mirror of
https://github.com/pgvector/pgvector.git
synced 2026-07-22 20:15:46 +08:00
Compare commits
23 Commits
windows-ar
...
intvec
| Author | SHA1 | Date | |
|---|---|---|---|
|
8d5631da72 | ||
|
|
54d91e1116 | ||
|
|
f298a72ada | ||
|
|
f5b225530f | ||
|
|
05944f5a6f | ||
|
|
a344ecb60d | ||
|
|
67de46487e | ||
|
|
f52abfc26c | ||
|
|
1066b4be60 | ||
|
|
6221779f74 | ||
|
|
4765df1c50 | ||
|
|
9514a7ffcd | ||
|
|
98deaa7407 | ||
|
|
faa25d31f7 | ||
|
|
39af767617 | ||
|
|
3424f46298 | ||
|
|
ab12f1622d | ||
|
|
2fb71a6183 | ||
|
|
67b4ba5a14 | ||
|
|
8597eb79aa | ||
|
|
ffca8e720c | ||
|
|
0296a08ccf | ||
|
|
b067b5f073 |
17
.github/workflows/build.yml
vendored
17
.github/workflows/build.yml
vendored
@@ -8,8 +8,8 @@ jobs:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
include:
|
||||
- postgres: 18
|
||||
os: ubuntu-24.04
|
||||
# - postgres: 18
|
||||
# os: ubuntu-24.04
|
||||
- postgres: 17
|
||||
os: ubuntu-24.04
|
||||
- postgres: 16
|
||||
@@ -46,8 +46,8 @@ jobs:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
include:
|
||||
- postgres: 17
|
||||
os: macos-15
|
||||
- postgres: 16
|
||||
os: macos-14
|
||||
- postgres: 14
|
||||
os: macos-13
|
||||
steps:
|
||||
@@ -70,13 +70,12 @@ jobs:
|
||||
tar xf $TAG.tar.gz
|
||||
mv postgres-$TAG postgres
|
||||
env:
|
||||
TAG: ${{ matrix.postgres == 17 && 'REL_17_2' || 'REL_14_15' }}
|
||||
TAG: ${{ matrix.postgres == 16 && 'REL_16_2' || 'REL_14_11' }}
|
||||
- run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl"
|
||||
env:
|
||||
PERL5LIB: /Users/runner/perl5/lib/perl5
|
||||
- run: make clean && $(brew --prefix llvm@$LLVM_VERSION)/bin/scan-build --status-bugs make
|
||||
- run: make clean && $(brew --prefix llvm@15)/bin/scan-build --status-bugs make
|
||||
env:
|
||||
LLVM_VERSION: ${{ matrix.os == 'macos-15' && 18 || 15 }}
|
||||
PG_CFLAGS: -DUSE_ASSERT_CHECKING
|
||||
windows:
|
||||
runs-on: windows-latest
|
||||
@@ -87,7 +86,7 @@ jobs:
|
||||
with:
|
||||
postgres-version: 14
|
||||
- run: |
|
||||
call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars32.bat" && ^
|
||||
call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
|
||||
nmake /NOLOGO /F Makefile.win && ^
|
||||
nmake /NOLOGO /F Makefile.win install && ^
|
||||
nmake /NOLOGO /F Makefile.win installcheck && ^
|
||||
@@ -126,7 +125,7 @@ jobs:
|
||||
- uses: actions/checkout@v4
|
||||
- uses: ankane/setup-postgres-valgrind@v1
|
||||
with:
|
||||
postgres-version: 17
|
||||
postgres-version: 16
|
||||
check-ub: yes
|
||||
- run: make OPTFLAGS=""
|
||||
- run: sudo --preserve-env=PG_CONFIG make install
|
||||
|
||||
@@ -1,10 +1,11 @@
|
||||
## 0.8.0 (2024-10-30)
|
||||
## 0.8.0 (unreleased)
|
||||
|
||||
- Added support for iterative index scans
|
||||
- Added `intvec` type
|
||||
- Added casts for arrays to `sparsevec`
|
||||
- Improved cost estimation for better index selection when filtering
|
||||
- Improved performance of HNSW index scans
|
||||
- Improved cost estimation
|
||||
- Improved performance of HNSW inserts and on-disk index builds
|
||||
- Reduced memory usage for HNSW index scans
|
||||
- Dropped support for Postgres 12
|
||||
|
||||
## 0.7.4 (2024-08-05)
|
||||
|
||||
2
LICENSE
2
LICENSE
@@ -1,4 +1,4 @@
|
||||
Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
|
||||
Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
|
||||
|
||||
Portions Copyright (c) 1994, The Regents of the University of California
|
||||
|
||||
|
||||
@@ -2,7 +2,7 @@
|
||||
"name": "vector",
|
||||
"abstract": "Open-source vector similarity search for Postgres",
|
||||
"description": "Supports L2 distance, inner product, and cosine distance",
|
||||
"version": "0.8.0",
|
||||
"version": "0.7.4",
|
||||
"maintainer": [
|
||||
"Andrew Kane <andrew@ankane.org>"
|
||||
],
|
||||
@@ -12,7 +12,7 @@
|
||||
"prereqs": {
|
||||
"runtime": {
|
||||
"requires": {
|
||||
"PostgreSQL": "13.0.0"
|
||||
"PostgreSQL": "12.0.0"
|
||||
}
|
||||
}
|
||||
},
|
||||
@@ -20,7 +20,7 @@
|
||||
"vector": {
|
||||
"file": "sql/vector.sql",
|
||||
"docfile": "README.md",
|
||||
"version": "0.8.0",
|
||||
"version": "0.7.4",
|
||||
"abstract": "Open-source vector similarity search for Postgres"
|
||||
}
|
||||
},
|
||||
|
||||
6
Makefile
6
Makefile
@@ -1,11 +1,11 @@
|
||||
EXTENSION = vector
|
||||
EXTVERSION = 0.8.0
|
||||
EXTVERSION = 0.7.4
|
||||
|
||||
MODULE_big = vector
|
||||
DATA = $(wildcard sql/*--*--*.sql)
|
||||
DATA_built = sql/$(EXTENSION)--$(EXTVERSION).sql
|
||||
OBJS = src/bitutils.o src/bitvec.o src/halfutils.o src/halfvec.o src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/sparsevec.o src/vector.o
|
||||
HEADERS = src/halfvec.h src/sparsevec.h src/vector.h
|
||||
OBJS = src/bitutils.o src/bitvec.o src/halfutils.o src/halfvec.o src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/intvec.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/sparsevec.o src/vector.o
|
||||
HEADERS = src/halfvec.h src/intvec.h src/sparsevec.h src/vector.h
|
||||
|
||||
TESTS = $(wildcard test/sql/*.sql)
|
||||
REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
|
||||
|
||||
@@ -1,9 +1,9 @@
|
||||
EXTENSION = vector
|
||||
EXTVERSION = 0.8.0
|
||||
EXTVERSION = 0.7.4
|
||||
|
||||
DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql
|
||||
OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
|
||||
HEADERS = src\halfvec.h src\sparsevec.h src\vector.h
|
||||
OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\intvec.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
|
||||
HEADERS = src\halfvec.h src\intvec.h src\sparsevec.h src\vector.h
|
||||
|
||||
REGRESS = bit btree cast copy halfvec hnsw_bit hnsw_halfvec hnsw_sparsevec hnsw_vector ivfflat_bit ivfflat_halfvec ivfflat_vector sparsevec vector_type
|
||||
REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
|
||||
|
||||
264
README.md
264
README.md
@@ -17,11 +17,11 @@ Plus [ACID](https://en.wikipedia.org/wiki/ACID) compliance, point-in-time recove
|
||||
|
||||
### Linux and Mac
|
||||
|
||||
Compile and install the extension (supports Postgres 13+)
|
||||
Compile and install the extension (supports Postgres 12+)
|
||||
|
||||
```sh
|
||||
cd /tmp
|
||||
git clone --branch v0.8.0 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
make
|
||||
make install # may need sudo
|
||||
@@ -46,12 +46,14 @@ Then use `nmake` to build:
|
||||
```cmd
|
||||
set "PGROOT=C:\Program Files\PostgreSQL\16"
|
||||
cd %TEMP%
|
||||
git clone --branch v0.8.0 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
nmake /F Makefile.win
|
||||
nmake /F Makefile.win install
|
||||
```
|
||||
|
||||
Note: Postgres 17 is not supported yet due to an upstream issue
|
||||
|
||||
See the [installation notes](#installation-notes---windows) if you run into issues
|
||||
|
||||
You can also install it with [Docker](#docker) or [conda-forge](#conda-forge).
|
||||
@@ -82,7 +84,7 @@ Get the nearest neighbors by L2 distance
|
||||
SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
|
||||
```
|
||||
|
||||
Also supports inner product (`<#>`), cosine distance (`<=>`), and L1 distance (`<+>`)
|
||||
Also supports inner product (`<#>`), cosine distance (`<=>`), and L1 distance (`<+>`, added in 0.7.0)
|
||||
|
||||
Note: `<#>` returns the negative inner product since Postgres only supports `ASC` order index scans on operators
|
||||
|
||||
@@ -146,9 +148,9 @@ Supported distance functions are:
|
||||
- `<->` - L2 distance
|
||||
- `<#>` - (negative) inner product
|
||||
- `<=>` - cosine distance
|
||||
- `<+>` - L1 distance
|
||||
- `<~>` - Hamming distance (binary vectors)
|
||||
- `<%>` - Jaccard distance (binary vectors)
|
||||
- `<+>` - L1 distance (added in 0.7.0)
|
||||
- `<~>` - Hamming distance (binary vectors, added in 0.7.0)
|
||||
- `<%>` - Jaccard distance (binary vectors, added in 0.7.0)
|
||||
|
||||
Get the nearest neighbors to a row
|
||||
|
||||
@@ -221,7 +223,7 @@ L2 distance
|
||||
CREATE INDEX ON items USING hnsw (embedding vector_l2_ops);
|
||||
```
|
||||
|
||||
Note: Use `halfvec_l2_ops` for `halfvec` and `sparsevec_l2_ops` for `sparsevec` (and similar with the other distance functions)
|
||||
Note: Use `halfvec_l2_ops` for `halfvec`, `intvec_l2_ops` for `intvec`, and `sparsevec_l2_ops` for `sparsevec` (and similar with the other distance functions)
|
||||
|
||||
Inner product
|
||||
|
||||
@@ -235,19 +237,19 @@ Cosine distance
|
||||
CREATE INDEX ON items USING hnsw (embedding vector_cosine_ops);
|
||||
```
|
||||
|
||||
L1 distance
|
||||
L1 distance - added in 0.7.0
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING hnsw (embedding vector_l1_ops);
|
||||
```
|
||||
|
||||
Hamming distance
|
||||
Hamming distance - added in 0.7.0
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING hnsw (embedding bit_hamming_ops);
|
||||
```
|
||||
|
||||
Jaccard distance
|
||||
Jaccard distance - added in 0.7.0
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING hnsw (embedding bit_jaccard_ops);
|
||||
@@ -256,9 +258,10 @@ CREATE INDEX ON items USING hnsw (embedding bit_jaccard_ops);
|
||||
Supported types are:
|
||||
|
||||
- `vector` - up to 2,000 dimensions
|
||||
- `halfvec` - up to 4,000 dimensions
|
||||
- `bit` - up to 64,000 dimensions
|
||||
- `sparsevec` - up to 1,000 non-zero elements
|
||||
- `halfvec` - up to 4,000 dimensions (added in 0.7.0)
|
||||
- `intvec` - up to 8,000 dimensions (added in 0.8.0)
|
||||
- `bit` - up to 64,000 dimensions (added in 0.7.0)
|
||||
- `sparsevec` - up to 1,000 non-zero elements (added in 0.7.0)
|
||||
|
||||
### Index Options
|
||||
|
||||
@@ -312,7 +315,7 @@ Note: Do not set `maintenance_work_mem` so high that it exhausts the memory on t
|
||||
|
||||
Like other index types, it’s faster to create an index after loading your initial data
|
||||
|
||||
You can also speed up index creation by increasing the number of parallel workers (2 by default)
|
||||
Starting with 0.6.0, you can also speed up index creation by increasing the number of parallel workers (2 by default)
|
||||
|
||||
```sql
|
||||
SET max_parallel_maintenance_workers = 7; -- plus leader
|
||||
@@ -322,7 +325,7 @@ For a large number of workers, you may also need to increase `max_parallel_worke
|
||||
|
||||
### Indexing Progress
|
||||
|
||||
Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING)
|
||||
Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING) with Postgres 12+
|
||||
|
||||
```sql
|
||||
SELECT phase, round(100.0 * blocks_done / nullif(blocks_total, 0), 1) AS "%" FROM pg_stat_progress_create_index;
|
||||
@@ -365,7 +368,7 @@ Cosine distance
|
||||
CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);
|
||||
```
|
||||
|
||||
Hamming distance
|
||||
Hamming distance - added in 0.7.0
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING ivfflat (embedding bit_hamming_ops) WITH (lists = 100);
|
||||
@@ -374,8 +377,8 @@ CREATE INDEX ON items USING ivfflat (embedding bit_hamming_ops) WITH (lists = 10
|
||||
Supported types are:
|
||||
|
||||
- `vector` - up to 2,000 dimensions
|
||||
- `halfvec` - up to 4,000 dimensions
|
||||
- `bit` - up to 64,000 dimensions
|
||||
- `halfvec` - up to 4,000 dimensions (added in 0.7.0)
|
||||
- `bit` - up to 64,000 dimensions (added in 0.7.0)
|
||||
|
||||
### Query Options
|
||||
|
||||
@@ -408,7 +411,7 @@ For a large number of workers, you may also need to increase `max_parallel_worke
|
||||
|
||||
### Indexing Progress
|
||||
|
||||
Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING)
|
||||
Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING) with Postgres 12+
|
||||
|
||||
```sql
|
||||
SELECT phase, round(100.0 * tuples_done / nullif(tuples_total, 0), 1) AS "%" FROM pg_stat_progress_create_index;
|
||||
@@ -425,128 +428,34 @@ Note: `%` is only populated during the `loading tuples` phase
|
||||
|
||||
## Filtering
|
||||
|
||||
There are a few ways to index nearest neighbor queries with a `WHERE` clause.
|
||||
There are a few ways to index nearest neighbor queries with a `WHERE` clause
|
||||
|
||||
```sql
|
||||
SELECT * FROM items WHERE category_id = 123 ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
|
||||
```
|
||||
|
||||
A good place to start is creating an index on the filter column. This can provide fast, exact nearest neighbor search in many cases. Postgres has a number of [index types](https://www.postgresql.org/docs/current/indexes-types.html) for this: B-tree (default), hash, GiST, SP-GiST, GIN, and BRIN.
|
||||
Create an index on one [or more](https://www.postgresql.org/docs/current/indexes-multicolumn.html) of the `WHERE` columns for exact search
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items (category_id);
|
||||
```
|
||||
|
||||
For multiple columns, consider a [multicolumn index](https://www.postgresql.org/docs/current/indexes-multicolumn.html).
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items (location_id, category_id);
|
||||
```
|
||||
|
||||
Exact indexes work well for conditions that match a low percentage of rows. Otherwise, [approximate indexes](#indexing) can work better.
