Compare commits

..

64 Commits

Author SHA1 Message Date
Andrew Kane
ea689b92b2 Removed duplicated macros [skip ci] 2024-10-10 09:42:03 -07:00
Andrew Kane
51dad9ff2f Merge branch 'master' into hnsw-streaming 2024-10-10 09:41:35 -07:00
Andrew Kane
2cbd08b6c0 Moved unions and macros [skip ci] 2024-10-10 09:41:26 -07:00
Andrew Kane
d06e75f78c Moved union and macros [skip ci] 2024-10-10 09:39:04 -07:00
Andrew Kane
8e1a2715ea Removed assertions [skip ci] 2024-10-10 09:36:17 -07:00
Andrew Kane
25fae23fc3 Removed bench code [skip ci] 2024-10-10 09:35:46 -07:00
Andrew Kane
46d3164a30 Improved variable name [skip ci] 2024-10-10 02:14:01 -07:00
Andrew Kane
f2b2040306 Fixed assertion 2024-10-10 01:29:34 -07:00
Andrew Kane
c46f078e3c Merge branch 'master' into hnsw-streaming 2024-10-10 01:14:33 -07:00
Andrew Kane
fa6782985a Added HnswQuery struct for query data 2024-10-09 23:45:47 -07:00
Andrew Kane
32ab27d72a Added HnswSupport struct for support functions 2024-10-09 23:10:26 -07:00
Andrew Kane
064db12de7 Moved procinfo initialization for inserts [skip ci] 2024-10-09 21:59:21 -07:00
Andrew Kane
45a6eef9e0 Improved variable name [skip ci] 2024-10-09 21:52:10 -07:00
Andrew Kane
17266ed409 Use inMemory for conditionals 2024-10-09 21:49:32 -07:00
Andrew Kane
a98534e5ab DRY HNSW procinfo 2024-10-09 21:03:18 -07:00
Andrew Kane
57c05c59a2 DRY code for forming index value 2024-10-09 20:50:17 -07:00
Andrew Kane
3126fbdb6f Use double for distance [skip ci] 2024-10-09 17:04:25 -07:00
Andrew Kane
f4b67b078f DRY HNSW distance calculations 2024-10-09 17:01:49 -07:00
Andrew Kane
c0f6570c4a Debug updates [skip ci] 2024-10-08 22:58:09 -07:00
Andrew Kane
1d3d0f46ac Use DEBUG1 for exceeding work_mem [skip ci] 2024-10-08 22:52:12 -07:00
Andrew Kane
77688b4309 Improve total cost for cost estimation (#686) 2024-10-08 12:42:03 -07:00
Andrew Kane
d5f4a0e435 Fixed memory context leak in HnswUpdateNeighborsOnDisk - fixes #692 2024-10-08 12:21:26 -07:00
Andrew Kane
57248ba128 Use separate memory context for updating neighbors, which improves performance around 10% for larger vectors 2024-09-30 11:15:27 -07:00
Andrew Kane
ff6da4fcea Moved logic to get update neighbor on disk to separate function 2024-09-30 10:30:01 -07:00
Andrew Kane
a8b4b6675a Moved logic to get update index to separate function 2024-09-30 10:14:52 -07:00
Andrew Kane
d148b4e61b Fixed insert logic 2024-09-30 09:59:12 -07:00
Andrew Kane
658d74e2f6 Use Size for memory [skip ci] 2024-09-29 23:48:58 -07:00
Andrew Kane
7ba593c492 Improved SelectNeighbors signature [skip ci] 2024-09-29 23:03:02 -07:00
Andrew Kane
caabac33b8 Fixed compilation [skip ci] 2024-09-29 21:33:07 -07:00
Andrew Kane
525e3b81e1 Improved HnswUpdateConnection parameters [skip ci] 2024-09-29 19:47:25 -07:00
Andrew Kane
8eb8cdf0f3 Moved insert-specific code to hnswinsert.c 2024-09-29 19:44:11 -07:00
Andrew Kane
4c72f91206 Improved variable name [skip ci] 2024-09-29 19:26:15 -07:00
Andrew Kane
