Compare commits

...

42 Commits

Author SHA1 Message Date
Andrew Kane
8fcf77f89a Added support for bigint attributes [skip ci] 2023-11-15 15:37:27 -08:00
Andrew Kane
08bd246529 Fixed vacuum test [skip ci] 2023-11-10 14:03:13 -08:00
Andrew Kane
f57f2b6821 Added support for inline filtering with HNSW 2023-11-10 13:28:48 -08:00
Andrew Kane
69a2ce0d43 Use datumIsEqual to compare 2023-11-10 10:46:48 -08:00
Andrew Kane
c5e8c46b80 Switched from VECTOR_SIZE to VARSIZE_ANY [skip ci] 2023-11-09 19:41:38 -08:00
Andrew Kane
94f7304ccd Keep vector for now to be overly cautious about packing [skip ci] 2023-11-09 18:43:55 -08:00
Andrew Kane
d078db3d25 Switched HnswElementTuple to generic data and zero full section 2023-11-09 18:28:25 -08:00
Andrew Kane
fbb904ae2f Use pointer for VARSIZE_ANY 2023-11-09 17:50:28 -08:00
Andrew Kane
3cf6f62900 Switched to datum for HnswElement 2023-11-09 17:35:39 -08:00
Andrew Kane
2a69e22ca4 Switched from VECTOR_SIZE to VARSIZE_ANY where possible (less vector-specific) 2023-11-09 17:16:43 -08:00
Andrew Kane
84e073888c Removed vector-specific code from HNSW_ELEMENT_TUPLE_SIZE [skip ci] 2023-11-09 16:57:01 -08:00
Andrew Kane
81a62d55d1 Switched from HNSW_ELEMENT_TUPLE_SIZE to ItemIdGetLength where possible (less vector-specific) 2023-11-09 16:32:00 -08:00
Andrew Kane
3f3463bde5 Improved memory calculation for HNSW and removed vector-specific code 2023-11-09 16:21:26 -08:00
Andrew Kane
a01a72d812 Updated comment [skip ci] 2023-11-05 08:42:06 -08:00
Andrew Kane
0c2fc18a80 Updated comment [skip ci] 2023-11-05 08:40:21 -08:00
Andrew Kane
e860042d3c Improved variable name [skip ci] 2023-11-05 08:35:54 -08:00
Andrew Kane
5986862bd2 Added note about check constraint [skip ci] 2023-11-04 15:01:37 -07:00
Andrew Kane
5d24f5d09a Improved header installation on Windows 2023-11-04 11:16:40 -07:00
Andrew Kane
7c43b0d8ee Updated example [skip ci] 2023-11-03 23:54:50 -07:00
Andrew Kane
7be40036f4 Updated readme [skip ci] 2023-11-03 23:46:23 -07:00
Andrew Kane
9b5a1a69db Updated readme [skip ci] 2023-11-03 23:43:47 -07:00
Andrew Kane
04b96506f5 Added info on storing vectors with more precision [skip ci] 2023-11-03 20:14:28 -07:00
Andrew Kane
35cd7b63cb Updated readme [skip ci] 2023-11-03 17:02:30 -07:00
Andrew Kane
b5416d6f10 Updated readme [skip ci] 2023-11-03 16:48:57 -07:00
Andrew Kane
f361bf2704 Improved docs on indexing vectors with different dimensions [skip ci] 2023-11-03 16:42:14 -07:00
Andrew Kane
3d8c1921aa Improved upgrading docs - #339 [skip ci] 2023-11-03 16:15:06 -07:00
Andrew Kane
154207bc17 Added info on columns with different dimensions [skip ci] 2023-11-03 16:02:00 -07:00
Andrew Kane
8e507f3bf5 Free remaining allocation from deconstruct_array - #332 2023-11-02 21:20:21 -07:00
Andrew Kane
e115773a55 Removed unneeded allocation 2023-11-02 21:16:06 -07:00
Andrew Kane
9333bef046 Added link to setup-pgvector [skip ci] 2023-11-02 13:22:19 -07:00
Andrew Kane
4851e47d9f Added Reciprocal Rank Fusion example to readme [skip ci] 2023-11-01 13:20:49 -07:00
Andrew Kane
12aecfb4f5 Added Nim and Zig to readme [skip ci] 2023-10-31 02:26:18 -07:00
Andrew Kane
800697fb14 Updated column alias [skip ci] 2023-10-29 16:47:55 -07:00
Andrew Kane
de1f2b09dd Improved indexing progress queries [skip ci] 2023-10-29 16:41:39 -07:00
Andrew Kane
bcccb7f5a5 Improved docs for indexing progress - closes #320 and closes #321 [skip ci] 2023-10-29 16:13:12 -07:00
Andrew Kane
bec3d30d68 Added TypeScript to readme [skip ci] 2023-10-29 12:49:01 -07:00
Andrew Kane
588de60445 Added Groovy to readme [skip ci] 2023-10-29 12:39:53 -07:00
Andrew Kane
c599f92b52 Updated readme [skip ci] 2023-10-27 13:22:37 -07:00
Andrew Kane
2a17b335da Added Kotlin to readme [skip ci] 2023-10-26 12:25:58 -07:00
Andrew Kane
6ede6ac301 Added link to pgvector-c [skip ci] 2023-10-26 00:30:06 -07:00
Andrew Kane
3f49b95f01 Added Postgres 17 to CI [skip ci] 2023-10-19 00:37:24 -07:00
Andrew Kane
ef1bea7163 Updated checkout action [skip ci] 2023-10-19 00:36:53 -07:00
15 changed files with 672 additions and 167 deletions

View File

@@ -8,6 +8,8 @@ jobs:
fail-fast: false
matrix:
include:
- postgres: 17
os: ubuntu-22.04
- postgres: 16
os: ubuntu-22.04
- postgres: 15
@@ -21,7 +23,7 @@ jobs:
- postgres: 11
os: ubuntu-20.04
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v4
- uses: ankane/setup-postgres@v1
with:
postgres-version: ${{ matrix.postgres }}
@@ -43,7 +45,7 @@ jobs:
runs-on: macos-latest
if: ${{ !startsWith(github.ref_name, 'windows') }}
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v4
- uses: ankane/setup-postgres@v1
with:
postgres-version: 14
@@ -65,7 +67,7 @@ jobs:
runs-on: windows-latest
if: ${{ !startsWith(github.ref_name, 'mac') }}
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v4
- uses: ankane/setup-postgres@v1
with:
postgres-version: 14

View File

@@ -1,3 +1,7 @@
## 0.6.0 (unreleased)
- Added support for inline filtering with HNSW
## 0.5.1 (2023-10-10)
- Improved performance of HNSW index builds

View File

@@ -56,7 +56,7 @@ install:
copy $(EXTENSION).control "$(SHAREDIR)\extension"
copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
copy $(HEADERS) "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
for %f in ($(HEADERS)) do copy %f "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
installcheck:
"$(BINDIR)\pg_regress" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)

102
README.md
View File

@@ -26,7 +26,7 @@ make install # may need sudo
See the [installation notes](#installation-notes) if you run into issues
You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres)
You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
## Getting Started
@@ -215,6 +215,23 @@ SELECT ...
COMMIT;
```
### Indexing Progress
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;
```
The phases for IVFFlat are:
1. `initializing`
2. `performing k-means`
3. `assigning tuples`
4. `loading tuples`
Note: `%` is only populated during the `loading tuples` phase
## HNSW
An HNSW index creates a multilayer graph. It has slower build times and uses more memory than IVFFlat, but has better query performance (in terms of speed-recall tradeoff). Theres no training step like IVFFlat, so the index can be created without any data in the table.
@@ -271,22 +288,18 @@ SELECT ...
COMMIT;
```
## Indexing Progress
### Indexing Progress
Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING) with Postgres 12+
```sql
SELECT phase, tuples_done, tuples_total FROM pg_stat_progress_create_index;
SELECT phase, round(100.0 * blocks_done / nullif(blocks_total, 0), 1) AS "%" FROM pg_stat_progress_create_index;
```
The phases are:
The phases for HNSW are:
1. `initializing`
2. `performing k-means` - IVFFlat only
3. `assigning tuples` - IVFFlat only
4. `loading tuples`
Note: `tuples_done` and `tuples_total` are only populated during the `loading tuples` phase
2. `loading tuples`
## Filtering
@@ -302,6 +315,12 @@ Create an index on one [or more](https://www.postgresql.org/docs/current/indexes
CREATE INDEX ON items (category_id);
```
Or a composite HNSW index for approximate search (added in 0.6.0)
```sql
CREATE INDEX ON items USING hnsw (embedding vector_l2_ops, category_id);
```
Or a [partial index](https://www.postgresql.org/docs/current/indexes-partial.html) on the vector column for approximate search
```sql
@@ -317,13 +336,15 @@ CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(cate
## Hybrid Search
Use together with Postgres [full-text search](https://www.postgresql.org/docs/current/textsearch-intro.html) for hybrid search ([Python example](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search.py)).
