Compare commits

..

28 Commits

Author SHA1 Message Date
Andrew Kane
fc0d3e7fdb Added search parameters to EXPLAIN [skip ci] 2024-10-28 01:20:59 -07:00
Andrew Kane
c530a3c490 Updated comment [skip ci] 2024-10-28 00:56:10 -07:00
Andrew Kane
d8b9e8ef73 Updated readme [skip ci] 2024-10-28 00:55:25 -07:00
Andrew Kane
00894efed5 Updated readme [skip ci] 2024-10-28 00:45:26 -07:00
Andrew Kane
0aa0f6619b Updated readme [skip ci] 2024-10-28 00:43:39 -07:00
Andrew Kane
38d053001e Updated readme [skip ci] 2024-10-28 00:32:12 -07:00
Andrew Kane
ccb95407e7 Updated readme [skip ci] 2024-10-28 00:09:26 -07:00
Andrew Kane
04d5e934a1 Scan 2000 more tuples with default work_mem 2024-10-27 22:23:28 -07:00
Andrew Kane
b163b5b196 Moved code [skip ci] 2024-10-27 22:21:43 -07:00
Andrew Kane
6a30c1e824 Fixed test [skip ci] 2024-10-27 21:10:51 -07:00
Andrew Kane
2db1b19644 Use greater than [skip ci] 2024-10-27 21:07:22 -07:00
Andrew Kane
305d62146e Updated comment [skip ci] 2024-10-27 21:05:32 -07:00
Andrew Kane
f9d627c9a9 Updated default value of hnsw.scan_mem_multiplier [skip ci] 2024-10-27 21:05:04 -07:00
Andrew Kane
38f42820be Added test for hnsw.scan_mem_multiplier [skip ci] 2024-10-27 20:05:58 -07:00
Andrew Kane
15c8245b42 Updated readme [skip ci] 2024-10-27 19:30:40 -07:00
Andrew Kane
572a9ab404 Updated readme [skip ci] 2024-10-27 18:58:50 -07:00
Andrew Kane
00492d7e57 Ensure max memory fits into Size for HNSW index scans 2024-10-27 14:21:15 -07:00
Andrew Kane
857d716d9e Renamed iterative_search to iterative_scan 2024-10-27 14:02:22 -07:00
Andrew Kane
c5dd2af750 Added comments [skip ci] 2024-10-25 21:39:03 -07:00
Andrew Kane
78b877bdaf Revert "Renamed iterative_search to iterative_scan"
This reverts commit 7043cce893.
2024-10-24 20:32:07 -07:00
Andrew Kane
7043cce893 Renamed iterative_search to iterative_scan 2024-10-24 20:31:43 -07:00
Andrew Kane
62039d74f6 Added iterative search section to readme [skip ci] 2024-10-24 18:05:29 -07:00
Andrew Kane
ac6576e53a Added hnsw.search_mem_multiplier option 2024-10-24 18:02:20 -07:00
Andrew Kane
67eff41c44 Updated changelog [skip ci] 2024-10-23 15:12:40 -07:00
Andrew Kane
1291b12090 Added Postgres 18 to CI [skip ci] 2024-10-22 00:38:19 -07:00
Andrew Kane
24522700b8 Improved hnswrescan 2024-10-21 23:41:32 -07:00
Andrew Kane
bfb3a45b31 Use consistent order [skip ci] 2024-10-21 21:47:03 -07:00
Andrew Kane
e718eb8da4 Updated range and defaults for iterative search parameters 2024-10-21 20:38:50 -07:00
39 changed files with 234 additions and 2396 deletions

View File

@@ -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

View File

@@ -1,7 +1,6 @@
## 0.8.0 (unreleased)
- Added support for iterative index scans
- Added `intvec` type
- Added casts for arrays to `sparsevec`
- Improved cost estimation
- Improved performance of HNSW inserts and on-disk index builds

View File

@@ -4,8 +4,8 @@ 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/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
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
TESTS = $(wildcard test/sql/*.sql)
REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))

View File

@@ -2,8 +2,8 @@ EXTENSION = vector
EXTVERSION = 0.7.4
DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql
OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\intvec.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
HEADERS = src\halfvec.h src\intvec.h src\sparsevec.h src\vector.h
OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
HEADERS = src\halfvec.h src\sparsevec.h src\vector.h
REGRESS = bit btree cast copy halfvec hnsw_bit hnsw_halfvec hnsw_sparsevec hnsw_vector ivfflat_bit ivfflat_halfvec ivfflat_vector sparsevec vector_type
REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)