4ac86f62a1 Improved variable names [skip ci] 2024-09-29 19:22:35 -07:00
Andrew Kane
648dd8af78 Moved LoadElementsForInsert to separate function and removed unused code path 2024-09-29 19:12:38 -07:00
Andrew Kane
ee43ee9b16 Use HnswLoadNeighborTids for inserts 2024-09-29 18:52:12 -07:00
Andrew Kane
5ce367e18b Removed lc from HnswUpdateConnection [skip ci] 2024-09-29 18:18:42 -07:00
Andrew Kane
f371eb119b Removed lc from SelectNeighbors [skip ci] 2024-09-29 18:14:28 -07:00
Andrew Kane
382a25aefb Split loading neighbor TIDs into separate function [skip ci] 2024-09-29 17:20:54 -07:00
Andrew Kane
0b6214aad6 Moved HnswLoadNeighbors to hnswinsert.c [skip ci] 2024-09-29 15:49:01 -07:00
Andrew Kane
8b9333d468 Merge branch 'master' into hnsw-streaming 2024-09-29 15:13:51 -07:00
Andrew Kane
f2afd11257 Use sc for search candidates [skip ci] 2024-09-29 15:09:54 -07:00
Andrew Kane
cae3458329 Updated distance to use double 2024-09-29 15:06:50 -07:00
Andrew Kane
87f5e40495 Updated name [skip ci] 2024-09-29 13:46:05 -07:00
Andrew Kane
8798c1474c Updated naming [skip ci] 2024-09-29 13:45:25 -07:00
Andrew Kane
1b337ad97d Changed option to enum [skip ci] 2024-09-29 13:39:53 -07:00
Andrew Kane
0047630baf Updated readme [skip ci] 2024-09-29 10:37:07 -07:00
Andrew Kane
dc23752618 Fixed uninitialized variable [skip ci] 2024-09-28 19:18:52 -07:00
Andrew Kane
8b253359ab Improved code [skip ci] 2024-09-28 16:33:30 -07:00
Andrew Kane
8de5f55b0b Fixed assertion [skip ci] 2024-09-28 16:18:53 -07:00
Andrew Kane
351db562af Improved code [skip ci] 2024-09-28 16:07:31 -07:00
Andrew Kane
e1c2d03dba Improved cost estimation [skip ci] 2024-09-28 16:04:07 -07:00
Andrew Kane
ba8e29600b Added todo [skip ci] 2024-09-28 15:24:21 -07:00
Andrew Kane
158d9340bc Added distance filters to cost tests [skip ci] 2024-09-28 14:50:23 -07:00
Andrew Kane
49e05fb5ba Updated readme [skip ci] 2024-09-28 13:13:10 -07:00
Andrew Kane
9b42662188 Only adjust cost if scanning less than half of the tuples [skip ci] 2024-09-28 12:21:20 -07:00
Andrew Kane
7265927fd6 Updated readme [skip ci] 2024-09-28 12:07:09 -07:00
Andrew Kane
5ee0471ead Updated readme [skip ci] 2024-09-28 09:23:41 -07:00
Andrew Kane
54f8d9733d Updated default Postgres version in Dockerfile [skip ci] 2024-09-27 16:19:57 -07:00
Andrew Kane
cf419f448b Updated Postgres version for Docker [skip ci] 2024-09-27 16:19:06 -07:00
Andrew Kane
8a2eebd6a4 Added note about Postgres 17 on Windows - #669 [skip ci] 2024-09-27 14:05:36 -07:00
Andrew Kane
daf9c5c743 Updated package versions in readme [skip ci] 2024-09-27 13:52:11 -07:00
Andrew Kane
2bca4e406b Restored quarterly package version for FreeBSD in readme [skip ci] 2024-09-27 13:50:57 -07:00
Andrew Kane
74020a90da Updated package versions in readme [skip ci] 2024-09-27 13:49:43 -07:00
Andrew Kane
44d8d28b40 Added note about postgresql@17 formula [skip ci] 2024-09-27 13:39:54 -07:00
16 changed files with 592 additions and 535 deletions