Use together with Postgres [full-text search](https://www.postgresql.org/docs/current/textsearch-intro.html) for hybrid search.
```sql
SELECT id, content FROM items, plainto_tsquery('hello search') query
WHERE textsearch @@ query ORDER BY ts_rank_cd(textsearch, query) DESC LIMIT 5;
```
You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search_rrf.py) or a [cross-encoder](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search.py) to combine results.
## Performance
Use `EXPLAIN ANALYZE` to debug performance.
@@ -360,6 +381,7 @@ Use pgvector from any language with a Postgres client. You can even generate and
Language | Libraries / Examples
--- | ---
C | [pgvector-c](https://github.com/pgvector/pgvector-c)
C++ | [pgvector-cpp](https://github.com/pgvector/pgvector-cpp)
C# | [pgvector-dotnet](https://github.com/pgvector/pgvector-dotnet)
Crystal | [pgvector-crystal](https://github.com/pgvector/pgvector-crystal)
@@ -367,10 +389,11 @@ Dart | [pgvector-dart](https://github.com/pgvector/pgvector-dart)
Elixir | [pgvector-elixir](https://github.com/pgvector/pgvector-elixir)
Go | [pgvector-go](https://github.com/pgvector/pgvector-go)
Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell)
Java, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
Java, Kotlin, Groovy, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
JavaScript, TypeScript | [pgvector-node](https://github.com/pgvector/pgvector-node)
Julia | [pgvector-julia](https://github.com/pgvector/pgvector-julia)
Lua | [pgvector-lua](https://github.com/pgvector/pgvector-lua)
Node.js | [pgvector-node](https://github.com/pgvector/pgvector-node)
Nim | [pgvector-nim](https://github.com/pgvector/pgvector-nim)
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)
@@ -378,6 +401,7 @@ R | [pgvector-r](https://github.com/pgvector/pgvector-r)
Ruby | [pgvector-ruby](https://github.com/pgvector/pgvector-ruby), [Neighbor](https://github.com/ankane/neighbor)
Rust | [pgvector-rust](https://github.com/pgvector/pgvector-rust)
Swift | [pgvector-swift](https://github.com/pgvector/pgvector-swift)
Zig | [pgvector-zig](https://github.com/pgvector/pgvector-zig)
## Frequently Asked Questions
@@ -393,6 +417,55 @@ Yes, pgvector uses the write-ahead log (WAL), which allows for replication and p
Youll need to use [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction) at the moment.
#### Can I store vectors with different dimensions in the same column?
You can use `vector` as the type (instead of `vector(3)`).
```sql
CREATE TABLE embeddings (model_id bigint, item_id bigint, embedding vector, PRIMARY KEY (model_id, item_id));
```
However, you can only create indexes on rows with the same number of dimensions (using [expression](https://www.postgresql.org/docs/current/indexes-expressional.html) and [partial](https://www.postgresql.org/docs/current/indexes-partial.html) indexing):
```sql
CREATE INDEX ON embeddings USING hnsw ((embedding::vector(3)) vector_l2_ops) WHERE (model_id = 123);
```
and query with:
```sql
SELECT * FROM embeddings WHERE model_id = 123 ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
```
#### Can I store vectors with more precision?
You can use the `double precision[]` or `numeric[]` type to store vectors with more precision.
```sql
CREATE TABLE items (id bigserial PRIMARY KEY, embedding double precision[]);
-- use {} instead of [] for Postgres arrays
INSERT INTO items (embedding) VALUES ('{1,2,3}'), ('{4,5,6}');
```
Optionally, add a [check constraint](https://www.postgresql.org/docs/current/ddl-constraints.html) to ensure data can be converted to the `vector` type and has the expected dimensions.
```sql
ALTER TABLE items ADD CHECK (vector_dims(embedding::vector) = 3);
```
Use [expression indexing](https://www.postgresql.org/docs/current/indexes-expressional.html) to index (at a lower precision):
```sql
CREATE INDEX ON items USING hnsw ((embedding::vector(3)) vector_l2_ops);
```
and query with:
```sql
SELECT * FROM items ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
```
## Troubleshooting
#### Why isnt a query using an index?
@@ -406,6 +479,8 @@ SELECT ...
COMMIT;
```
Also, if the table is small, a table scan may be faster.
#### Why isnt a query using a parallel table scan?
The planner doesnt consider [out-of-line storage](https://www.postgresql.org/docs/current/storage-toast.html) in cost estimates, which can make a serial scan look cheaper. You can reduce the cost of a parallel scan for a query with:
@@ -595,7 +670,7 @@ pgvector is available on [these providers](https://github.com/pgvector/pgvector/
## Upgrading
Install the latest version. Then in each database you want to upgrade, run:
[Install](#installation) the latest version (use the same method as the original installation). Then in each database you want to upgrade, run:
```sql
ALTER EXTENSION vector UPDATE;
@@ -643,6 +718,7 @@ Thanks to:
- [k-means++: The Advantage of Careful Seeding](https://theory.stanford.edu/~sergei/papers/kMeansPP-soda.pdf)
- [Concept Decompositions for Large Sparse Text Data using Clustering](https://www.cs.utexas.edu/users/inderjit/public_papers/concept_mlj.pdf)
- [Efficient and Robust Approximate Nearest Neighbor Search using Hierarchical Navigable Small World Graphs](https://arxiv.org/ftp/arxiv/papers/1603/1603.09320.pdf)
- [HQANN: Efficient and Robust Similarity Search for Hybrid Queries with Structured and Unstructured Constraints](https://arxiv.org/pdf/2207.07940.pdf)
## History

View File

@@ -0,0 +1,18 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.0'" to load this file. \quit
CREATE FUNCTION hnsw_attribute_distance(integer, integer) RETURNS float8
AS 'MODULE_PATHNAME', 'hnsw_int4_attribute_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION hnsw_attribute_distance(bigint, bigint) RETURNS float8
AS 'MODULE_PATHNAME', 'hnsw_int8_attribute_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE OPERATOR CLASS vector_integer_ops
DEFAULT FOR TYPE integer USING hnsw AS
OPERATOR 2 = (integer, integer),
FUNCTION 3 hnsw_attribute_distance(integer, integer);
CREATE OPERATOR CLASS vector_bigint_ops
DEFAULT FOR TYPE bigint USING hnsw AS
OPERATOR 2 = (bigint, bigint),
FUNCTION 3 hnsw_attribute_distance(bigint, bigint);

View File

@@ -290,3 +290,21 @@ CREATE OPERATOR CLASS vector_cosine_ops
OPERATOR 1 <=> (vector, vector) FOR ORDER BY float_ops,
FUNCTION 1 vector_negative_inner_product(vector, vector),
FUNCTION 2 vector_norm(vector);
-- hnsw attributes
CREATE FUNCTION hnsw_attribute_distance(integer, integer) RETURNS float8
AS 'MODULE_PATHNAME', 'hnsw_int4_attribute_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION hnsw_attribute_distance(bigint, bigint) RETURNS float8
AS 'MODULE_PATHNAME', 'hnsw_int8_attribute_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE OPERATOR CLASS vector_integer_ops
DEFAULT FOR TYPE integer USING hnsw AS
OPERATOR 2 = (integer, integer),
FUNCTION 3 hnsw_attribute_distance(integer, integer);
CREATE OPERATOR CLASS vector_bigint_ops
DEFAULT FOR TYPE bigint USING hnsw AS
OPERATOR 2 = (bigint, bigint),
FUNCTION 3 hnsw_attribute_distance(bigint, bigint);

View File

@@ -167,7 +167,7 @@ hnswhandler(PG_FUNCTION_ARGS)
IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
amroutine->amstrategies = 0;
amroutine->amsupport = 2;
amroutine->amsupport = 3;
#if PG_VERSION_NUM >= 130000
amroutine->amoptsprocnum = 0;
#endif
@@ -175,7 +175,7 @@ hnswhandler(PG_FUNCTION_ARGS)
amroutine->amcanorderbyop = true;
amroutine->amcanbackward = false; /* can change direction mid-scan */
amroutine->amcanunique = false;
amroutine->amcanmulticol = false;
amroutine->amcanmulticol = true;
amroutine->amoptionalkey = true;
amroutine->amsearcharray = false;
amroutine->amsearchnulls = false;
@@ -222,3 +222,31 @@ hnswhandler(PG_FUNCTION_ARGS)
