21:52:31 -08:00
Andrew Kane
9658d3c1ad Made recall test less flaky 2022-02-14 21:49:08 -08:00
Andrew Kane
6b9c6516f4 Improved code 2022-02-14 21:41:27 -08:00
Andrew Kane
88be03a3fa Removed unused code 2022-02-14 21:37:53 -08:00
Andrew Kane
bf5b2c8d7e Use tuple id directly 2022-02-14 21:31:08 -08:00
Andrew Kane
ac65ec2856 Improved code [skip ci] 2022-02-14 19:38:59 -08:00
Andrew Kane
7bba0e2a01 Test recall for all operators 2022-02-14 15:22:53 -08:00
Andrew Kane
c35e9f3b84 Removed comment [skip ci] 2022-02-13 03:53:30 -08:00
Andrew Kane
9549d93260 Improved recall test 2022-02-13 03:08:39 -08:00
Andrew Kane
310809d0e5 Fixed warnings 2022-02-13 03:03:17 -08:00
Andrew Kane
ce72ca8620 Switched to heap for nearest lists for performance 2022-02-13 02:59:27 -08:00
Andrew Kane
4f2c937a1f Improved recall test 2022-02-13 02:35:28 -08:00
Andrew Kane
ff400ce5f1 Use macro for UpdateProgress 2022-02-12 23:56:28 -08:00
Andrew Kane
41d11c62d6 Fixed style [skip ci] 2022-02-12 22:14:57 -08:00
Andrew Kane
fa5e90585d Added inertia 2022-02-12 21:23:53 -08:00
Andrew Kane
71d2908be1 Fixed style [skip ci] 2022-02-12 20:47:25 -08:00
Andrew Kane
d0a1c1d0ed Added test for centers 2022-02-12 20:44:05 -08:00
Andrew Kane
8063201cdf Fixed indentation [skip ci] 2022-02-12 20:43:48 -08:00
Andrew Kane
2f52127bee Version bump to 0.2.5 [skip ci] 2022-02-11 18:50:04 -08:00
Andrew Kane
4778039ffa Added benchmarking to readme [skip ci] 2022-02-11 16:09:45 -08:00
Andrew Kane
e3c9d70acc Added to existing PG_CFLAGS 2022-02-11 16:06:25 -08:00
Andrew Kane
5971e28f8c Added more benchmarking [skip ci] 2022-02-11 16:05:45 -08:00
Andrew Kane
2da27e2f01 Updated name [skip ci] 2022-02-11 15:52:59 -08:00
Andrew Kane
7dd934913d Improved output [skip ci] 2022-02-11 15:51:14 -08:00
Andrew Kane
9b46987e73 Added support for benchmarking 2022-02-11 15:48:13 -08:00
Andrew Kane
868e1f99a4 Added todo [skip ci] 2022-02-11 15:14:09 -08:00
Andrew Kane
354271178f Added todo [skip ci] 2022-02-11 14:31:42 -08:00
Andrew Kane
fe9b218903 Reduced memory for unlogged tables [skip ci] 2022-02-11 13:57:53 -08:00
Andrew Kane
3cdd1c16fb Improved code [skip ci] 2022-02-11 13:55:39 -08:00
Andrew Kane
4f7a5c92da Use maxlen for check [skip ci] 2022-02-11 13:54:03 -08:00
Andrew Kane
9399575e02 Include samples size and centers size in check 2022-02-11 12:43:01 -08:00
Andrew Kane
e7539fe548 Added new centers size 2022-02-11 11:58:58 -08:00
Andrew Kane
61860ee6d9 Updated changelog [skip ci] 2022-02-11 11:49:25 -08:00
Andrew Kane
eb99730d87 Removed unnecessary cast [skip ci] 2022-02-11 03:43:31 -08:00
Andrew Kane
e5da5cc532 Add one to error message to ceil [skip ci] 2022-02-11 03:41:54 -08:00
Andrew Kane
f59b1184fc Calculate all allocation sizes 2022-02-11 03:33:43 -08:00
Andrew Kane
cab103a2e7 Fixed index creation exceeding maintenance_work_mem and fixed error when lists > 1600 2022-02-11 03:10:19 -08:00
Andrew Kane
0094b70a36 Updated changelog [skip ci] 2022-02-11 03:02:59 -08:00
Andrew Kane
46ed56d80f Use float instead of double for k-means to save memory 2022-02-11 01:56:28 -08:00
Andrew Kane
d390cc9fa6 Increased recall queries in test 2022-02-11 00:45:39 -08:00
Andrew Kane
02ae98791c Improved recall test 2022-02-11 00:43:01 -08:00
Andrew Kane
b64c7d6228 Improved test [skip ci] 2022-02-11 00:15:37 -08:00
Andrew Kane
76f117c81c Added test for array of vectors [skip ci] 2022-02-07 15:25:37 -08:00
Andrew Kane
5414b2bc9e Added link to pgvector-ruby [skip ci] 2022-02-07 00:52:46 -08:00
Andrew Kane
6ba583b587 Improved test style [skip ci] 2022-02-06 20:24:50 -08:00
Andrew Kane
da88217fd8 Fixed test style 2022-02-06 20:19:20 -08:00
Andrew Kane
d5da22f24b Fixed tests 2022-02-06 20:14:21 -08:00
Andrew Kane
062e220936 Test probes 2022-02-06 20:08:53 -08:00
Andrew Kane
3d2b867a58 Added recall test 2022-02-06 19:59:46 -08:00
Andrew Kane
5aeabc3648 Added comments on index handler [skip ci] 2022-02-06 15:26:18 -08:00
Andrew Kane
687ec6d8d7 Added amadjustmembers 2022-02-06 15:21:59 -08:00
Andrew Kane
51dc18d5b7 Added compiler error for unsupported versions of Postgres 2022-02-06 15:15:46 -08:00
Andrew Kane
cb108ebfd2 Removed reindexing section [skip ci] 2022-02-06 12:45:49 -08:00
53 changed files with 1021 additions and 331 deletions

View File

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

View File

@@ -1,38 +1,65 @@
name: build name: build
on: [push, pull_request] on: [push, pull_request]
jobs: jobs:
build: ubuntu:
runs-on: ${{ matrix.os }} runs-on: ubuntu-latest
if: ${{ !startsWith(github.ref_name, 'windows') }}
strategy: strategy:
fail-fast: false fail-fast: false
matrix: matrix:
os: [ubuntu-latest] postgres: [15, 14, 13, 12, 11, 10]
postgres: [14, 13, 12, 11, 10, 9.6]
include:
- os: macos-latest
postgres: 14
steps: steps:
- uses: actions/checkout@v2 - uses: actions/checkout@v3
- uses: ankane/setup-postgres@v1 - uses: ankane/setup-postgres@v1
with: with:
postgres-version: ${{ matrix.postgres }} postgres-version: ${{ matrix.postgres }}
- if: ${{ startsWith(matrix.os, 'ubuntu') }} dev-files: true
run: sudo apt-get update && sudo apt-get install postgresql-server-dev-${{ matrix.postgres }} libipc-run-perl - run: make
- run: make - run: |
- if: ${{ startsWith(matrix.os, 'ubuntu') }} export PG_CONFIG=`which pg_config`
run: | sudo --preserve-env=PG_CONFIG make install
export PG_CONFIG=`which pg_config` - run: make installcheck
sudo --preserve-env=PG_CONFIG make install - if: ${{ failure() }}
- if: ${{ startsWith(matrix.os, 'macos') }} run: cat regression.diffs
run: make install - run: |
- run: make installcheck sudo apt-get update
- if: ${{ failure() }} sudo apt-get install libipc-run-perl
run: cat regression.diffs make prove_installcheck
- if: ${{ startsWith(matrix.os, 'ubuntu') }} mac:
run: make prove_installcheck runs-on: macos-latest
- if: ${{ startsWith(matrix.os, 'macos') }} if: ${{ !startsWith(github.ref_name, 'windows') }}
run: | steps:
brew install cpanm && cpanm IPC::Run - uses: actions/checkout@v3
wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_1.tar.gz - uses: ankane/setup-postgres@v1
tar xf REL_14_1.tar.gz with:
make prove_installcheck PROVE=prove PERL5LIB=postgres-REL_14_1/src/test/perl postgres-version: 14
- run: make
- run: make install
- run: make installcheck
- if: ${{ failure() }}
run: cat regression.diffs
- run: |
brew install cpanm
cpanm IPC::Run
wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_5.tar.gz
tar xf REL_14_5.tar.gz
make prove_installcheck PROVE=prove PROVE_FLAGS="-I ./postgres-REL_14_5/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
windows:
runs-on: windows-latest
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
curl -Ls -o REL_14_5.tar.gz https://github.com/postgres/postgres/archive/refs/tags/REL_14_5.tar.gz
7z x REL_14_5.tar.gz
7z x REL_14_5.tar
ls ./postgres-REL_14_5/src/test/perl
set PROVE=prove
set PROVE_FLAGS="-I ./postgres-REL_14_5/src/test/perl"
nmake /NOLOGO /F Makefile.win prove_installcheck
shell: cmd

1
.gitignore vendored
View File

@@ -5,3 +5,4 @@
regression.* regression.*
*.o *.o
*.so *.so
*.bc

View File

@@ -1,3 +1,39 @@
## 0.4.0 (unreleased)
- Changed text representation for vector elements to match `real`
- Improved accuracy of text parsing for certain inputs
- Added experimental support for Windows
## 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
- Fixed index creation exceeding `maintenance_work_mem`
- Fixed error with index creation when lists > 1600
## 0.2.4 (2022-02-06) ## 0.2.4 (2022-02-06)
- Added support for parallel vacuum - Added support for parallel vacuum

View File

@@ -1,9 +1,9 @@
FROM postgres:14 FROM postgres:15
COPY . /tmp/pgvector COPY . /tmp/pgvector
RUN apt-get update && \ RUN apt-get update && \
apt-get install -y --no-install-recommends build-essential postgresql-server-dev-14 && \ apt-get install -y --no-install-recommends build-essential postgresql-server-dev-15 && \
cd /tmp/pgvector && \ cd /tmp/pgvector && \
make clean && \ make clean && \
make OPTFLAGS="" && \ make OPTFLAGS="" && \
@@ -11,6 +11,6 @@ RUN apt-get update && \
mkdir /usr/share/doc/pgvector && \ mkdir /usr/share/doc/pgvector && \
cp LICENSE README.md /usr/share/doc/pgvector && \ cp LICENSE README.md /usr/share/doc/pgvector && \
rm -r /tmp/pgvector && \ rm -r /tmp/pgvector && \
apt-get remove -y build-essential postgresql-server-dev-14 && \ apt-get remove -y build-essential postgresql-server-dev-15 && \
apt-get autoremove -y && \ apt-get autoremove -y && \
rm -rf /var/lib/apt/lists/* rm -rf /var/lib/apt/lists/*

View File

@@ -2,7 +2,7 @@
"name": "vector", "name": "vector",
"abstract": "Open-source vector similarity search for Postgres", "abstract": "Open-source vector similarity search for Postgres",
"description": "Supports L2 distance, inner product, and cosine distance", "description": "Supports L2 distance, inner product, and cosine distance",
"version": "0.2.4", "version": "0.3.2",
"maintainer": [ "maintainer": [
"Andrew Kane <andrew@ankane.org>" "Andrew Kane <andrew@ankane.org>"
], ],
@@ -12,7 +12,7 @@
"prereqs": { "prereqs": {
"runtime": { "runtime": {
"requires": { "requires": {
"PostgreSQL": "9.6.0" "PostgreSQL": "10.0.0"
} }
} }
}, },
@@ -20,7 +20,7 @@
"vector": { "vector": {
"file": "sql/vector.sql", "file": "sql/vector.sql",
"docfile": "README.md", "docfile": "README.md",
"version": "0.2.4", "version": "0.3.2",
"abstract": "Open-source vector similarity search for Postgres" "abstract": "Open-source vector similarity search for Postgres"
} }
}, },