View File

@@ -3,6 +3,7 @@
- Added support for iterative index scans
- Added casts for arrays to `sparsevec`
- Improved cost estimation
- Improved performance of HNSW inserts and on-disk index builds
- Reduced memory usage for HNSW index scans
- Dropped support for Postgres 12

View File

@@ -1,4 +1,4 @@
ARG PG_MAJOR=16
ARG PG_MAJOR=17
FROM postgres:$PG_MAJOR
ARG PG_MAJOR

View File

@@ -66,7 +66,7 @@ dist:
git archive --format zip --prefix=$(EXTENSION)-$(EXTVERSION)/ --output dist/$(EXTENSION)-$(EXTVERSION).zip master
# for Docker
PG_MAJOR ?= 16
PG_MAJOR ?= 17
.PHONY: docker

View File

@@ -52,6 +52,8 @@ nmake /F Makefile.win
nmake /F Makefile.win install
```
Note: Postgres 17 is not supported yet due to an upstream issue
See the [installation notes](#installation-notes---windows) if you run into issues
You can also install it with [Docker](#docker) or [conda-forge](#conda-forge).
@@ -100,6 +102,8 @@ Or add a vector column to an existing table
ALTER TABLE items ADD COLUMN embedding vector(3);
```
Also supports [half-precision](#half-precision-vectors), [binary](#binary-vectors), and [sparse](#sparse-vectors) vectors
Insert vectors
```sql
@@ -145,6 +149,8 @@ Supported distance functions are:
- `<#>` - (negative) inner product
- `<=>` - cosine distance
- `<+>` - L1 distance (added in 0.7.0)
- `<~>` - Hamming distance (binary vectors, added in 0.7.0)
- `<%>` - Jaccard distance (binary vectors, added in 0.7.0)
Get the nearest neighbors to a row
@@ -445,13 +451,13 @@ 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);
```
## Streaming Queries [unreleased]
## Iterative Search [unreleased]
*Added in 0.8.0*
With approximate indexes, you can end up with less results than expected due to filtering conditions in the query.
Starting with 0.8.0, you can enable streaming queries. If too few results from the initial index scan match the query filters, it will resume scanning until enough results are found. This can significantly improve recall (at the cost of speed).
Starting with 0.8.0, you can enable iterative search. If too few results from the initial index scan match the query filters, it will resume scanning until enough results are found. This can significantly improve recall (at the cost of speed).
```tsql
SET hnsw.streaming = on;
@@ -459,7 +465,27 @@ SET hnsw.streaming = on;
SET ivfflat.streaming = on;
```
### Streaming Options
However, there are some important caveats.
### Iterative Caveats
With iterative search, its possible for rows to be slightly out of order by distance. For strict ordering, use:
```sql
WITH approx_order AS MATERIALIZED (
SELECT *, embedding <-> '[1,2,3]' AS distance FROM items WHERE ... ORDER BY distance LIMIT 5
) SELECT * FROM approx_order ORDER BY distance;
```
For distance filters, use a CTE and place the filter outside it.
```sql
WITH approx_order AS MATERIALIZED (
SELECT *, embedding <-> '[1,2,3]' AS distance FROM items WHERE ... ORDER BY distance LIMIT 5
) SELECT * FROM approx_order WHERE distance < 0.1 ORDER BY distance;
```
### Iterative Options
Since scanning a large portion of the index is expensive, there are options to control when the scan ends.
@@ -471,14 +497,18 @@ Specify the max number of additional tuples visited
SET hnsw.ef_stream = 10000;
```
The scan will also end if reaches `work_mem`, at which point a notice is shown
The scan will also end if reaches `work_mem`. You can see when this happens by enabling debug messages.
```sql
SET client_min_messages = debug1;
```
```text
NOTICE: hnsw index scan exceeded work_mem after 50000 tuples
DEBUG: hnsw index scan exceeded work_mem after 10000 tuples
HINT: Increase work_mem to scan more tuples.
```
Adjust this with:
If the server has enough memory, you can adjust this with:
```sql
SET work_mem = '8MB';