PG_RETURN_POINTER(amroutine);
}
/*
* Get the distance between two int4 attributes
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(hnsw_int4_attribute_distance);
Datum
hnsw_int4_attribute_distance(PG_FUNCTION_ARGS)
{
int32 a = PG_GETARG_INT32(0);
int32 b = PG_GETARG_INT32(1);
double distance = ((double) a) - ((double) b);
PG_RETURN_FLOAT8(distance);
}
/*
* Get the distance between two int8 attributes
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(hnsw_int8_attribute_distance);
Datum
hnsw_int8_attribute_distance(PG_FUNCTION_ARGS)
{
int64 a = PG_GETARG_INT64(0);
int64 b = PG_GETARG_INT64(1);
double distance = ((double) a) - ((double) b);
PG_RETURN_FLOAT8(distance);
}

View File

@@ -19,6 +19,7 @@
/* Support functions */
#define HNSW_DISTANCE_PROC 1
#define HNSW_NORM_PROC 2
#define HNSW_ATTRIBUTE_DISTANCE_PROC 3
#define HNSW_VERSION 1
#define HNSW_MAGIC_NUMBER 0xA953A953
@@ -59,7 +60,7 @@
#define HNSW_MAX_SIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - sizeof(ItemIdData))
#define HNSW_ELEMENT_TUPLE_SIZE(_dim) MAXALIGN(offsetof(HnswElementTupleData, vec) + VECTOR_SIZE(_dim))
#define HNSW_ELEMENT_TUPLE_SIZE(size) MAXALIGN(offsetof(HnswElementTupleData, data) + (size))
#define HNSW_NEIGHBOR_TUPLE_SIZE(level, m) MAXALIGN(offsetof(HnswNeighborTupleData, indextids) + ((level) + 2) * (m) * sizeof(ItemPointerData))
#define HnswPageGetOpaque(page) ((HnswPageOpaque) PageGetSpecialPointer(page))
@@ -103,7 +104,8 @@ typedef struct HnswElementData
OffsetNumber offno;
OffsetNumber neighborOffno;
BlockNumber neighborPage;
Vector *vec;
Datum value;
IndexTuple itup;
} HnswElementData;
typedef HnswElementData * HnswElement;
@@ -154,17 +156,18 @@ typedef struct HnswBuildState
double reltuples;
/* Support functions */
FmgrInfo *procinfo;
FmgrInfo **procinfos;
FmgrInfo *normprocinfo;
Oid collation;
Oid *collations;
/* Variables */
List *elements;
HnswElement entryPoint;
double ml;
int maxLevel;
double maxInMemoryElements;
long memoryLeft;
bool flushed;
bool useIndexTuple;
Vector *normvec;
/* Memory */
@@ -204,7 +207,7 @@ typedef struct HnswElementTupleData
ItemPointerData heaptids[HNSW_HEAPTIDS];
ItemPointerData neighbortid;
uint16 unused2;
Vector vec;
Vector data;
} HnswElementTupleData;
typedef HnswElementTupleData * HnswElementTuple;
@@ -226,9 +229,9 @@ typedef struct HnswScanOpaqueData
MemoryContext tmpCtx;
/* Support functions */
FmgrInfo *procinfo;
FmgrInfo **procinfos;
FmgrInfo *normprocinfo;
Oid collation;
Oid *collations;
} HnswScanOpaqueData;
typedef HnswScanOpaqueData * HnswScanOpaque;
@@ -246,8 +249,8 @@ typedef struct HnswVacuumState
int efConstruction;
/* Support functions */
FmgrInfo *procinfo;
Oid collation;
FmgrInfo **procinfos;
Oid *collations;
/* Variables */
HTAB *deleted;
@@ -269,26 +272,27 @@ Buffer HnswNewBuffer(Relation index, ForkNumber forkNum);
void HnswInitPage(Buffer buf, Page page);
void HnswInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state);
void HnswInit(void);
List *HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement);
List *HnswSearchLayer(Datum q, IndexTuple qtup, ScanKeyData *keyData, List *ep, int ef, int lc, Relation index, FmgrInfo **procinfos, Oid *collations, int m, bool loadVec, HnswElement skipElement, bool inMemory);
HnswElement HnswGetEntryPoint(Relation index);
void HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
HnswElement HnswInitElement(ItemPointer tid, int m, double ml, int maxLevel);
void HnswFreeElement(HnswElement element);
HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
void HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing);
HnswElement HnswFindDuplicate(HnswElement e);
HnswCandidate *HnswEntryCandidate(HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
void HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo **procinfos, Oid *collations, int m, int efConstruction, bool existing, bool inMemory);
HnswElement HnswFindDuplicate(HnswElement e, Relation index);
HnswCandidate *HnswEntryCandidate(HnswElement em, Datum q, IndexTuple qtup, ScanKeyData *keyData, Relation rel, FmgrInfo **procinfos, Oid *collations, bool loadVec, bool inMemory);
void HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum);
void HnswSetNeighborTuple(HnswNeighborTuple ntup, HnswElement e, int m);
void HnswAddHeapTid(HnswElement element, ItemPointer heaptid);
void HnswInitNeighbors(HnswElement element, int m);
bool HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel);
void HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting);
void HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec);
void HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec);
void HnswSetElementTuple(HnswElementTuple etup, HnswElement element);
void HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
void HnswUpdateNeighborPages(Relation index, FmgrInfo **procinfos, Oid *collations, HnswElement e, int m, bool checkExisting);
void HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec, Relation index);
void HnswLoadElement(HnswElement element, float *distance, Datum *q, IndexTuple qtup, ScanKeyData *keyData, Relation index, FmgrInfo **procinfos, Oid *collations, bool loadVec);
void HnswSetElementTuple(HnswElementTuple etup, HnswElement element, bool useIndexTuple);
void HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int *updateIdx, Relation index, FmgrInfo **procinfos, Oid *collations, bool inMemory);
void HnswLoadNeighbors(HnswElement element, Relation index, int m);
void HnswElementSetData(HnswElement element, Relation index, Datum value, Datum *values, bool *isnull);
/* Index access methods */
IndexBuildResult *hnswbuild(Relation heap, Relation index, IndexInfo *indexInfo);
@@ -305,5 +309,6 @@ IndexScanDesc hnswbeginscan(Relation index, int nkeys, int norderbys);
void hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int norderbys);
bool hnswgettuple(IndexScanDesc scan, ScanDirection dir);
void hnswendscan(IndexScanDesc scan);
FmgrInfo **HnswInitProcinfos(Relation index);
#endif

View File

@@ -8,6 +8,7 @@
#include "lib/pairingheap.h"
#include "nodes/pg_list.h"
#include "storage/bufmgr.h"
#include "utils/datum.h"
#include "utils/memutils.h"
#if PG_VERSION_NUM >= 140000
@@ -105,8 +106,8 @@ CreateElementPages(HnswBuildState * buildstate)
{
Relation index = buildstate->index;
ForkNumber forkNum = buildstate->forkNum;
int dimensions = buildstate->dimensions;
Size etupSize;
bool useIndexTuple = buildstate->useIndexTuple;
Size etupAllocSize;
Size maxSize;
HnswElementTuple etup;
HnswNeighborTuple ntup;
@@ -117,11 +118,11 @@ CreateElementPages(HnswBuildState * buildstate)
ListCell *lc;
/* Calculate sizes */
etupAllocSize = BLCKSZ;
maxSize = HNSW_MAX_SIZE;
etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
/* Allocate once */
etup = palloc0(etupSize);
etup = palloc0(etupAllocSize);
ntup = palloc0(BLCKSZ);
/* Prepare first page */
@@ -133,15 +134,24 @@ CreateElementPages(HnswBuildState * buildstate)
foreach(lc, buildstate->elements)
{
HnswElement element = lfirst(lc);
Size etupSize;
Size ntupSize;
Size combinedSize;
HnswSetElementTuple(etup, element);
/* Zero memory for each element */
MemSet(etup, 0, etupAllocSize);
/* Calculate sizes */
etupSize = HNSW_ELEMENT_TUPLE_SIZE(useIndexTuple ? IndexTupleSize(element->itup) : VARSIZE_ANY(DatumGetPointer(element->value)));
ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, buildstate->m);
combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
/* Initial size check */
if (etupSize > etupAllocSize)
elog(ERROR, "index tuple too large");
HnswSetElementTuple(etup, element, useIndexTuple);
/* Keep element and neighbors on the same page if possible */
if (PageGetFreeSpace(page) < etupSize || (combinedSize <= maxSize && PageGetFreeSpace(page) < combinedSize))
HnswBuildAppendPage(index, &buf, &page, &state, forkNum);