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING hnsw (embedding vector_l2_ops);
|
||||
```
|
||||
|
||||
With approximate indexes, filtering is applied *after* the index is scanned. If a condition matches 10% of rows, with HNSW and the default `hnsw.ef_search` of 40, only 4 rows will match on average. For more rows, increase `hnsw.ef_search`.
|
||||
|
||||
```sql
|
||||
SET hnsw.ef_search = 200;
|
||||
```
|
||||
|
||||
Starting with 0.8.0, you can enable [iterative index scans](#iterative-index-scans), which will automatically scan more of the index when needed.
|
||||
|
||||
```sql
|
||||
SET hnsw.iterative_scan = strict_order;
|
||||
```
|
||||
|
||||
If filtering by only a few distinct values, consider [partial indexing](https://www.postgresql.org/docs/current/indexes-partial.html).
|
||||
Or a [partial index](https://www.postgresql.org/docs/current/indexes-partial.html) on the vector column for approximate search
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING hnsw (embedding vector_l2_ops) WHERE (category_id = 123);
|
||||
```
|
||||
|
||||
If filtering by many different values, consider [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html).
|
||||
Use [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) for approximate search on many different values of the `WHERE` columns
|
||||
|
||||
```sql
|
||||
CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
|
||||
```
|
||||
|
||||
## Iterative Index Scans
|
||||
|
||||
*Added in 0.8.0*
|
||||
|
||||
With approximate indexes, queries with filtering can return less results since filtering is applied *after* the index is scanned. Starting with 0.8.0, you can enable iterative index scans, which will automatically scan more of the index until enough results are found (or it reaches `hnsw.max_scan_tuples` or `ivfflat.max_probes`).
|
||||
|
||||
Iterative scans can use strict or relaxed ordering.
|
||||
|
||||
Strict ensures results are in the exact order by distance
|
||||
|
||||
```sql
|
||||
SET hnsw.iterative_scan = strict_order;
|
||||
```
|
||||
|
||||
Relaxed allows results to be slightly out of order by distance, but provides better recall
|
||||
|
||||
```sql
|
||||
SET hnsw.iterative_scan = relaxed_order;
|
||||
# or
|
||||
SET ivfflat.iterative_scan = relaxed_order;
|
||||
```
|
||||
|
||||
With relaxed ordering, you can use a [materialized CTE](https://www.postgresql.org/docs/current/queries-with.html#QUERIES-WITH-CTE-MATERIALIZATION) to get strict ordering
|
||||
|
||||
```sql
|
||||
WITH relaxed_results AS MATERIALIZED (
|
||||
SELECT id, embedding <-> '[1,2,3]' AS distance FROM items WHERE category_id = 123 ORDER BY distance LIMIT 5
|
||||
) SELECT * FROM relaxed_results ORDER BY distance;
|
||||
```
|
||||
|
||||
For queries that filter by distance, use a materialized CTE and place the distance filter outside of it for best performance (due to the [current behavior](https://www.postgresql.org/message-id/flat/CAOdR5yGUoMQ6j7M5hNUXrySzaqZVGf_Ne%2B8fwZMRKTFxU1nbJg%40mail.gmail.com) of the Postgres executor)
|
||||
|
||||
```sql
|
||||
WITH nearest_results AS MATERIALIZED (
|
||||
SELECT id, embedding <-> '[1,2,3]' AS distance FROM items ORDER BY distance LIMIT 5
|
||||
) SELECT * FROM nearest_results WHERE distance < 5 ORDER BY distance;
|
||||
```
|
||||
|
||||
Note: Place any other filters inside the CTE
|
||||
|
||||
### Iterative Scan Options
|
||||
|
||||
Since scanning a large portion of an approximate index is expensive, there are options to control when a scan ends.
|
||||
|
||||
#### HNSW
|
||||
|
||||
Specify the max number of tuples to visit (20,000 by default)
|
||||
|
||||
```sql
|
||||
SET hnsw.max_scan_tuples = 20000;
|
||||
```
|
||||
|
||||
Note: This is approximate and does not affect the initial scan
|
||||
|
||||
Specify the max amount of memory to use, as a multiple of `work_mem` (1 by default)
|
||||
|
||||
```sql
|
||||
SET hnsw.scan_mem_multiplier = 2;
|
||||
```
|
||||
|
||||
Note: Try increasing this if increasing `hnsw.max_scan_tuples` does not improve recall
|
||||
|
||||
#### IVFFlat
|
||||
|
||||
Specify the max number of probes
|
||||
|
||||
```sql
|
||||
SET ivfflat.max_probes = 100;
|
||||
```
|
||||
|
||||
Note: If this is lower than `ivfflat.probes`, `ivfflat.probes` will be used
|
||||
|
||||
## Half-Precision Vectors
|
||||
|
||||
*Added in 0.7.0*
|
||||
|
||||
Use the `halfvec` type to store half-precision vectors
|
||||
|
||||
```sql
|
||||
@@ -555,6 +464,8 @@ CREATE TABLE items (id bigserial PRIMARY KEY, embedding halfvec(3));
|
||||
|
||||
## Half-Precision Indexing
|
||||
|
||||
*Added in 0.7.0*
|
||||
|
||||
Index vectors at half precision for smaller indexes
|
||||
|
||||
```sql
|
||||
@@ -567,6 +478,16 @@ Get the nearest neighbors
|
||||
SELECT * FROM items ORDER BY embedding::halfvec(3) <-> '[1,2,3]' LIMIT 5;
|
||||
```
|
||||
|
||||
## Integer Vectors
|
||||
|
||||
*Added in 0.8.0*
|
||||
|
||||
Use the `intvec` type to store 8-bit integer vectors
|
||||
|
||||
```sql
|
||||
CREATE TABLE items (id bigserial PRIMARY KEY, embedding intvec(3));
|
||||
```
|
||||
|
||||
## Binary Vectors
|
||||
|
||||
Use the `bit` type to store binary vectors ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/imagehash/example.py))
|
||||
@@ -576,16 +497,24 @@ CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
|
||||
INSERT INTO items (embedding) VALUES ('000'), ('111');
|
||||
```
|
||||
|
||||
Get the nearest neighbors by Hamming distance
|
||||
Get the nearest neighbors by Hamming distance (added in 0.7.0)
|
||||
|
||||
```sql
|
||||
SELECT * FROM items ORDER BY embedding <~> '101' LIMIT 5;
|
||||
```
|
||||
|
||||
Or (before 0.7.0)
|
||||
|
||||
```sql
|
||||
SELECT * FROM items ORDER BY bit_count(embedding # '101') LIMIT 5;
|
||||
```
|
||||
|
||||
Also supports Jaccard distance (`<%>`)
|
||||
|
||||
## Binary Quantization
|
||||
|
||||
*Added in 0.7.0*
|
||||
|
||||
Use expression indexing for binary quantization
|
||||
|
||||
```sql
|
||||
@@ -608,6 +537,8 @@ SELECT * FROM (
|
||||
|
||||
## Sparse Vectors
|
||||
|
||||
*Added in 0.7.0*
|
||||
|
||||
Use the `sparsevec` type to store sparse vectors
|
||||
|
||||
```sql
|
||||
@@ -641,6 +572,8 @@ You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python
|
||||
|
||||
## Indexing Subvectors
|
||||
|
||||
*Added in 0.7.0*
|
||||
|
||||
Use expression indexing to index subvectors
|
||||
|
||||
```sql
|
||||
@@ -783,12 +716,8 @@ C | [pgvector-c](https://github.com/pgvector/pgvector-c)
|
||||
C++ | [pgvector-cpp](https://github.com/pgvector/pgvector-cpp)
|
||||
C#, F#, Visual Basic | [pgvector-dotnet](https://github.com/pgvector/pgvector-dotnet)
|
||||
Crystal | [pgvector-crystal](https://github.com/pgvector/pgvector-crystal)
|
||||
D | [pgvector-d](https://github.com/pgvector/pgvector-d)
|
||||
Dart | [pgvector-dart](https://github.com/pgvector/pgvector-dart)
|
||||
Elixir | [pgvector-elixir](https://github.com/pgvector/pgvector-elixir)
|
||||
Erlang | [pgvector-erlang](https://github.com/pgvector/pgvector-erlang)
|
||||
Fortran | [pgvector-fortran](https://github.com/pgvector/pgvector-fortran)
|
||||
Gleam | [pgvector-gleam](https://github.com/pgvector/pgvector-gleam)
|
||||
Go | [pgvector-go](https://github.com/pgvector/pgvector-go)
|
||||
Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell)
|
||||
Java, Kotlin, Groovy, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
|
||||
@@ -802,7 +731,6 @@ Perl | [pgvector-perl](https://github.com/pgvector/pgvector-perl)
|
||||
PHP | [pgvector-php](https://github.com/pgvector/pgvector-php)
|
||||
Python | [pgvector-python](https://github.com/pgvector/pgvector-python)
|
||||
R | [pgvector-r](https://github.com/pgvector/pgvector-r)
|
||||
Raku | [pgvector-raku](https://github.com/pgvector/pgvector-raku)
|
||||
Ruby | [pgvector-ruby](https://github.com/pgvector/pgvector-ruby), [Neighbor](https://github.com/ankane/neighbor)
|
||||
Rust | [pgvector-rust](https://github.com/pgvector/pgvector-rust)
|
||||
Swift | [pgvector-swift](https://github.com/pgvector/pgvector-swift)
|
||||
@@ -924,7 +852,7 @@ ALTER TABLE items ALTER COLUMN embedding SET STORAGE PLAIN;
|
||||
|
||||
#### Why are there less results for a query after adding an HNSW index?
|
||||
|
||||
Results are limited by the size of the dynamic candidate list (`hnsw.ef_search`), which is 40 by default. There may be even less results due to dead tuples or filtering conditions in the query. Enabling [iterative index scans](#iterative-index-scans) can help address this.
|
||||
Results are limited by the size of the dynamic candidate list (`hnsw.ef_search`). There may be even less results due to dead tuples or filtering conditions in the query. We recommend setting `hnsw.ef_search` to at least twice the `LIMIT` of the query. If you need more than 500 results, use an IVFFlat index instead.
|
||||
|
||||
Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
|
||||
|
||||
@@ -936,7 +864,7 @@ The index was likely created with too little data for the number of lists. Drop
|
||||
DROP INDEX index_name;
|
||||
```
|
||||
|
||||
Results can also be limited by the number of probes (`ivfflat.probes`). Enabling [iterative index scans](#iterative-index-scans) can address this.
|
||||
Results can also be limited by the number of probes (`ivfflat.probes`).
|
||||
|
||||
Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
|
||||
|
||||
@@ -1023,6 +951,30 @@ Function | Description | Added
|
||||
avg(halfvec) → halfvec | average | 0.7.0
|
||||
sum(halfvec) → halfvec | sum | 0.7.0
|
||||
|
||||
### Intvec Type
|
||||
|
||||
Each int vector takes `dimensions + 8` bytes of storage. Each element is a single byte signed integer. Int vectors can have up to 16,000 dimensions.
|
||||
|
||||
### Intvec Operators
|
||||
|
||||
Operator | Description | Added
|
||||
--- | --- | ---
|
||||
<-> | Euclidean distance | 0.8.0
|
||||
<#> | negative inner product | 0.8.0
|
||||
<=> | cosine distance | 0.8.0
|
||||
<+> | taxicab distance | 0.8.0
|
||||
|
||||
### Intvec Functions
|
||||
|
||||
Function | Description | Added
|
||||
--- | --- | ---
|
||||
cosine_distance(intvec, intvec) → double precision | cosine distance | 0.8.0
|
||||
inner_product(intvec, intvec) → double precision | inner product | 0.8.0
|
||||
l1_distance(intvec, intvec) → double precision | taxicab distance | 0.8.0
|
||||
l2_distance(intvec, intvec) → double precision | Euclidean distance | 0.8.0
|
||||
l2_norm(intvec) → double precision | Euclidean norm | 0.8.0
|
||||
vector_dims(intvec) → integer | number of dimensions | 0.8.0
|
||||
|
||||
### Bit Type
|
||||
|
||||
Each bit vector takes `dimensions / 8 + 8` bytes of storage. See the [Postgres docs](https://www.postgresql.org/docs/current/datatype-bit.html) for more info.
|
||||
@@ -1103,13 +1055,7 @@ Note: Replace `17` with your Postgres server version
|
||||
|
||||
### Missing SDK
|
||||
|
||||
If compilation fails and the output includes `warning: no such sysroot directory` on Mac, your Postgres installation points to a path that no longer exists.
|
||||
|
||||
```sh
|
||||
pg_config --cppflags
|
||||
```
|
||||
|
||||
Reinstall Postgres to fix this.
|
||||
If compilation fails and the output includes `warning: no such sysroot directory` on Mac, reinstall Xcode Command Line Tools.
|
||||
|
||||
### Portability
|
||||
|
||||
@@ -1127,14 +1073,6 @@ make OPTFLAGS=""
|
||||
|
||||
If compilation fails with `Cannot open include file: 'postgres.h': No such file or directory`, make sure `PGROOT` is correct.
|
||||
|
||||
### Mismatched Architecture
|
||||
|
||||
If compilation fails with `error C2196: case value '4' already used`, make sure `vcvars64.bat` was called.
|
||||
|
||||
### Missing Symbol
|
||||
|
||||
If linking fails with `unresolved external symbol float_to_shortest_decimal_bufn` with Postgres 17.0-17.2, upgrade to Postgres 17.3+.
|
||||
|
||||
### Permissions
|
||||
|
||||
If installation fails with `Access is denied`, re-run the installation instructions as an administrator.
|
||||
@@ -1154,17 +1092,11 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
|
||||
You can also build the image manually:
|
||||
|
||||
```sh
|
||||
git clone --branch v0.8.0 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector .
|
||||
```
|
||||
|
||||
If you increase `maintenance_work_mem`, make sure `--shm-size` is at least that size to avoid an error with parallel HNSW index builds.
|
||||
|
||||
```sh
|
||||
docker run --shm-size=1g ...
|
||||
```
|
||||
|
||||
### Homebrew
|
||||
|
||||
With Homebrew Postgres, you can use:
|
||||
@@ -1210,7 +1142,7 @@ Note: Replace `17` with your Postgres server version
|
||||
Install the FreeBSD package with:
|
||||
|
||||
```sh
|
||||
pkg install postgresql16-pgvector
|
||||
pkg install postgresql15-pgvector
|
||||
```
|
||||
|
||||
or the port with:
|
||||
@@ -1252,6 +1184,36 @@ You can check the version in the current database with:
|
||||
SELECT extversion FROM pg_extension WHERE extname = 'vector';
|
||||
```
|
||||
|
||||
## Upgrade Notes
|
||||
|
||||
### 0.6.0
|
||||
|
||||
#### Postgres 12
|
||||
|
||||
If upgrading with Postgres 12, remove this line from `sql/vector--0.5.1--0.6.0.sql`:
|
||||
|
||||
```sql
|
||||
ALTER TYPE vector SET (STORAGE = external);
|
||||
```
|
||||
|
||||
Then run `make install` and `ALTER EXTENSION vector UPDATE;`.
|
||||
|
||||
#### Docker
|
||||
|
||||
The Docker image is now published in the `pgvector` org, and there are tags for each supported version of Postgres (rather than a `latest` tag).
|
||||
|
||||
```sh
|
||||
docker pull pgvector/pgvector:pg16
|
||||
# or
|
||||
docker pull pgvector/pgvector:0.6.0-pg16
|
||||
```
|
||||
|
||||
Also, if you’ve increased `maintenance_work_mem`, make sure `--shm-size` is at least that size to avoid an error with parallel HNSW index builds.
|
||||
|
||||
```sh
|
||||
docker run --shm-size=1g ...