@@ -492,16 +522,6 @@ Specify the max number of probes
SET ivfflat.max_probes = 100;
```
### Streaming Order
With streaming queries, its possible for rows to be slightly out of order by distance. For strict ordering, use:
```sql
WITH approx_order AS MATERIALIZED (
SELECT *, embedding <-> '[1,2,3]' AS distance FROM items WHERE ... ORDER BY distance LIMIT 5
) SELECT * FROM approx_order ORDER BY distance;
```
## Half-Precision Vectors
*Added in 0.7.0*
@@ -1040,7 +1060,7 @@ l2_normalize(sparsevec) → sparsevec | Normalize with Euclidean norm | 0.7.0
If your machine has multiple Postgres installations, specify the path to [pg_config](https://www.postgresql.org/docs/current/app-pgconfig.html) with:
```sh
export PG_CONFIG=/Library/PostgreSQL/16/bin/pg_config
export PG_CONFIG=/Library/PostgreSQL/17/bin/pg_config
```
Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
@@ -1051,11 +1071,11 @@ sudo --preserve-env=PG_CONFIG make install
A few common paths on Mac are:
- EDB installer - `/Library/PostgreSQL/16/bin/pg_config`
- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@16/bin/pg_config`
- Homebrew (x86-64) - `/usr/local/opt/postgresql@16/bin/pg_config`
- EDB installer - `/Library/PostgreSQL/17/bin/pg_config`
- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@17/bin/pg_config`
- Homebrew (x86-64) - `/usr/local/opt/postgresql@17/bin/pg_config`
Note: Replace `16` with your Postgres server version
Note: Replace `17` with your Postgres server version
### Missing Header
@@ -1064,10 +1084,10 @@ If compilation fails with `fatal error: postgres.h: No such file or directory`,
For Ubuntu and Debian, use:
```sh
sudo apt install postgresql-server-dev-16
sudo apt install postgresql-server-dev-17
```
Note: Replace `16` with your Postgres server version
Note: Replace `17` with your Postgres server version
### Missing SDK
@@ -1100,17 +1120,17 @@ If installation fails with `Access is denied`, re-run the installation instructi
Get the [Docker image](https://hub.docker.com/r/pgvector/pgvector) with:
```sh
docker pull pgvector/pgvector:pg16
docker pull pgvector/pgvector:pg17
```
This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `16` with your Postgres server version, and run it the same way).
This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `17` with your Postgres server version, and run it the same way).
You can also build the image manually:
```sh
git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
cd pgvector
docker build --pull --build-arg PG_MAJOR=16 -t myuser/pgvector .
docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector .
```
### Homebrew
@@ -1121,7 +1141,7 @@ With Homebrew Postgres, you can use:
brew install pgvector
```
Note: This only adds it to the `postgresql@14` formula
Note: This only adds it to the `postgresql@17` and `postgresql@14` formulas
### PGXN
@@ -1136,22 +1156,22 @@ pgxn install vector
Debian and Ubuntu packages are available from the [PostgreSQL APT Repository](https://wiki.postgresql.org/wiki/Apt). Follow the [setup instructions](https://wiki.postgresql.org/wiki/Apt#Quickstart) and run:
```sh
sudo apt install postgresql-16-pgvector
sudo apt install postgresql-17-pgvector
```
Note: Replace `16` with your Postgres server version
Note: Replace `17` with your Postgres server version
### Yum
RPM packages are available from the [PostgreSQL Yum Repository](https://yum.postgresql.org/). Follow the [setup instructions](https://www.postgresql.org/download/linux/redhat/) for your distribution and run:
```sh
sudo yum install pgvector_16
sudo yum install pgvector_17
# or
sudo dnf install pgvector_16
sudo dnf install pgvector_17
```
Note: Replace `16` with your Postgres server version
Note: Replace `17` with your Postgres server version
### pkg