@@ -264,13 +274,15 @@ FlushPages(HnswBuildState * buildstate)
* Insert tuple
*/
static bool
InsertTuple(Relation index, Datum *values, HnswElement element, HnswBuildState * buildstate, HnswElement * dup)
InsertTuple(Relation index, Datum *values, bool *isnull, HnswElement element, HnswBuildState * buildstate, HnswElement * dup, MemoryContext outerCtx)
{
FmgrInfo *procinfo = buildstate->procinfo;
Oid collation = buildstate->collation;
FmgrInfo **procinfos = buildstate->procinfos;
Oid *collations = buildstate->collations;
HnswElement entryPoint = buildstate->entryPoint;
int efConstruction = buildstate->efConstruction;
int m = buildstate->m;
bool inMemory = true;
MemoryContext oldCtx;
/* Detoast once for all calls */
Datum value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
@@ -278,18 +290,20 @@ InsertTuple(Relation index, Datum *values, HnswElement element, HnswBuildState *
/* Normalize if needed */
if (buildstate->normprocinfo != NULL)
{
if (!HnswNormValue(buildstate->normprocinfo, collation, &value, buildstate->normvec))
if (!HnswNormValue(buildstate->normprocinfo, collations[0], &value, buildstate->normvec))
return false;
}
/* Copy value to element so accessible outside of memory context */
memcpy(element->vec, DatumGetVector(value), VECTOR_SIZE(buildstate->dimensions));
oldCtx = MemoryContextSwitchTo(outerCtx);
HnswElementSetData(element, index, value, values, isnull);
MemoryContextSwitchTo(oldCtx);
/* Insert element in graph */
HnswInsertElement(element, entryPoint, NULL, procinfo, collation, m, efConstruction, false);
HnswInsertElement(element, entryPoint, index, procinfos, collations, m, efConstruction, false, inMemory);
/* Look for duplicate */
*dup = HnswFindDuplicate(element);
*dup = HnswFindDuplicate(element, index);
/* Update neighbors if needed */
if (*dup == NULL)
@@ -300,7 +314,7 @@ InsertTuple(Relation index, Datum *values, HnswElement element, HnswBuildState *
HnswNeighborArray *neighbors = &element->neighbors[lc];
for (int i = 0; i < neighbors->length; i++)
HnswUpdateConnection(element, &neighbors->items[i], lm, lc, NULL, NULL, procinfo, collation);
HnswUpdateConnection(element, &neighbors->items[i], lm, lc, NULL, index, procinfos, collations, inMemory);
}
}
@@ -313,6 +327,21 @@ InsertTuple(Relation index, Datum *values, HnswElement element, HnswBuildState *
return *dup == NULL;
}
/*
* Get the memory used by an element
*/
static long
HnswElementMemory(HnswElement e, int m)
{
long elementSize = sizeof(HnswElementData);
elementSize += sizeof(HnswNeighborArray) * (e->level + 1);
elementSize += sizeof(HnswCandidate) * (m * (e->level + 2));
elementSize += sizeof(ItemPointerData);
elementSize += IndexTupleSize(e->itup);
return elementSize;
}
/*
* Callback for table_index_build_scan
*/
@@ -334,7 +363,7 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
if (isnull[0])
return;
if (buildstate->indtuples >= buildstate->maxInMemoryElements)
if (buildstate->memoryLeft <= 0)
{
if (!buildstate->flushed)
{
@@ -360,13 +389,12 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
/* Allocate necessary memory outside of memory context */
element = HnswInitElement(tid, buildstate->m, buildstate->ml, buildstate->maxLevel);
element->vec = palloc(VECTOR_SIZE(buildstate->dimensions));
/* Use memory context since detoast can allocate */
oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
/* Insert tuple */
inserted = InsertTuple(index, values, element, buildstate, &dup);
inserted = InsertTuple(index, values, isnull, element, buildstate, &dup, oldCtx);
/* Reset memory context */
MemoryContextSwitchTo(oldCtx);
@@ -374,31 +402,21 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
/* Add outside memory context */
if (dup != NULL)
{
HnswAddHeapTid(dup, tid);
buildstate->memoryLeft -= sizeof(ItemPointerData);
}
/* Add to buildstate or free */
if (inserted)
{
buildstate->elements = lappend(buildstate->elements, element);
buildstate->memoryLeft -= HnswElementMemory(element, buildstate->m);
}
else
HnswFreeElement(element);
}
/*
* Get the max number of elements that fit into maintenance_work_mem
*/
static double
HnswGetMaxInMemoryElements(int m, double ml, int dimensions)
{
Size elementSize = sizeof(HnswElementData);
double avgLevel = -log(0.5) * ml;
elementSize += sizeof(HnswNeighborArray) * (avgLevel + 1);
elementSize += sizeof(HnswCandidate) * (m * (avgLevel + 2));
elementSize += sizeof(ItemPointerData);
elementSize += VECTOR_SIZE(dimensions);
return (maintenance_work_mem * 1024L) / elementSize;
}
/*
* Initialize the build state
*/
@@ -414,6 +432,19 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
buildstate->efConstruction = HnswGetEfConstruction(index);
buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
/* TODO See if needed */
if (IndexRelationGetNumberOfKeyAttributes(index) > 2)
elog(ERROR, "index cannot have more than two columns");
if (!OidIsValid(index_getprocid(index, 1, HNSW_DISTANCE_PROC)))
elog(ERROR, "first column must be a vector");
for (int i = 1; i < IndexRelationGetNumberOfKeyAttributes(index); i++)
{
if (!OidIsValid(index_getprocid(index, i + 1, HNSW_ATTRIBUTE_DISTANCE_PROC)))
elog(ERROR, "column %d cannot be a vector", i + 1);
}
/* Require column to have dimensions to be indexed */
if (buildstate->dimensions < 0)
elog(ERROR, "column does not have dimensions");
@@ -428,16 +459,17 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
buildstate->indtuples = 0;
/* Get support functions */
buildstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
buildstate->procinfos = HnswInitProcinfos(index);
buildstate->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
buildstate->collation = index->rd_indcollation[0];
buildstate->collations = index->rd_indcollation;
buildstate->elements = NIL;
buildstate->entryPoint = NULL;
buildstate->ml = HnswGetMl(buildstate->m);
buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
buildstate->maxInMemoryElements = HnswGetMaxInMemoryElements(buildstate->m, buildstate->ml, buildstate->dimensions);
buildstate->memoryLeft = maintenance_work_mem * 1024L;
buildstate->flushed = false;
buildstate->useIndexTuple = IndexRelationGetNumberOfAttributes(index) > 1;
/* Reuse for each tuple */
buildstate->normvec = InitVector(buildstate->dimensions);
@@ -453,6 +485,7 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
static void
FreeBuildState(HnswBuildState * buildstate)
{
pfree(buildstate->procinfos);
pfree(buildstate->normvec);
MemoryContextDelete(buildstate->tmpCtx);
}

View File

@@ -123,16 +123,16 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
Size minCombinedSize;
HnswElementTuple etup;
BlockNumber currentPage = insertPage;
int dimensions = e->vec->dim;
HnswNeighborTuple ntup;
Buffer nbuf;
Page npage;
OffsetNumber freeOffno = InvalidOffsetNumber;
OffsetNumber freeNeighborOffno = InvalidOffsetNumber;
BlockNumber newInsertPage = InvalidBlockNumber;
bool useIndexTuple = IndexRelationGetNumberOfAttributes(index) > 1;
/* Calculate sizes */
etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
etupSize = HNSW_ELEMENT_TUPLE_SIZE(useIndexTuple ? IndexTupleSize(e->itup) : VARSIZE_ANY(DatumGetPointer(e->value)));
ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
maxSize = HNSW_MAX_SIZE;
@@ -140,7 +140,7 @@ WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPag
/* Prepare element tuple */
etup = palloc0(etupSize);
HnswSetElementTuple(etup, e);
HnswSetElementTuple(etup, e, useIndexTuple);
/* Prepare neighbor tuple */
ntup = palloc0(ntupSize);
@@ -302,7 +302,7 @@ ConnectionExists(HnswElement e, HnswNeighborTuple ntup, int startIdx, int lm)
* Update neighbors
*/
void
HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting)
HnswUpdateNeighborPages(Relation index, FmgrInfo **procinfos, Oid *collations, HnswElement e, int m, bool checkExisting)