View File

@@ -1,5 +1,5 @@
EXTENSION = vector EXTENSION = vector
EXTVERSION = 0.2.4 EXTVERSION = 0.3.2
MODULE_big = vector MODULE_big = vector
DATA = $(wildcard sql/*--*.sql) DATA = $(wildcard sql/*--*.sql)
@@ -7,7 +7,7 @@ OBJS = src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.
TESTS = $(wildcard test/sql/*.sql) TESTS = $(wildcard test/sql/*.sql)
REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS)) REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
REGRESS_OPTS = --inputdir=test REGRESS_OPTS = --inputdir=test --load-extension=vector
OPTFLAGS = -march=native OPTFLAGS = -march=native
@@ -21,7 +21,7 @@ endif
# For auto-vectorization: # For auto-vectorization:
# - GCC (needs -ftree-vectorize OR -O3) - https://gcc.gnu.org/projects/tree-ssa/vectorization.html # - GCC (needs -ftree-vectorize OR -O3) - https://gcc.gnu.org/projects/tree-ssa/vectorization.html
# - Clang (could use pragma instead) - https://llvm.org/docs/Vectorizers.html # - Clang (could use pragma instead) - https://llvm.org/docs/Vectorizers.html
PG_CFLAGS = $(OPTFLAGS) -ftree-vectorize -fassociative-math -fno-signed-zeros -fno-trapping-math PG_CFLAGS += $(OPTFLAGS) -ftree-vectorize -fassociative-math -fno-signed-zeros -fno-trapping-math
# Debug GCC auto-vectorization # Debug GCC auto-vectorization
# PG_CFLAGS += -fopt-info-vec # PG_CFLAGS += -fopt-info-vec
@@ -40,6 +40,9 @@ PG_CONFIG ?= pg_config
PGXS := $(shell $(PG_CONFIG) --pgxs) PGXS := $(shell $(PG_CONFIG) --pgxs)
include $(PGXS) include $(PGXS)
# for Postgres 15
PROVE_FLAGS += -I ./test/perl
prove_installcheck: prove_installcheck:
rm -rf $(CURDIR)/tmp_check rm -rf $(CURDIR)/tmp_check
cd $(srcdir) && TESTDIR='$(CURDIR)' PATH="$(bindir):$$PATH" PGPORT='6$(DEF_PGPORT)' PG_REGRESS='$(top_builddir)/src/test/regress/pg_regress' $(PROVE) $(PG_PROVE_FLAGS) $(PROVE_FLAGS) $(if $(PROVE_TESTS),$(PROVE_TESTS),test/t/*.pl) cd $(srcdir) && TESTDIR='$(CURDIR)' PATH="$(bindir):$$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
Makefile.win Normal file
View File

@@ -0,0 +1,62 @@
EXTENSION = vector
EXTVERSION = 0.3.2
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
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) $(PG_CFLAGS)
SHLIB = src\$(EXTENSION).dll
LIBS = "$(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) $(REGRESS)
prove_installcheck:
rm -rf tmp_check
set PGPORT=65432
set PG_REGRESS="$(BINDIR)\pg_regress"
$(PROVE) $(PG_PROVE_FLAGS) $(PROVE_FLAGS) test/t/*.pl