|
||||
```
|
||||
|
||||
## Thanks
|
||||
|
||||
Thanks to:
|
||||
|
||||
@@ -1,6 +1,188 @@
|
||||
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
|
||||
\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.0'" to load this file. \quit
|
||||
|
||||
CREATE FUNCTION hnsw_intvec_support(internal) RETURNS internal
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C;
|
||||
|
||||
CREATE TYPE intvec;
|
||||
|
||||
CREATE FUNCTION intvec_in(cstring, oid, integer) RETURNS intvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_out(intvec) RETURNS cstring
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_typmod_in(cstring[]) RETURNS integer
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_recv(internal, oid, integer) RETURNS intvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_send(intvec) RETURNS bytea
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE TYPE intvec (
|
||||
INPUT = intvec_in,
|
||||
OUTPUT = intvec_out,
|
||||
TYPMOD_IN = intvec_typmod_in,
|
||||
RECEIVE = intvec_recv,
|
||||
SEND = intvec_send,
|
||||
STORAGE = external
|
||||
);
|
||||
|
||||
CREATE FUNCTION l2_distance(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION inner_product(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION cosine_distance(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION l1_distance(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION vector_dims(intvec) RETURNS integer
|
||||
AS 'MODULE_PATHNAME', 'intvec_vector_dims' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION l2_norm(intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_lt(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_le(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_eq(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_ne(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_ge(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_gt(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_cmp(intvec, intvec) RETURNS int4
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_l2_squared_distance(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_negative_inner_product(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec(intvec, integer, boolean) RETURNS intvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_intvec(integer[], integer, boolean) RETURNS intvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_to_integer(intvec, integer, boolean) RETURNS integer[]
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE CAST (intvec AS intvec)
|
||||
WITH FUNCTION intvec(intvec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (intvec AS integer[])
|
||||
WITH FUNCTION intvec_to_integer(intvec, integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE CAST (integer[] AS intvec)
|
||||
WITH FUNCTION array_to_intvec(integer[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE OPERATOR <-> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = l2_distance,
|
||||
COMMUTATOR = '<->'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <#> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_negative_inner_product,
|
||||
COMMUTATOR = '<#>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <=> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = cosine_distance,
|
||||
COMMUTATOR = '<=>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <+> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = l1_distance,
|
||||
COMMUTATOR = '<+>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR < (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_lt,
|
||||
COMMUTATOR = > , NEGATOR = >= ,
|
||||
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR <= (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_le,
|
||||
COMMUTATOR = >= , NEGATOR = > ,
|
||||
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR = (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_eq,
|
||||
COMMUTATOR = = , NEGATOR = <> ,
|
||||
RESTRICT = eqsel, JOIN = eqjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR <> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_ne,
|
||||
COMMUTATOR = <> , NEGATOR = = ,
|
||||
RESTRICT = eqsel, JOIN = eqjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR >= (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_ge,
|
||||
COMMUTATOR = <= , NEGATOR = < ,
|
||||
RESTRICT = scalargesel, JOIN = scalargejoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR > (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_gt,
|
||||
COMMUTATOR = < , NEGATOR = <= ,
|
||||
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_ops
|
||||
DEFAULT FOR TYPE intvec USING btree AS
|
||||
OPERATOR 1 < ,
|
||||
OPERATOR 2 <= ,
|
||||
OPERATOR 3 = ,
|
||||
OPERATOR 4 >= ,
|
||||
OPERATOR 5 > ,
|
||||
FUNCTION 1 intvec_cmp(intvec, intvec);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_l2_ops
|
||||
FOR TYPE intvec USING hnsw AS
|
||||
OPERATOR 1 <-> (intvec, intvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 intvec_l2_squared_distance(intvec, intvec),
|
||||
FUNCTION 3 hnsw_intvec_support(internal);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_ip_ops
|
||||
FOR TYPE intvec USING hnsw AS
|
||||
OPERATOR 1 <#> (intvec, intvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 intvec_negative_inner_product(intvec, intvec),
|
||||
FUNCTION 3 hnsw_intvec_support(internal);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_cosine_ops
|
||||
FOR TYPE intvec USING hnsw AS
|
||||
OPERATOR 1 <=> (intvec, intvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 cosine_distance(intvec, intvec),
|
||||
FUNCTION 2 l2_norm(intvec),
|
||||
FUNCTION 3 hnsw_intvec_support(internal);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_l1_ops
|
||||
FOR TYPE intvec USING hnsw AS
|
||||
OPERATOR 1 <+> (intvec, intvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 l1_distance(intvec, intvec),
|
||||
FUNCTION 3 hnsw_intvec_support(internal);
|
||||
|
||||
CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
|
||||
196
sql/vector.sql
196
sql/vector.sql
@@ -272,6 +272,9 @@ CREATE FUNCTION ivfflat_bit_support(internal) RETURNS internal
|
||||
CREATE FUNCTION hnsw_halfvec_support(internal) RETURNS internal
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C;
|
||||
|
||||
CREATE FUNCTION hnsw_intvec_support(internal) RETURNS internal
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C;
|
||||
|
||||
CREATE FUNCTION hnsw_bit_support(internal) RETURNS internal
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C;
|
||||
|
||||
@@ -647,6 +650,199 @@ CREATE OPERATOR CLASS halfvec_l1_ops
|
||||
FUNCTION 1 l1_distance(halfvec, halfvec),
|
||||
FUNCTION 3 hnsw_halfvec_support(internal);
|
||||
|
||||
-- intvec type
|
||||
|
||||
CREATE TYPE intvec;
|
||||
|
||||
CREATE FUNCTION intvec_in(cstring, oid, integer) RETURNS intvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_out(intvec) RETURNS cstring
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_typmod_in(cstring[]) RETURNS integer
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_recv(internal, oid, integer) RETURNS intvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_send(intvec) RETURNS bytea
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE TYPE intvec (
|
||||
INPUT = intvec_in,
|
||||
OUTPUT = intvec_out,
|
||||
TYPMOD_IN = intvec_typmod_in,
|
||||
RECEIVE = intvec_recv,
|
||||
SEND = intvec_send,
|
||||
STORAGE = external
|
||||
);
|
||||
|
||||
-- intvec functions
|
||||
|
||||
CREATE FUNCTION l2_distance(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION inner_product(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION cosine_distance(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION l1_distance(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION vector_dims(intvec) RETURNS integer
|
||||
AS 'MODULE_PATHNAME', 'intvec_vector_dims' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION l2_norm(intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'intvec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- intvec private functions
|
||||
|
||||
CREATE FUNCTION intvec_lt(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_le(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_eq(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_ne(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_ge(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_gt(intvec, intvec) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_cmp(intvec, intvec) RETURNS int4
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_l2_squared_distance(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_negative_inner_product(intvec, intvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- intvec cast functions
|
||||
|
||||
CREATE FUNCTION intvec(intvec, integer, boolean) RETURNS intvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_intvec(integer[], integer, boolean) RETURNS intvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION intvec_to_integer(intvec, integer, boolean) RETURNS integer[]
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- intvec casts
|
||||
|
||||
CREATE CAST (intvec AS intvec)
|
||||
WITH FUNCTION intvec(intvec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (intvec AS integer[])
|
||||
WITH FUNCTION intvec_to_integer(intvec, integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE CAST (integer[] AS intvec)
|
||||
WITH FUNCTION array_to_intvec(integer[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
-- intvec operators
|
||||
|
||||
CREATE OPERATOR <-> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = l2_distance,
|
||||
COMMUTATOR = '<->'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <#> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_negative_inner_product,
|
||||
COMMUTATOR = '<#>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <=> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = cosine_distance,
|
||||
COMMUTATOR = '<=>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <+> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = l1_distance,
|
||||
COMMUTATOR = '<+>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR < (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_lt,
|
||||
COMMUTATOR = > , NEGATOR = >= ,
|
||||
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR <= (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_le,
|
||||
COMMUTATOR = >= , NEGATOR = > ,
|
||||
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR = (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_eq,
|
||||
COMMUTATOR = = , NEGATOR = <> ,
|
||||
RESTRICT = eqsel, JOIN = eqjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR <> (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_ne,
|
||||
COMMUTATOR = <> , NEGATOR = = ,
|
||||
RESTRICT = eqsel, JOIN = eqjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR >= (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_ge,
|
||||
COMMUTATOR = <= , NEGATOR = < ,
|
||||
RESTRICT = scalargesel, JOIN = scalargejoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR > (
|
||||
LEFTARG = intvec, RIGHTARG = intvec, PROCEDURE = intvec_gt,
|
||||
COMMUTATOR = < , NEGATOR = <= ,
|
||||
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
|
||||
);
|
||||
|
||||
-- intvec opclasses
|
||||
|
||||
CREATE OPERATOR CLASS intvec_ops
|
||||
DEFAULT FOR TYPE intvec USING btree AS
|
||||
OPERATOR 1 < ,
|
||||
OPERATOR 2 <= ,
|
||||
OPERATOR 3 = ,
|
||||
OPERATOR 4 >= ,
|
||||
OPERATOR 5 > ,
|
||||
FUNCTION 1 intvec_cmp(intvec, intvec);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_l2_ops
|
||||
FOR TYPE intvec USING hnsw AS
|
||||
OPERATOR 1 <-> (intvec, intvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 intvec_l2_squared_distance(intvec, intvec),
|
||||
FUNCTION 3 hnsw_intvec_support(internal);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_ip_ops
|
||||
FOR TYPE intvec USING hnsw AS
|
||||
OPERATOR 1 <#> (intvec, intvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 intvec_negative_inner_product(intvec, intvec),
|
||||
FUNCTION 3 hnsw_intvec_support(internal);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_cosine_ops
|
||||
FOR TYPE intvec USING hnsw AS
|
||||
OPERATOR 1 <=> (intvec, intvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 cosine_distance(intvec, intvec),
|
||||
FUNCTION 2 l2_norm(intvec),
|
||||
FUNCTION 3 hnsw_intvec_support(internal);
|
||||
|
||||
CREATE OPERATOR CLASS intvec_l1_ops
|
||||
FOR TYPE intvec USING hnsw AS
|
||||
OPERATOR 1 <+> (intvec, intvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 l1_distance(intvec, intvec),
|
||||
FUNCTION 3 hnsw_intvec_support(internal);
|
||||
|
||||
-- bit functions
|
||||
|
||||
CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8
|
||||
|
||||
@@ -8,9 +8,6 @@
|
||||
#error "Requires PostgreSQL 13+"
|
||||
#endif
|
||||
|
||||
/* Check architecture in first header */
|
||||
StaticAssertDecl(sizeof(Datum) == SIZEOF_DATUM, "Architecture mismatch");
|
||||
|
||||
extern uint64 (*BitHammingDistance) (uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 distance);
|
||||
extern double (*BitJaccardDistance) (uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 ab, uint64 aa, uint64 bb);
|
||||
|
||||
|
||||
36
src/hnsw.c
36
src/hnsw.c
@@ -18,17 +18,16 @@
|
||||
#define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
|
||||
#endif
|
||||
|
||||
static const struct config_enum_entry hnsw_iterative_scan_options[] = {
|
||||
{"off", HNSW_ITERATIVE_SCAN_OFF, false},
|
||||
{"relaxed_order", HNSW_ITERATIVE_SCAN_RELAXED, false},
|
||||
{"strict_order", HNSW_ITERATIVE_SCAN_STRICT, false},
|
||||
static const struct config_enum_entry hnsw_iterative_search_options[] = {
|
||||
{"off", HNSW_ITERATIVE_SEARCH_OFF, false},
|
||||
{"relaxed_order", HNSW_ITERATIVE_SEARCH_RELAXED, false},
|
||||
{"strict_order", HNSW_ITERATIVE_SEARCH_STRICT, false},
|
||||
{NULL, 0, false}
|
||||
};
|
||||
|
||||
int hnsw_ef_search;
|
||||
int hnsw_iterative_scan;
|
||||
int hnsw_max_scan_tuples;
|
||||
double hnsw_scan_mem_multiplier;
|
||||
int hnsw_max_search_tuples;
|
||||
int hnsw_iterative_search;
|
||||
int hnsw_lock_tranche_id;
|
||||
static relopt_kind hnsw_relopt_kind;
|
||||
|
||||
@@ -79,19 +78,14 @@ HnswInit(void)
|
||||
"Valid range is 1..1000.", &hnsw_ef_search,
|
||||
HNSW_DEFAULT_EF_SEARCH, HNSW_MIN_EF_SEARCH, HNSW_MAX_EF_SEARCH, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
|
||||
DefineCustomEnumVariable("hnsw.iterative_scan", "Sets the mode for iterative scans",
|
||||
NULL, &hnsw_iterative_scan,
|
||||
HNSW_ITERATIVE_SCAN_OFF, hnsw_iterative_scan_options, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
DefineCustomEnumVariable("hnsw.iterative_search", "Sets the iterative search mode",
|
||||
NULL, &hnsw_iterative_search,
|
||||
HNSW_ITERATIVE_SEARCH_OFF, hnsw_iterative_search_options, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
|
||||
/* This is approximate and does not affect the initial scan */
|
||||
DefineCustomIntVariable("hnsw.max_scan_tuples", "Sets the max number of tuples to visit for iterative scans",
|
||||
NULL, &hnsw_max_scan_tuples,
|
||||
20000, 1, INT_MAX, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
|
||||
/* Same range as hash_mem_multiplier */
|
||||
DefineCustomRealVariable("hnsw.scan_mem_multiplier", "Sets the multiple of work_mem to use for iterative scans",
|
||||
NULL, &hnsw_scan_mem_multiplier,
|
||||
1, 1, 1000, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
/* This is approximate and does not apply to the initial scan */
|
||||
DefineCustomIntVariable("hnsw.max_search_tuples", "Sets the max number of candidates to visit for iterative search",
|
||||
"-1 means no limit", &hnsw_max_search_tuples,
|
||||
-1, -1, INT_MAX, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
|
||||
MarkGUCPrefixReserved("hnsw");
|
||||
}
|
||||
@@ -137,10 +131,6 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
|
||||
*indexSelectivity = 0;
|
||||
*indexCorrelation = 0;
|
||||
*indexPages = 0;
|
||||
#if PG_VERSION_NUM >= 180000
|
||||
/* See "On disable_cost" thread on pgsql-hackers */
|
||||
path->path.disabled_nodes = 2;
|
||||
#endif
|
||||
return;
|
||||
}
|
||||
|
||||
|
||||
18
src/hnsw.h
18
src/hnsw.h
@@ -109,24 +109,23 @@
|
||||
|
||||
/* Variables */
|
||||
extern int hnsw_ef_search;
|
||||
extern int hnsw_iterative_scan;
|
||||
extern int hnsw_max_scan_tuples;
|
||||
extern double hnsw_scan_mem_multiplier;
|
||||
extern int hnsw_iterative_search;
|
||||
extern int hnsw_max_search_tuples;
|
||||
extern int hnsw_lock_tranche_id;
|
||||
|
||||
typedef enum HnswIterativeScanMode
|
||||
typedef enum HnswIterativeSearchMode
|
||||
{
|
||||
HNSW_ITERATIVE_SCAN_OFF,
|
||||
HNSW_ITERATIVE_SCAN_RELAXED,
|
||||
HNSW_ITERATIVE_SCAN_STRICT
|
||||
} HnswIterativeScanMode;
|
||||
HNSW_ITERATIVE_SEARCH_OFF,
|
||||
HNSW_ITERATIVE_SEARCH_RELAXED,
|
||||
HNSW_ITERATIVE_SEARCH_STRICT
|
||||
} HnswIterativeSearchMode;
|
||||
|
||||
typedef struct HnswElementData HnswElementData;
|
||||
typedef struct HnswNeighborArray HnswNeighborArray;
|
||||
|
||||
#define HnswPtrDeclare(type, relptrtype, ptrtype) \
|
||||
relptr_declare(type, relptrtype); \
|
||||
typedef union { type *ptr; relptrtype relptr; } ptrtype
|
||||
typedef union { type *ptr; relptrtype relptr; } ptrtype;
|
||||
|
||||
/* Pointers that can be absolute or relative */
|
||||
/* Use char for DatumPtr so works with Pointer */
|
||||
@@ -373,7 +372,6 @@ typedef struct HnswScanOpaqueData
|
||||
int m;
|
||||
int64 tuples;
|
||||
double previousDistance;
|
||||
Size maxMemory;
|
||||
MemoryContext tmpCtx;
|
||||
|
||||
/* Support functions */