View File

@@ -18,9 +18,18 @@
#define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
#endif
static const struct config_enum_entry hnsw_iterative_search_options[] = {
{"off", HNSW_ITERATIVE_SEARCH_OFF, false},
{"strict", HNSW_ITERATIVE_SEARCH_STRICT, false},
{"relaxed", HNSW_ITERATIVE_SEARCH_RELAXED, false},
/* TODO Change to strict before merging */
{"on", HNSW_ITERATIVE_SEARCH_RELAXED, false},
{NULL, 0, false}
};
int hnsw_ef_search;
int hnsw_ef_stream;
bool hnsw_streaming;
int hnsw_max_iterative_tuples;
int hnsw_iterative_search;
int hnsw_lock_tranche_id;
static relopt_kind hnsw_relopt_kind;
@@ -71,15 +80,15 @@ HnswInit(void)
"Valid range is 1..1000.", &hnsw_ef_search,
HNSW_DEFAULT_EF_SEARCH, HNSW_MIN_EF_SEARCH, HNSW_MAX_EF_SEARCH, PGC_USERSET, 0, NULL, NULL, NULL);
/* TODO Figure out name */
DefineCustomBoolVariable("hnsw.streaming", "Use streaming mode",
NULL, &hnsw_streaming,
HNSW_DEFAULT_STREAMING, PGC_USERSET, 0, NULL, NULL, NULL);
/* TODO Change name */
DefineCustomEnumVariable("hnsw.streaming", "Iterative search mode",
NULL, &hnsw_iterative_search,
HNSW_ITERATIVE_SEARCH_OFF, hnsw_iterative_search_options, PGC_USERSET, 0, NULL, NULL, NULL);
/* TODO Figure out name */
/* TODO Use same value as ivfflat.max_probes for "all" */
/* TODO Change name */
/* TODO Ensure ivfflat.max_probes uses same value for "all" */
DefineCustomIntVariable("hnsw.ef_stream", "Sets the max number of additional candidates to visit for streaming search",
"-1 means all", &hnsw_ef_stream,
"-1 means all", &hnsw_max_iterative_tuples,
HNSW_DEFAULT_EF_STREAM, HNSW_MIN_EF_STREAM, HNSW_MAX_EF_STREAM, PGC_USERSET, 0, NULL, NULL, NULL);
MarkGUCPrefixReserved("hnsw");
@@ -102,33 +111,6 @@ hnswbuildphasename(int64 phasenum)
}
}
/*
* Estimate ef needed for iterative scans
*/
static int
EstimateEf(PlannerInfo *root, IndexPath *path)
{
double selectivity = 1;
ListCell *lc;
/* Cannot estimate without limit */
/* limit_tuples includes offset */
if (root->limit_tuples < 0)
return 0;
/* Get the selectivity of non-index conditions */
foreach(lc, path->indexinfo->indrestrictinfo)
{
RestrictInfo *rinfo = lfirst(lc);
/* Skip DEFAULT_INEQ_SEL since it may be a distance filter */
if (rinfo->norm_selec >= 0 && rinfo->norm_selec <= 1 && rinfo->norm_selec != (Selectivity) DEFAULT_INEQ_SEL)
selectivity *= rinfo->norm_selec;
}
return root->limit_tuples / Max(selectivity, 0.00001);
}
/*
* Estimate the cost of an index scan
*/
@@ -140,11 +122,8 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
{
GenericCosts costs;
int m;
int ef;
int entryLevel;
int layer0TuplesMax;
double layer0Selectivity;
double scalingFactor = 0.55;
double ratio;
double startupPages;
double spc_seq_page_cost;
Relation index;
@@ -161,12 +140,12 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
MemSet(&costs, 0, sizeof(costs));
genericcostestimate(root, path, loop_count, &costs);
index = index_open(path->indexinfo->indexoid, NoLock);
HnswGetMetaPageInfo(index, &m, NULL);
index_close(index, NoLock);
ef = hnsw_streaming ? Max(hnsw_ef_search, EstimateEf(root, path)) : hnsw_ef_search;
/*
* HNSW cost estimation follows a formula that accounts for the total
* number of tuples indexed combined with the parameters that most
@@ -194,30 +173,38 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
* at L0, accounting for previously visited tuples, multiplied by the
* "scalingFactor" (currently hardcoded).
*/
entryLevel = (int) (log(path->indexinfo->tuples + 1) * HnswGetMl(m));
layer0TuplesMax = HnswGetLayerM(m, 0) * ef;
layer0Selectivity = (scalingFactor * log(path->indexinfo->tuples + 1)) /
(log(m) * (1 + log(ef)));
if (path->indexinfo->tuples > 0)
{
double scalingFactor = 0.55;
int entryLevel = (int) (log(path->indexinfo->tuples) * HnswGetMl(m));
int layer0TuplesMax = HnswGetLayerM(m, 0) * hnsw_ef_search;
double layer0Selectivity = scalingFactor * log(path->indexinfo->tuples) / (log(m) * (1 + log(hnsw_ef_search)));
costs.numIndexTuples = (entryLevel * m) +
(layer0TuplesMax * layer0Selectivity);
ratio = (entryLevel * m + layer0TuplesMax * layer0Selectivity) / path->indexinfo->tuples;
genericcostestimate(root, path, loop_count, &costs);
if (ratio > 1)
ratio = 1;
}
else
ratio = 1;
get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
/* Startup cost is cost before returning the first row */
costs.indexStartupCost = costs.indexTotalCost * ratio;
/* Adjust cost if needed since TOAST not included in seq scan cost */
if (costs.numIndexPages > path->indexinfo->rel->pages)
startupPages = costs.numIndexPages * ratio;
if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
{
/* Change all page cost from random to sequential */
costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
costs.indexStartupCost -= startupPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Remove cost of extra pages */
costs.indexTotalCost -= (costs.numIndexPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
costs.indexStartupCost -= (startupPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
}
/* Use total cost since most work happens before first tuple is returned */
*indexStartupCost = costs.indexTotalCost;
*indexStartupCost = costs.indexStartupCost;
*indexTotalCost = costs.indexTotalCost;
*indexSelectivity = costs.indexSelectivity;
*indexCorrelation = costs.indexCorrelation;