View File

@@ -3,9 +3,9 @@
Open-source vector similarity search for Postgres Open-source vector similarity search for Postgres
```sql ```sql
CREATE TABLE table (column vector(3)); CREATE TABLE items (embedding vector(3));
CREATE INDEX ON table USING ivfflat (column vector_l2_ops); CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops);
SELECT * FROM table ORDER BY column <-> '[1,2,3]' LIMIT 5; SELECT * FROM items ORDER BY embedding <-> '[1,2,3]' LIMIT 5;
``` ```
Supports L2 distance, inner product, and cosine distance Supports L2 distance, inner product, and cosine distance
@@ -14,10 +14,10 @@ Supports L2 distance, inner product, and cosine distance
## Installation ## Installation
Compile and install the extension (supports Postgres 9.6+) Compile and install the extension (supports Postgres 10+)
```sh ```sh
git clone --branch v0.2.4 https://github.com/pgvector/pgvector.git git clone --branch v0.3.2 https://github.com/pgvector/pgvector.git
cd pgvector cd pgvector
make make
make install # may need sudo make install # may need sudo
@@ -33,22 +33,22 @@ You can also install it with [Docker](#docker), [Homebrew](#homebrew), or [PGXN]
## Getting Started ## Getting Started
Create a vector column with 3 dimensions (replace `table` and `column` with non-reserved names) Create a vector column with 3 dimensions
```sql ```sql
CREATE TABLE table (column vector(3)); CREATE TABLE items (embedding vector(3));
``` ```
Insert values Insert values
```sql ```sql
INSERT INTO table VALUES ('[1,2,3]'), ('[4,5,6]'); INSERT INTO items VALUES ('[1,2,3]'), ('[4,5,6]');
``` ```
Get the nearest neighbor by L2 distance Get the nearest neighbor by L2 distance
```sql ```sql
SELECT * FROM table ORDER BY column <-> '[3,1,2]' LIMIT 1; SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 1;
``` ```
Also supports inner product (`<#>`) and cosine distance (`<=>`) Also supports inner product (`<#>`) and cosine distance (`<=>`)
@@ -62,41 +62,29 @@ Speed up queries with an approximate index. Add an index for each distance funct
L2 distance L2 distance
```sql ```sql
CREATE INDEX ON table USING ivfflat (column vector_l2_ops); CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops);
``` ```
Inner product Inner product
```sql ```sql
CREATE INDEX ON table USING ivfflat (column vector_ip_ops); CREATE INDEX ON items USING ivfflat (embedding vector_ip_ops);
``` ```
Cosine distance Cosine distance
```sql ```sql
CREATE INDEX ON table USING ivfflat (column vector_cosine_ops); CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops);
``` ```
Indexes should be created after the table has data for optimal clustering. If the distribution of data changes significantly, you can reindex without downtime: 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.
```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;
```
Also, unlike typical indexes which only affect performance, you may see different results for queries after adding an approximate index.
### Index Options ### Index Options
Specify the number of inverted lists (100 by default) Specify the number of inverted lists (100 by default)
```sql ```sql
CREATE INDEX ON table USING ivfflat (column opclass) WITH (lists = 100); CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100);
``` ```
A [good place to start](https://github.com/facebookresearch/faiss/issues/112) is `4 * sqrt(rows)` A [good place to start](https://github.com/facebookresearch/faiss/issues/112) is `4 * sqrt(rows)`
@@ -143,10 +131,20 @@ 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 Consider [partial indexes](https://www.postgresql.org/docs/current/indexes-partial.html) for queries with a `WHERE` clause
```sql ```sql
CREATE INDEX ON table USING ivfflat (column opclass) WHERE (other_column = 123); SELECT * FROM items WHERE category_id = 123 ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
``` ```
To index many different values of `other_column`, consider [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) on `other_column`. can be indexed with:
```sql
CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) 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);
```
## Performance ## Performance
@@ -159,14 +157,14 @@ SET max_parallel_workers_per_gather = 4;
To speed up queries with an index, increase the number of inverted lists (at the expense of recall). To speed up queries with an index, increase the number of inverted lists (at the expense of recall).
```sql ```sql
CREATE INDEX ON table USING ivfflat (column opclass) WITH (lists = 1000); CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 1000);
``` ```
## Reference ## Reference
### Vector Type ### Vector Type
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. 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 1024 dimensions.
### Vector Operators ### Vector Operators
@@ -194,10 +192,13 @@ Libraries that use pgvector:
- [pgvector-python](https://github.com/pgvector/pgvector-python) (Python) - [pgvector-python](https://github.com/pgvector/pgvector-python) (Python)
- [Neighbor](https://github.com/ankane/neighbor) (Ruby) - [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-node](https://github.com/pgvector/pgvector-node) (Node.js)
- [pgvector-go](https://github.com/pgvector/pgvector-go) (Go) - [pgvector-go](https://github.com/pgvector/pgvector-go) (Go)
- [pgvector-php](https://github.com/pgvector/pgvector-php) (PHP)
- [pgvector-rust](https://github.com/pgvector/pgvector-rust) (Rust) - [pgvector-rust](https://github.com/pgvector/pgvector-rust) (Rust)
- [pgvector-cpp](https://github.com/pgvector/pgvector-cpp) (C++) - [pgvector-cpp](https://github.com/pgvector/pgvector-cpp) (C++)
- [pgvector-elixir](https://github.com/pgvector/pgvector-elixir) (Elixir)
## Frequently Asked Questions ## Frequently Asked Questions
@@ -231,14 +232,14 @@ 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 ```sh
git clone --branch v0.2.4 https://github.com/pgvector/pgvector.git git clone --branch v0.3.2 https://github.com/pgvector/pgvector.git
cd pgvector cd pgvector
docker build -t pgvector . docker build -t pgvector .
``` ```
### Homebrew ### Homebrew
On Mac with Homebrew Postgres, you can use: With Homebrew Postgres, you can use:
```sh ```sh
brew install pgvector/brew/pgvector brew install pgvector/brew/pgvector
@@ -269,6 +270,22 @@ Install the latest version and run:
ALTER EXTENSION vector UPDATE; ALTER EXTENSION vector UPDATE;
``` ```
## Upgrade Notes
### 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
Thanks to: Thanks to:
@@ -315,6 +332,12 @@ make installcheck REGRESS=functions # regression test
make prove_installcheck PROVE_TESTS=test/t/001_wal.pl # TAP test make prove_installcheck PROVE_TESTS=test/t/001_wal.pl # TAP test
``` ```
To enable benchmarking:
```sh
make clean && PG_CFLAGS=-DIVFFLAT_BENCH make && make install
```
Resources for contributors Resources for contributors
- [Extension Building Infrastructure](https://www.postgresql.org/docs/current/extend-pgxs.html) - [Extension Building Infrastructure](https://www.postgresql.org/docs/current/extend-pgxs.html)

View File

@@ -0,0 +1,2 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.2.5'" to load this file. \quit

View File

@@ -0,0 +1,2 @@
-- 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

@@ -0,0 +1,2 @@
-- 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

@@ -0,0 +1,2 @@
-- 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

@@ -0,0 +1,2 @@
-- 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

@@ -0,0 +1,2 @@
-- 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

@@ -36,16 +36,11 @@
#define CALLBACK_ITEM_POINTER HeapTuple hup #define CALLBACK_ITEM_POINTER HeapTuple hup
#endif #endif
/*
* Update build phase progress
*/
static inline void
UpdateProgress(int index, int64 val)
{
#if PG_VERSION_NUM >= 120000 #if PG_VERSION_NUM >= 120000
pgstat_progress_update_param(index, val); #define UpdateProgress(index, val) pgstat_progress_update_param(index, val)
#else
#define UpdateProgress(index, val) ((void)val)
#endif #endif
}
/* /*
* Callback for sampling * Callback for sampling
@@ -85,7 +80,11 @@ SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
if (buildstate->rowstoskip <= 0) 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)); int k = (int) (targsamples * sampler_random_fract(buildstate->rstate.randstate));
#endif
Assert(k >= 0 && k < targsamples); Assert(k >= 0 && k < targsamples);
VectorArraySet(samples, k, DatumGetVector(value)); VectorArraySet(samples, k, DatumGetVector(value));
@@ -108,7 +107,7 @@ SampleRows(IvfflatBuildState * buildstate)
buildstate->rowstoskip = -1; buildstate->rowstoskip = -1;
BlockSampler_Init(&buildstate->bs, totalblocks, targsamples, random()); BlockSampler_Init(&buildstate->bs, totalblocks, targsamples, RandomInt());
reservoir_init_selection_state(&buildstate->rstate, targsamples); reservoir_init_selection_state(&buildstate->rstate, targsamples);
while (BlockSampler_HasMore(&buildstate->bs)) while (BlockSampler_HasMore(&buildstate->bs))
@@ -117,13 +116,13 @@ SampleRows(IvfflatBuildState * buildstate)
#if PG_VERSION_NUM >= 120000 #if PG_VERSION_NUM >= 120000
table_index_build_range_scan(buildstate->heap, buildstate->index, buildstate->indexInfo, table_index_build_range_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
false, true, true, targblock, 1, SampleCallback, (void *) buildstate, NULL); false, true, false, targblock, 1, SampleCallback, (void *) buildstate, NULL);
#elif PG_VERSION_NUM >= 110000 #elif PG_VERSION_NUM >= 110000
IndexBuildHeapRangeScan(buildstate->heap, buildstate->index, buildstate->indexInfo, IndexBuildHeapRangeScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
true, true, targblock, 1, SampleCallback, (void *) buildstate, NULL); false, true, targblock, 1, SampleCallback, (void *) buildstate, NULL);
#else #else
IndexBuildHeapRangeScan(buildstate->heap, buildstate->index, buildstate->indexInfo, IndexBuildHeapRangeScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
true, true, targblock, 1, SampleCallback, (void *) buildstate); false, true, targblock, 1, SampleCallback, (void *) buildstate);
#endif #endif
} }
} }
@@ -170,16 +169,20 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
} }
} }
#ifdef IVFFLAT_KMEANS_DEBUG
buildstate->inertia += minDistance;
buildstate->listSums[closestCenter] += minDistance;
buildstate->listCounts[closestCenter]++;
#endif
/* Create a virtual tuple */ /* Create a virtual tuple */
ExecClearTuple(slot); ExecClearTuple(slot);
slot->tts_values[0] = Int32GetDatum(closestCenter); slot->tts_values[0] = Int32GetDatum(closestCenter);
slot->tts_isnull[0] = false; slot->tts_isnull[0] = false;
slot->tts_values[1] = Int32GetDatum(ItemPointerGetBlockNumberNoCheck(tid)); slot->tts_values[1] = PointerGetDatum(tid);
slot->tts_isnull[1] = false; slot->tts_isnull[1] = false;
slot->tts_values[2] = Int32GetDatum(ItemPointerGetOffsetNumberNoCheck(tid)); slot->tts_values[2] = value;
slot->tts_isnull[2] = false; slot->tts_isnull[2] = false;
slot->tts_values[3] = value;
slot->tts_isnull[3] = false;
ExecStoreVirtualTuple(slot); ExecStoreVirtualTuple(slot);
/* /*
@@ -201,8 +204,6 @@ GetNextTuple(Tuplesortstate *sortstate, TupleDesc tupdesc, TupleTableSlot *slot,
{ {
Datum value; Datum value;
bool isnull; bool isnull;
int tupblk;
int tupoff;
#if PG_VERSION_NUM >= 100000 #if PG_VERSION_NUM >= 100000
if (tuplesort_gettupleslot(sortstate, true, false, slot, NULL)) if (tuplesort_gettupleslot(sortstate, true, false, slot, NULL))