|
||||
|
||||
@@ -41,7 +41,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
|
||||
ep = w;
|
||||
}
|
||||
|
||||
return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, &so->v, hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF ? &so->discarded : NULL, true, &so->tuples);
|
||||
return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, &so->v, hnsw_iterative_search != HNSW_ITERATIVE_SEARCH_OFF ? &so->discarded : NULL, true, &so->tuples);
|
||||
}
|
||||
|
||||
/*
|
||||
@@ -121,15 +121,14 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
|
||||
{
|
||||
IndexScanDesc scan;
|
||||
HnswScanOpaque so;
|
||||
double maxMemory;
|
||||
|
||||
scan = RelationGetIndexScan(index, nkeys, norderbys);
|
||||
|
||||
so = (HnswScanOpaque) palloc(sizeof(HnswScanOpaqueData));
|
||||
so->typeInfo = HnswGetTypeInfo(index);
|
||||
|
||||
/* Set support functions */
|
||||
HnswInitSupport(&so->support, index);
|
||||
so->first = true;
|
||||
so->v.tids = NULL;
|
||||
so->discarded = NULL;
|
||||
|
||||
/*
|
||||
* Use a lower max allocation size than default to allow scanning more
|
||||
@@ -137,12 +136,10 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
|
||||
*/
|
||||
so->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
|
||||
"Hnsw scan temporary context",
|
||||
0, 8 * 1024, 256 * 1024);
|
||||
0, 8 * 1024, 512 * 1024);
|
||||
|
||||
/* Calculate max memory */
|
||||
/* Add 256 extra bytes to fill last block when close */
|
||||
maxMemory = (double) work_mem * hnsw_scan_mem_multiplier * 1024.0 + 256;
|
||||
so->maxMemory = Min(maxMemory, (double) SIZE_MAX);
|
||||
/* Set support functions */
|
||||
HnswInitSupport(&so->support, index);
|
||||
|
||||
scan->opaque = so;
|
||||
|
||||
@@ -157,10 +154,13 @@ hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int no
|
||||
{
|
||||
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
|
||||
|
||||
if (so->v.tids != NULL)
|
||||
tidhash_reset(so->v.tids);
|
||||
|
||||
if (so->discarded != NULL)
|
||||
pairingheap_reset(so->discarded);
|
||||
|
||||
so->first = true;
|
||||
/* v and discarded are allocated in tmpCtx */
|
||||
so->v.tids = NULL;
|
||||
so->discarded = NULL;
|
||||
so->tuples = 0;
|
||||
so->previousDistance = -get_float8_infinity();
|
||||
MemoryContextReset(so->tmpCtx);
|
||||
@@ -233,15 +233,15 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
|
||||
if (list_length(so->w) == 0)
|
||||
{
|
||||
if (hnsw_iterative_scan == HNSW_ITERATIVE_SCAN_OFF)
|
||||
if (hnsw_iterative_search == HNSW_ITERATIVE_SEARCH_OFF)
|
||||
break;
|
||||
|
||||
/* Empty index */
|
||||
if (so->discarded == NULL)
|
||||
break;
|
||||
|
||||
/* Reached max number of tuples or memory limit */
|
||||
if (so->tuples >= hnsw_max_scan_tuples || MemoryContextMemAllocated(so->tmpCtx, false) > so->maxMemory)
|
||||
/* Reached max number of tuples */
|
||||
if (hnsw_max_search_tuples != -1 && so->tuples >= hnsw_max_search_tuples)
|
||||
{
|
||||
if (pairingheap_is_empty(so->discarded))
|
||||
break;
|
||||
@@ -249,6 +249,21 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
/* Return remaining tuples */
|
||||
so->w = lappend(so->w, HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded)));
|
||||
}
|
||||
/* Prevent scans from consuming too much memory */
|
||||
else if (MemoryContextMemAllocated(so->tmpCtx, false) > (Size) work_mem * 1024L)
|
||||
{
|
||||
if (pairingheap_is_empty(so->discarded))
|
||||
{
|
||||
ereport(DEBUG1,
|
||||
(errmsg("hnsw index scan exceeded work_mem after " INT64_FORMAT " tuples", so->tuples),
|
||||
errhint("Increase work_mem to scan more tuples.")));
|
||||
|
||||
break;
|
||||
}
|
||||
|
||||
/* Return remaining tuples */
|
||||
so->w = lappend(so->w, HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded)));
|
||||
}
|
||||
else
|
||||
{
|
||||
/*
|
||||
@@ -284,7 +299,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
so->w = list_delete_last(so->w);
|
||||
|
||||
/* Mark memory as free for next iteration */
|
||||
if (hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF)
|
||||
if (hnsw_iterative_search != HNSW_ITERATIVE_SEARCH_OFF)
|
||||
{
|
||||
pfree(element);
|
||||
pfree(sc);
|
||||
@@ -295,7 +310,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
|
||||
heaptid = &element->heaptids[--element->heaptidsLength];
|
||||
|
||||
if (hnsw_iterative_scan == HNSW_ITERATIVE_SCAN_STRICT)
|
||||
if (hnsw_iterative_search == HNSW_ITERATIVE_SEARCH_STRICT)
|
||||
{
|
||||
if (sc->distance < so->previousDistance)
|
||||
continue;
|
||||
|
||||
@@ -1,5 +1,6 @@
|
||||
#include "postgres.h"
|
||||
|
||||
#include <float.h>
|
||||
#include <math.h>
|
||||
|
||||
#include "access/generic_xlog.h"
|
||||
@@ -157,6 +158,9 @@ HnswInitSupport(HnswSupport * support, Relation index)
|
||||
Datum
|
||||
HnswNormValue(const HnswTypeInfo * typeInfo, Oid collation, Datum value)
|
||||
{
|
||||
if (!typeInfo->normalize)
|
||||
return value;
|
||||
|
||||
return DirectFunctionCall1Coll(typeInfo->normalize, collation, value);
|
||||
}
|
||||
|
||||
@@ -546,7 +550,14 @@ HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, Hns
|
||||
if (DatumGetPointer(q->value) == NULL)
|
||||
*distance = 0;
|
||||
else
|
||||
{
|
||||
*distance = HnswGetDistance(q->value, PointerGetDatum(&etup->data), support);
|
||||
|
||||
/* Needed for intvec cosine distance */
|
||||
/* TODO Improve */
|
||||
if (isnan(*distance))
|
||||
*distance = DBL_MAX;
|
||||
}
|
||||
}
|
||||
|
||||
/* Load element */
|
||||
@@ -857,7 +868,6 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
|
||||
{
|
||||
AddToVisited(base, v, sc->element, inMemory, &found);
|
||||
|
||||
/* OK to count elements instead of tuples */
|
||||
if (tuples != NULL)
|
||||
(*tuples)++;
|
||||
}
|
||||
@@ -890,7 +900,6 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
|
||||
else
|
||||
HnswLoadUnvisitedFromDisk(cElement, unvisited, &unvisitedLength, v, index, m, lm, lc);
|
||||
|
||||
/* OK to count elements instead of tuples */
|
||||
if (tuples != NULL)
|
||||
(*tuples) += unvisitedLength;
|
||||
|
||||
@@ -1393,7 +1402,20 @@ hnsw_halfvec_support(PG_FUNCTION_ARGS)
|
||||
};
|
||||
|
||||
PG_RETURN_POINTER(&typeInfo);
|
||||
}
|
||||
};
|
||||
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnsw_intvec_support);
|
||||
Datum
|
||||
hnsw_intvec_support(PG_FUNCTION_ARGS)
|
||||
{
|
||||
static const HnswTypeInfo typeInfo = {
|
||||
.maxDimensions = HNSW_MAX_DIM * 4,
|
||||
.normalize = NULL,
|
||||
.checkValue = NULL
|
||||
};
|
||||
|
||||
PG_RETURN_POINTER(&typeInfo);
|
||||
};
|
||||
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnsw_bit_support);
|
||||
Datum
|
||||
@@ -1406,7 +1428,7 @@ hnsw_bit_support(PG_FUNCTION_ARGS)
|
||||
};
|
||||
|
||||
PG_RETURN_POINTER(&typeInfo);
|
||||
}
|
||||
};
|
||||
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnsw_sparsevec_support);
|
||||
Datum
|
||||
@@ -1419,4 +1441,4 @@ hnsw_sparsevec_support(PG_FUNCTION_ARGS)
|
||||
};
|
||||
|
||||
PG_RETURN_POINTER(&typeInfo);
|
||||
}
|
||||
};
|
||||
|
||||
@@ -9,10 +9,6 @@
|
||||
#include "storage/lmgr.h"
|
||||
#include "utils/memutils.h"
|
||||
|
||||
#if PG_VERSION_NUM >= 180000
|
||||
#define vacuum_delay_point() vacuum_delay_point(false)
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Check if deleted list contains an index TID
|
||||
*/
|
||||
|
||||
754
src/intvec.c
Normal file
754
src/intvec.c
Normal file
@@ -0,0 +1,754 @@
|
||||
#include "postgres.h"
|
||||
|
||||
#include <limits.h>
|
||||
#include <math.h>
|
||||
|
||||
#include "catalog/pg_type.h"
|
||||
#include "fmgr.h"
|
||||
#include "intvec.h"
|
||||
#include "lib/stringinfo.h"
|
||||
#include "libpq/pqformat.h"
|
||||
#include "utils/array.h"
|
||||
#include "utils/builtins.h"
|
||||
#include "utils/lsyscache.h"
|
||||
|
||||
/*
|
||||
* Ensure same dimensions
|
||||
*/
|
||||
static inline void
|
||||
CheckDims(IntVector * a, IntVector * b)
|
||||
{
|
||||
if (a->dim != b->dim)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("different intvec dimensions %d and %d", a->dim, b->dim)));
|
||||
}
|
||||
|
||||
/*
|
||||
* Ensure expected dimensions
|
||||
*/
|
||||
static inline void
|
||||
CheckExpectedDim(int32 typmod, int dim)
|
||||
{
|
||||
if (typmod != -1 && typmod != dim)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("expected %d dimensions, not %d", typmod, dim)));
|
||||
}
|
||||
|
||||
/*
|
||||
* Ensure valid dimensions
|
||||
*/
|
||||
static inline void
|
||||
CheckDim(int dim)
|
||||
{
|
||||
if (dim < 1)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("intvec must have at least 1 dimension")));
|
||||
|
||||
if (dim > INTVEC_MAX_DIM)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
|
||||
errmsg("intvec cannot have more than %d dimensions", INTVEC_MAX_DIM)));
|
||||
}
|
||||
|
||||
/*
|
||||
* Ensure element in range
|
||||
*/
|
||||
static inline void
|
||||
CheckElement(long value)
|
||||
{
|
||||
if (value < SCHAR_MIN || value > SCHAR_MAX)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
|
||||
errmsg("value \"%ld\" is out of range for type intvec", value)));
|
||||
}
|
||||
|
||||
/*
|
||||
* Allocate and initialize a new int vector
|
||||
*/
|
||||
IntVector *
|
||||
InitIntVector(int dim)
|
||||
{
|
||||
IntVector *result;
|
||||
int size;
|
||||
|
||||
size = INTVEC_SIZE(dim);
|
||||
result = (IntVector *) palloc0(size);
|
||||
SET_VARSIZE(result, size);
|
||||
result->dim = dim;
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
/*
|
||||
* Check for whitespace, since array_isspace() is static
|
||||
*/
|
||||
static inline bool
|
||||
intvec_isspace(char ch)
|
||||
{
|
||||
if (ch == ' ' ||
|
||||
ch == '\t' ||
|
||||
ch == '\n' ||
|
||||
ch == '\r' ||
|
||||
ch == '\v' ||
|
||||
ch == '\f')
|
||||
return true;
|
||||
return false;
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert textual representation to internal representation
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_in);
|
||||
Datum
|
||||
intvec_in(PG_FUNCTION_ARGS)
|
||||
{
|
||||
char *lit = PG_GETARG_CSTRING(0);
|
||||
int32 typmod = PG_GETARG_INT32(2);
|
||||
int8 x[INTVEC_MAX_DIM];
|
||||
int dim = 0;
|
||||
char *pt = lit;
|
||||
IntVector *result;
|
||||
|
||||
while (intvec_isspace(*pt))
|
||||
pt++;
|
||||
|
||||
if (*pt != '[')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type intvec: \"%s\"", lit),
|
||||
errdetail("Vector contents must start with \"[\".")));
|
||||
|
||||
pt++;
|
||||
|
||||
while (intvec_isspace(*pt))
|
||||
pt++;
|
||||
|
||||
if (*pt == ']')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("intvec must have at least 1 dimension")));
|
||||
|
||||
for (;;)
|
||||
{
|
||||
long val;
|
||||
char *stringEnd;
|
||||
|
||||
if (dim == INTVEC_MAX_DIM)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
|
||||
errmsg("intvec cannot have more than %d dimensions", VECTOR_MAX_DIM)));
|
||||
|
||||
while (intvec_isspace(*pt))
|
||||
pt++;
|
||||
|
||||
/* Check for empty string like float4in */
|
||||
if (*pt == '\0')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type intvec: \"%s\"", lit)));
|
||||
|
||||
errno = 0;
|
||||
|
||||
/* Use similar logic as int2vectorin */
|
||||
val = strtol(pt, &stringEnd, 10);
|
||||
|
||||
if (stringEnd == pt)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type intvec: \"%s\"", lit)));
|
||||
|
||||
/* Check for range error like float4in */
|
||||
if (errno == ERANGE || val < SCHAR_MIN || val > SCHAR_MAX)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
|
||||
errmsg("\"%s\" is out of range for type intvec", pnstrdup(pt, stringEnd - pt))));
|
||||
|
||||
CheckElement(val);
|
||||
x[dim++] = val;
|
||||
|
||||
pt = stringEnd;
|
||||
|
||||
while (intvec_isspace(*pt))
|
||||
pt++;
|
||||
|
||||
if (*pt == ',')
|
||||
pt++;
|
||||
else if (*pt == ']')
|
||||
{
|
||||
pt++;
|
||||
break;
|
||||
}
|
||||
else
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type intvec: \"%s\"", lit)));
|
||||
}
|
||||
|
||||
/* Only whitespace is allowed after the closing brace */
|
||||
while (intvec_isspace(*pt))
|
||||
pt++;
|
||||
|
||||
if (*pt != '\0')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type intvec: \"%s\"", lit),
|
||||
errdetail("Junk after closing right brace.")));
|
||||
|
||||
CheckDim(dim);
|
||||
CheckExpectedDim(typmod, dim);
|
||||
|
||||
result = InitIntVector(dim);
|
||||
for (int i = 0; i < dim; i++)
|
||||
result->x[i] = x[i];
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert internal representation to textual representation
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_out);
|
||||
Datum
|
||||
intvec_out(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *vector = PG_GETARG_INTVEC_P(0);
|
||||
int dim = vector->dim;
|
||||
char *buf;
|
||||
char *ptr;
|
||||
|
||||
/*
|
||||
* Need:
|
||||
*
|
||||
* dim * 4 bytes for elements (-128 to 127)
|
||||
*
|
||||
* dim - 1 bytes for separator
|
||||
*
|
||||
* 3 bytes for [, ], and \0
|
||||
*/
|
||||
buf = (char *) palloc(5 * dim + 2);
|
||||
ptr = buf;
|
||||
|
||||
*ptr = '[';
|
||||
ptr++;
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
if (i > 0)
|
||||
{
|
||||
*ptr = ',';
|
||||
ptr++;
|
||||
}
|
||||
|
||||
#if PG_VERSION_NUM >= 140000
|
||||
ptr += pg_ltoa(vector->x[i], ptr);
|
||||
#else
|
||||
pg_ltoa(vector->x[i], ptr);
|
||||
while (*ptr != '\0')
|
||||
ptr++;
|
||||
#endif
|
||||
}
|
||||
*ptr = ']';
|
||||
ptr++;
|
||||
*ptr = '\0';
|
||||
|
||||
PG_FREE_IF_COPY(vector, 0);
|
||||
PG_RETURN_CSTRING(buf);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert type modifier
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_typmod_in);
|
||||
Datum
|
||||
intvec_typmod_in(PG_FUNCTION_ARGS)
|
||||
{
|
||||
ArrayType *ta = PG_GETARG_ARRAYTYPE_P(0);
|
||||
int32 *tl;
|
||||
int n;
|
||||
|
||||
tl = ArrayGetIntegerTypmods(ta, &n);
|
||||
|
||||
if (n != 1)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
|
||||
errmsg("invalid type modifier")));
|
||||
|
||||
if (*tl < 1)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
|
||||
errmsg("dimensions for type intvec must be at least 1")));
|
||||
|
||||
if (*tl > INTVEC_MAX_DIM)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
|
||||
errmsg("dimensions for type intvec cannot exceed %d", INTVEC_MAX_DIM)));
|
||||
|
||||
PG_RETURN_INT32(*tl);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert external binary representation to internal representation
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_recv);
|
||||
Datum
|
||||
intvec_recv(PG_FUNCTION_ARGS)
|
||||
{
|
||||
StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
|
||||
int32 typmod = PG_GETARG_INT32(2);
|
||||
IntVector *result;
|
||||
int16 dim;
|
||||
int16 unused;
|
||||
|
||||
dim = pq_getmsgint(buf, sizeof(int16));
|
||||
unused = pq_getmsgint(buf, sizeof(int16));
|
||||
|
||||
CheckDim(dim);
|
||||
CheckExpectedDim(typmod, dim);
|
||||
|
||||
if (unused != 0)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("expected unused to be 0, not %d", unused)));
|
||||
|
||||
result = InitIntVector(dim);
|
||||
for (int i = 0; i < dim; i++)
|
||||
result->x[i] = pq_getmsgint(buf, sizeof(int8));
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert internal representation to the external binary representation
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_send);
|
||||
Datum
|
||||
intvec_send(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *vec = PG_GETARG_INTVEC_P(0);
|
||||
StringInfoData buf;
|
||||
|
||||
pq_begintypsend(&buf);
|
||||
pq_sendint(&buf, vec->dim, sizeof(int16));
|
||||
pq_sendint(&buf, vec->unused, sizeof(int16));
|
||||
for (int i = 0; i < vec->dim; i++)
|
||||
pq_sendint8(&buf, vec->x[i]);
|
||||
|
||||
PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert int vector to int vector
|
||||
* This is needed to check the type modifier
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec);
|
||||
Datum
|
||||
intvec(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *vec = PG_GETARG_INTVEC_P(0);
|
||||
int32 typmod = PG_GETARG_INT32(1);
|
||||
|
||||
CheckExpectedDim(typmod, vec->dim);
|
||||
|
||||
PG_RETURN_POINTER(vec);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert array to intvec vector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(array_to_intvec);
|
||||
Datum
|
||||
array_to_intvec(PG_FUNCTION_ARGS)
|
||||
{
|
||||
ArrayType *array = PG_GETARG_ARRAYTYPE_P(0);
|
||||
int32 typmod = PG_GETARG_INT32(1);
|
||||
IntVector *result;
|
||||
int16 typlen;
|
||||
bool typbyval;
|
||||
char typalign;
|
||||
Datum *elemsp;
|
||||
int nelemsp;
|
||||
|
||||
if (ARR_NDIM(array) > 1)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("array must be 1-D")));
|
||||
|
||||
if (ARR_HASNULL(array) && array_contains_nulls(array))
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
|
||||
errmsg("array must not contain nulls")));
|
||||
|
||||
get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
|
||||
deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, NULL, &nelemsp);
|
||||
|
||||
CheckDim(nelemsp);
|
||||
CheckExpectedDim(typmod, nelemsp);
|
||||
|
||||
result = InitIntVector(nelemsp);
|
||||
|
||||
if (ARR_ELEMTYPE(array) == INT4OID)
|
||||
{
|
||||
for (int i = 0; i < nelemsp; i++)
|
||||
{
|
||||
long l = DatumGetInt32(elemsp[i]);
|
||||
|
||||
CheckElement(l);
|
||||
|
||||
result->x[i] = l;
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("unsupported array type")));
|
||||
}
|
||||
|
||||
/*
|
||||
* Free allocation from deconstruct_array. Do not free individual elements
|
||||
* when pass-by-reference since they point to original array.