View File

@@ -12,10 +12,6 @@
#include "utils/sampling.h"
#include "vector.h"
#ifdef HNSW_BENCH
#include "portability/instr_time.h"
#endif
#define HNSW_MAX_DIM 2000
#define HNSW_MAX_NNZ 1000
@@ -46,7 +42,6 @@
#define HNSW_DEFAULT_EF_SEARCH 40
#define HNSW_MIN_EF_SEARCH 1
#define HNSW_MAX_EF_SEARCH 1000
#define HNSW_DEFAULT_STREAMING false
#define HNSW_DEFAULT_EF_STREAM -1
#define HNSW_MIN_EF_STREAM -1
#define HNSW_MAX_EF_STREAM INT_MAX
@@ -76,21 +71,6 @@
#define HnswPageGetOpaque(page) ((HnswPageOpaque) PageGetSpecialPointer(page))
#define HnswPageGetMeta(page) ((HnswMetaPageData *) PageGetContents(page))
#ifdef HNSW_BENCH
#define HnswBench(name, code) \
do { \
instr_time start; \
instr_time duration; \
INSTR_TIME_SET_CURRENT(start); \
(code); \
INSTR_TIME_SET_CURRENT(duration); \
INSTR_TIME_SUBTRACT(duration, start); \
elog(INFO, "%s: %.3f ms", name, INSTR_TIME_GET_MILLISEC(duration)); \
} while (0)
#else
#define HnswBench(name, code) (code)
#endif
#if PG_VERSION_NUM >= 150000
#define RandomDouble() pg_prng_double(&pg_global_prng_state)
#define SeedRandom(seed) pg_prng_seed(&pg_global_prng_state, seed)
@@ -111,6 +91,9 @@
/* Ensure fits on page and in uint8 */
#define HnswGetMaxLevel(m) Min(((BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - offsetof(HnswNeighborTupleData, indextids) - sizeof(ItemIdData)) / (sizeof(ItemPointerData)) / (m)) - 2, 255)
#define HnswGetSearchCandidate(membername, ptr) pairingheap_container(HnswSearchCandidate, membername, ptr)
#define HnswGetSearchCandidateConst(membername, ptr) pairingheap_const_container(HnswSearchCandidate, membername, ptr)
#define HnswGetValue(base, element) PointerGetDatum(HnswPtrAccess(base, (element)->value))
#if PG_VERSION_NUM < 140005
@@ -129,10 +112,17 @@
/* Variables */
extern int hnsw_ef_search;
extern int hnsw_ef_stream;
extern bool hnsw_streaming;
extern int hnsw_max_iterative_tuples;
extern int hnsw_iterative_search;
extern int hnsw_lock_tranche_id;
typedef enum HnswIterativeSearchType
{
HNSW_ITERATIVE_SEARCH_OFF,
HNSW_ITERATIVE_SEARCH_STRICT,
HNSW_ITERATIVE_SEARCH_RELAXED
} HnswIterativeSearchType;
typedef struct HnswElementData HnswElementData;
typedef struct HnswNeighborArray HnswNeighborArray;
@@ -186,12 +176,9 @@ typedef struct HnswSearchCandidate
pairingheap_node c_node;
pairingheap_node w_node;
HnswElementPtr element;
float distance;
double distance;
} HnswSearchCandidate;
#define HnswGetSearchCandidate(membername, ptr) pairingheap_container(HnswSearchCandidate, membername, ptr)
#define HnswGetSearchCandidateConst(membername, ptr) pairingheap_const_container(HnswSearchCandidate, membername, ptr)
/* HNSW index options */
typedef struct HnswOptions
{
@@ -214,8 +201,8 @@ typedef struct HnswGraph
/* Allocations state */
LWLock allocatorLock;
long memoryUsed;
long memoryTotal;
Size memoryUsed;
Size memoryTotal;
/* Flushed state */
LWLock flushLock;
@@ -266,6 +253,18 @@ typedef struct HnswTypeInfo
void (*checkValue) (Pointer v);
} HnswTypeInfo;
typedef struct HnswSupport