@@ -211,13 +212,11 @@ GetNextTuple(Tuplesortstate *sortstate, TupleDesc tupdesc, TupleTableSlot *slot,
#endif #endif
{ {
*list = DatumGetInt32(slot_getattr(slot, 1, &isnull)); *list = DatumGetInt32(slot_getattr(slot, 1, &isnull));
tupblk = DatumGetInt32(slot_getattr(slot, 2, &isnull)); value = slot_getattr(slot, 3, &isnull);
tupoff = DatumGetInt32(slot_getattr(slot, 3, &isnull));
value = slot_getattr(slot, 4, &isnull);
/* Form the index tuple */ /* Form the index tuple */
*itup = index_form_tuple(tupdesc, &value, &isnull); *itup = index_form_tuple(tupdesc, &value, &isnull);
ItemPointerSet(&(*itup)->t_tid, tupblk, tupoff); (*itup)->t_tid = *((ItemPointer) DatumGetPointer(slot_getattr(slot, 2, &isnull)));
} }
else else
*list = -1; *list = -1;
@@ -234,8 +233,8 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
GenericXLogState *state; GenericXLogState *state;
int list; int list;
IndexTuple itup = NULL; /* silence compiler warning */ IndexTuple itup = NULL; /* silence compiler warning */
BlockNumber startPage = InvalidBlockNumber; BlockNumber startPage;
BlockNumber insertPage = InvalidBlockNumber; BlockNumber insertPage;
Size itemsz; Size itemsz;
int i; int i;
int64 inserted = 0; int64 inserted = 0;
@@ -260,7 +259,7 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
CHECK_FOR_INTERRUPTS(); CHECK_FOR_INTERRUPTS();
buf = IvfflatNewBuffer(index, forkNum); buf = IvfflatNewBuffer(index, forkNum);
IvfflatInitPage(index, &buf, &page, &state); IvfflatInitRegisterPage(index, &buf, &page, &state);
startPage = BufferGetBlockNumber(buf); startPage = BufferGetBlockNumber(buf);
@@ -326,17 +325,16 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
/* Create tuple description for sorting */ /* Create tuple description for sorting */
#if PG_VERSION_NUM >= 120000 #if PG_VERSION_NUM >= 120000
buildstate->tupdesc = CreateTemplateTupleDesc(4); buildstate->tupdesc = CreateTemplateTupleDesc(3);
#else #else
buildstate->tupdesc = CreateTemplateTupleDesc(4, false); buildstate->tupdesc = CreateTemplateTupleDesc(3, false);
#endif #endif
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 1, "list", INT4OID, -1, 0); TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 2, "blkno", INT4OID, -1, 0); TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 3, "offset", INT4OID, -1, 0);
#if PG_VERSION_NUM >= 110000 #if PG_VERSION_NUM >= 110000
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 4, "vector", RelationGetDescr(index)->attrs[0].atttypid, -1, 0); TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 3, "vector", RelationGetDescr(index)->attrs[0].atttypid, -1, 0);
#else #else
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 4, "vector", RelationGetDescr(index)->attrs[0]->atttypid, -1, 0); TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 3, "vector", RelationGetDescr(index)->attrs[0]->atttypid, -1, 0);
#endif #endif
#if PG_VERSION_NUM >= 120000 #if PG_VERSION_NUM >= 120000
@@ -350,6 +348,12 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
/* Reuse for each tuple */ /* Reuse for each tuple */
buildstate->normvec = InitVector(buildstate->dimensions); buildstate->normvec = InitVector(buildstate->dimensions);
#ifdef IVFFLAT_KMEANS_DEBUG
buildstate->inertia = 0;
buildstate->listSums = palloc0(sizeof(double) * buildstate->lists);
buildstate->listCounts = palloc0(sizeof(int) * buildstate->lists);
#endif
} }
/* /*
@@ -358,9 +362,14 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
static void static void
FreeBuildState(IvfflatBuildState * buildstate) FreeBuildState(IvfflatBuildState * buildstate)
{ {
pfree(buildstate->centers); VectorArrayFree(buildstate->centers);
pfree(buildstate->listInfo); pfree(buildstate->listInfo);
pfree(buildstate->normvec); pfree(buildstate->normvec);
#ifdef IVFFLAT_KMEANS_DEBUG
pfree(buildstate->listSums);
pfree(buildstate->listCounts);
#endif
} }
/* /*
@@ -377,17 +386,22 @@ ComputeCenters(IvfflatBuildState * buildstate)
if (numSamples < 10000) if (numSamples < 10000)
numSamples = 10000; numSamples = 10000;
/* Sample samples */ /* Skip samples for unlogged table */
if (buildstate->heap == NULL)
numSamples = 1;
/* Sample rows */
/* TODO Ensure within maintenance_work_mem */
buildstate->samples = VectorArrayInit(numSamples, buildstate->dimensions); buildstate->samples = VectorArrayInit(numSamples, buildstate->dimensions);
if (buildstate->heap != NULL) if (buildstate->heap != NULL)
SampleRows(buildstate); SampleRows(buildstate);
/* Calculate centers */ /* Calculate centers */
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_KMEANS); UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_KMEANS);
IvfflatKmeans(buildstate->index, buildstate->samples, buildstate->centers); IvfflatBench("k-means", IvfflatKmeans(buildstate->index, buildstate->samples, buildstate->centers));
/* Free samples before we allocate more memory */ /* Free samples before we allocate more memory */
pfree(buildstate->samples); VectorArrayFree(buildstate->samples);
} }
/* /*
@@ -402,7 +416,7 @@ CreateMetaPage(Relation index, int dimensions, int lists, ForkNumber forkNum)
IvfflatMetaPage metap; IvfflatMetaPage metap;
buf = IvfflatNewBuffer(index, forkNum); buf = IvfflatNewBuffer(index, forkNum);
IvfflatInitPage(index, &buf, &page, &state); IvfflatInitRegisterPage(index, &buf, &page, &state);
/* Set metapage data */ /* Set metapage data */
metap = IvfflatPageGetMeta(page); metap = IvfflatPageGetMeta(page);
@@ -435,7 +449,7 @@ CreateListPages(Relation index, VectorArray centers, int dimensions,
list = palloc(itemsz); list = palloc(itemsz);
buf = IvfflatNewBuffer(index, forkNum); buf = IvfflatNewBuffer(index, forkNum);
IvfflatInitPage(index, &buf, &page, &state); IvfflatInitRegisterPage(index, &buf, &page, &state);
for (i = 0; i < lists; i++) for (i = 0; i < lists; i++)
{ {
@@ -463,6 +477,51 @@ CreateListPages(Relation index, VectorArray centers, int dimensions,
pfree(list); 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 * Create entry pages
*/ */
@@ -497,8 +556,14 @@ CreateEntryPages(IvfflatBuildState * buildstate, ForkNumber forkNum)
#endif #endif
} }
/* Sort and insert */ /* Sort */
tuplesort_performsort(buildstate->sortstate); tuplesort_performsort(buildstate->sortstate);
#ifdef IVFFLAT_KMEANS_DEBUG
PrintKmeansMetrics(buildstate);
#endif
/* Insert */
InsertTuples(buildstate->index, buildstate, forkNum); InsertTuples(buildstate->index, buildstate, forkNum);
tuplesort_end(buildstate->sortstate); tuplesort_end(buildstate->sortstate);
} }
@@ -517,7 +582,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
/* Create pages */ /* Create pages */
CreateMetaPage(index, buildstate->dimensions, buildstate->lists, forkNum); CreateMetaPage(index, buildstate->dimensions, buildstate->lists, forkNum);
CreateListPages(index, buildstate->centers, buildstate->dimensions, buildstate->lists, forkNum, &buildstate->listInfo); CreateListPages(index, buildstate->centers, buildstate->dimensions, buildstate->lists, forkNum, &buildstate->listInfo);
CreateEntryPages(buildstate, forkNum); IvfflatBench("CreateEntryPages", CreateEntryPages(buildstate, forkNum));
FreeBuildState(buildstate); FreeBuildState(buildstate);
} }

View File

@@ -159,7 +159,12 @@ ivfflatvalidate(Oid opclassoid)
return true; return true;
} }
PG_FUNCTION_INFO_V1(ivfflathandler); /*
* Define index handler
*
* See https://www.postgresql.org/docs/current/index-api.html
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(ivfflathandler);
Datum Datum
ivfflathandler(PG_FUNCTION_ARGS) ivfflathandler(PG_FUNCTION_ARGS)
{ {
@@ -193,6 +198,7 @@ ivfflathandler(PG_FUNCTION_ARGS)
#endif #endif
amroutine->amkeytype = InvalidOid; amroutine->amkeytype = InvalidOid;
/* Interface functions */
amroutine->ambuild = ivfflatbuild; amroutine->ambuild = ivfflatbuild;
amroutine->ambuildempty = ivfflatbuildempty; amroutine->ambuildempty = ivfflatbuildempty;
amroutine->aminsert = ivfflatinsert; amroutine->aminsert = ivfflatinsert;
@@ -206,6 +212,9 @@ ivfflathandler(PG_FUNCTION_ARGS)
amroutine->ambuildphasename = ivfflatbuildphasename; amroutine->ambuildphasename = ivfflatbuildphasename;
#endif #endif
amroutine->amvalidate = ivfflatvalidate; amroutine->amvalidate = ivfflatvalidate;
#if PG_VERSION_NUM >= 140000
amroutine->amadjustmembers = NULL;
#endif
amroutine->ambeginscan = ivfflatbeginscan; amroutine->ambeginscan = ivfflatbeginscan;
amroutine->amrescan = ivfflatrescan; amroutine->amrescan = ivfflatrescan;
amroutine->amgettuple = ivfflatgettuple; amroutine->amgettuple = ivfflatgettuple;
@@ -213,6 +222,8 @@ ivfflathandler(PG_FUNCTION_ARGS)
amroutine->amendscan = ivfflatendscan; amroutine->amendscan = ivfflatendscan;
amroutine->ammarkpos = NULL; amroutine->ammarkpos = NULL;
amroutine->amrestrpos = NULL; amroutine->amrestrpos = NULL;
/* Interface functions to support parallel index scans */
#if PG_VERSION_NUM >= 100000 #if PG_VERSION_NUM >= 100000
amroutine->amestimateparallelscan = NULL; amroutine->amestimateparallelscan = NULL;
amroutine->aminitparallelscan = NULL; amroutine->aminitparallelscan = NULL;

View File

@@ -3,13 +3,26 @@
#include "postgres.h" #include "postgres.h"
#if PG_VERSION_NUM < 100000
#error "Requires PostgreSQL 10+"
#endif
#include "access/generic_xlog.h" #include "access/generic_xlog.h"
#include "access/reloptions.h" #include "access/reloptions.h"
#include "nodes/execnodes.h" #include "nodes/execnodes.h"
#include "port.h" /* for strtof() and random() */
#include "utils/sampling.h" #include "utils/sampling.h"
#include "utils/tuplesort.h" #include "utils/tuplesort.h"
#include "vector.h" #include "vector.h"
#if PG_VERSION_NUM >= 150000
#include "common/pg_prng.h"
#endif
#ifdef IVFFLAT_BENCH
#include "portability/instr_time.h"
#endif
/* Support functions */ /* Support functions */
#define IVFFLAT_DISTANCE_PROC 1 #define IVFFLAT_DISTANCE_PROC 1
#define IVFFLAT_NORM_PROC 2 #define IVFFLAT_NORM_PROC 2
@@ -39,20 +52,41 @@
#define IvfflatPageGetOpaque(page) ((IvfflatPageOpaque) PageGetSpecialPointer(page)) #define IvfflatPageGetOpaque(page) ((IvfflatPageOpaque) PageGetSpecialPointer(page))
#define IvfflatPageGetMeta(page) ((IvfflatMetaPageData *) PageGetContents(page)) #define IvfflatPageGetMeta(page) ((IvfflatMetaPageData *) PageGetContents(page))
#if PG_VERSION_NUM < 100000 #ifdef IVFFLAT_BENCH
#define ItemPointerGetBlockNumberNoCheck ItemPointerGetBlockNumber #define IvfflatBench(name, code) \
#define ItemPointerGetOffsetNumberNoCheck ItemPointerGetOffsetNumber do { \
instr_time start; \
instr_time duration; \
INSTR_TIME_SET_CURRENT(start); \
(code); \
INSTR_TIME_SET_CURRENT(duration); \
INSTR_TIME_SUBTRACT(duration, start); \
elog(INFO, "%s: %.3f ms", name, INSTR_TIME_GET_MILLISEC(duration)); \
} while (0)
#else
#define 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()
#endif #endif
/* Variables */ /* Variables */
extern int ivfflat_probes; extern int ivfflat_probes;
/* Exported functions */
PGDLLEXPORT void _PG_init(void);
typedef struct VectorArrayData typedef struct VectorArrayData
{ {
int length; int length;
int maxlen; int maxlen;
int dim; int dim;
Vector items[FLEXIBLE_ARRAY_MEMBER]; Vector *items;
} VectorArrayData; } VectorArrayData;
typedef VectorArrayData * VectorArray; typedef VectorArrayData * VectorArray;
@@ -97,6 +131,12 @@ typedef struct IvfflatBuildState
ListInfo *listInfo; ListInfo *listInfo;
Vector *normvec; Vector *normvec;
#ifdef IVFFLAT_KMEANS_DEBUG
double inertia;
double *listSums;
int *listCounts;
#endif
/* Sampling */ /* Sampling */
BlockSamplerData bs; BlockSamplerData bs;