|
||||
*/
|
||||
pfree(elemsp);
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert int vector to integer[]
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_to_integer);
|
||||
Datum
|
||||
intvec_to_integer(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *vec = PG_GETARG_INTVEC_P(0);
|
||||
Datum *datums;
|
||||
ArrayType *result;
|
||||
|
||||
datums = (Datum *) palloc(sizeof(Datum) * vec->dim);
|
||||
|
||||
for (int i = 0; i < vec->dim; i++)
|
||||
datums[i] = Int32GetDatum((int) vec->x[i]);
|
||||
|
||||
result = construct_array(datums, vec->dim, INT4OID, sizeof(int32), true, TYPALIGN_INT);
|
||||
|
||||
pfree(datums);
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
static int
|
||||
IntvecL2SquaredDistance(int dim, int8 *ax, int8 *bx)
|
||||
{
|
||||
int distance = 0;
|
||||
|
||||
/* Auto-vectorized */
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
int diff = (int) ax[i] - (int) bx[i];
|
||||
|
||||
distance += diff * diff;
|
||||
}
|
||||
|
||||
return distance;
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L2 distance between int vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_l2_distance);
|
||||
Datum
|
||||
intvec_l2_distance(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_FLOAT8(sqrt((double) IntvecL2SquaredDistance(a->dim, a->x, b->x)));
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L2 squared distance between int vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_l2_squared_distance);
|
||||
Datum
|
||||
intvec_l2_squared_distance(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_FLOAT8((double) IntvecL2SquaredDistance(a->dim, a->x, b->x));
|
||||
}
|
||||
|
||||
static int
|
||||
IntvecInnerProduct(int dim, int8 *ax, int8 *bx)
|
||||
{
|
||||
int distance = 0;
|
||||
|
||||
/* Auto-vectorized */
|
||||
for (int i = 0; i < dim; i++)
|
||||
distance += (int) ax[i] * (int) bx[i];
|
||||
|
||||
return distance;
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the inner product of two int vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_inner_product);
|
||||
Datum
|
||||
intvec_inner_product(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_FLOAT8((double) IntvecInnerProduct(a->dim, a->x, b->x));
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the negative inner product of two int vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_negative_inner_product);
|
||||
Datum
|
||||
intvec_negative_inner_product(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_FLOAT8((double) -IntvecInnerProduct(a->dim, a->x, b->x));
|
||||
}
|
||||
|
||||
static double
|
||||
IntvecCosineSimilarity(int dim, int8 *ax, int8 *bx)
|
||||
{
|
||||
int similarity = 0;
|
||||
int norma = 0;
|
||||
int normb = 0;
|
||||
|
||||
/* Auto-vectorized */
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
int axi = ax[i];
|
||||
int bxi = bx[i];
|
||||
|
||||
similarity += axi * bxi;
|
||||
norma += axi * axi;
|
||||
normb += bxi * bxi;
|
||||
}
|
||||
|
||||
/* Use sqrt(a * b) over sqrt(a) * sqrt(b) */
|
||||
return (double) similarity / sqrt((double) norma * (double) normb);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the cosine distance between two int vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_cosine_distance);
|
||||
Datum
|
||||
intvec_cosine_distance(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
double similarity;
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
similarity = IntvecCosineSimilarity(a->dim, a->x, b->x);
|
||||
|
||||
#ifdef _MSC_VER
|
||||
/* /fp:fast may not propagate NaN */
|
||||
if (isnan(similarity))
|
||||
PG_RETURN_FLOAT8(NAN);
|
||||
#endif
|
||||
|
||||
/* Keep in range */
|
||||
if (similarity > 1)
|
||||
similarity = 1;
|
||||
else if (similarity < -1)
|
||||
similarity = -1;
|
||||
|
||||
PG_RETURN_FLOAT8(1 - similarity);
|
||||
}
|
||||
|
||||
static int
|
||||
IntvecL1Distance(int dim, int8 *ax, int8 *bx)
|
||||
{
|
||||
int distance = 0;
|
||||
|
||||
/* Auto-vectorized */
|
||||
for (int i = 0; i < dim; i++)
|
||||
distance += abs((int) ax[i] - (int) bx[i]);
|
||||
|
||||
return distance;
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L1 distance between two int vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_l1_distance);
|
||||
Datum
|
||||
intvec_l1_distance(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_FLOAT8((double) IntvecL1Distance(a->dim, a->x, b->x));
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the dimensions of an int vector
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_vector_dims);
|
||||
Datum
|
||||
intvec_vector_dims(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
|
||||
PG_RETURN_INT32(a->dim);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L2 norm of an int vector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(intvec_l2_norm);
|
||||
Datum
|
||||
intvec_l2_norm(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
int8 *ax = a->x;
|
||||
int norm = 0;
|
||||
|
||||
/* Auto-vectorized */
|
||||
for (int i = 0; i < a->dim; i++)
|
||||
norm += (int) ax[i] * (int) ax[i];
|
||||
|
||||
PG_RETURN_FLOAT8(sqrt((double) norm));
|
||||
}
|
||||
|
||||
/*
|
||||
* Internal helper to compare int vectors
|
||||
*/
|
||||
static int
|
||||
intvec_cmp_internal(IntVector * a, IntVector * b)
|
||||
{
|
||||
int dim = Min(a->dim, b->dim);
|
||||
|
||||
/* Check values before dimensions to be consistent with Postgres arrays */
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
if (a->x[i] < b->x[i])
|
||||
return -1;
|
||||
|
||||
if (a->x[i] > b->x[i])
|
||||
return 1;
|
||||
}
|
||||
|
||||
if (a->dim < b->dim)
|
||||
return -1;
|
||||
|
||||
if (a->dim > b->dim)
|
||||
return 1;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*
|
||||
* Less than
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_lt);
|
||||
Datum
|
||||
intvec_lt(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
PG_RETURN_BOOL(intvec_cmp_internal(a, b) < 0);
|
||||
}
|
||||
|
||||
/*
|
||||
* Less than or equal
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_le);
|
||||
Datum
|
||||
intvec_le(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
PG_RETURN_BOOL(intvec_cmp_internal(a, b) <= 0);
|
||||
}
|
||||
|
||||
/*
|
||||
* Equal
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_eq);
|
||||
Datum
|
||||
intvec_eq(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
PG_RETURN_BOOL(intvec_cmp_internal(a, b) == 0);
|
||||
}
|
||||
|
||||
/*
|
||||
* Not equal
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_ne);
|
||||
Datum
|
||||
intvec_ne(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
PG_RETURN_BOOL(intvec_cmp_internal(a, b) != 0);
|
||||
}
|
||||
|
||||
/*
|
||||
* Greater than or equal
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_ge);
|
||||
Datum
|
||||
intvec_ge(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
PG_RETURN_BOOL(intvec_cmp_internal(a, b) >= 0);
|
||||
}
|
||||
|
||||
/*
|
||||
* Greater than
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_gt);
|
||||
Datum
|
||||
intvec_gt(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
PG_RETURN_BOOL(intvec_cmp_internal(a, b) > 0);
|
||||
}
|
||||
|
||||
/*
|
||||
* Compare int vectors
|
||||
*/
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(intvec_cmp);
|
||||
Datum
|
||||
intvec_cmp(PG_FUNCTION_ARGS)
|
||||
{
|
||||
IntVector *a = PG_GETARG_INTVEC_P(0);
|
||||
IntVector *b = PG_GETARG_INTVEC_P(1);
|
||||
|
||||
PG_RETURN_INT32(intvec_cmp_internal(a, b));
|
||||
}
|
||||
23
src/intvec.h
Normal file
23
src/intvec.h
Normal file
@@ -0,0 +1,23 @@
|
||||
#ifndef INTVEC_H
|
||||
#define INTVEC_H
|
||||
|
||||
#include "vector.h"
|
||||
|
||||
#define INTVEC_MAX_DIM VECTOR_MAX_DIM
|
||||
|
||||
#define INTVEC_SIZE(_dim) (offsetof(IntVector, x) + sizeof(int8)*(_dim))
|
||||
#define DatumGetIntVector(x) ((IntVector *) PG_DETOAST_DATUM(x))
|
||||
#define PG_GETARG_INTVEC_P(x) DatumGetIntVector(PG_GETARG_DATUM(x))
|
||||
#define PG_RETURN_INTVEC_P(x) PG_RETURN_POINTER(x)
|
||||
|
||||
typedef struct IntVector
|
||||
{
|
||||
int32 vl_len_; /* varlena header (do not touch directly!) */
|
||||
int16 dim; /* number of dimensions */
|
||||
int16 unused;
|
||||
int8 x[FLEXIBLE_ARRAY_MEMBER];
|
||||
} IntVector;
|
||||
|
||||
IntVector *InitIntVector(int dim);
|
||||
|
||||
#endif
|
||||
@@ -360,7 +360,7 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
|
||||
buildstate->sortdesc = CreateTemplateTupleDesc(3);
|
||||
TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
|
||||
TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
|
||||
TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 3, "vector", TupleDescAttr(buildstate->tupdesc, 0)->atttypid, -1, 0);
|
||||
TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 3, "vector", buildstate->tupdesc->attrs[0].atttypid, -1, 0);
|
||||
|
||||
buildstate->slot = MakeSingleTupleTableSlot(buildstate->sortdesc, &TTSOpsVirtual);
|
||||
|
||||
|
||||
@@ -17,13 +17,13 @@
|
||||
#endif
|
||||
|
||||
int ivfflat_probes;
|
||||
int ivfflat_iterative_scan;
|
||||
int ivfflat_iterative_search;
|
||||
int ivfflat_max_probes;
|
||||
static relopt_kind ivfflat_relopt_kind;
|
||||
|
||||
static const struct config_enum_entry ivfflat_iterative_scan_options[] = {
|
||||
{"off", IVFFLAT_ITERATIVE_SCAN_OFF, false},
|
||||
{"relaxed_order", IVFFLAT_ITERATIVE_SCAN_RELAXED, false},
|
||||
static const struct config_enum_entry ivfflat_iterative_search_options[] = {
|
||||
{"off", IVFFLAT_ITERATIVE_SEARCH_OFF, false},
|
||||
{"relaxed_order", IVFFLAT_ITERATIVE_SEARCH_RELAXED, false},
|
||||
{NULL, 0, false}
|
||||
};
|
||||
|
||||
@@ -41,14 +41,14 @@ IvfflatInit(void)
|
||||
"Valid range is 1..lists.", &ivfflat_probes,
|
||||
IVFFLAT_DEFAULT_PROBES, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
|
||||
DefineCustomEnumVariable("ivfflat.iterative_scan", "Sets the mode for iterative scans",
|
||||
NULL, &ivfflat_iterative_scan,
|
||||
IVFFLAT_ITERATIVE_SCAN_OFF, ivfflat_iterative_scan_options, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
DefineCustomEnumVariable("ivfflat.iterative_search", "Sets the iterative search mode",
|
||||
NULL, &ivfflat_iterative_search,
|
||||
IVFFLAT_ITERATIVE_SEARCH_OFF, ivfflat_iterative_search_options, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
|
||||
/* If this is less than probes, probes is used */
|
||||
DefineCustomIntVariable("ivfflat.max_probes", "Sets the max number of probes for iterative scans",
|
||||
NULL, &ivfflat_max_probes,
|
||||
IVFFLAT_MAX_LISTS, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
DefineCustomIntVariable("ivfflat.max_probes", "Sets the max number of probes for iterative search",
|
||||
"-1 means no limit", &ivfflat_max_probes,
|
||||
-1, -1, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
|
||||
|
||||
MarkGUCPrefixReserved("ivfflat");
|
||||
}
|
||||
@@ -99,10 +99,6 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
|
||||
*indexSelectivity = 0;
|
||||
*indexCorrelation = 0;
|
||||
*indexPages = 0;
|
||||
#if PG_VERSION_NUM >= 180000
|
||||
/* See "On disable_cost" thread on pgsql-hackers */
|
||||
path->path.disabled_nodes = 2;
|
||||
#endif
|
||||
return;
|
||||
}
|
||||
|
||||
|
||||
@@ -80,14 +80,14 @@
|
||||
|
||||
/* Variables */
|
||||
extern int ivfflat_probes;
|
||||
extern int ivfflat_iterative_scan;
|
||||
extern int ivfflat_iterative_search;
|
||||
extern int ivfflat_max_probes;
|
||||
|
||||
typedef enum IvfflatIterativeScanMode
|
||||
typedef enum IvfflatIterativeSearchMode
|
||||
{
|
||||
IVFFLAT_ITERATIVE_SCAN_OFF,
|
||||
IVFFLAT_ITERATIVE_SCAN_RELAXED
|
||||
} IvfflatIterativeScanMode;
|
||||
IVFFLAT_ITERATIVE_SEARCH_OFF,
|
||||
IVFFLAT_ITERATIVE_SEARCH_RELAXED
|
||||
} IvfflatIterativeSearchMode;
|
||||
|
||||
typedef struct VectorArrayData
|
||||
{
|
||||
|
||||
@@ -114,6 +114,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
|
||||
{
|
||||
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
|
||||
TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
|
||||
double tuples = 0;
|
||||
TupleTableSlot *slot = so->vslot;
|
||||
int batchProbes = 0;
|
||||
|
||||
@@ -160,6 +161,8 @@ GetScanItems(IndexScanDesc scan, Datum value)
|
||||
ExecStoreVirtualTuple(slot);
|
||||
|
||||
tuplesort_puttupleslot(so->sortstate, slot);
|
||||
|
||||
tuples++;
|
||||
}
|
||||
|
||||
searchPage = IvfflatPageGetOpaque(page)->nextblkno;
|
||||
@@ -168,6 +171,12 @@ GetScanItems(IndexScanDesc scan, Datum value)
|
||||
}
|
||||
}
|
||||
|
||||
if (tuples < 100 && ivfflat_iterative_search == IVFFLAT_ITERATIVE_SEARCH_OFF)
|
||||
ereport(DEBUG1,
|
||||
(errmsg("index scan found few tuples"),
|
||||
errdetail("Index may have been created with little data."),
|
||||
errhint("Recreate the index and possibly decrease lists.")));
|
||||
|
||||
tuplesort_performsort(so->sortstate);
|
||||
|
||||
#if defined(IVFFLAT_MEMORY)
|
||||
@@ -254,8 +263,18 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
|
||||
/* Get lists and dimensions from metapage */
|
||||
IvfflatGetMetaPageInfo(index, &lists, &dimensions);
|
||||
|
||||
if (ivfflat_iterative_scan != IVFFLAT_ITERATIVE_SCAN_OFF)
|
||||
maxProbes = Max(ivfflat_max_probes, probes);
|
||||
if (ivfflat_iterative_search != IVFFLAT_ITERATIVE_SEARCH_OFF)
|
||||
{
|
||||
maxProbes = ivfflat_max_probes;
|
||||
|
||||
if (maxProbes < 0)
|
||||
maxProbes = lists;
|
||||
else if (maxProbes < probes)
|
||||
{
|
||||
/* TODO Show notice */
|
||||
maxProbes = probes;
|
||||
}
|
||||
}
|
||||
else
|
||||
maxProbes = probes;
|
||||
|
||||
|
||||
@@ -355,7 +355,7 @@ ivfflat_halfvec_support(PG_FUNCTION_ARGS)
|
||||
};
|
||||
|
||||
PG_RETURN_POINTER(&typeInfo);
|
||||
}
|
||||
};
|
||||
|
||||
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_bit_support);
|
||||
Datum
|
||||
@@ -370,4 +370,4 @@ ivfflat_bit_support(PG_FUNCTION_ARGS)
|
||||
};
|
||||
|
||||
PG_RETURN_POINTER(&typeInfo);
|
||||
}
|
||||
};
|
||||
|
||||
@@ -5,10 +5,6 @@
|
||||
#include "ivfflat.h"
|
||||
#include "storage/bufmgr.h"
|
||||
|
||||
#if PG_VERSION_NUM >= 180000
|
||||
#define vacuum_delay_point() vacuum_delay_point(false)
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Bulk delete tuples from the index
|
||||
*/
|
||||
|
||||
@@ -4,7 +4,6 @@
|
||||
#include <math.h>
|
||||
|
||||
#include "catalog/pg_type.h"