{
for (int lc = e->level; lc >= 0; lc--)
{
@@ -334,7 +334,7 @@ HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswE
*/
/* Select neighbors */
HnswUpdateConnection(e, hc, lm, lc, &idx, index, procinfo, collation);
HnswUpdateConnection(e, hc, lm, lc, &idx, index, procinfos, collations, false);
/* New element was not selected as a neighbor */
if (idx == -1)
@@ -405,8 +405,9 @@ HnswAddDuplicate(Relation index, HnswElement element, HnswElement dup)
Buffer buf;
Page page;
GenericXLogState *state;
Size etupSize = HNSW_ELEMENT_TUPLE_SIZE(dup->vec->dim);
ItemId itemid;
HnswElementTuple etup;
Size etupSize;
int i;
/* Read page */
@@ -416,7 +417,9 @@ HnswAddDuplicate(Relation index, HnswElement element, HnswElement dup)
page = GenericXLogRegisterBuffer(state, buf, 0);
/* Find space */
etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, dup->offno));
itemid = PageGetItemId(page, dup->offno);
etup = (HnswElementTuple) PageGetItem(page, itemid);
etupSize = ItemIdGetLength(itemid);
for (i = 0; i < HNSW_HEAPTIDS; i++)
{
if (!ItemPointerIsValid(&etup->heaptids[i]))
@@ -449,7 +452,7 @@ HnswAddDuplicate(Relation index, HnswElement element, HnswElement dup)
* Write changes to disk
*/
static void
WriteElement(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement dup, HnswElement entryPoint)
WriteElement(Relation index, FmgrInfo **procinfos, Oid *collations, HnswElement element, int m, int efConstruction, HnswElement dup, HnswElement entryPoint)
{
BlockNumber newInsertPage = InvalidBlockNumber;
@@ -468,7 +471,7 @@ WriteElement(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement elem
HnswUpdateMetaPage(index, 0, NULL, newInsertPage, MAIN_FORKNUM);
/* Update neighbors */
HnswUpdateNeighborPages(index, procinfo, collation, element, m, false);
HnswUpdateNeighborPages(index, procinfos, collations, element, m, false);
/* Update metapage if needed */
if (entryPoint == NULL || element->level > entryPoint->level)
@@ -487,8 +490,8 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
HnswElement element;
int m;
int efConstruction = HnswGetEfConstruction(index);
FmgrInfo *procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
Oid collation = index->rd_indcollation[0];
FmgrInfo **procinfos = HnswInitProcinfos(index);
Oid *collations = index->rd_indcollation;
HnswElement dup;
LOCKMODE lockmode = ShareLock;
@@ -499,7 +502,7 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
if (normprocinfo != NULL)
{
if (!HnswNormValue(normprocinfo, collation, &value, NULL))
if (!HnswNormValue(normprocinfo, collations[0], &value, NULL))
return false;
}
@@ -515,7 +518,7 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
/* Create an element */
element = HnswInitElement(heap_tid, m, HnswGetMl(m), HnswGetMaxLevel(m));
element->vec = DatumGetVector(value);
HnswElementSetData(element, index, value, values, isnull);
/* Prevent concurrent inserts when likely updating entry point */
if (entryPoint == NULL || element->level > entryPoint->level)
@@ -532,13 +535,13 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
}
/* Insert element in graph */
HnswInsertElement(element, entryPoint, index, procinfo, collation, m, efConstruction, false);
HnswInsertElement(element, entryPoint, index, procinfos, collations, m, efConstruction, false, false);
/* Look for duplicate */
dup = HnswFindDuplicate(element);
dup = HnswFindDuplicate(element, index);
/* Write to disk */
WriteElement(index, procinfo, collation, element, m, efConstruction, dup, entryPoint);
WriteElement(index, procinfos, collations, element, m, efConstruction, dup, entryPoint);
/* Release lock */
UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);

View File

@@ -15,12 +15,13 @@ GetScanItems(IndexScanDesc scan, Datum q)
{
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
Relation index = scan->indexRelation;
FmgrInfo *procinfo = so->procinfo;
Oid collation = so->collation;
FmgrInfo **procinfos = so->procinfos;
Oid *collations = so->collations;
List *ep;
List *w;
int m;
HnswElement entryPoint;
ScanKeyData *keyData = scan->keyData;
/* Get m and entry point */
HnswGetMetaPageInfo(index, &m, &entryPoint);
@@ -28,15 +29,15 @@ GetScanItems(IndexScanDesc scan, Datum q)
if (entryPoint == NULL)
return NIL;
ep = list_make1(HnswEntryCandidate(entryPoint, q, index, procinfo, collation, false));
ep = list_make1(HnswEntryCandidate(entryPoint, q, NULL, keyData, index, procinfos, collations, false, false));
for (int lc = entryPoint->level; lc >= 1; lc--)
{
w = HnswSearchLayer(q, ep, 1, lc, index, procinfo, collation, m, false, NULL);
w = HnswSearchLayer(q, NULL, keyData, ep, 1, lc, index, procinfos, collations, m, false, NULL, false);
ep = w;
}
return HnswSearchLayer(q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
return HnswSearchLayer(q, NULL, keyData, ep, hnsw_ef_search, 0, index, procinfos, collations, m, false, NULL, false);
}
/*
@@ -83,7 +84,7 @@ GetScanValue(IndexScanDesc scan)
/* Fine if normalization fails */
if (so->normprocinfo != NULL)
HnswNormValue(so->normprocinfo, so->collation, &value, NULL);
HnswNormValue(so->normprocinfo, so->collations[0], &value, NULL);
}
return value;
@@ -107,9 +108,9 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
ALLOCSET_DEFAULT_SIZES);
/* Set support functions */
so->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
so->procinfos = HnswInitProcinfos(index);
so->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
so->collation = index->rd_indcollation[0];
so->collations = index->rd_indcollation;
scan->opaque = so;
@@ -206,6 +207,9 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
scan->xs_ctup.t_self = *heaptid;
#endif
/* TODO Check during scan */
scan->xs_recheck = scan->numberOfKeys > 0;
scan->xs_recheckorderby = false;
return true;
}
@@ -222,6 +226,7 @@ hnswendscan(IndexScanDesc scan)
{
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
pfree(so->procinfos);
MemoryContextDelete(so->tmpCtx);
pfree(so);

View File

@@ -4,8 +4,13 @@
#include "hnsw.h"
#include "storage/bufmgr.h"
#include "utils/datum.h"
#include "vector.h"
#if PG_VERSION_NUM < 130000
#define TYPSTORAGE_PLAIN 'p'
#endif
/*
* Get the max number of connections in an upper layer for each element in the index
*/
@@ -46,6 +51,22 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
return index_getprocinfo(index, 1, procnum);
}
/*
* Init procs
*/
FmgrInfo **
HnswInitProcinfos(Relation index)
{
int keyAttributes = IndexRelationGetNumberOfKeyAttributes(index);
FmgrInfo **procinfos = palloc(keyAttributes * sizeof(FmgrInfo *));
procinfos[0] = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
for (int i = 1; i < keyAttributes; i++)
procinfos[i] = index_getprocinfo(index, i + 1, HNSW_ATTRIBUTE_DISTANCE_PROC);
return procinfos;
}
/*
* Divide by the norm
*
@@ -173,6 +194,7 @@ HnswInitElement(ItemPointer heaptid, int m, double ml, int maxLevel)
element->level = level;
element->deleted = 0;
element->itup = NULL;
HnswInitNeighbors(element, m);
@@ -187,7 +209,8 @@ HnswFreeElement(HnswElement element)
{
HnswFreeNeighbors(element);
list_free_deep(element->heaptids);
pfree(element->vec);
if (element->itup)
pfree(element->itup);
pfree(element);
}
@@ -214,7 +237,8 @@ HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno)
element->blkno = blkno;
element->offno = offno;
element->neighbors = NULL;
element->vec = NULL;
element->value = PointerGetDatum(NULL);
element->itup = NULL;
return element;
}
@@ -312,7 +336,7 @@ HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, Bloc
* Set element tuple, except for neighbor info
*/
void
HnswSetElementTuple(HnswElementTuple etup, HnswElement element)
HnswSetElementTuple(HnswElementTuple etup, HnswElement element, bool useIndexTuple)
{
etup->type = HNSW_ELEMENT_TUPLE_TYPE;
etup->level = element->level;