ReservoirStateData rstate; ReservoirStateData rstate;
@@ -138,6 +178,7 @@ typedef IvfflatListData * IvfflatList;
typedef struct IvfflatScanList typedef struct IvfflatScanList
{ {
pairingheap_node ph_node;
BlockNumber startPage; BlockNumber startPage;
double distance; double distance;
} IvfflatScanList; } IvfflatScanList;
@@ -159,19 +200,21 @@ typedef struct IvfflatScanOpaqueData
FmgrInfo *normprocinfo; FmgrInfo *normprocinfo;
Oid collation; Oid collation;
/* Lists */
pairingheap *listQueue;
IvfflatScanList lists[FLEXIBLE_ARRAY_MEMBER]; /* must come last */ IvfflatScanList lists[FLEXIBLE_ARRAY_MEMBER]; /* must come last */
} IvfflatScanOpaqueData; } IvfflatScanOpaqueData;
typedef IvfflatScanOpaqueData * IvfflatScanOpaque; typedef IvfflatScanOpaqueData * IvfflatScanOpaque;
#define VECTOR_ARRAY_SIZE(_length, _dim) (offsetof(VectorArrayData, items) + _length * VECTOR_SIZE(_dim)) #define VECTOR_ARRAY_SIZE(_length, _dim) (sizeof(VectorArrayData) + (_length) * VECTOR_SIZE(_dim))
#define VECTOR_ARRAY_OFFSET(_arr, _offset) ((char*) _arr + offsetof(VectorArrayData, items) + (_offset) * VECTOR_SIZE(_arr->dim)) #define VECTOR_ARRAY_OFFSET(_arr, _offset) ((char*) (_arr)->items + (_offset) * VECTOR_SIZE((_arr)->dim))
#define VectorArrayGet(_arr, _offset) ((Vector *) VECTOR_ARRAY_OFFSET(_arr, _offset)) #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 */ /* Methods */
void _PG_init(void);
VectorArray VectorArrayInit(int maxlen, int dimensions); VectorArray VectorArrayInit(int maxlen, int dimensions);
void VectorArrayFree(VectorArray arr);
void PrintVectorArray(char *msg, VectorArray arr); void PrintVectorArray(char *msg, VectorArray arr);
void IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers); void IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
FmgrInfo *IvfflatOptionalProcInfo(Relation rel, uint16 procnum); FmgrInfo *IvfflatOptionalProcInfo(Relation rel, uint16 procnum);
@@ -181,7 +224,8 @@ void IvfflatUpdateList(Relation index, GenericXLogState *state, ListInfo listIn
void IvfflatCommitBuffer(Buffer buf, GenericXLogState *state); void IvfflatCommitBuffer(Buffer buf, GenericXLogState *state);
void IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum); void IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum);
Buffer IvfflatNewBuffer(Relation index, ForkNumber forkNum); Buffer IvfflatNewBuffer(Relation index, ForkNumber forkNum);
void IvfflatInitPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state); void IvfflatInitPage(Buffer buf, Page page);
void IvfflatInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state);
/* Index access methods */ /* Index access methods */
IndexBuildResult *ivfflatbuild(Relation heap, Relation index, IndexInfo *indexInfo); IndexBuildResult *ivfflatbuild(Relation heap, Relation index, IndexInfo *indexInfo);

View File

@@ -53,18 +53,6 @@ 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 * Insert a tuple into the index
*/ */
@@ -87,11 +75,18 @@ InsertTuple(Relation rel, IndexTuple itup, Relation heapRel, Datum *values)
itemsz = MAXALIGN(IndexTupleSize(itup)); itemsz = MAXALIGN(IndexTupleSize(itup));
Assert(itemsz <= BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(IvfflatPageOpaqueData))); Assert(itemsz <= BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(IvfflatPageOpaqueData)));
LoadInsertPage(rel, &buf, &page, &state, insertPage);
/* Find a page to insert the item */ /* Find a page to insert the item */
while (PageGetFreeSpace(page) < itemsz) for (;;)
{ {
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; insertPage = IvfflatPageGetOpaque(page)->nextblkno;
if (BlockNumberIsValid(insertPage)) if (BlockNumberIsValid(insertPage))
@@ -99,15 +94,45 @@ InsertTuple(Relation rel, IndexTuple itup, Relation heapRel, Datum *values)
/* Move to next page */ /* Move to next page */
GenericXLogAbort(state); GenericXLogAbort(state);
UnlockReleaseBuffer(buf); UnlockReleaseBuffer(buf);
LoadInsertPage(rel, &buf, &page, &state, insertPage);
} }
else else
{ {
/* Add a new page */ Buffer metabuf;
IvfflatAppendPage(rel, &buf, &page, &state, MAIN_FORKNUM); Buffer newbuf;
Page newpage;
insertPage = BufferGetBlockNumber(buf); /*
* 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);
/* 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 rest */
UnlockReleaseBuffer(newbuf);
UnlockReleaseBuffer(buf);
} }
} }

View File

@@ -11,17 +11,17 @@
* https://theory.stanford.edu/~sergei/papers/kMeansPP-soda.pdf * https://theory.stanford.edu/~sergei/papers/kMeansPP-soda.pdf
*/ */
static void static void
InitCenters(Relation index, VectorArray samples, VectorArray centers, double *lowerBound) InitCenters(Relation index, VectorArray samples, VectorArray centers, float *lowerBound)
{ {
FmgrInfo *procinfo; FmgrInfo *procinfo;
Oid collation; Oid collation;
int i; int i;
int j; int64 j;
double distance; double distance;
double sum; double sum;
double choice; double choice;
Vector *vec; Vector *vec;
double *weight = palloc(samples->length * sizeof(double)); float *weight = palloc(samples->length * sizeof(float));
int numCenters = centers->maxlen; int numCenters = centers->maxlen;
int numSamples = samples->length; int numSamples = samples->length;
@@ -29,7 +29,7 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, double *lo
collation = index->rd_indcollation[0]; collation = index->rd_indcollation[0];
/* Choose an initial center uniformly at random */ /* Choose an initial center uniformly at random */
VectorArraySet(centers, 0, VectorArrayGet(samples, random() % samples->length)); VectorArraySet(centers, 0, VectorArrayGet(samples, RandomInt() % samples->length));
centers->length++; centers->length++;
for (j = 0; j < numSamples; j++) for (j = 0; j < numSamples; j++)
@@ -66,7 +66,7 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, double *lo
break; break;
/* Choose new center using weighted probability distribution. */ /* Choose new center using weighted probability distribution. */
choice = sum * (((double) random()) / MAX_RANDOM_VALUE); choice = sum * RandomDouble();
for (j = 0; j < numSamples - 1; j++) for (j = 0; j < numSamples - 1; j++)
{ {
choice -= weight[j]; choice -= weight[j];
@@ -121,15 +121,18 @@ QuickCenters(Relation index, VectorArray samples, VectorArray centers)
FmgrInfo *normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_KMEANS_NORM_PROC); FmgrInfo *normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_KMEANS_NORM_PROC);
/* Copy existing vectors while avoiding duplicates */ /* Copy existing vectors while avoiding duplicates */
qsort(samples->items, samples->length, VECTOR_SIZE(samples->dim), CompareVectors); if (samples->length > 0)
for (i = 0; i < samples->length; i++)
{ {
vec = VectorArrayGet(samples, i); qsort(samples->items, samples->length, VECTOR_SIZE(samples->dim), CompareVectors);
for (i = 0; i < samples->length; i++)
if (i == 0 || CompareVectors(vec, VectorArrayGet(samples, i - 1)) != 0)
{ {
VectorArraySet(centers, centers->length, vec); vec = VectorArrayGet(samples, i);
centers->length++;
if (i == 0 || CompareVectors(vec, VectorArrayGet(samples, i - 1)) != 0)
{
VectorArraySet(centers, centers->length, vec);
centers->length++;
}
} }
} }
@@ -142,7 +145,7 @@ QuickCenters(Relation index, VectorArray samples, VectorArray centers)
vec->dim = dimensions; vec->dim = dimensions;
for (j = 0; j < dimensions; j++) for (j = 0; j < dimensions; j++)
vec->x[j] = ((double) random()) / MAX_RANDOM_VALUE; vec->x[j] = RandomDouble();
/* Normalize if needed (only needed for random centers) */ /* Normalize if needed (only needed for random centers) */
if (normprocinfo != NULL) if (normprocinfo != NULL)
@@ -169,19 +172,19 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
Vector *vec; Vector *vec;
Vector *newCenter; Vector *newCenter;
int iteration; int iteration;
int j; int64 j;
int k; int64 k;
int dimensions = centers->dim; int dimensions = centers->dim;
int numCenters = centers->maxlen; int numCenters = centers->maxlen;
int numSamples = samples->length; int numSamples = samples->length;
VectorArray newCenters; VectorArray newCenters;
int *centerCounts; int *centerCounts;
int *closestCenters; int *closestCenters;
double *lowerBound; float *lowerBound;
double *upperBound; float *upperBound;
double *s; float *s;
double *halfcdist; float *halfcdist;
double *newcdist; float *newcdist;
int changes; int changes;
double minDistance; double minDistance;
int closestCenter; int closestCenter;
@@ -191,19 +194,47 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
double dxcx; double dxcx;
double dxc; double dxc;
/* Calculate allocation sizes */
Size samplesSize = VECTOR_ARRAY_SIZE(samples->maxlen, samples->dim);
Size centersSize = VECTOR_ARRAY_SIZE(centers->maxlen, centers->dim);
Size newCentersSize = VECTOR_ARRAY_SIZE(numCenters, dimensions);
Size centerCountsSize = sizeof(int) * numCenters;
Size closestCentersSize = sizeof(int) * numSamples;
Size lowerBoundSize = sizeof(float) * numSamples * numCenters;
Size upperBoundSize = sizeof(float) * numSamples;
Size sSize = sizeof(float) * numCenters;
Size halfcdistSize = sizeof(float) * numCenters * numCenters;
Size newcdistSize = sizeof(float) * numCenters;
/* Calculate total size */
Size totalSize = samplesSize + centersSize + newCentersSize + centerCountsSize + closestCentersSize + lowerBoundSize + upperBoundSize + sSize + halfcdistSize + newcdistSize;
/* Check memory requirements */
/* Add one to error message to ceil */
if (totalSize / 1024 > maintenance_work_mem)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
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 */ /* Set support functions */
procinfo = index_getprocinfo(index, 1, IVFFLAT_KMEANS_DISTANCE_PROC); procinfo = index_getprocinfo(index, 1, IVFFLAT_KMEANS_DISTANCE_PROC);
normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_KMEANS_NORM_PROC); normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_KMEANS_NORM_PROC);
collation = index->rd_indcollation[0]; collation = index->rd_indcollation[0];
/* Allocate space */ /* Allocate space */
centerCounts = palloc(sizeof(int) * numCenters); /* Use float instead of double to save memory */
closestCenters = palloc(sizeof(int) * numSamples); centerCounts = palloc(centerCountsSize);
lowerBound = palloc(sizeof(double) * numSamples * numCenters); closestCenters = palloc(closestCentersSize);
upperBound = palloc(sizeof(double) * numSamples); lowerBound = palloc_extended(lowerBoundSize, MCXT_ALLOC_HUGE);
s = palloc(sizeof(double) * numCenters); upperBound = palloc(upperBoundSize);
halfcdist = palloc(sizeof(double) * numCenters * numCenters); s = palloc(sSize);
newcdist = palloc(sizeof(double) * numCenters); halfcdist = palloc_extended(halfcdistSize, MCXT_ALLOC_HUGE);
newcdist = palloc(newcdistSize);
newCenters = VectorArrayInit(numCenters, dimensions); newCenters = VectorArrayInit(numCenters, dimensions);
for (j = 0; j < numCenters; j++) for (j = 0; j < numCenters; j++)
@@ -222,8 +253,6 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
minDistance = DBL_MAX; minDistance = DBL_MAX;
closestCenter = -1; closestCenter = -1;
vec = VectorArrayGet(samples, j);
/* Find closest center */ /* Find closest center */
for (k = 0; k < numCenters; k++) for (k = 0; k < numCenters; k++)
{ {
@@ -376,7 +405,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
{ {
/* TODO Handle empty centers properly */ /* TODO Handle empty centers properly */
for (k = 0; k < dimensions; k++) for (k = 0; k < dimensions; k++)
vec->x[k] = ((double) random()) / MAX_RANDOM_VALUE; vec->x[k] = RandomDouble();