|
||||
#include "common/shortest_dec.h"
|
||||
#include "common/string.h"
|
||||
#include "fmgr.h"
|
||||
#include "halfutils.h"
|
||||
@@ -13,10 +12,17 @@
|
||||
#include "sparsevec.h"
|
||||
#include "utils/array.h"
|
||||
#include "utils/builtins.h"
|
||||
#include "utils/float.h"
|
||||
#include "utils/lsyscache.h"
|
||||
#include "vector.h"
|
||||
|
||||
#if PG_VERSION_NUM >= 120000
|
||||
#include "common/shortest_dec.h"
|
||||
#include "utils/float.h"
|
||||
#else
|
||||
#include <float.h>
|
||||
#include "utils/builtins.h"
|
||||
#endif
|
||||
|
||||
typedef struct SparseInputElement
|
||||
{
|
||||
int32 index;
|
||||
|
||||
@@ -140,6 +140,34 @@ SELECT '{1e-8,-1e-8}'::real[]::halfvec;
|
||||
[0,-0]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec::integer[];
|
||||
int4
|
||||
---------
|
||||
{1,2,3}
|
||||
(1 row)
|
||||
|
||||
SELECT '{1,2,3}'::integer[]::intvec;
|
||||
intvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '{1,2,3}'::integer[]::intvec(3);
|
||||
intvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '{1,2,3}'::integer[]::intvec(2);
|
||||
ERROR: expected 2 dimensions, not 3
|
||||
SELECT '{127,-128}'::integer[]::intvec;
|
||||
intvec
|
||||
------------
|
||||
[127,-128]
|
||||
(1 row)
|
||||
|
||||
SELECT '{128,-129}'::integer[]::intvec;
|
||||
ERROR: value "128" is out of range for type intvec
|
||||
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
|
||||
sparsevec
|
||||
-----------------
|
||||
|
||||
@@ -30,6 +30,23 @@ SELECT * FROM t2 ORDER BY val;
|
||||
|
||||
(4 rows)
|
||||
|
||||
DROP TABLE t;
|
||||
DROP TABLE t2;
|
||||
-- intvec
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE TABLE t2 (val intvec(3));
|
||||
\copy t TO 'results/intvec.bin' WITH (FORMAT binary)
|
||||
\copy t2 FROM 'results/intvec.bin' WITH (FORMAT binary)
|
||||
SELECT * FROM t2 ORDER BY val;
|
||||
val
|
||||
---------
|
||||
[0,0,0]
|
||||
[1,1,1]
|
||||
[1,2,3]
|
||||
|
||||
(4 rows)
|
||||
|
||||
DROP TABLE t;
|
||||
DROP TABLE t2;
|
||||
-- sparsevec
|
||||
|
||||
102
test/expected/hnsw_intvec.out
Normal file
102
test/expected/hnsw_intvec.out
Normal file
@@ -0,0 +1,102 @@
|
||||
SET enable_seqscan = off;
|
||||
-- L2
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val intvec_l2_ops);
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
[1,2,3]
|
||||
[1,2,4]
|
||||
[1,1,1]
|
||||
[0,0,0]
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::intvec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
SELECT COUNT(*) FROM t;
|
||||
count
|
||||
-------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
-----
|
||||
(0 rows)
|
||||
|
||||
DROP TABLE t;
|
||||
-- inner product
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val intvec_ip_ops);
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
[1,2,4]
|
||||
[1,2,3]
|
||||
[1,1,1]
|
||||
[0,0,0]
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::intvec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
-- cosine
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val intvec_cosine_ops);
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
[1,1,1]
|
||||
[1,2,3]
|
||||
[1,2,4]
|
||||
(3 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
|
||||
count
|
||||
-------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::intvec)) t2;
|
||||
count
|
||||
-------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
-- L1
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val intvec_l1_ops);
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
SELECT * FROM t ORDER BY val <+> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
[1,2,3]
|
||||
[1,2,4]
|
||||
[1,1,1]
|
||||
[0,0,0]
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <+> (SELECT NULL::intvec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -104,7 +104,7 @@ DROP TABLE t;
|
||||
CREATE TABLE t (val vector(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val vector_l2_ops);
|
||||
SET hnsw.iterative_scan = strict_order;
|
||||
SET hnsw.iterative_search = strict_order;
|
||||
SET hnsw.ef_search = 1;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
@@ -114,7 +114,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
[0,0,0]
|
||||
(3 rows)
|
||||
|
||||
SET hnsw.iterative_scan = relaxed_order;
|
||||
SET hnsw.iterative_search = relaxed_order;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
@@ -123,13 +123,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
[0,0,0]
|
||||
(3 rows)
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
-----
|
||||
(0 rows)
|
||||
|
||||
RESET hnsw.iterative_scan;
|
||||
RESET hnsw.iterative_search;
|
||||
RESET hnsw.ef_search;
|
||||
DROP TABLE t;
|
||||
-- unlogged
|
||||
@@ -171,31 +165,21 @@ SET hnsw.ef_search = 0;
|
||||
ERROR: 0 is outside the valid range for parameter "hnsw.ef_search" (1 .. 1000)
|
||||
SET hnsw.ef_search = 1001;
|
||||
ERROR: 1001 is outside the valid range for parameter "hnsw.ef_search" (1 .. 1000)
|
||||
SHOW hnsw.iterative_scan;
|
||||
hnsw.iterative_scan
|
||||
---------------------
|
||||
SHOW hnsw.iterative_search;
|
||||
hnsw.iterative_search
|
||||
-----------------------
|
||||
off
|
||||
(1 row)
|
||||
|
||||
SET hnsw.iterative_scan = on;
|
||||
ERROR: invalid value for parameter "hnsw.iterative_scan": "on"
|
||||
SET hnsw.iterative_search = on;
|
||||
ERROR: invalid value for parameter "hnsw.iterative_search": "on"
|
||||
HINT: Available values: off, relaxed_order, strict_order.
|
||||
SHOW hnsw.max_scan_tuples;
|
||||
hnsw.max_scan_tuples
|
||||
----------------------
|
||||
20000
|
||||
SHOW hnsw.max_search_tuples;
|
||||
hnsw.max_search_tuples
|
||||
------------------------
|
||||
-1
|
||||
(1 row)
|
||||
|
||||
SET hnsw.max_scan_tuples = 0;
|
||||
ERROR: 0 is outside the valid range for parameter "hnsw.max_scan_tuples" (1 .. 2147483647)
|
||||
SHOW hnsw.scan_mem_multiplier;
|
||||
hnsw.scan_mem_multiplier
|
||||
--------------------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SET hnsw.scan_mem_multiplier = 0;
|
||||
ERROR: 0 is outside the valid range for parameter "hnsw.scan_mem_multiplier" (1 .. 1000)
|
||||
SET hnsw.scan_mem_multiplier = 1001;
|
||||
ERROR: 1001 is outside the valid range for parameter "hnsw.scan_mem_multiplier" (1 .. 1000)
|
||||
SET hnsw.max_search_tuples = -2;
|
||||
ERROR: -2 is outside the valid range for parameter "hnsw.max_search_tuples" (-1 .. 2147483647)
|
||||
DROP TABLE t;
|
||||
|
||||
328
test/expected/intvec.out
Normal file
328
test/expected/intvec.out
Normal file
@@ -0,0 +1,328 @@
|
||||
SELECT '[1,2,3]'::intvec;
|
||||
intvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[-1,-2,-3]'::intvec;
|
||||
intvec
|
||||
------------
|
||||
[-1,-2,-3]
|
||||
(1 row)
|
||||
|
||||
SELECT ' [ 1, 2 , 3 ] '::intvec;
|
||||
intvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1.23456]'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[1.23456]"
|
||||
LINE 1: SELECT '[1.23456]'::intvec;
|
||||
^
|
||||
SELECT '[hello,1]'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[hello,1]"
|
||||
LINE 1: SELECT '[hello,1]'::intvec;
|
||||
^
|
||||
SELECT '[127,-128]'::intvec;
|
||||
intvec
|
||||
------------
|
||||
[127,-128]
|
||||
(1 row)
|
||||
|
||||
SELECT '[128,-129]'::intvec;
|
||||
ERROR: "128" is out of range for type intvec
|
||||
LINE 1: SELECT '[128,-129]'::intvec;
|
||||
^
|
||||
SELECT '[1,2,3'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[1,2,3"
|
||||
LINE 1: SELECT '[1,2,3'::intvec;
|
||||
^
|
||||
SELECT '[1,2,3]9'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[1,2,3]9"
|
||||
LINE 1: SELECT '[1,2,3]9'::intvec;
|
||||
^
|
||||
DETAIL: Junk after closing right brace.
|
||||
SELECT '1,2,3'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "1,2,3"
|
||||
LINE 1: SELECT '1,2,3'::intvec;
|
||||
^
|
||||
DETAIL: Vector contents must start with "[".
|
||||
SELECT ''::intvec;
|
||||
ERROR: invalid input syntax for type intvec: ""
|
||||
LINE 1: SELECT ''::intvec;
|
||||
^
|
||||
DETAIL: Vector contents must start with "[".
|
||||
SELECT '['::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "["
|
||||
LINE 1: SELECT '['::intvec;
|
||||
^
|
||||
SELECT '[,'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[,"
|
||||
LINE 1: SELECT '[,'::intvec;
|
||||
^
|
||||
SELECT '[]'::intvec;
|
||||
ERROR: intvec must have at least 1 dimension
|
||||
LINE 1: SELECT '[]'::intvec;
|
||||
^
|
||||
SELECT '[1,]'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[1,]"
|
||||
LINE 1: SELECT '[1,]'::intvec;
|
||||
^
|
||||
SELECT '[1a]'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[1a]"
|
||||
LINE 1: SELECT '[1a]'::intvec;
|
||||
^
|
||||
SELECT '[1,,3]'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[1,,3]"
|
||||
LINE 1: SELECT '[1,,3]'::intvec;
|
||||
^
|
||||
SELECT '[1, ,3]'::intvec;
|
||||
ERROR: invalid input syntax for type intvec: "[1, ,3]"
|
||||
LINE 1: SELECT '[1, ,3]'::intvec;
|
||||
^
|
||||
SELECT '[1,2,3]'::intvec(3);
|
||||
intvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec(2);
|
||||
ERROR: expected 2 dimensions, not 3
|
||||
SELECT '[1,2,3]'::intvec(3, 2);
|
||||
ERROR: invalid type modifier
|
||||
LINE 1: SELECT '[1,2,3]'::intvec(3, 2);
|
||||
^
|
||||
SELECT '[1,2,3]'::intvec('a');
|
||||
ERROR: invalid input syntax for type integer: "a"
|
||||
LINE 1: SELECT '[1,2,3]'::intvec('a');
|
||||
^
|
||||
SELECT '[1,2,3]'::intvec(0);
|
||||
ERROR: dimensions for type intvec must be at least 1
|
||||
LINE 1: SELECT '[1,2,3]'::intvec(0);
|
||||
^
|
||||
SELECT '[1,2,3]'::intvec(16001);
|
||||
ERROR: dimensions for type intvec cannot exceed 16000
|
||||
LINE 1: SELECT '[1,2,3]'::intvec(16001);
|
||||
^
|
||||
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::intvec[]);
|
||||
unnest
|
||||
---------
|
||||
[1,2,3]
|
||||
[4,5,6]
|
||||
(2 rows)
|
||||
|
||||
SELECT '{"[1,2,3]"}'::intvec(2)[];
|
||||
ERROR: expected 2 dimensions, not 3
|
||||
SELECT '[1,2,3]'::intvec < '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec < '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec <= '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec <= '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec = '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec = '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec != '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec != '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec >= '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec >= '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec > '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::intvec > '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT intvec_cmp('[1,2,3]', '[1,2,3]');
|
||||
intvec_cmp
|
||||
------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT intvec_cmp('[1,2,3]', '[0,0,0]');
|
||||
intvec_cmp
|
||||
------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT intvec_cmp('[0,0,0]', '[1,2,3]');
|
||||
intvec_cmp
|
||||
------------
|
||||
-1
|
||||
(1 row)
|
||||
|
||||
SELECT intvec_cmp('[1,2]', '[1,2,3]');
|
||||
intvec_cmp
|
||||
------------
|
||||
-1
|
||||
(1 row)
|
||||
|
||||
SELECT intvec_cmp('[1,2,3]', '[1,2]');
|
||||
intvec_cmp
|
||||
------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT intvec_cmp('[1,2]', '[2,3,4]');
|
||||
intvec_cmp
|
||||
------------
|
||||
-1
|
||||
(1 row)
|
||||
|
||||
SELECT intvec_cmp('[2,3]', '[1,2,3]');
|
||||
intvec_cmp
|
||||
------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT vector_dims('[1,2,3]'::intvec);
|
||||
vector_dims
|
||||
-------------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('[0,0]'::intvec, '[3,4]');
|
||||
l2_distance
|
||||
-------------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('[0,0]'::intvec, '[0,1]');
|
||||
l2_distance
|
||||
-------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('[1,2]'::intvec, '[3]');
|
||||
ERROR: different intvec dimensions 2 and 1
|
||||
SELECT '[0,0]'::intvec <-> '[3,4]';
|
||||
?column?
|
||||
----------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT inner_product('[1,2]'::intvec, '[3,4]');
|
||||
inner_product
|
||||
---------------
|
||||
11
|
||||
(1 row)
|
||||
|
||||
SELECT inner_product('[1,2]'::intvec, '[3]');
|
||||
ERROR: different intvec dimensions 2 and 1
|
||||
SELECT inner_product('[127]'::intvec, '[127]');
|
||||
inner_product
|
||||
---------------
|
||||
16129
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2]'::intvec <#> '[3,4]';
|
||||
?column?
|
||||
----------
|
||||
-11
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::intvec, '[2,4]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::intvec, '[0,0]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
NaN
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,1]'::intvec, '[1,1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,0]'::intvec, '[0,2]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,1]'::intvec, '[-1,-1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::intvec, '[3]');
|
||||
ERROR: different intvec dimensions 2 and 1
|
||||
SELECT '[1,2]'::intvec <=> '[2,4]';
|
||||
?column?
|
||||
----------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[0,0]'::intvec, '[3,4]');
|
||||
l1_distance
|
||||
-------------
|
||||
7
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[0,0]'::intvec, '[0,1]');
|
||||
l1_distance
|
||||
-------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[1,2]'::intvec, '[3]');
|
||||
ERROR: different intvec dimensions 2 and 1
|
||||
@@ -86,7 +86,7 @@ DROP TABLE t;
|
||||
CREATE TABLE t (val vector(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 3);
|
||||
SET ivfflat.iterative_scan = relaxed_order;
|
||||
SET ivfflat.iterative_search = relaxed_order;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
@@ -95,6 +95,13 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
[0,0,0]
|
||||
(3 rows)
|
||||
|
||||
SET ivfflat.max_probes = 0;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SET ivfflat.max_probes = 1;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
@@ -110,16 +117,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
[1,1,1]
|
||||
(2 rows)
|
||||
|
||||
TRUNCATE t;
|
||||
NOTICE: ivfflat index created with little data
|
||||
DETAIL: This will cause low recall.
|
||||
HINT: Drop the index until the table has more data.