120
README.md
View File

@@ -223,7 +223,7 @@ L2 distance
CREATE INDEX ON items USING hnsw (embedding vector_l2_ops);
```
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)
Note: Use `halfvec_l2_ops` for `halfvec` and `sparsevec_l2_ops` for `sparsevec` (and similar with the other distance functions)
Inner product
@@ -259,7 +259,6 @@ Supported types are:
- `vector` - up to 2,000 dimensions
- `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)
@@ -452,6 +451,89 @@ Use [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html
CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
```
## Iterative Index Scans
*Unreleased*
With approximate indexes, queries with filtering can return less results (due to post-filtering). Starting with 0.8.0, you can enable iterative index scans. If too few results from the initial scan match the filters, the scan will resume until enough results are found (or it reaches `hnsw.max_scan_tuples` or `ivfflat.max_probes`). This can significantly improve recall.
There are two modes for iterative scans: strict and relaxed.
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;
```
You can see when increasing this is needed by enabling debug messages
```sql
SET client_min_messages = debug1;
```
which will show when a scan reaches the memory limit
```text
DEBUG: hnsw index scan reached memory limit after 20000 tuples
HINT: Increase hnsw.scan_mem_multiplier to scan more tuples.
```
#### 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*
@@ -478,16 +560,6 @@ 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))
@@ -951,30 +1023,6 @@ 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.

View File

@@ -1,188 +1,6 @@
-- 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;

View File

@@ -272,9 +272,6 @@ 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;
@@ -650,199 +647,6 @@ 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

View File

@@ -18,16 +18,17 @@
#define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
#endif
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},
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},
{NULL, 0, false}
};
int hnsw_ef_search;
int hnsw_max_search_tuples;
int hnsw_iterative_search;
int hnsw_iterative_scan;
int hnsw_max_scan_tuples;
double hnsw_scan_mem_multiplier;
int hnsw_lock_tranche_id;
static relopt_kind hnsw_relopt_kind;
@@ -76,16 +77,21 @@ HnswInit(void)
DefineCustomIntVariable("hnsw.ef_search", "Sets the size of the dynamic candidate list for search",
"Valid range is 1..1000.", &hnsw_ef_search,
HNSW_DEFAULT_EF_SEARCH, HNSW_MIN_EF_SEARCH, HNSW_MAX_EF_SEARCH, PGC_USERSET, 0, NULL, NULL, NULL);
HNSW_DEFAULT_EF_SEARCH, HNSW_MIN_EF_SEARCH, HNSW_MAX_EF_SEARCH, PGC_USERSET, GUC_EXPLAIN, 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);
DefineCustomEnumVariable("hnsw.iterative_scan", "Sets the mode for iterative scans",
NULL, &hnsw_iterative_scan,
HNSW_ITERATIVE_SCAN_OFF, hnsw_iterative_scan_options, PGC_USERSET, GUC_EXPLAIN, 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);
/* 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, GUC_EXPLAIN, 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, GUC_EXPLAIN, NULL, NULL, NULL);
MarkGUCPrefixReserved("hnsw");
}
@@ -131,6 +137,10 @@ 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;
}

View File

@@ -109,16 +109,17 @@
/* Variables */
extern int hnsw_ef_search;
extern int hnsw_iterative_search;
extern int hnsw_max_search_tuples;
extern int hnsw_iterative_scan;
extern int hnsw_max_scan_tuples;
extern double hnsw_scan_mem_multiplier;
extern int hnsw_lock_tranche_id;
typedef enum HnswIterativeSearchMode
typedef enum HnswIterativeScanMode
{
HNSW_ITERATIVE_SEARCH_OFF,
HNSW_ITERATIVE_SEARCH_RELAXED,
HNSW_ITERATIVE_SEARCH_STRICT
} HnswIterativeSearchMode;
HNSW_ITERATIVE_SCAN_OFF,
HNSW_ITERATIVE_SCAN_RELAXED,
HNSW_ITERATIVE_SCAN_STRICT
} HnswIterativeScanMode;
typedef struct HnswElementData HnswElementData;
typedef struct HnswNeighborArray HnswNeighborArray;
@@ -372,6 +373,7 @@ typedef struct HnswScanOpaqueData
int m;
int64 tuples;
double previousDistance;
Size maxMemory;
MemoryContext tmpCtx;
/* Support functions */

View File

@@ -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_search != HNSW_ITERATIVE_SEARCH_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_scan != HNSW_ITERATIVE_SCAN_OFF ? &so->discarded : NULL, true, &so->tuples);
}
/*
@@ -121,14 +121,15 @@ 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);
so->first = true;
so->v.tids = NULL;
so->discarded = NULL;
/* Set support functions */
HnswInitSupport(&so->support, index);
/*