} }
/* Normalize if needed */ /* Normalize if needed */
@@ -414,7 +443,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
break; break;
} }
pfree(newCenters); VectorArrayFree(newCenters);
pfree(centerCounts); pfree(centerCounts);
pfree(closestCenters); pfree(closestCenters);
pfree(lowerBound); pfree(lowerBound);

View File

@@ -1,5 +1,7 @@
#include "postgres.h" #include "postgres.h"
#include <float.h>
#include "access/relscan.h" #include "access/relscan.h"
#include "ivfflat.h" #include "ivfflat.h"
#include "miscadmin.h" #include "miscadmin.h"
@@ -17,14 +19,12 @@
* Compare list distances * Compare list distances
*/ */
static int static int
CompareLists(const void *a, const void *b) CompareLists(const pairingheap_node *a, const pairingheap_node *b, void *arg)
{ {
double diff = (((IvfflatScanList *) a)->distance - ((IvfflatScanList *) b)->distance); if (((const IvfflatScanList *) a)->distance > ((const IvfflatScanList *) b)->distance)
if (diff > 0)
return 1; return 1;
if (diff < 0) if (((const IvfflatScanList *) a)->distance < ((const IvfflatScanList *) b)->distance)
return -1; return -1;
return 0; return 0;
@@ -45,6 +45,8 @@ GetScanLists(IndexScanDesc scan, Datum value)
int listCount = 0; int listCount = 0;
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque; IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
double distance; double distance;
IvfflatScanList *scanlist;
double maxDistance = DBL_MAX;
/* Search all list pages */ /* Search all list pages */
while (BlockNumberIsValid(nextblkno)) while (BlockNumberIsValid(nextblkno))
@@ -62,21 +64,39 @@ GetScanLists(IndexScanDesc scan, Datum value)
/* Use procinfo from the index instead of scan key for performance */ /* Use procinfo from the index instead of scan key for performance */
distance = DatumGetFloat8(FunctionCall2Coll(so->procinfo, so->collation, PointerGetDatum(&list->center), value)); distance = DatumGetFloat8(FunctionCall2Coll(so->procinfo, so->collation, PointerGetDatum(&list->center), value));
so->lists[listCount].startPage = list->startPage; if (listCount < so->probes)
so->lists[listCount].distance = distance; {
listCount++; 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;
}
} }
nextblkno = IvfflatPageGetOpaque(cpage)->nextblkno; nextblkno = IvfflatPageGetOpaque(cpage)->nextblkno;
UnlockReleaseBuffer(cbuf); UnlockReleaseBuffer(cbuf);
} }
/* Sort by distance */
qsort(so->lists, listCount, sizeof(IvfflatScanList), CompareLists);
if (so->probes > listCount)
so->probes = listCount;
} }
/* /*
@@ -94,7 +114,6 @@ GetScanItems(IndexScanDesc scan, Datum value)
OffsetNumber maxoffno; OffsetNumber maxoffno;
Datum datum; Datum datum;
bool isnull; bool isnull;
int i;
TupleDesc tupdesc = RelationGetDescr(scan->indexRelation); TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
#if PG_VERSION_NUM >= 120000 #if PG_VERSION_NUM >= 120000
@@ -111,9 +130,9 @@ GetScanItems(IndexScanDesc scan, Datum value)
BufferAccessStrategy bas = GetAccessStrategy(BAS_BULKREAD); BufferAccessStrategy bas = GetAccessStrategy(BAS_BULKREAD);
/* Search closest probes lists */ /* Search closest probes lists */
for (i = 0; i < so->probes; i++) while (!pairingheap_is_empty(so->listQueue))
{ {
searchPage = so->lists[i].startPage; searchPage = ((IvfflatScanList *) pairingheap_remove_first(so->listQueue))->startPage;
/* Search all entry pages for list */ /* Search all entry pages for list */
while (BlockNumberIsValid(searchPage)) while (BlockNumberIsValid(searchPage))
@@ -137,12 +156,10 @@ GetScanItems(IndexScanDesc scan, Datum value)
ExecClearTuple(slot); ExecClearTuple(slot);
slot->tts_values[0] = FunctionCall2Coll(so->procinfo, so->collation, datum, value); slot->tts_values[0] = FunctionCall2Coll(so->procinfo, so->collation, datum, value);
slot->tts_isnull[0] = false; slot->tts_isnull[0] = false;
slot->tts_values[1] = Int32GetDatum((int) ItemPointerGetBlockNumberNoCheck(&itup->t_tid)); slot->tts_values[1] = PointerGetDatum(&itup->t_tid);
slot->tts_isnull[1] = false; slot->tts_isnull[1] = false;
slot->tts_values[2] = Int32GetDatum((int) ItemPointerGetOffsetNumberNoCheck(&itup->t_tid)); slot->tts_values[2] = Int32GetDatum((int) searchPage);
slot->tts_isnull[2] = false; slot->tts_isnull[2] = false;
slot->tts_values[3] = Int32GetDatum((int) searchPage);
slot->tts_isnull[3] = false;
ExecStoreVirtualTuple(slot); ExecStoreVirtualTuple(slot);
tuplesort_puttupleslot(so->sortstate, slot); tuplesort_puttupleslot(so->sortstate, slot);
@@ -153,6 +170,8 @@ GetScanItems(IndexScanDesc scan, Datum value)
UnlockReleaseBuffer(buf); UnlockReleaseBuffer(buf);
} }
} }
tuplesort_performsort(so->sortstate);
} }
/* /*
@@ -168,13 +187,18 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
Oid sortOperators[] = {Float8LessOperator}; Oid sortOperators[] = {Float8LessOperator};
Oid sortCollations[] = {InvalidOid}; Oid sortCollations[] = {InvalidOid};
bool nullsFirstFlags[] = {false}; bool nullsFirstFlags[] = {false};
int probes = ivfflat_probes;
scan = RelationGetIndexScan(index, nkeys, norderbys); scan = RelationGetIndexScan(index, nkeys, norderbys);
lists = IvfflatGetLists(scan->indexRelation); lists = IvfflatGetLists(scan->indexRelation);
so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + lists * sizeof(IvfflatScanList)); if (probes > lists)
probes = lists;
so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + probes * sizeof(IvfflatScanList));
so->buf = InvalidBuffer; so->buf = InvalidBuffer;
so->first = true; so->first = true;
so->probes = probes;
/* Set support functions */ /* Set support functions */
so->procinfo = index_getprocinfo(index, 1, IVFFLAT_DISTANCE_PROC); so->procinfo = index_getprocinfo(index, 1, IVFFLAT_DISTANCE_PROC);
@@ -183,14 +207,13 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
/* Create tuple description for sorting */ /* Create tuple description for sorting */
#if PG_VERSION_NUM >= 120000 #if PG_VERSION_NUM >= 120000
so->tupdesc = CreateTemplateTupleDesc(4); so->tupdesc = CreateTemplateTupleDesc(3);
#else #else
so->tupdesc = CreateTemplateTupleDesc(4, false); so->tupdesc = CreateTemplateTupleDesc(3, false);
#endif #endif
TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0); TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "blkno", INT4OID, -1, 0); TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
TupleDescInitEntry(so->tupdesc, (AttrNumber) 3, "offset", INT4OID, -1, 0); TupleDescInitEntry(so->tupdesc, (AttrNumber) 3, "indexblkno", INT4OID, -1, 0);
TupleDescInitEntry(so->tupdesc, (AttrNumber) 4, "indexblkno", INT4OID, -1, 0);
/* Prep sort */ /* Prep sort */
#if PG_VERSION_NUM >= 110000 #if PG_VERSION_NUM >= 110000
@@ -205,6 +228,8 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
so->slot = MakeSingleTupleTableSlot(so->tupdesc); so->slot = MakeSingleTupleTableSlot(so->tupdesc);
#endif #endif
so->listQueue = pairingheap_allocate(CompareLists, scan);
scan->opaque = so; scan->opaque = so;
return scan; return scan;
@@ -224,7 +249,7 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
#endif #endif
so->first = true; so->first = true;
so->probes = ivfflat_probes; pairingheap_reset(so->listQueue);
if (keys && scan->numberOfKeys > 0) if (keys && scan->numberOfKeys > 0)
memmove(scan->keyData, keys, scan->numberOfKeys * sizeof(ScanKeyData)); memmove(scan->keyData, keys, scan->numberOfKeys * sizeof(ScanKeyData));
@@ -268,9 +293,8 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
return false; return false;
} }
GetScanLists(scan, value); IvfflatBench("GetScanLists", GetScanLists(scan, value));
GetScanItems(scan, value); IvfflatBench("GetScanItems", GetScanItems(scan, value));
tuplesort_performsort(so->sortstate);
so->first = false; so->first = false;
/* Clean up if we allocated a new value */ /* Clean up if we allocated a new value */
@@ -284,14 +308,13 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
if (tuplesort_gettupleslot(so->sortstate, true, so->slot, NULL)) if (tuplesort_gettupleslot(so->sortstate, true, so->slot, NULL))
#endif #endif
{ {
BlockNumber blkno = DatumGetInt32(slot_getattr(so->slot, 2, &so->isnull)); ItemPointer tid = (ItemPointer) DatumGetPointer(slot_getattr(so->slot, 2, &so->isnull));
OffsetNumber offset = DatumGetInt32(slot_getattr(so->slot, 3, &so->isnull)); BlockNumber indexblkno = DatumGetInt32(slot_getattr(so->slot, 3, &so->isnull));
BlockNumber indexblkno = DatumGetInt32(slot_getattr(so->slot, 4, &so->isnull));
#if PG_VERSION_NUM >= 120000 #if PG_VERSION_NUM >= 120000
ItemPointerSet(&scan->xs_heaptid, blkno, offset); scan->xs_heaptid = *tid;
#else #else
ItemPointerSet(&scan->xs_ctup.t_self, blkno, offset); scan->xs_ctup.t_self = *tid;
#endif #endif
if (BufferIsValid(so->buf)) if (BufferIsValid(so->buf))
@@ -324,6 +347,7 @@ ivfflatendscan(IndexScanDesc scan)
if (BufferIsValid(so->buf)) if (BufferIsValid(so->buf))
ReleaseBuffer(so->buf); ReleaseBuffer(so->buf);
pairingheap_free(so->listQueue);
tuplesort_end(so->sortstate); tuplesort_end(so->sortstate);
pfree(so); pfree(so);

View File

@@ -10,14 +10,25 @@
VectorArray VectorArray
VectorArrayInit(int maxlen, int dimensions) VectorArrayInit(int maxlen, int dimensions)
{ {
VectorArray res = palloc0(VECTOR_ARRAY_SIZE(maxlen, dimensions)); VectorArray res = palloc(sizeof(VectorArrayData));
res->length = 0; res->length = 0;
res->maxlen = maxlen; res->maxlen = maxlen;
res->dim = dimensions; res->dim = dimensions;
res->items = palloc_extended(maxlen * VECTOR_SIZE(dimensions), MCXT_ALLOC_ZERO | MCXT_ALLOC_HUGE);
return res; return res;
} }
/*
* Free a vector array
*/
void
VectorArrayFree(VectorArray arr)
{
pfree(arr->items);
pfree(arr);
}
/* /*
* Print vector array - useful for debugging * Print vector array - useful for debugging
*/ */
@@ -107,13 +118,22 @@ IvfflatNewBuffer(Relation index, ForkNumber forkNum)
* Init page * Init page
*/ */
void void
IvfflatInitPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state) 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)
{ {
*state = GenericXLogStart(index); *state = GenericXLogStart(index);
*page = GenericXLogRegisterBuffer(*state, *buf, GENERIC_XLOG_FULL_IMAGE); *page = GenericXLogRegisterBuffer(*state, *buf, GENERIC_XLOG_FULL_IMAGE);
PageInit(*page, BufferGetPageSize(*buf), sizeof(IvfflatPageOpaqueData)); IvfflatInitPage(*buf, *page);
IvfflatPageGetOpaque(*page)->nextblkno = InvalidBlockNumber;
IvfflatPageGetOpaque(*page)->page_id = IVFFLAT_PAGE_ID;
} }
/* /*
@@ -135,17 +155,27 @@ IvfflatCommitBuffer(Buffer buf, GenericXLogState *state)
void void
IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum) IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum)
{ {
Buffer prevbuf = *buf;
/* Get new buffer */ /* Get new buffer */
*buf = IvfflatNewBuffer(index, forkNum); Buffer newbuf = IvfflatNewBuffer(index, forkNum);
Page newpage = GenericXLogRegisterBuffer(*state, newbuf, GENERIC_XLOG_FULL_IMAGE);
/* Update and commit previous buffer */ /* Update the previous buffer */
IvfflatPageGetOpaque(*page)->nextblkno = BufferGetBlockNumber(*buf); IvfflatPageGetOpaque(*page)->nextblkno = BufferGetBlockNumber(newbuf);
IvfflatCommitBuffer(prevbuf, *state);
/* Init new page */ /* Init new page */
IvfflatInitPage(index, buf, page, state); 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;
} }
/* /*

View File

@@ -13,7 +13,10 @@
#include "utils/numeric.h" #include "utils/numeric.h"
#if PG_VERSION_NUM >= 120000 #if PG_VERSION_NUM >= 120000
#include "common/shortest_dec.h"
#include "utils/float.h" #include "utils/float.h"