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
-----
|
||||
(0 rows)
|
||||
|
||||
RESET ivfflat.iterative_scan;
|
||||
RESET ivfflat.iterative_search;
|
||||
RESET ivfflat.max_probes;
|
||||
DROP TABLE t;
|
||||
-- unlogged
|
||||
@@ -153,23 +151,23 @@ SET ivfflat.probes = 0;
|
||||
ERROR: 0 is outside the valid range for parameter "ivfflat.probes" (1 .. 32768)
|
||||
SET ivfflat.probes = 32769;
|
||||
ERROR: 32769 is outside the valid range for parameter "ivfflat.probes" (1 .. 32768)
|
||||
SHOW ivfflat.iterative_scan;
|
||||
ivfflat.iterative_scan
|
||||
------------------------
|
||||
SHOW ivfflat.iterative_search;
|
||||
ivfflat.iterative_search
|
||||
--------------------------
|
||||
off
|
||||
(1 row)
|
||||
|
||||
SET ivfflat.iterative_scan = on;
|
||||
ERROR: invalid value for parameter "ivfflat.iterative_scan": "on"
|
||||
SET ivfflat.iterative_search = on;
|
||||
ERROR: invalid value for parameter "ivfflat.iterative_search": "on"
|
||||
HINT: Available values: off, relaxed_order.
|
||||
SHOW ivfflat.max_probes;
|
||||
ivfflat.max_probes
|
||||
--------------------
|
||||
32768
|
||||
-1
|
||||
(1 row)
|
||||
|
||||
SET ivfflat.max_probes = 0;
|
||||
ERROR: 0 is outside the valid range for parameter "ivfflat.max_probes" (1 .. 32768)
|
||||
SET ivfflat.max_probes = -2;
|
||||
ERROR: -2 is outside the valid range for parameter "ivfflat.max_probes" (-1 .. 32768)
|
||||
SET ivfflat.max_probes = 32769;
|
||||
ERROR: 32769 is outside the valid range for parameter "ivfflat.max_probes" (1 .. 32768)
|
||||
ERROR: 32769 is outside the valid range for parameter "ivfflat.max_probes" (-1 .. 32768)
|
||||
DROP TABLE t;
|
||||
|
||||
@@ -38,6 +38,14 @@ SELECT '{1,2,3}'::real[]::halfvec(2);
|
||||
SELECT '{65520,-65520}'::real[]::halfvec;
|
||||
SELECT '{1e-8,-1e-8}'::real[]::halfvec;
|
||||
|
||||
SELECT '[1,2,3]'::intvec::integer[];
|
||||
|
||||
SELECT '{1,2,3}'::integer[]::intvec;
|
||||
SELECT '{1,2,3}'::integer[]::intvec(3);
|
||||
SELECT '{1,2,3}'::integer[]::intvec(2);
|
||||
SELECT '{127,-128}'::integer[]::intvec;
|
||||
SELECT '{128,-129}'::integer[]::intvec;
|
||||
|
||||
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
|
||||
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec(5);
|
||||
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec(4);
|
||||
|
||||
@@ -28,6 +28,21 @@ SELECT * FROM t2 ORDER BY val;
|
||||
DROP TABLE t;
|
||||
DROP TABLE t2;
|
||||
|
||||
-- intvec
|
||||
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
|
||||
CREATE TABLE t2 (val intvec(3));
|
||||
|
||||
\copy t TO 'results/intvec.bin' WITH (FORMAT binary)
|
||||
\copy t2 FROM 'results/intvec.bin' WITH (FORMAT binary)
|
||||
|
||||
SELECT * FROM t2 ORDER BY val;
|
||||
|
||||
DROP TABLE t;
|
||||
DROP TABLE t2;
|
||||
|
||||
-- sparsevec
|
||||
|
||||
CREATE TABLE t (val sparsevec(3));
|
||||
|
||||
58
test/sql/hnsw_intvec.sql
Normal file
58
test/sql/hnsw_intvec.sql
Normal file
@@ -0,0 +1,58 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
-- L2
|
||||
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val intvec_l2_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::intvec)) t2;
|
||||
SELECT COUNT(*) FROM t;
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
-- inner product
|
||||
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val intvec_ip_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
|
||||
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::intvec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
-- cosine
|
||||
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val intvec_cosine_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
|
||||
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::intvec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
-- L1
|
||||
|
||||
CREATE TABLE t (val intvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val intvec_l1_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
|
||||
SELECT * FROM t ORDER BY val <+> '[3,3,3]';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <+> (SELECT NULL::intvec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -63,17 +63,14 @@ CREATE TABLE t (val vector(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val vector_l2_ops);
|
||||
|
||||
SET hnsw.iterative_scan = strict_order;
|
||||
SET hnsw.iterative_search = strict_order;
|
||||
SET hnsw.ef_search = 1;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
SET hnsw.iterative_scan = relaxed_order;
|
||||
SET hnsw.iterative_search = relaxed_order;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
RESET hnsw.iterative_scan;
|
||||
RESET hnsw.iterative_search;
|
||||
RESET hnsw.ef_search;
|
||||
DROP TABLE t;
|
||||
|
||||
@@ -101,17 +98,12 @@ SHOW hnsw.ef_search;
|
||||
SET hnsw.ef_search = 0;
|
||||
SET hnsw.ef_search = 1001;
|
||||
|
||||
SHOW hnsw.iterative_scan;
|
||||
SHOW hnsw.iterative_search;
|
||||
|
||||
SET hnsw.iterative_scan = on;
|
||||
SET hnsw.iterative_search = on;
|
||||
|
||||
SHOW hnsw.max_scan_tuples;
|
||||
SHOW hnsw.max_search_tuples;
|
||||
|
||||
SET hnsw.max_scan_tuples = 0;
|
||||
|
||||
SHOW hnsw.scan_mem_multiplier;
|
||||
|
||||
SET hnsw.scan_mem_multiplier = 0;
|
||||
SET hnsw.scan_mem_multiplier = 1001;
|
||||
SET hnsw.max_search_tuples = -2;
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
73
test/sql/intvec.sql
Normal file
73
test/sql/intvec.sql
Normal file
@@ -0,0 +1,73 @@
|
||||
SELECT '[1,2,3]'::intvec;
|
||||
SELECT '[-1,-2,-3]'::intvec;
|
||||
SELECT ' [ 1, 2 , 3 ] '::intvec;
|
||||
SELECT '[1.23456]'::intvec;
|
||||
SELECT '[hello,1]'::intvec;
|
||||
SELECT '[127,-128]'::intvec;
|
||||
SELECT '[128,-129]'::intvec;
|
||||
SELECT '[1,2,3'::intvec;
|
||||
SELECT '[1,2,3]9'::intvec;
|
||||
SELECT '1,2,3'::intvec;
|
||||
SELECT ''::intvec;
|
||||
SELECT '['::intvec;
|
||||
SELECT '[,'::intvec;
|
||||
SELECT '[]'::intvec;
|
||||
SELECT '[1,]'::intvec;
|
||||
SELECT '[1a]'::intvec;
|
||||
SELECT '[1,,3]'::intvec;
|
||||
SELECT '[1, ,3]'::intvec;
|
||||
|
||||
SELECT '[1,2,3]'::intvec(3);
|
||||
SELECT '[1,2,3]'::intvec(2);
|
||||
SELECT '[1,2,3]'::intvec(3, 2);
|
||||
SELECT '[1,2,3]'::intvec('a');
|
||||
SELECT '[1,2,3]'::intvec(0);
|
||||
SELECT '[1,2,3]'::intvec(16001);
|
||||
|
||||
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::intvec[]);
|
||||
SELECT '{"[1,2,3]"}'::intvec(2)[];
|
||||
|
||||
SELECT '[1,2,3]'::intvec < '[1,2,3]';
|
||||
SELECT '[1,2,3]'::intvec < '[1,2]';
|
||||
SELECT '[1,2,3]'::intvec <= '[1,2,3]';
|
||||
SELECT '[1,2,3]'::intvec <= '[1,2]';
|
||||
SELECT '[1,2,3]'::intvec = '[1,2,3]';
|
||||
SELECT '[1,2,3]'::intvec = '[1,2]';
|
||||
SELECT '[1,2,3]'::intvec != '[1,2,3]';
|
||||
SELECT '[1,2,3]'::intvec != '[1,2]';
|
||||
SELECT '[1,2,3]'::intvec >= '[1,2,3]';
|
||||
SELECT '[1,2,3]'::intvec >= '[1,2]';
|
||||
SELECT '[1,2,3]'::intvec > '[1,2,3]';
|
||||
SELECT '[1,2,3]'::intvec > '[1,2]';
|
||||
|
||||
SELECT intvec_cmp('[1,2,3]', '[1,2,3]');
|
||||
SELECT intvec_cmp('[1,2,3]', '[0,0,0]');
|
||||
SELECT intvec_cmp('[0,0,0]', '[1,2,3]');
|
||||
SELECT intvec_cmp('[1,2]', '[1,2,3]');
|
||||
SELECT intvec_cmp('[1,2,3]', '[1,2]');
|
||||
SELECT intvec_cmp('[1,2]', '[2,3,4]');
|
||||
SELECT intvec_cmp('[2,3]', '[1,2,3]');
|
||||
|
||||
SELECT vector_dims('[1,2,3]'::intvec);
|
||||
|
||||
SELECT l2_distance('[0,0]'::intvec, '[3,4]');
|
||||
SELECT l2_distance('[0,0]'::intvec, '[0,1]');
|
||||
SELECT l2_distance('[1,2]'::intvec, '[3]');
|
||||
SELECT '[0,0]'::intvec <-> '[3,4]';
|
||||
|
||||
SELECT inner_product('[1,2]'::intvec, '[3,4]');
|
||||
SELECT inner_product('[1,2]'::intvec, '[3]');
|
||||
SELECT inner_product('[127]'::intvec, '[127]');
|
||||
SELECT '[1,2]'::intvec <#> '[3,4]';
|
||||
|
||||
SELECT cosine_distance('[1,2]'::intvec, '[2,4]');
|
||||
SELECT cosine_distance('[1,2]'::intvec, '[0,0]');
|
||||
SELECT cosine_distance('[1,1]'::intvec, '[1,1]');
|
||||
SELECT cosine_distance('[1,0]'::intvec, '[0,2]');
|
||||
SELECT cosine_distance('[1,1]'::intvec, '[-1,-1]');
|
||||
SELECT cosine_distance('[1,2]'::intvec, '[3]');
|
||||
SELECT '[1,2]'::intvec <=> '[2,4]';
|
||||
|
||||
SELECT l1_distance('[0,0]'::intvec, '[3,4]');
|
||||
SELECT l1_distance('[0,0]'::intvec, '[0,1]');
|
||||
SELECT l1_distance('[1,2]'::intvec, '[3]');
|
||||
@@ -50,7 +50,10 @@ CREATE TABLE t (val vector(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 3);
|
||||
|
||||
SET ivfflat.iterative_scan = relaxed_order;
|
||||
SET ivfflat.iterative_search = relaxed_order;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
SET ivfflat.max_probes = 0;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
SET ivfflat.max_probes = 1;
|
||||
@@ -59,10 +62,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
SET ivfflat.max_probes = 2;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
RESET ivfflat.iterative_scan;
|
||||
RESET ivfflat.iterative_search;
|
||||
RESET ivfflat.max_probes;
|
||||
DROP TABLE t;
|
||||
|
||||
@@ -87,13 +87,13 @@ SHOW ivfflat.probes;
|
||||
SET ivfflat.probes = 0;
|
||||
SET ivfflat.probes = 32769;
|
||||
|
||||
SHOW ivfflat.iterative_scan;
|
||||
SHOW ivfflat.iterative_search;
|
||||
|
||||
SET ivfflat.iterative_scan = on;
|
||||
SET ivfflat.iterative_search = on;
|
||||
|
||||
SHOW ivfflat.max_probes;
|
||||
|
||||
SET ivfflat.max_probes = 0;
|
||||
SET ivfflat.max_probes = -2;
|
||||
SET ivfflat.max_probes = 32769;
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
@@ -40,6 +40,10 @@ for (1 .. 50)
|
||||
$actual = $node->safe_psql("postgres", "SELECT halfvec_cmp(v::halfvec, '$query'::real[]::halfvec) FROM tst");
|
||||
is($expected, $actual);
|
||||
|
||||
# Test intvec
|
||||
$actual = $node->safe_psql("postgres", "SELECT intvec_cmp(v::integer[]::intvec, '$query'::integer[]::intvec) FROM tst");
|
||||
is($expected, $actual);
|
||||
|
||||
# Test sparsevec
|
||||
$actual = $node->safe_psql("postgres", "SELECT sparsevec_cmp(v::vector::sparsevec, '$query'::real[]::vector::sparsevec) FROM tst");
|
||||
is($expected, $actual);
|
||||
|
||||
@@ -45,6 +45,10 @@ for my $function (@functions)
|
||||
my $actual = $node->safe_psql("postgres", "SELECT $function(v::halfvec, '$query'::vector::halfvec) FROM tst");
|
||||
is($expected, $actual, "halfvec $function");
|
||||
|
||||
# Test intvec
|
||||
$actual = $node->safe_psql("postgres", "SELECT $function(v::real[]::integer[]::intvec, '$query'::vector::real[]::integer[]::intvec) FROM tst");
|
||||
is($expected, $actual, "intvec $function");
|
||||
|
||||
# Test sparsevec
|
||||
$actual = $node->safe_psql("postgres", "SELECT $function(v::sparsevec, '$query'::vector::sparsevec) FROM tst");
|
||||
is($expected, $actual, "sparsevec $function");
|
||||
|
||||
@@ -12,8 +12,8 @@ $node->start;
|
||||
# Create extension
|
||||
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
|
||||
|
||||
my @types = ("vector", "halfvec", "sparsevec");
|
||||
my @inputs = ("[1.23,4.56,7.89]", "[1.23,4.56,7.89]", "{1:1.23,2:4.56,3:7.89}/3");
|
||||
my @types = ("vector", "halfvec", "intvec", "sparsevec");
|
||||
my @inputs = ("[1.23,4.56,7.89]", "[1.23,4.56,7.89]", "[1,2,3]", "{1:1.23,2:4.56,3:7.89}/3");
|
||||
my @subs = (" ", " ", ",", ":", "-", "1", "9", "\0", "2147483648", "-2147483649");
|
||||
|
||||
for my $i (0 .. $#types)
|
||||
|
||||
@@ -23,7 +23,7 @@ $node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v vector_l2_ops
|
||||
my $count = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET ivfflat.probes = 10;
|
||||
SET ivfflat.iterative_scan = relaxed_order;
|
||||
SET ivfflat.iterative_search = relaxed_order;
|
||||
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
|
||||
));
|
||||
is($count, 10);
|
||||
@@ -39,7 +39,7 @@ foreach ((30, 50, 70))
|
||||
$count = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET ivfflat.probes = 10;
|
||||
SET ivfflat.iterative_scan = relaxed_order;
|
||||
SET ivfflat.iterative_search = relaxed_order;
|
||||
SET ivfflat.max_probes = $max_probes;
|
||||
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst WHERE i = $i) LIMIT 11) t;
|
||||
));
|
||||
@@ -19,7 +19,7 @@ sub test_recall
|
||||
my $explain = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET ivfflat.probes = $probes;
|
||||
SET ivfflat.iterative_scan = relaxed_order;
|
||||
SET ivfflat.iterative_search = relaxed_order;
|
||||
EXPLAIN ANALYZE SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[0]' LIMIT $limit;
|
||||
));
|
||||
like($explain, qr/Index Scan using idx on tst/);
|
||||
@@ -29,7 +29,7 @@ sub test_recall
|
||||
my $actual = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET ivfflat.probes = $probes;
|
||||
SET ivfflat.iterative_scan = relaxed_order;
|
||||
SET ivfflat.iterative_search = relaxed_order;
|
||||
SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[$i]' LIMIT $limit;
|
||||
));
|
||||
my @actual_ids = split("\n", $actual);