@@ -324,7 +348,11 @@ HnswSetElementTuple(HnswElementTuple etup, HnswElement element)
else
ItemPointerSetInvalid(&etup->heaptids[i]);
}
memcpy(&etup->vec, element->vec, VECTOR_SIZE(element->vec->dim));
if (useIndexTuple)
memcpy(&etup->data, element->itup, IndexTupleSize(element->itup));
else
memcpy(&etup->data, DatumGetPointer(element->value), VARSIZE_ANY(DatumGetPointer(element->value)));
}
/*
@@ -425,7 +453,7 @@ HnswLoadNeighbors(HnswElement element, Relation index, int m)
* Load an element from a tuple
*/
void
HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec)
HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec, Relation index)
{
element->level = etup->level;
element->deleted = etup->deleted;
@@ -447,16 +475,149 @@ HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHe
if (loadVec)
{
element->vec = palloc(VECTOR_SIZE(etup->vec.dim));
memcpy(element->vec, &etup->vec, VECTOR_SIZE(etup->vec.dim));
if (IndexRelationGetNumberOfAttributes(index) > 1)
{
TupleDesc tupdesc = RelationGetDescr(index);
bool unused;
element->itup = CopyIndexTuple((IndexTuple) &etup->data);
element->value = index_getattr(element->itup, 1, tupdesc, &unused);
}
else
{
Vector *vec = palloc(VARSIZE_ANY(&etup->data));
memcpy(vec, &etup->data, VARSIZE_ANY(&etup->data));
element->value = PointerGetDatum(vec);
}
}
}
/*
* Get the tuple descriptor
*/
static TupleDesc
HnswTupleDesc(Relation index)
{
TupleDesc tupdesc = CreateTupleDescCopyConstr(RelationGetDescr(index));
/* Prevent compression */
TupleDescAttr(tupdesc, 0)->attstorage = TYPSTORAGE_PLAIN;
return tupdesc;
}
/*
* Set element data
*/
void
HnswElementSetData(HnswElement element, Relation index, Datum value, Datum *values, bool *isnull)
{
/* TODO Create once per index build */
TupleDesc tupdesc = HnswTupleDesc(index);
bool unused;
Datum tmp;
tmp = values[0];
values[0] = value;
element->itup = index_form_tuple(tupdesc, values, isnull);
values[0] = tmp;
element->value = index_getattr(element->itup, 1, tupdesc, &unused);
FreeTupleDesc(tupdesc);
}
/*
* Get the attribute distance
*/
static inline double
AttributeDistance(double e)
{
/* TODO Better bias */
/* must be >> max(w * g) + 1 / log10(2) */
double bias = 4.32;
return e > 0 ? bias - 1.0 / log10(e + 1) : 0;
}
/*
* Get the distance
*/
static double
GetDistance(IndexTuple itup, Datum vec, Datum q, IndexTuple qtup, ScanKeyData *keyData, Relation index, FmgrInfo **procinfos, Oid *collations)
{
double g = DatumGetFloat8(FunctionCall2Coll(procinfos[0], collations[0], q, vec));
if (IndexRelationGetNumberOfKeyAttributes(index) > 1)
{
double w = 0.25;
double e = 0.0;
TupleDesc tupdesc = RelationGetDescr(index);
if (keyData)
{
/* TODO need to pass length of key data */
int keyCount = 1;
for (int i = 0; i < keyCount; i++)
{
ScanKey key = &keyData[i];
bool isnull;
Datum value = index_getattr(itup, key->sk_attno, tupdesc, &isnull);
bool attnull = key->sk_flags & SK_ISNULL;
if (isnull || attnull)
{
if (isnull != attnull)
e += 1000;
}
else if (!DatumGetBool(FunctionCall2Coll(&key->sk_func, key->sk_collation, value, key->sk_argument)))
{
double ei = fabs(DatumGetFloat8(FunctionCall2Coll(procinfos[key->sk_attno - 1], collations[key->sk_attno - 1], value, key->sk_argument)));
if (ei > 0)
e += ei;
else
/* Distance is zero for inequality */
e += 1000;
}
}
return w * g + AttributeDistance(e);
}
else if (qtup)
{
int keyCount = IndexRelationGetNumberOfKeyAttributes(index) - 1;
for (int i = 0; i < keyCount; i++)
{
bool isnull;
bool attnull;
Datum value = index_getattr(itup, i + 2, tupdesc, &isnull);
Datum value2 = index_getattr(qtup, i + 2, tupdesc, &attnull);
if (isnull || attnull)
{
if (isnull != attnull)
e += 1000;
}
else
e += fabs(DatumGetFloat8(FunctionCall2Coll(procinfos[i + 1], collations[i + 1], value, value2)));
}
return w * g + AttributeDistance(e);
}
}
return g;
}
/*
* Load an element and optionally get its distance from q
*/
void
HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
HnswLoadElement(HnswElement element, float *distance, Datum *q, IndexTuple qtup, ScanKeyData *keyData, Relation index, FmgrInfo **procinfos, Oid *collations, bool loadVec)
{
Buffer buf;
Page page;
@@ -472,11 +633,27 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
Assert(HnswIsElementTuple(etup));
/* Load element */
HnswLoadElementFromTuple(element, etup, true, loadVec);
HnswLoadElementFromTuple(element, etup, true, loadVec, index);
/* Calculate distance */
if (distance != NULL)
*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->vec)));
{
IndexTuple itup = NULL;
Datum value;
if (IndexRelationGetNumberOfAttributes(index) > 1)
{
TupleDesc tupdesc = RelationGetDescr(index);
bool unused;
itup = (IndexTuple) &etup->data;
value = index_getattr(itup, 1, tupdesc, &unused);
}
else
value = PointerGetDatum(&etup->data);
*distance = GetDistance(itup, value, *q, qtup, keyData, index, procinfos, collations);
}
UnlockReleaseBuffer(buf);
}
@@ -485,24 +662,24 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
* Get the distance for a candidate
*/
static float
GetCandidateDistance(HnswCandidate * hc, Datum q, FmgrInfo *procinfo, Oid collation)
GetCandidateDistance(HnswCandidate * hc, Datum q, IndexTuple qtup, ScanKeyData *keyData, Relation index, FmgrInfo **procinfos, Oid *collations)
{
return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q, PointerGetDatum(hc->element->vec)));
return GetDistance(hc->element->itup, hc->element->value, q, qtup, keyData, index, procinfos, collations);
}
/*
* Create a candidate for the entry point
*/
HnswCandidate *
HnswEntryCandidate(HnswElement entryPoint, Datum q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
HnswEntryCandidate(HnswElement entryPoint, Datum q, IndexTuple qtup, ScanKeyData *keyData, Relation index, FmgrInfo **procinfos, Oid *collations, bool loadVec, bool inMemory)
{
HnswCandidate *hc = palloc(sizeof(HnswCandidate));
hc->element = entryPoint;
if (index == NULL)
hc->distance = GetCandidateDistance(hc, q, procinfo, collation);
if (inMemory)
hc->distance = GetCandidateDistance(hc, q, qtup, keyData, index, procinfos, collations);
else
HnswLoadElement(hc->element, &hc->distance, &q, index, procinfo, collation, loadVec);
HnswLoadElement(hc->element, &hc->distance, &q, qtup, keyData, index, procinfos, collations, loadVec);
return hc;
}
@@ -552,9 +729,9 @@ CreatePairingHeapNode(HnswCandidate * c)
* Add to visited
*/
static inline void
AddToVisited(HTAB *v, HnswCandidate * hc, Relation index, bool *found)
AddToVisited(HTAB *v, HnswCandidate * hc, bool inMemory, bool *found)
{
if (index == NULL)
if (inMemory)
hash_search(v, &hc->element, HASH_ENTER, found);
else
{
@@ -569,7 +746,7 @@ AddToVisited(HTAB *v, HnswCandidate * hc, Relation index, bool *found)
* Algorithm 2 from paper
*/
List *
HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement)
HnswSearchLayer(Datum q, IndexTuple qtup, ScanKeyData *keyData, List *ep, int ef, int lc, Relation index, FmgrInfo **procinfos, Oid *collations, int m, bool loadVec, HnswElement skipElement, bool inMemory)
{
ListCell *lc2;
@@ -581,7 +758,7 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
HTAB *v;
/* Create hash table */
if (index == NULL)
if (inMemory)
{
hash_ctl.keysize = sizeof(HnswElement *);
hash_ctl.entrysize = sizeof(HnswElement *);
@@ -600,7 +777,7 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
{
HnswCandidate *hc = (HnswCandidate *) lfirst(lc2);
AddToVisited(v, hc, index, NULL);
AddToVisited(v, hc, inMemory, NULL);
pairingheap_add(C, &(CreatePairingHeapNode(hc)->ph_node));
pairingheap_add(W, &(CreatePairingHeapNode(hc)->ph_node));
@@ -634,7 +811,7 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
HnswCandidate *e = &neighborhood->items[i];
bool visited;
AddToVisited(v, e, index, &visited);
AddToVisited(v, e, inMemory, &visited);
if (!visited)
{
@@ -642,10 +819,10 @@ HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *pro