#else
#include <float.h>
#endif #endif
#if PG_VERSION_NUM < 130000 #if PG_VERSION_NUM < 130000
@@ -106,14 +109,14 @@ PrintVector(char *msg, Vector * vector)
/* /*
* Convert textual representation to internal representation * Convert textual representation to internal representation
*/ */
PG_FUNCTION_INFO_V1(vector_in); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_in);
Datum Datum
vector_in(PG_FUNCTION_ARGS) vector_in(PG_FUNCTION_ARGS)
{ {
char *str = PG_GETARG_CSTRING(0); char *str = PG_GETARG_CSTRING(0);
int32 typmod = PG_GETARG_INT32(2); int32 typmod = PG_GETARG_INT32(2);
int i; int i;
double x[VECTOR_MAX_DIM]; float x[VECTOR_MAX_DIM];
int dim = 0; int dim = 0;
char *pt; char *pt;
char *stringEnd; char *stringEnd;
@@ -136,7 +139,8 @@ vector_in(PG_FUNCTION_ARGS)
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("vector cannot have more than %d dimensions", VECTOR_MAX_DIM))); errmsg("vector cannot have more than %d dimensions", VECTOR_MAX_DIM)));
x[dim] = strtod(pt, &stringEnd); /* Use strtof like float4in to avoid a double-rounding problem */
x[dim] = strtof(pt, &stringEnd);
CheckElement(x[dim]); CheckElement(x[dim]);
dim++; dim++;
@@ -182,35 +186,68 @@ vector_in(PG_FUNCTION_ARGS)
/* /*
* Convert internal representation to textual representation * Convert internal representation to textual representation
*/ */
PG_FUNCTION_INFO_V1(vector_out); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_out);
Datum Datum
vector_out(PG_FUNCTION_ARGS) vector_out(PG_FUNCTION_ARGS)
{ {
Vector *vector = PG_GETARG_VECTOR_P(0); Vector *vector = PG_GETARG_VECTOR_P(0);
StringInfoData buf;
int dim = vector->dim; int dim = vector->dim;
char *buf;
char *ptr;
int i; int i;
int n;
initStringInfo(&buf); #if PG_VERSION_NUM < 120000
int ndig = FLT_DIG + extra_float_digits;
appendStringInfoChar(&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++;
for (i = 0; i < dim; i++) for (i = 0; i < dim; i++)
{ {
if (i > 0) if (i > 0)
appendStringInfoString(&buf, ","); {
*ptr = ',';
ptr++;
}
appendStringInfoString(&buf, float8out_internal(vector->x[i])); #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;
} }
appendStringInfoChar(&buf, ']'); *ptr = ']';
ptr++;
*ptr = '\0';
PG_FREE_IF_COPY(vector, 0); PG_FREE_IF_COPY(vector, 0);
PG_RETURN_CSTRING(buf.data); PG_RETURN_CSTRING(buf);
} }
/* /*
* Convert type modifier * Convert type modifier
*/ */
PG_FUNCTION_INFO_V1(vector_typmod_in); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_typmod_in);
Datum Datum
vector_typmod_in(PG_FUNCTION_ARGS) vector_typmod_in(PG_FUNCTION_ARGS)
{ {
@@ -241,7 +278,7 @@ vector_typmod_in(PG_FUNCTION_ARGS)
/* /*
* Convert external binary representation to internal representation * Convert external binary representation to internal representation
*/ */
PG_FUNCTION_INFO_V1(vector_recv); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_recv);
Datum Datum
vector_recv(PG_FUNCTION_ARGS) vector_recv(PG_FUNCTION_ARGS)
{ {
@@ -273,7 +310,7 @@ vector_recv(PG_FUNCTION_ARGS)
/* /*
* Convert internal representation to the external binary representation * Convert internal representation to the external binary representation
*/ */
PG_FUNCTION_INFO_V1(vector_send); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_send);
Datum Datum
vector_send(PG_FUNCTION_ARGS) vector_send(PG_FUNCTION_ARGS)
{ {
@@ -293,7 +330,7 @@ vector_send(PG_FUNCTION_ARGS)
/* /*
* Convert vector to vector * Convert vector to vector
*/ */
PG_FUNCTION_INFO_V1(vector); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector);
Datum Datum
vector(PG_FUNCTION_ARGS) vector(PG_FUNCTION_ARGS)
{ {
@@ -308,7 +345,7 @@ vector(PG_FUNCTION_ARGS)
/* /*
* Convert array to vector * Convert array to vector
*/ */
PG_FUNCTION_INFO_V1(array_to_vector); PGDLLEXPORT PG_FUNCTION_INFO_V1(array_to_vector);
Datum Datum
array_to_vector(PG_FUNCTION_ARGS) array_to_vector(PG_FUNCTION_ARGS)
{ {
@@ -366,7 +403,7 @@ array_to_vector(PG_FUNCTION_ARGS)
/* /*
* Convert vector to float4[] * Convert vector to float4[]
*/ */
PG_FUNCTION_INFO_V1(vector_to_float4); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_to_float4);
Datum Datum
vector_to_float4(PG_FUNCTION_ARGS) vector_to_float4(PG_FUNCTION_ARGS)
{ {
@@ -389,12 +426,14 @@ vector_to_float4(PG_FUNCTION_ARGS)
/* /*
* Get the L2 distance between vectors * Get the L2 distance between vectors
*/ */
PG_FUNCTION_INFO_V1(l2_distance); PGDLLEXPORT PG_FUNCTION_INFO_V1(l2_distance);
Datum Datum
l2_distance(PG_FUNCTION_ARGS) l2_distance(PG_FUNCTION_ARGS)
{ {
Vector *a = PG_GETARG_VECTOR_P(0); Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1); Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0; double distance = 0.0;
double diff; double diff;
@@ -402,7 +441,7 @@ l2_distance(PG_FUNCTION_ARGS)
for (int i = 0; i < a->dim; i++) for (int i = 0; i < a->dim; i++)
{ {
diff = a->x[i] - b->x[i]; diff = ax[i] - bx[i];
distance += diff * diff; distance += diff * diff;
} }
@@ -413,12 +452,14 @@ l2_distance(PG_FUNCTION_ARGS)
* Get the L2 squared distance between vectors * Get the L2 squared distance between vectors
* This saves a sqrt calculation * This saves a sqrt calculation
*/ */
PG_FUNCTION_INFO_V1(vector_l2_squared_distance); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_l2_squared_distance);
Datum Datum
vector_l2_squared_distance(PG_FUNCTION_ARGS) vector_l2_squared_distance(PG_FUNCTION_ARGS)
{ {
Vector *a = PG_GETARG_VECTOR_P(0); Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1); Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0; double distance = 0.0;
double diff; double diff;
@@ -426,7 +467,7 @@ vector_l2_squared_distance(PG_FUNCTION_ARGS)
for (int i = 0; i < a->dim; i++) for (int i = 0; i < a->dim; i++)
{ {
diff = a->x[i] - b->x[i]; diff = ax[i] - bx[i];
distance += diff * diff; distance += diff * diff;
} }
@@ -436,18 +477,20 @@ vector_l2_squared_distance(PG_FUNCTION_ARGS)
/* /*
* Get the inner product of two vectors * Get the inner product of two vectors
*/ */
PG_FUNCTION_INFO_V1(inner_product); PGDLLEXPORT PG_FUNCTION_INFO_V1(inner_product);
Datum Datum
inner_product(PG_FUNCTION_ARGS) inner_product(PG_FUNCTION_ARGS)
{ {
Vector *a = PG_GETARG_VECTOR_P(0); Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1); Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0; double distance = 0.0;
CheckDims(a, b); CheckDims(a, b);
for (int i = 0; i < a->dim; i++) for (int i = 0; i < a->dim; i++)
distance += a->x[i] * b->x[i]; distance += ax[i] * bx[i];
PG_RETURN_FLOAT8(distance); PG_RETURN_FLOAT8(distance);
} }
@@ -455,18 +498,20 @@ inner_product(PG_FUNCTION_ARGS)
/* /*
* Get the negative inner product of two vectors * Get the negative inner product of two vectors
*/ */
PG_FUNCTION_INFO_V1(vector_negative_inner_product); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_negative_inner_product);
Datum Datum
vector_negative_inner_product(PG_FUNCTION_ARGS) vector_negative_inner_product(PG_FUNCTION_ARGS)
{ {
Vector *a = PG_GETARG_VECTOR_P(0); Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1); Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0; double distance = 0.0;
CheckDims(a, b); CheckDims(a, b);
for (int i = 0; i < a->dim; i++) for (int i = 0; i < a->dim; i++)
distance += a->x[i] * b->x[i]; distance += ax[i] * bx[i];
PG_RETURN_FLOAT8(distance * -1); PG_RETURN_FLOAT8(distance * -1);
} }
@@ -474,12 +519,14 @@ vector_negative_inner_product(PG_FUNCTION_ARGS)
/* /*
* Get the cosine distance between two vectors * Get the cosine distance between two vectors
*/ */
PG_FUNCTION_INFO_V1(cosine_distance); PGDLLEXPORT PG_FUNCTION_INFO_V1(cosine_distance);
Datum Datum
cosine_distance(PG_FUNCTION_ARGS) cosine_distance(PG_FUNCTION_ARGS)
{ {
Vector *a = PG_GETARG_VECTOR_P(0); Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1); Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
double distance = 0.0; double distance = 0.0;
double norma = 0.0; double norma = 0.0;
double normb = 0.0; double normb = 0.0;
@@ -488,9 +535,9 @@ cosine_distance(PG_FUNCTION_ARGS)
for (int i = 0; i < a->dim; i++) for (int i = 0; i < a->dim; i++)
{ {
distance += a->x[i] * b->x[i]; distance += ax[i] * bx[i];
norma += a->x[i] * a->x[i]; norma += ax[i] * ax[i];
normb += b->x[i] * b->x[i]; normb += bx[i] * bx[i];
} }
PG_RETURN_FLOAT8(1 - (distance / (sqrt(norma) * sqrt(normb)))); PG_RETURN_FLOAT8(1 - (distance / (sqrt(norma) * sqrt(normb))));
@@ -501,7 +548,7 @@ cosine_distance(PG_FUNCTION_ARGS)
* Currently uses angular distance since needs to satisfy triangle inequality * Currently uses angular distance since needs to satisfy triangle inequality
* Assumes inputs are unit vectors (skips norm) * Assumes inputs are unit vectors (skips norm)
*/ */
PG_FUNCTION_INFO_V1(vector_spherical_distance); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_spherical_distance);
Datum Datum
vector_spherical_distance(PG_FUNCTION_ARGS) vector_spherical_distance(PG_FUNCTION_ARGS)
{ {
@@ -526,7 +573,7 @@ vector_spherical_distance(PG_FUNCTION_ARGS)
/* /*
* Get the dimensions of a vector * Get the dimensions of a vector
*/ */
PG_FUNCTION_INFO_V1(vector_dims); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_dims);
Datum Datum
vector_dims(PG_FUNCTION_ARGS) vector_dims(PG_FUNCTION_ARGS)
{ {
@@ -538,15 +585,16 @@ vector_dims(PG_FUNCTION_ARGS)
/* /*
* Get the L2 norm of a vector * Get the L2 norm of a vector
*/ */
PG_FUNCTION_INFO_V1(vector_norm); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_norm);
Datum Datum
vector_norm(PG_FUNCTION_ARGS) vector_norm(PG_FUNCTION_ARGS)
{ {
Vector *a = PG_GETARG_VECTOR_P(0); Vector *a = PG_GETARG_VECTOR_P(0);
float *ax = a->x;
double norm = 0.0; double norm = 0.0;
for (int i = 0; i < a->dim; i++) for (int i = 0; i < a->dim; i++)
norm += a->x[i] * a->x[i]; norm += ax[i] * ax[i];
PG_RETURN_FLOAT8(sqrt(norm)); PG_RETURN_FLOAT8(sqrt(norm));
} }
@@ -554,20 +602,23 @@ vector_norm(PG_FUNCTION_ARGS)
/* /*
* Add vectors * Add vectors
*/ */
PG_FUNCTION_INFO_V1(vector_add); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_add);
Datum Datum
vector_add(PG_FUNCTION_ARGS) vector_add(PG_FUNCTION_ARGS)
{ {
Vector *a = PG_GETARG_VECTOR_P(0); Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1); Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
Vector *result; Vector *result;
int i; float *rx;
CheckDims(a, b); CheckDims(a, b);
result = InitVector(a->dim); result = InitVector(a->dim);
for (i = 0; i < a->dim; i++) rx = result->x;
result->x[i] = a->x[i] + b->x[i]; for (int i = 0, imax = a->dim; i < imax; i++)
rx[i] = ax[i] + bx[i];
PG_RETURN_POINTER(result); PG_RETURN_POINTER(result);
} }
@@ -575,20 +626,23 @@ vector_add(PG_FUNCTION_ARGS)
/* /*
* Subtract vectors * Subtract vectors
*/ */
PG_FUNCTION_INFO_V1(vector_sub); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_sub);
Datum Datum
vector_sub(PG_FUNCTION_ARGS) vector_sub(PG_FUNCTION_ARGS)
{ {
Vector *a = PG_GETARG_VECTOR_P(0); Vector *a = PG_GETARG_VECTOR_P(0);
Vector *b = PG_GETARG_VECTOR_P(1); Vector *b = PG_GETARG_VECTOR_P(1);
float *ax = a->x;
float *bx = b->x;
Vector *result; Vector *result;
int i; float *rx;
CheckDims(a, b); CheckDims(a, b);
result = InitVector(a->dim); result = InitVector(a->dim);
for (i = 0; i < a->dim; i++) rx = result->x;
result->x[i] = a->x[i] - b->x[i]; for (int i = 0, imax = a->dim; i < imax; i++)
rx[i] = ax[i] - bx[i];
PG_RETURN_POINTER(result); PG_RETURN_POINTER(result);
} }
@@ -617,7 +671,7 @@ vector_cmp_internal(Vector * a, Vector * b)
/* /*
* Less than * Less than
*/ */
PG_FUNCTION_INFO_V1(vector_lt); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_lt);