* Use a lower max allocation size than default to allow scanning more
@@ -136,10 +137,12 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
*/
so->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
"Hnsw scan temporary context",
0, 8 * 1024, 512 * 1024);
0, 8 * 1024, 256 * 1024);
/* Set support functions */
HnswInitSupport(&so->support, index);
/* 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);
scan->opaque = so;
@@ -154,13 +157,10 @@ 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);
@@ -233,7 +233,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
if (list_length(so->w) == 0)
{
if (hnsw_iterative_search == HNSW_ITERATIVE_SEARCH_OFF)
if (hnsw_iterative_scan == HNSW_ITERATIVE_SCAN_OFF)
break;
/* Empty index */
@@ -241,7 +241,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
break;
/* Reached max number of tuples */
if (hnsw_max_search_tuples != -1 && so->tuples >= hnsw_max_search_tuples)
if (so->tuples >= hnsw_max_scan_tuples)
{
if (pairingheap_is_empty(so->discarded))
break;
@@ -250,13 +250,13 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
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)
else if (MemoryContextMemAllocated(so->tmpCtx, false) > so->maxMemory)
{
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.")));
(errmsg("hnsw index scan reached memory limit after " INT64_FORMAT " tuples", so->tuples),
errhint("Increase hnsw.scan_mem_multiplier to scan more tuples.")));
break;
}
@@ -299,7 +299,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
so->w = list_delete_last(so->w);
/* Mark memory as free for next iteration */
if (hnsw_iterative_search != HNSW_ITERATIVE_SEARCH_OFF)
if (hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF)
{
pfree(element);
pfree(sc);
@@ -310,7 +310,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
heaptid = &element->heaptids[--element->heaptidsLength];
if (hnsw_iterative_search == HNSW_ITERATIVE_SEARCH_STRICT)
if (hnsw_iterative_scan == HNSW_ITERATIVE_SCAN_STRICT)
{
if (sc->distance < so->previousDistance)
continue;

View File

@@ -1,6 +1,5 @@
#include "postgres.h"
#include <float.h>
#include <math.h>
#include "access/generic_xlog.h"
@@ -158,9 +157,6 @@ 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);
}
@@ -550,14 +546,7 @@ 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 */
@@ -868,6 +857,7 @@ 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)++;
}
@@ -900,6 +890,7 @@ 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;
@@ -1404,19 +1395,6 @@ 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
hnsw_bit_support(PG_FUNCTION_ARGS)

View File

@@ -1,754 +0,0 @@
#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));
}

View File

@@ -1,23 +0,0 @@
#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

View File

@@ -17,13 +17,13 @@
#endif
int ivfflat_probes;
int ivfflat_iterative_search;
int ivfflat_iterative_scan;
int ivfflat_max_probes;
static relopt_kind ivfflat_relopt_kind;
static const struct config_enum_entry ivfflat_iterative_search_options[] = {
{"off", IVFFLAT_ITERATIVE_SEARCH_OFF, false},
{"relaxed_order", IVFFLAT_ITERATIVE_SEARCH_RELAXED, false},
static const struct config_enum_entry ivfflat_iterative_scan_options[] = {
{"off", IVFFLAT_ITERATIVE_SCAN_OFF, false},
{"relaxed_order", IVFFLAT_ITERATIVE_SCAN_RELAXED, false},
{NULL, 0, false}
};
@@ -39,16 +39,16 @@ IvfflatInit(void)
DefineCustomIntVariable("ivfflat.probes", "Sets the number of probes",
"Valid range is 1..lists.", &ivfflat_probes,
IVFFLAT_DEFAULT_PROBES, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
IVFFLAT_DEFAULT_PROBES, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, PGC_USERSET, GUC_EXPLAIN, 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);
DefineCustomEnumVariable("ivfflat.iterative_scan", "Sets the mode for iterative scans",
NULL, &ivfflat_iterative_scan,
IVFFLAT_ITERATIVE_SCAN_OFF, ivfflat_iterative_scan_options, PGC_USERSET, GUC_EXPLAIN, NULL, NULL, NULL);
/* If this is less than probes, probes is used */
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);
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, GUC_EXPLAIN, NULL, NULL, NULL);
MarkGUCPrefixReserved("ivfflat");
}
@@ -99,6 +99,10 @@ 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;
}

View File

@@ -80,14 +80,14 @@
/* Variables */
extern int ivfflat_probes;
extern int ivfflat_iterative_search;
extern int ivfflat_iterative_scan;
extern int ivfflat_max_probes;
typedef enum IvfflatIterativeSearchMode
typedef enum IvfflatIterativeScanMode
{
IVFFLAT_ITERATIVE_SEARCH_OFF,
IVFFLAT_ITERATIVE_SEARCH_RELAXED
} IvfflatIterativeSearchMode;
IVFFLAT_ITERATIVE_SCAN_OFF,