f = ((HnswPairingHeapNode *) pairingheap_first(W))->inner;
if (index == NULL)
eDistance = GetCandidateDistance(e, q, procinfo, collation);
if (inMemory)
eDistance = GetCandidateDistance(e, q, qtup, keyData, index, procinfos, collations);
else
HnswLoadElement(e->element, &eDistance, &q, index, procinfo, collation, inserting);
HnswLoadElement(e->element, &eDistance, &q, qtup, keyData, index, procinfos, collations, loadVec);
Assert(!e->element->deleted);
@@ -725,7 +902,7 @@ CompareCandidateDistances(const void *a, const void *b)
* Calculate the distance between elements
*/
static float
HnswGetDistance(HnswElement a, HnswElement b, int lc, FmgrInfo *procinfo, Oid collation)
HnswGetCachedDistance(HnswElement a, HnswElement b, int lc, Relation index, FmgrInfo **procinfos, Oid *collations)
{
/* Look for cached distance */
if (a->neighbors != NULL)
@@ -750,21 +927,21 @@ HnswGetDistance(HnswElement a, HnswElement b, int lc, FmgrInfo *procinfo, Oid co
}
}
return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(a->vec), PointerGetDatum(b->vec)));
return GetDistance(a->itup, a->value, b->value, b->itup, NULL, index, procinfos, collations);
}
/*
* Check if an element is closer to q than any element from R
*/
static bool
CheckElementCloser(HnswCandidate * e, List *r, int lc, FmgrInfo *procinfo, Oid collation)
CheckElementCloser(HnswCandidate * e, List *r, int lc, Relation index, FmgrInfo **procinfos, Oid *collations)
{
ListCell *lc2;
foreach(lc2, r)
{
HnswCandidate *ri = lfirst(lc2);
float distance = HnswGetDistance(e->element, ri->element, lc, procinfo, collation);
float distance = HnswGetCachedDistance(e->element, ri->element, lc, index, procinfos, collations);
if (distance <= e->distance)
return false;
@@ -777,7 +954,7 @@ CheckElementCloser(HnswCandidate * e, List *r, int lc, FmgrInfo *procinfo, Oid c
* Algorithm 4 from paper
*/
static List *
SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswElement e2, HnswCandidate * newCandidate, HnswCandidate * *pruned, bool sortCandidates)
SelectNeighbors(List *c, int m, int lc, Relation index, FmgrInfo **procinfos, Oid *collations, HnswElement e2, HnswCandidate * newCandidate, HnswCandidate * *pruned, bool sortCandidates)
{
List *r = NIL;
List *w = list_copy(c);
@@ -804,7 +981,7 @@ SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswE
/* Use previous state of r and wd to skip work when possible */
if (mustCalculate)
e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
e->closer = CheckElementCloser(e, r, lc, index, procinfos, collations);
else if (list_length(added) > 0)
{
/*
@@ -813,7 +990,7 @@ SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswE
*/
if (e->closer)
{
e->closer = CheckElementCloser(e, added, lc, procinfo, collation);
e->closer = CheckElementCloser(e, added, lc, index, procinfos, collations);
if (!e->closer)
removedAny = true;
@@ -826,7 +1003,7 @@ SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswE
*/
if (removedAny)
{
e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
e->closer = CheckElementCloser(e, r, lc, index, procinfos, collations);
if (e->closer)
added = lappend(added, e);
}
@@ -834,7 +1011,7 @@ SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswE
}
else if (e == newCandidate)
{
e->closer = CheckElementCloser(e, r, lc, procinfo, collation);
e->closer = CheckElementCloser(e, r, lc, index, procinfos, collations);
if (e->closer)
added = lappend(added, e);
}
@@ -868,16 +1045,20 @@ SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswE
* Find duplicate element
*/
HnswElement
HnswFindDuplicate(HnswElement e)
HnswFindDuplicate(HnswElement e, Relation index)
{
HnswNeighborArray *neighbors = &e->neighbors[0];
/* TODO Implement */
if (IndexRelationGetNumberOfAttributes(index) > 1)
return NULL;
for (int i = 0; i < neighbors->length; i++)
{
HnswCandidate *neighbor = &neighbors->items[i];
/* Exit early since ordered by distance */
if (vector_cmp_internal(e->vec, neighbor->element->vec) != 0)
if (!datumIsEqual(e->value, neighbor->element->value, false, -1))
break;
/* Check for space */
@@ -905,7 +1086,7 @@ AddConnections(HnswElement element, List *neighbors, int m, int lc)
* Update connections
*/
void
HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation)
HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int *updateIdx, Relation index, FmgrInfo **procinfos, Oid *collations, bool inMemory)
{
HnswNeighborArray *currentNeighbors = &hc->element->neighbors[lc];
@@ -928,18 +1109,20 @@ HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int
HnswCandidate *pruned = NULL;
/* Load elements on insert */
if (index != NULL)
if (!inMemory)
{
Datum q = PointerGetDatum(hc->element->vec);
Datum q = hc->element->value;
IndexTuple qtup = hc->element->itup;
ScanKeyData *keyData = NULL;
for (int i = 0; i < currentNeighbors->length; i++)
{
HnswCandidate *hc3 = &currentNeighbors->items[i];
if (hc3->element->vec == NULL)
HnswLoadElement(hc3->element, &hc3->distance, &q, index, procinfo, collation, true);
if (DatumGetPointer(hc3->element->value) == NULL)
HnswLoadElement(hc3->element, &hc3->distance, &q, qtup, keyData, index, procinfos, collations, true);
else
hc3->distance = GetCandidateDistance(hc3, q, procinfo, collation);
hc3->distance = GetCandidateDistance(hc3, q, qtup, keyData, index, procinfos, collations);
/* Prune element if being deleted */
if (list_length(hc3->element->heaptids) == 0)
@@ -959,7 +1142,7 @@ HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int
c = lappend(c, &currentNeighbors->items[i]);
c = lappend(c, &hc2);
SelectNeighbors(c, m, lc, procinfo, collation, hc->element, &hc2, &pruned, true);
SelectNeighbors(c, m, lc, index, procinfos, collations, hc->element, &hc2, &pruned, true);
/* Should not happen */
if (pruned == NULL)
@@ -1011,13 +1194,15 @@ RemoveElements(List *w, HnswElement skipElement)
* Algorithm 1 from paper
*/
void
HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing)
HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo **procinfos, Oid *collations, int m, int efConstruction, bool existing, bool inMemory)
{
List *ep;
List *w;
int level = element->level;
int entryLevel;
Datum q = PointerGetDatum(element->vec);
Datum q = element->value;
IndexTuple qtup = element->itup;
ScanKeyData *keyData = NULL;
HnswElement skipElement = existing ? element : NULL;
/* No neighbors if no entry point */
@@ -1025,13 +1210,13 @@ HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, F
return;
/* Get entry point and level */
ep = list_make1(HnswEntryCandidate(entryPoint, q, index, procinfo, collation, true));
ep = list_make1(HnswEntryCandidate(entryPoint, q, qtup, keyData, index, procinfos, collations, true, inMemory));
entryLevel = entryPoint->level;
/* 1st phase: greedy search to insert level */
for (int lc = entryLevel; lc >= level + 1; lc--)
{
w = HnswSearchLayer(q, ep, 1, lc, index, procinfo, collation, m, true, skipElement);
w = HnswSearchLayer(q, qtup, keyData, ep, 1, lc, index, procinfos, collations, m, true, skipElement, inMemory);
ep = w;
}
@@ -1049,11 +1234,11 @@ HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, F
List *neighbors;
List *lw;
w = HnswSearchLayer(q, ep, efConstruction, lc, index, procinfo, collation, m, true, skipElement);
w = HnswSearchLayer(q, qtup, keyData, ep, efConstruction, lc, index, procinfos, collations, m, true, skipElement, inMemory);
/* Elements being deleted or skipped can help with search */
/* but should be removed before selecting neighbors */
if (index != NULL)
if (!inMemory)
lw = RemoveElements(w, skipElement);
else
lw = w;
@@ -1063,7 +1248,7 @@ HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, F
* sortCandidates to true for in-memory builds to enable closer
* caching, but there does not seem to be a difference in performance.