Datum Datum
vector_lt(PG_FUNCTION_ARGS) vector_lt(PG_FUNCTION_ARGS)
{ {
@@ -630,7 +684,7 @@ vector_lt(PG_FUNCTION_ARGS)
/* /*
* Less than or equal * Less than or equal
*/ */
PG_FUNCTION_INFO_V1(vector_le); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_le);
Datum Datum
vector_le(PG_FUNCTION_ARGS) vector_le(PG_FUNCTION_ARGS)
{ {
@@ -643,7 +697,7 @@ vector_le(PG_FUNCTION_ARGS)
/* /*
* Equal * Equal
*/ */
PG_FUNCTION_INFO_V1(vector_eq); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_eq);
Datum Datum
vector_eq(PG_FUNCTION_ARGS) vector_eq(PG_FUNCTION_ARGS)
{ {
@@ -656,7 +710,7 @@ vector_eq(PG_FUNCTION_ARGS)
/* /*
* Not equal * Not equal
*/ */
PG_FUNCTION_INFO_V1(vector_ne); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_ne);
Datum Datum
vector_ne(PG_FUNCTION_ARGS) vector_ne(PG_FUNCTION_ARGS)
{ {
@@ -669,7 +723,7 @@ vector_ne(PG_FUNCTION_ARGS)
/* /*
* Greater than or equal * Greater than or equal
*/ */
PG_FUNCTION_INFO_V1(vector_ge); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_ge);
Datum Datum
vector_ge(PG_FUNCTION_ARGS) vector_ge(PG_FUNCTION_ARGS)
{ {
@@ -682,7 +736,7 @@ vector_ge(PG_FUNCTION_ARGS)
/* /*
* Greater than * Greater than
*/ */
PG_FUNCTION_INFO_V1(vector_gt); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_gt);
Datum Datum
vector_gt(PG_FUNCTION_ARGS) vector_gt(PG_FUNCTION_ARGS)
{ {
@@ -695,7 +749,7 @@ vector_gt(PG_FUNCTION_ARGS)
/* /*
* Compare vectors * Compare vectors
*/ */
PG_FUNCTION_INFO_V1(vector_cmp); PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_cmp);
Datum Datum
vector_cmp(PG_FUNCTION_ARGS) vector_cmp(PG_FUNCTION_ARGS)
{ {

View File

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

View File

@@ -1,5 +1,3 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT ARRAY[1,2,3]::vector; SELECT ARRAY[1,2,3]::vector;
array array
--------- ---------

View File

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

View File

@@ -1,5 +1,3 @@
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]';
?column? ?column?
---------- ----------

View File

@@ -1,5 +1,3 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT '[1,2,3]'::vector; SELECT '[1,2,3]'::vector;
vector vector
--------- ---------
@@ -12,6 +10,12 @@ SELECT '[-1,2,3]'::vector;
[-1,2,3] [-1,2,3]
(1 row) (1 row)
SELECT '[1.23456]'::vector;
vector
-----------
[1.23456]
(1 row)
SELECT '[hello,1]'::vector; SELECT '[hello,1]'::vector;
ERROR: invalid input syntax for type vector: "hello" ERROR: invalid input syntax for type vector: "hello"
LINE 1: SELECT '[hello,1]'::vector; LINE 1: SELECT '[hello,1]'::vector;
@@ -53,3 +57,12 @@ LINE 1: SELECT '[1,]'::vector;
^ ^
SELECT '[1,2,3]'::vector(2); SELECT '[1,2,3]'::vector(2);
ERROR: expected 2 dimensions, not 3 ERROR: expected 2 dimensions, not 3
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
unnest
---------
[1,2,3]
[4,5,6]
(2 rows)
SELECT '{"[1,2,3]"}'::vector(2)[];
ERROR: expected 2 dimensions, not 3

View File

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

View File

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

View File

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

View File

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

View File

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

View File

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

3
test/perl/TestLib.pm Normal file
View File

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

View File

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

View File

@@ -1,6 +1,3 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT ARRAY[1,2,3]::vector; SELECT ARRAY[1,2,3]::vector;
SELECT ARRAY[1.0,2.0,3.0]::vector; SELECT ARRAY[1.0,2.0,3.0]::vector;
SELECT ARRAY[1,2,3]::float4[]::vector; SELECT ARRAY[1,2,3]::float4[]::vector;

View File

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

View File

@@ -1,6 +1,3 @@
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 '[1,2,3]'::vector - '[4,5,6]'; SELECT '[1,2,3]'::vector - '[4,5,6]';

View File

@@ -1,8 +1,6 @@
SET client_min_messages = warning;
CREATE EXTENSION IF NOT EXISTS vector;
SELECT '[1,2,3]'::vector; SELECT '[1,2,3]'::vector;
SELECT '[-1,2,3]'::vector; SELECT '[-1,2,3]'::vector;
SELECT '[1.23456]'::vector;
SELECT '[hello,1]'::vector; SELECT '[hello,1]'::vector;
SELECT '[NaN,1]'::vector; SELECT '[NaN,1]'::vector;
SELECT '[Infinity,1]'::vector; SELECT '[Infinity,1]'::vector;
@@ -13,3 +11,6 @@ SELECT '1,2,3'::vector;
SELECT '[]'::vector; SELECT '[]'::vector;
SELECT '[1,]'::vector; SELECT '[1,]'::vector;
SELECT '[1,2,3]'::vector(2); SELECT '[1,2,3]'::vector(2);
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
SELECT '{"[1,2,3]"}'::vector(2)[];

View File

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

View File

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

View File

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

View File

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

View File

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

View File

@@ -7,6 +7,8 @@ use PostgresNode;
use TestLib; use TestLib;
use Test::More tests => 31; use Test::More tests => 31;
my $dim = 32;
my $node_primary; my $node_primary;
my $node_replica; my $node_replica;
@@ -30,13 +32,16 @@ sub test_index_replay
$node_primary->poll_query_until('postgres', $caughtup_query) $node_primary->poll_query_until('postgres', $caughtup_query)
or die "Timed out while waiting for replica 1 to catch up"; or die "Timed out while waiting for replica 1 to catch up";
my $r1 = rand(); my @r = ();
my $r2 = rand(); for (1 .. $dim) {
my $r3 = rand(); push(@r, rand());
}
my $sql = join(",", @r);
my $queries = qq(SET enable_seqscan=off; my $queries = qq(
SELECT * FROM tst ORDER BY v <-> '[$r1,$r2,$r3]' LIMIT 10; SET enable_seqscan = off;
); SELECT * FROM tst ORDER BY v <-> '[$sql]' LIMIT 10;
);
# Run test queries and compare their result # Run test queries and compare their result
my $primary_result = $node_primary->safe_psql("postgres", $queries); my $primary_result = $node_primary->safe_psql("postgres", $queries);
@@ -46,9 +51,18 @@ SELECT * FROM tst ORDER BY v <-> '[$r1,$r2,$r3]' LIMIT 10;
return; 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 # Initialize primary node
$node_primary = get_new_node('primary'); $node_primary = get_new_node('primary');
$node_primary->init(allows_streaming => 1); $node_primary->init(allows_streaming => 1);
if ($dim > 32) {
# TODO use wal_keep_segments for Postgres < 13
$node_primary->append_conf('postgresql.conf', qq(wal_keep_size = 1GB));
}
$node_primary->start; $node_primary->start;
my $backup_name = 'my_backup'; my $backup_name = 'my_backup';
@@ -63,12 +77,11 @@ $node_replica->start;
# Create ivfflat index on primary # Create ivfflat index on primary
$node_primary->safe_psql("postgres", "CREATE EXTENSION vector;"); $node_primary->safe_psql("postgres", "CREATE EXTENSION vector;");
$node_primary->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));"); $node_primary->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
$node_primary->safe_psql("postgres", $node_primary->safe_psql("postgres",
"INSERT INTO tst SELECT i%10, ARRAY[random(), random(), random()] FROM generate_series(1,100000) i;" "INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
); );
$node_primary->safe_psql("postgres", $node_primary->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v);");
"CREATE INDEX ON tst USING ivfflat (v);");
# Test that queries give same result # Test that queries give same result
test_index_replay('initial'); test_index_replay('initial');
@@ -82,7 +95,7 @@ for my $i (1 .. 10)
test_index_replay("vacuum $i"); test_index_replay("vacuum $i");
my ($start, $end) = (100001 + ($i - 1) * 10000, 100000 + $i * 10000); my ($start, $end) = (100001 + ($i - 1) * 10000, 100000 + $i * 10000);
$node_primary->safe_psql("postgres", $node_primary->safe_psql("postgres",
"INSERT INTO tst SELECT i%10, ARRAY[random(), random(), random()] FROM generate_series($start,$end) i;" "INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series($start, $end) i;"
); );
test_index_replay("insert $i"); test_index_replay("insert $i");
} }

View File

@@ -13,7 +13,7 @@ $node->start;
$node->safe_psql("postgres", "CREATE EXTENSION vector;"); $node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));"); $node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));");
$node->safe_psql("postgres", $node->safe_psql("postgres",
"INSERT INTO tst SELECT i%10, ARRAY[i%1000, i%333, i%55] 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);"); $node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v);");
@@ -24,7 +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", "DELETE FROM tst;");
$node->safe_psql("postgres", "VACUUM tst;"); $node->safe_psql("postgres", "VACUUM tst;");
$node->safe_psql("postgres", $node->safe_psql("postgres",
"INSERT INTO tst SELECT i%10, ARRAY[i%1000, i%333, i%55] 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 # Check size

88
test/t/003_recall.pl Normal file
View File

@@ -0,0 +1,88 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More tests => 9;
my $node;
my @queries = ();
my @expected;
my $limit = 20;
sub test_recall
{
my ($probes, $min, $operator) = @_;
my $correct = 0;
my $total = 0;
for my $i (0 .. $#queries) {
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;
));
my @actual_ids = split("\n", $actual);
my %actual_set = map { $_ => 1 } @actual_ids;
my @expected_ids = split("\n", $expected[$i]);
foreach (@expected_ids) {
if (exists($actual_set{$_})) {
$correct++;
}
$total++;
}
}
cmp_ok($correct / $total, ">=", $min, $operator);
}
# Initialize node
$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, v vector(3));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 100000) i;"
);
# Generate queries
for (1..20) {
my $r1 = rand();
my $r2 = rand();
my $r3 = rand();
push(@queries, "[$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);
}

36
test/t/004_centers.pl Normal file
View File

@@ -0,0 +1,36 @@
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 (i int4, v vector(3));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, '[1,2,3]' FROM generate_series(1, 10) i;"
);
sub test_centers
{
my ($lists, $min) = @_;
my ($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING ivfflat (v) WITH (lists = $lists);");
is($ret, 0, $stderr);
}
# Test no error for duplicate centers
test_centers(5);
test_centers(10);
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, '[4,5,6]' FROM generate_series(1, 10) i;"
);
# Test no error for duplicate centers
test_centers(10);

View File

@@ -0,0 +1,45 @@
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);
}
}

31
test/t/006_lists.pl Normal file
View File

@@ -0,0 +1,31 @@
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/);

45
test/t/007_inserts.pl Normal file
View File

@@ -0,0 +1,45 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More tests => 5;
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;"
}
);
my $expected = 10000 + 5 * 100 * 10;
my $count = $node->safe_psql("postgres", "SELECT COUNT(*) FROM tst;");
is($count, $expected);
$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);

View File

@@ -1,4 +1,4 @@
comment = 'vector data type and ivfflat access method' comment = 'vector data type and ivfflat access method'
default_version = '0.2.4' default_version = '0.3.2'
module_pathname = '$libdir/vector' module_pathname = '$libdir/vector'
relocatable = true relocatable = true