IVFFLAT_ITERATIVE_SCAN_RELAXED
} IvfflatIterativeScanMode;
typedef struct VectorArrayData
{

View File

@@ -171,7 +171,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
}
}
if (tuples < 100 && ivfflat_iterative_search == IVFFLAT_ITERATIVE_SEARCH_OFF)
if (tuples < 100 && ivfflat_iterative_scan == IVFFLAT_ITERATIVE_SCAN_OFF)
ereport(DEBUG1,
(errmsg("index scan found few tuples"),
errdetail("Index may have been created with little data."),
@@ -263,18 +263,8 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
/* Get lists and dimensions from metapage */
IvfflatGetMetaPageInfo(index, &lists, &dimensions);
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;
}
}
if (ivfflat_iterative_scan != IVFFLAT_ITERATIVE_SCAN_OFF)
maxProbes = Max(ivfflat_max_probes, probes);
else
maxProbes = probes;

View File

@@ -140,34 +140,6 @@ 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
-----------------

View File

@@ -30,23 +30,6 @@ 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

View File

@@ -1,102 +0,0 @@
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;

View File

@@ -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_search = strict_order;
SET hnsw.iterative_scan = 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_search = relaxed_order;
SET hnsw.iterative_scan = relaxed_order;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
@@ -123,7 +123,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
[0,0,0]
(3 rows)
RESET hnsw.iterative_search;
RESET hnsw.iterative_scan;
RESET hnsw.ef_search;
DROP TABLE t;
-- unlogged
@@ -165,21 +165,29 @@ 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_search;
hnsw.iterative_search
-----------------------
SHOW hnsw.iterative_scan;
hnsw.iterative_scan
---------------------
off
(1 row)
SET hnsw.iterative_search = on;
ERROR: invalid value for parameter "hnsw.iterative_search": "on"
SET hnsw.iterative_scan = on;
ERROR: invalid value for parameter "hnsw.iterative_scan": "on"
HINT: Available values: off, relaxed_order, strict_order.
SHOW hnsw.max_search_tuples;
hnsw.max_search_tuples
------------------------
-1
SHOW hnsw.max_scan_tuples;
hnsw.max_scan_tuples
----------------------
20000
(1 row)
SET hnsw.max_search_tuples = -2;
ERROR: -2 is outside the valid range for parameter "hnsw.max_search_tuples" (-1 .. 2147483647)
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)
DROP TABLE t;

View File

@@ -1,328 +0,0 @@
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

View File

@@ -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_search = relaxed_order;
SET ivfflat.iterative_scan = relaxed_order;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
@@ -95,13 +95,6 @@ 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
@@ -117,7 +110,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
[1,1,1]
(2 rows)
RESET ivfflat.iterative_search;
RESET ivfflat.iterative_scan;
RESET ivfflat.max_probes;
DROP TABLE t;
-- unlogged
@@ -151,23 +144,23 @@ 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_search;
ivfflat.iterative_search
--------------------------
SHOW ivfflat.iterative_scan;
ivfflat.iterative_scan
------------------------
off
(1 row)
SET ivfflat.iterative_search = on;
ERROR: invalid value for parameter "ivfflat.iterative_search": "on"
SET ivfflat.iterative_scan = on;
ERROR: invalid value for parameter "ivfflat.iterative_scan": "on"
HINT: Available values: off, relaxed_order.
SHOW ivfflat.max_probes;
ivfflat.max_probes
--------------------
-1
32768
(1 row)
SET ivfflat.max_probes = -2;
ERROR: -2 is outside the valid range for parameter "ivfflat.max_probes" (-1 .. 32768)
SET ivfflat.max_probes = 0;
ERROR: 0 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;

View File

@@ -38,14 +38,6 @@ 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);

View File

@@ -28,21 +28,6 @@ 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));

View File

@@ -1,58 +0,0 @@
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;

View File

@@ -63,14 +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_search = strict_order;
SET hnsw.iterative_scan = strict_order;
SET hnsw.ef_search = 1;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
SET hnsw.iterative_search = relaxed_order;
SET hnsw.iterative_scan = relaxed_order;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
RESET hnsw.iterative_search;
RESET hnsw.iterative_scan;
RESET hnsw.ef_search;
DROP TABLE t;
@@ -98,12 +98,16 @@ SHOW hnsw.ef_search;
SET hnsw.ef_search = 0;
SET hnsw.ef_search = 1001;
SHOW hnsw.iterative_search;
SHOW hnsw.iterative_scan;
SET hnsw.iterative_search = on;
SET hnsw.iterative_scan = on;