*/
neighbors = SelectNeighbors(lw, lm, lc, procinfo, collation, element, NULL, NULL, false);
neighbors = SelectNeighbors(lw, lm, lc, index, procinfos, collations, element, NULL, NULL, false);
AddConnections(element, neighbors, lm, lc);

View File

@@ -62,7 +62,8 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
/* Iterate over nodes */
for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
{
HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
ItemId itemid = PageGetItemId(page, offno);
HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, itemid);
int idx = 0;
bool itemUpdated = false;
@@ -93,7 +94,7 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
if (itemUpdated)
{
Size etupSize = HNSW_ELEMENT_TUPLE_SIZE(etup->vec.dim);
Size etupSize = ItemIdGetLength(itemid);
/* Mark rest as invalid */
for (int i = idx; i < HNSW_HEAPTIDS; i++)
@@ -194,8 +195,8 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
GenericXLogState *state;
int m = vacuumstate->m;
int efConstruction = vacuumstate->efConstruction;
FmgrInfo *procinfo = vacuumstate->procinfo;
Oid collation = vacuumstate->collation;
FmgrInfo **procinfos = vacuumstate->procinfos;
Oid *collations = vacuumstate->collations;
BufferAccessStrategy bas = vacuumstate->bas;
HnswNeighborTuple ntup = vacuumstate->ntup;
Size ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, m);
@@ -209,7 +210,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
element->heaptids = NIL;
/* Add element to graph, skipping itself */
HnswInsertElement(element, entryPoint, index, procinfo, collation, m, efConstruction, true);
HnswInsertElement(element, entryPoint, index, procinfos, collations, m, efConstruction, true, false);
/* Update neighbor tuple */
/* Do this before getting page to minimize locking */
@@ -230,7 +231,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
UnlockReleaseBuffer(buf);
/* Update neighbors */
HnswUpdateNeighborPages(index, procinfo, collation, element, m, true);
HnswUpdateNeighborPages(index, procinfos, collations, element, m, true);
}
/*
@@ -257,7 +258,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
LockPage(index, HNSW_UPDATE_LOCK, ShareLock);
/* Load element */
HnswLoadElement(highestPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true);
HnswLoadElement(highestPoint, NULL, NULL, NULL, NULL, index, vacuumstate->procinfos, vacuumstate->collations, true);
/* Repair if needed */
if (NeedsUpdated(vacuumstate, highestPoint))
@@ -295,7 +296,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
* is outdated, this can remove connections at higher levels in
* the graph until they are repaired, but this should be fine.
*/
HnswLoadElement(entryPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true);
HnswLoadElement(entryPoint, NULL, NULL, NULL, NULL, index, vacuumstate->procinfos, vacuumstate->collations, true);
if (NeedsUpdated(vacuumstate, entryPoint))
{
@@ -371,7 +372,7 @@ RepairGraph(HnswVacuumState * vacuumstate)
/* Create an element */
element = HnswInitElementFromBlock(blkno, offno);
HnswLoadElementFromTuple(element, etup, false, true);
HnswLoadElementFromTuple(element, etup, false, true, index);
elements = lappend(elements, element);
}
@@ -441,6 +442,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
BlockNumber insertPage = InvalidBlockNumber;
Relation index = vacuumstate->index;
BufferAccessStrategy bas = vacuumstate->bas;
bool useIndexTuple = IndexRelationGetNumberOfAttributes(index);
/*
* Wait for index scans to complete. Scans before this point may contain
@@ -477,7 +479,8 @@ MarkDeleted(HnswVacuumState * vacuumstate)
/* Update element and neighbors together */
for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
{
HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
ItemId itemid = PageGetItemId(page, offno);
HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, itemid);
HnswNeighborTuple ntup;
Size etupSize;
Size ntupSize;
@@ -505,7 +508,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
continue;
/* Calculate sizes */
etupSize = HNSW_ELEMENT_TUPLE_SIZE(etup->vec.dim);
etupSize = ItemIdGetLength(itemid);
ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(etup->level, vacuumstate->m);
/* Get neighbor page */
@@ -528,7 +531,18 @@ MarkDeleted(HnswVacuumState * vacuumstate)
/* Overwrite element */
etup->deleted = 1;
MemSet(&etup->vec.x, 0, etup->vec.dim * sizeof(float));
if (useIndexTuple)
{
IndexTuple itup = (IndexTuple) &etup->data;
MemSet(itup, 0, IndexTupleSize(itup));
}
else
{
Vector *vec = (Vector *) (&etup->data);
MemSet(vec, 0, VARSIZE_ANY(vec));
}
/* Overwrite neighbors */
for (int i = 0; i < ntup->count; i++)
@@ -584,8 +598,8 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
vacuumstate->callback_state = callback_state;
vacuumstate->efConstruction = HnswGetEfConstruction(index);
vacuumstate->bas = GetAccessStrategy(BAS_BULKREAD);
vacuumstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
vacuumstate->collation = index->rd_indcollation[0];
vacuumstate->procinfos = HnswInitProcinfos(index);
vacuumstate->collations = index->rd_indcollation;
vacuumstate->ntup = palloc0(BLCKSZ);
vacuumstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
"Hnsw vacuum temporary context",
@@ -609,6 +623,7 @@ FreeVacuumState(HnswVacuumState * vacuumstate)
{
hash_destroy(vacuumstate->deleted);
FreeAccessStrategy(vacuumstate->bas);
pfree(vacuumstate->procinfos);
pfree(vacuumstate->ntup);
MemoryContextDelete(vacuumstate->tmpCtx);
}

View File

@@ -89,7 +89,7 @@ CheckDim(int dim)
}
/*
* Ensure finite elements
* Ensure finite element
*/
static inline void
CheckElement(float value)
@@ -437,17 +437,18 @@ vector_send(PG_FUNCTION_ARGS)
/*
* Convert vector to vector
* This is needed to check the type modifier
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector);
Datum
vector(PG_FUNCTION_ARGS)
{
Vector *arg = PG_GETARG_VECTOR_P(0);
Vector *vec = PG_GETARG_VECTOR_P(0);
int32 typmod = PG_GETARG_INT32(1);
CheckExpectedDim(typmod, arg->dim);
CheckExpectedDim(typmod, vec->dim);
PG_RETURN_POINTER(arg);
PG_RETURN_POINTER(vec);
}
/*
@@ -464,7 +465,6 @@ array_to_vector(PG_FUNCTION_ARGS)
bool typbyval;
char typalign;
Datum *elemsp;
bool *nullsp;
int nelemsp;
if (ARR_NDIM(array) > 1)
@@ -478,7 +478,7 @@ array_to_vector(PG_FUNCTION_ARGS)
errmsg("array must not contain nulls")));
get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, &nullsp, &nelemsp);
deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, NULL, &nelemsp);
CheckDim(nelemsp);
CheckExpectedDim(typmod, nelemsp);
@@ -512,6 +512,12 @@ array_to_vector(PG_FUNCTION_ARGS)
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);
/* Check elements */
for (int i = 0; i < result->dim; i++)
CheckElement(result->x[i]);

View File

@@ -0,0 +1,107 @@
use strict;
use warnings;
use PostgresNode;
use TestLib;
use Test::More;
my $node;
my @queries = ();
my @cs = ();
my @expected;
my $limit = 20;
my $dim = 3;
my $array_sql = join(",", ('random()') x $dim);
my $nc = 50;
sub test_recall
{
my ($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 WHERE c = $cs[0] ORDER BY v $operator '$queries[0]' LIMIT $limit;
));
like($explain, qr/Index Cond/);
for my $i (0 .. $#queries)
{
my $actual = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SELECT i FROM tst WHERE c = $cs[$i] ORDER BY v $operator '$queries[$i]' LIMIT $limit;
));
my @actual_ids = split("\n", $actual);
my %actual_set = map { $_ => 1 } @actual_ids;
my @expected_ids = split("\n", $expected[$i]);
foreach (@expected_ids)
{
if (exists($actual_set{$_}))
{
$correct++;
}
$total++;
}
}
cmp_ok($correct / $total, ">=", $min, $operator);
}
# Initialize node
$node = get_new_node('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int8);");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc FROM generate_series(1, 20000) i;"
);
# Generate queries
for (1 .. 20)
{
my @r = ();
for (1 .. $dim)
{
push(@r, rand());
}
push(@queries, "[" . join(",", @r) . "]");
push(@cs, int(rand() * $nc));
}
# Get exact results
@expected = ();
for my $i (0 .. $#queries)
{
my $res = $node->safe_psql("postgres", "SELECT i FROM tst WHERE c = $cs[$i] ORDER BY v <-> '$queries[$i]' LIMIT $limit;");
push(@expected, $res);
}
# Add index
$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, c);");
# Test recall
test_recall(0.99, '<->');
# Test vacuum
$node->safe_psql("postgres", "DELETE FROM tst WHERE c > 5;");
$node->safe_psql("postgres", "VACUUM tst;");
# Test columns
my ($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (c);");
like($stderr, qr/first column must be a vector/);
($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (c, v vector_l2_ops);");
like($stderr, qr/first column must be a vector/);
($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, c, c);");
like($stderr, qr/index cannot have more than two columns/);
($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, v vector_l2_ops);");
like($stderr, qr/column 2 cannot be a vector/);
done_testing();