|
||||
@@ -26,9 +26,8 @@ $node->safe_psql("postgres", qq(
|
||||
|
||||
my $count = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.iterative_scan = relaxed_order;
|
||||
SET hnsw.max_scan_tuples = 100000;
|
||||
SET hnsw.scan_mem_multiplier = 2;
|
||||
SET hnsw.iterative_search = relaxed_order;
|
||||
SET work_mem = '8MB';
|
||||
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
|
||||
));
|
||||
is($count, 10);
|
||||
@@ -43,9 +42,9 @@ foreach ((30000, 50000, 70000))
|
||||
{
|
||||
$count = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.iterative_scan = relaxed_order;
|
||||
SET hnsw.max_scan_tuples = $max_tuples;
|
||||
SET hnsw.scan_mem_multiplier = 2;
|
||||
SET hnsw.iterative_search = relaxed_order;
|
||||
SET hnsw.max_search_tuples = $max_tuples;
|
||||
SET work_mem = '8MB';
|
||||
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst WHERE i = $i) LIMIT 11) t;
|
||||
));
|
||||
$sum += $count;
|
||||
@@ -56,4 +55,13 @@ foreach ((30000, 50000, 70000))
|
||||
cmp_ok($avg, '<', $expected + 2);
|
||||
}
|
||||
|
||||
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.iterative_search = relaxed_order;
|
||||
SET client_min_messages = debug1;
|
||||
SET work_mem = '2MB';
|
||||
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
|
||||
));
|
||||
like($stderr, qr/hnsw index scan exceeded work_mem after \d+ tuples/);
|
||||
|
||||
done_testing();
|
||||
@@ -21,7 +21,7 @@ sub test_recall
|
||||
my $explain = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.ef_search = $ef_search;
|
||||
SET hnsw.iterative_scan = $mode;
|
||||
SET hnsw.iterative_search = $mode;
|
||||
EXPLAIN ANALYZE SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[0]' LIMIT $limit;
|
||||
));
|
||||
like($explain, qr/Index Scan using idx on tst/);
|
||||
@@ -31,7 +31,7 @@ sub test_recall
|
||||
my $actual = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.ef_search = $ef_search;
|
||||
SET hnsw.iterative_scan = $mode;
|
||||
SET hnsw.iterative_search = $mode;
|
||||
SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[$i]' LIMIT $limit;
|
||||
));
|
||||
my @actual_ids = split("\n", $actual);
|
||||
132
test/t/045_hnsw_intvec_build_recall.pl
Normal file
132
test/t/045_hnsw_intvec_build_recall.pl
Normal file
@@ -0,0 +1,132 @@
|
||||
use strict;
|
||||
use warnings FATAL => 'all';
|
||||
use PostgreSQL::Test::Cluster;
|
||||
use PostgreSQL::Test::Utils;
|
||||
use Test::More;
|
||||
|
||||
my $node;
|
||||
my @queries = ();
|
||||
my @expected;
|
||||
my $limit = 20;
|
||||
my $dim = 10;
|
||||
my $array_sql = join(",", ('(random() * 255)::int - 128') x $dim);
|
||||
|
||||
sub test_recall
|
||||
{
|
||||
my ($min, $operator) = @_;
|
||||
my $correct = 0;
|
||||
my $total = 0;
|
||||
|
||||
my $explain = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
|
||||
));
|
||||
like($explain, qr/Index Scan/);
|
||||
|
||||
for my $i (0 .. $#queries)
|
||||
{
|
||||
my $actual = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
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 = PostgreSQL::Test::Cluster->new('node');
|
||||
$node->init;
|
||||
$node->start;
|
||||
|
||||
# Create table
|
||||
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
|
||||
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v intvec($dim));");
|
||||
$node->safe_psql("postgres",
|
||||
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
|
||||
);
|
||||
|
||||
# Generate queries
|
||||
for (1 .. 20)
|
||||
{
|
||||
my @r = ();
|
||||
for (1 .. $dim)
|
||||
{
|
||||
push(@r, int(rand(256)) - 128);
|
||||
}
|
||||
push(@queries, "[" . join(",", @r) . "]");
|
||||
}
|
||||
|
||||
# Check each index type
|
||||
my @operators = ("<->", "<#>", "<=>", "<+>");
|
||||
my @opclasses = ("intvec_l2_ops", "intvec_ip_ops", "intvec_cosine_ops", "intvec_l1_ops");
|
||||
|
||||
for my $i (0 .. $#operators)
|
||||
{
|
||||
my $operator = $operators[$i];
|
||||
my $opclass = $opclasses[$i];
|
||||
|
||||
# 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);
|
||||
}
|
||||
|
||||
# Build index serially
|
||||
$node->safe_psql("postgres", qq(
|
||||
SET max_parallel_maintenance_workers = 0;
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
));
|
||||
|
||||
# Test approximate results
|
||||
my $min = 0.98;
|
||||
test_recall($min, $operator);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
|
||||
# Build index in parallel in memory
|
||||
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
|
||||
SET client_min_messages = DEBUG;
|
||||
SET min_parallel_table_scan_size = 1;
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
));
|
||||
is($ret, 0, $stderr);
|
||||
like($stderr, qr/using \d+ parallel workers/);
|
||||
|
||||
# Test approximate results
|
||||
test_recall($min, $operator);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
|
||||
# Build index in parallel on disk
|
||||
# Set parallel_workers on table to use workers with low maintenance_work_mem
|
||||
($ret, $stdout, $stderr) = $node->psql("postgres", qq(
|
||||
ALTER TABLE tst SET (parallel_workers = 2);
|
||||
SET client_min_messages = DEBUG;
|
||||
SET maintenance_work_mem = '4MB';
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
ALTER TABLE tst RESET (parallel_workers);
|
||||
));
|
||||
is($ret, 0, $stderr);
|
||||
like($stderr, qr/using \d+ parallel workers/);
|
||||
like($stderr, qr/hnsw graph no longer fits into maintenance_work_mem/);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
}
|
||||
|
||||
done_testing();
|
||||
113
test/t/046_hnsw_intvec_insert_recall.pl
Normal file
113
test/t/046_hnsw_intvec_insert_recall.pl
Normal file
@@ -0,0 +1,113 @@
|
||||
use strict;
|
||||
use warnings FATAL => 'all';
|
||||
use PostgreSQL::Test::Cluster;
|
||||
use PostgreSQL::Test::Utils;
|
||||
use Test::More;
|
||||
|
||||
my $node;
|
||||
my @queries = ();
|
||||
my @expected;
|
||||
my $limit = 20;
|
||||
my $dim = 10;
|
||||
my $array_sql = join(",", ('(random() * 255)::int - 128') x $dim);
|
||||
|
||||
sub test_recall
|
||||
{
|
||||
my ($min, $operator) = @_;
|
||||
my $correct = 0;
|
||||
my $total = 0;
|
||||
|
||||
my $explain = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
|
||||
));
|
||||
like($explain, qr/Index Scan/);
|
||||
|
||||
for my $i (0 .. $#queries)
|
||||
{
|
||||
my $actual = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
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 = PostgreSQL::Test::Cluster->new('node');
|
||||
$node->init;
|
||||
$node->start;
|
||||
|
||||
# Create table
|
||||
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
|
||||
$node->safe_psql("postgres", "CREATE TABLE tst (i serial, v intvec($dim));");
|
||||
|
||||
# Generate queries
|
||||
for (1 .. 20)
|
||||
{
|
||||
my @r = ();
|
||||
for (1 .. $dim)
|
||||
{
|
||||
push(@r, int(rand(256)) - 128);
|
||||
}
|
||||
push(@queries, "[" . join(",", @r) . "]");
|
||||
}
|
||||
|
||||
# Check each index type
|
||||
my @operators = ("<->", "<#>", "<=>", "<+>");
|
||||
my @opclasses = ("intvec_l2_ops", "intvec_ip_ops", "intvec_cosine_ops", "intvec_l1_ops");
|
||||
|
||||
for my $i (0 .. $#operators)
|
||||
{
|
||||
my $operator = $operators[$i];
|
||||
my $opclass = $opclasses[$i];
|
||||
|
||||
# Add index
|
||||
$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v $opclass);");
|
||||
|
||||
# Use concurrent inserts
|
||||
$node->pgbench(
|
||||
"--no-vacuum --client=10 --transactions=1000",
|
||||
0,
|
||||
[qr{actually processed}],
|
||||
[qr{^$}],
|
||||
"concurrent INSERTs",
|
||||
{
|
||||
"040_hnsw_intvec_insert_recall_$opclass" => "INSERT INTO tst (v) VALUES (ARRAY[$array_sql]);"
|
||||
}
|
||||
);
|
||||
|
||||
# Get exact results
|
||||
@expected = ();
|
||||
foreach (@queries)
|
||||
{
|
||||
my $res = $node->safe_psql("postgres", qq(
|
||||
SET enable_indexscan = off;
|
||||
SELECT i FROM tst ORDER BY v $operator '$_' LIMIT $limit;
|
||||
));
|
||||
push(@expected, $res);
|
||||
}
|
||||
|
||||
# Test approximate results
|
||||
my $min = 0.98;
|
||||
test_recall($min, $operator);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
$node->safe_psql("postgres", "TRUNCATE tst;");
|
||||
}
|
||||
|
||||
done_testing();
|
||||
101
test/t/047_hnsw_intvec_vacuum_recall.pl
Normal file
101
test/t/047_hnsw_intvec_vacuum_recall.pl
Normal file
@@ -0,0 +1,101 @@
|
||||
use strict;
|
||||
use warnings FATAL => 'all';
|
||||
use PostgreSQL::Test::Cluster;
|
||||
use PostgreSQL::Test::Utils;
|
||||
use Test::More;
|
||||
|
||||
my $node;
|
||||
my @queries = ();
|
||||
my @expected;
|
||||
my $limit = 20;
|
||||
my $dim = 10;
|
||||
my $array_sql = join(",", ('(random() * 255)::int - 128') x $dim);
|
||||
|
||||
sub test_recall
|
||||
{
|
||||
my ($min, $ef_search, $test_name) = @_;
|
||||
my $correct = 0;
|
||||
my $total = 0;
|
||||
|
||||
my $explain = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.ef_search = $ef_search;
|
||||
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v <-> '$queries[0]' LIMIT $limit;
|
||||
));
|
||||
like($explain, qr/Index Scan/);
|
||||
|
||||
for my $i (0 .. $#queries)
|
||||
{
|
||||
my $actual = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.ef_search = $ef_search;
|
||||
SELECT i FROM tst ORDER BY v <-> '$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, $test_name);
|
||||
}
|
||||
|
||||
# Initialize node
|
||||
$node = PostgreSQL::Test::Cluster->new('node');
|
||||
$node->init;
|
||||
$node->start;
|
||||
|
||||
# Create table
|
||||
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
|
||||
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v intvec($dim));");
|
||||
$node->safe_psql("postgres", "ALTER TABLE tst SET (autovacuum_enabled = false);");
|
||||
$node->safe_psql("postgres",
|
||||
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
|
||||
);
|
||||
|
||||
# Add index
|
||||
$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v intvec_l2_ops) WITH (m = 4, ef_construction = 8);");
|
||||
|
||||
# Delete data
|
||||
$node->safe_psql("postgres", "DELETE FROM tst WHERE i > 2500;");
|
||||
|
||||
# Generate queries
|
||||
for (1 .. 20)
|
||||
{
|
||||
my @r = ();
|
||||
for (1 .. $dim)
|
||||
{
|
||||
push(@r, int(rand(256)) - 128);
|
||||
}
|
||||
push(@queries, "[" . join(",", @r) . "]");
|
||||
}
|
||||
|
||||
# Get exact results
|
||||
@expected = ();
|
||||
foreach (@queries)
|
||||
{
|
||||
my $res = $node->safe_psql("postgres", qq(
|
||||
SET enable_indexscan = off;
|
||||
SELECT i FROM tst ORDER BY v <-> '$_' LIMIT $limit;
|
||||
));
|
||||
push(@expected, $res);
|
||||
}
|
||||
|
||||
test_recall(0.18, $limit, "before vacuum");
|
||||
test_recall(0.84, 100, "before vacuum");
|
||||
|
||||
# TODO Test concurrent inserts with vacuum
|
||||
$node->safe_psql("postgres", "VACUUM tst;");
|
||||
|
||||
test_recall(0.84, $limit, "after vacuum");
|
||||
|
||||
done_testing();
|
||||
58
test/t/048_hnsw_intvec_duplicates.pl
Normal file
58
test/t/048_hnsw_intvec_duplicates.pl
Normal file
@@ -0,0 +1,58 @@
|
||||
use strict;
|
||||
use warnings FATAL => 'all';
|
||||
use PostgreSQL::Test::Cluster;
|
||||
use PostgreSQL::Test::Utils;
|
||||
use Test::More;
|
||||
|
||||
# Initialize node
|
||||
my $node = PostgreSQL::Test::Cluster->new('node');
|
||||
$node->init;
|
||||
$node->start;
|
||||
|
||||
# Create table
|
||||
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
|
||||
$node->safe_psql("postgres", "CREATE TABLE tst (v intvec(3));");
|
||||
|
||||
sub insert_vectors
|
||||
{
|
||||
for my $i (1 .. 20)
|
||||
{
|
||||
$node->safe_psql("postgres", "INSERT INTO tst VALUES ('[1,1,1]');");
|
||||
}
|
||||
}
|
||||
|
||||
sub test_duplicates
|
||||
{
|
||||
my $res = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.ef_search = 1;
|
||||
SELECT COUNT(*) FROM (SELECT * FROM tst ORDER BY v <-> '[1,1,1]') t;
|
||||
));
|
||||
is($res, 10);
|
||||
}
|
||||
|
||||
# Test duplicates with build
|
||||
insert_vectors();
|
||||
$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v intvec_l2_ops);");
|
||||
test_duplicates();
|
||||
|
||||
# Reset
|
||||
$node->safe_psql("postgres", "TRUNCATE tst;");
|
||||
|
||||
# Test duplicates with inserts
|
||||
insert_vectors();
|
||||
test_duplicates();
|
||||
|
||||
# Test fallback path for inserts
|
||||
$node->pgbench(
|
||||
"--no-vacuum --client=5 --transactions=100",
|
||||
0,
|
||||
[qr{actually processed}],
|
||||
[qr{^$}],
|
||||
"concurrent INSERTs",
|
||||
{
|
||||
"042_hnsw_intvec_duplicates" => "INSERT INTO tst VALUES ('[1,1,1]');"
|
||||
}
|
||||
);
|
||||
|
||||
done_testing();
|
||||
@@ -1,4 +1,4 @@
|
||||
comment = 'vector data type and ivfflat and hnsw access methods'
|
||||
default_version = '0.8.0'
|
||||
default_version = '0.7.4'
|
||||
module_pathname = '$libdir/vector'
|
||||
relocatable = true
|
||||
|
||||
Reference in New Issue
Block a user