SHOW hnsw.max_search_tuples;
SHOW hnsw.max_scan_tuples;
SET hnsw.max_search_tuples = -2;
SET hnsw.max_scan_tuples = 0;
SHOW hnsw.scan_mem_multiplier;
SET hnsw.scan_mem_multiplier = 0;
DROP TABLE t;

View File

@@ -1,73 +0,0 @@
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]');

View File

@@ -50,10 +50,7 @@ CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 3);
SET ivfflat.iterative_search = relaxed_order;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
SET ivfflat.max_probes = 0;
SET ivfflat.iterative_scan = relaxed_order;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
SET ivfflat.max_probes = 1;
@@ -62,7 +59,7 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
SET ivfflat.max_probes = 2;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
RESET ivfflat.iterative_search;
RESET ivfflat.iterative_scan;
RESET ivfflat.max_probes;
DROP TABLE t;
@@ -87,13 +84,13 @@ SHOW ivfflat.probes;
SET ivfflat.probes = 0;
SET ivfflat.probes = 32769;
SHOW ivfflat.iterative_search;
SHOW ivfflat.iterative_scan;
SET ivfflat.iterative_search = on;
SET ivfflat.iterative_scan = on;
SHOW ivfflat.max_probes;
SET ivfflat.max_probes = -2;
SET ivfflat.max_probes = 0;
SET ivfflat.max_probes = 32769;
DROP TABLE t;

View File

@@ -40,10 +40,6 @@ 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);

View File

@@ -45,10 +45,6 @@ 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");

View File

@@ -12,8 +12,8 @@ $node->start;
# Create extension
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
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 @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 @subs = (" ", " ", ",", ":", "-", "1", "9", "\0", "2147483648", "-2147483649");
for my $i (0 .. $#types)

View File

@@ -23,7 +23,7 @@ $node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v vector_l2_ops
my $count = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET ivfflat.probes = 10;
SET ivfflat.iterative_search = relaxed_order;
SET ivfflat.iterative_scan = relaxed_order;
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
));
is($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_search = relaxed_order;
SET ivfflat.iterative_scan = 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;
));

View File

@@ -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_search = relaxed_order;
SET ivfflat.iterative_scan = 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_search = relaxed_order;
SET ivfflat.iterative_scan = relaxed_order;
SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[$i]' LIMIT $limit;
));
my @actual_ids = split("\n", $actual);

View File

@@ -26,8 +26,9 @@ $node->safe_psql("postgres", qq(
my $count = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET hnsw.iterative_search = relaxed_order;
SET work_mem = '8MB';
SET hnsw.iterative_scan = relaxed_order;
SET hnsw.max_scan_tuples = 100000;
SET hnsw.scan_mem_multiplier = 2;
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
));
is($count, 10);
@@ -42,9 +43,9 @@ foreach ((30000, 50000, 70000))
{
$count = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET hnsw.iterative_search = relaxed_order;
SET hnsw.max_search_tuples = $max_tuples;
SET work_mem = '8MB';
SET hnsw.iterative_scan = relaxed_order;
SET hnsw.max_scan_tuples = $max_tuples;
SET hnsw.scan_mem_multiplier = 2;
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 += $count;
@@ -57,11 +58,11 @@ foreach ((30000, 50000, 70000))
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
SET enable_seqscan = off;
SET hnsw.iterative_search = relaxed_order;
SET hnsw.iterative_scan = relaxed_order;
SET client_min_messages = debug1;
SET work_mem = '2MB';
SET work_mem = '1MB';
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
));
like($stderr, qr/hnsw index scan exceeded work_mem after \d+ tuples/);
like($stderr, qr/hnsw index scan reached memory limit after \d+ tuples/);
done_testing();

View File

@@ -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_search = $mode;
SET hnsw.iterative_scan = $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_search = $mode;
SET hnsw.iterative_scan = $mode;
SELECT i FROM tst WHERE i % $c = 0 ORDER BY v $operator '$queries[$i]' LIMIT $limit;
));
my @actual_ids = split("\n", $actual);

View File

@@ -1,132 +0,0 @@
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();

View File

@@ -1,113 +0,0 @@
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();

View File

@@ -1,101 +0,0 @@
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();

View File

@@ -1,58 +0,0 @@
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();