{
FmgrInfo *procinfo;
FmgrInfo *normprocinfo;
Oid collation;
} HnswSupport;
typedef struct HnswQuery
{
Datum value;
} HnswQuery;
typedef struct HnswBuildState
{
/* Info */
@@ -285,9 +284,7 @@ typedef struct HnswBuildState
double reltuples;
/* Support functions */
FmgrInfo *procinfo;
FmgrInfo *normprocinfo;
Oid collation;
HnswSupport support;
/* Variables */
HnswGraph graphData;
@@ -361,6 +358,12 @@ typedef union
struct tidhash_hash *tids;
} visited_hash;
typedef union
{
HnswElement element;
ItemPointerData indextid;
} HnswUnvisited;
typedef struct HnswScanOpaqueData
{
const HnswTypeInfo *typeInfo;
@@ -368,16 +371,14 @@ typedef struct HnswScanOpaqueData
List *w;
visited_hash v;
pairingheap *discarded;
Datum q;
HnswQuery q;
int m;
int64 tuples;
double previousDistance;
MemoryContext tmpCtx;
/* Support functions */
FmgrInfo *procinfo;
FmgrInfo *normprocinfo;
Oid collation;
HnswSupport support;
} HnswScanOpaqueData;
typedef HnswScanOpaqueData * HnswScanOpaque;
@@ -395,8 +396,7 @@ typedef struct HnswVacuumState
int efConstruction;
/* Support functions */
FmgrInfo *procinfo;
Oid collation;
HnswSupport support;
/* Variables */
struct tidhash_hash *deleted;
@@ -412,30 +412,33 @@ typedef struct HnswVacuumState
int HnswGetM(Relation index);
int HnswGetEfConstruction(Relation index);
FmgrInfo *HnswOptionalProcInfo(Relation index, uint16 procnum);
void HnswInitSupport(HnswSupport * support, Relation index);
Datum HnswNormValue(const HnswTypeInfo * typeInfo, Oid collation, Datum value);
bool HnswCheckNorm(FmgrInfo *procinfo, Oid collation, Datum value);
bool HnswCheckNorm(HnswSupport * support, Datum value);
Buffer HnswNewBuffer(Relation index, ForkNumber forkNum);
void HnswInitPage(Buffer buf, Page page);
void HnswInit(void);
List *HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples);
List *HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples);
HnswElement HnswGetEntryPoint(Relation index);
void HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
void *HnswAlloc(HnswAllocator * allocator, Size size);
HnswElement HnswInitElement(char *base, ItemPointer tid, int m, double ml, int maxLevel, HnswAllocator * alloc);
HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
void HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing);
HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
void HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing);
HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, HnswQuery * q, Relation rel, HnswSupport * support, bool loadVec);
void HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building);
void HnswSetNeighborTuple(char *base, HnswNeighborTuple ntup, HnswElement e, int m);
void HnswAddHeapTid(HnswElement element, ItemPointer heaptid);
HnswNeighborArray *HnswInitNeighborArray(int lm, HnswAllocator * allocator);
void HnswInitNeighbors(char *base, HnswElement element, int m, HnswAllocator * alloc);
bool HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull, ItemPointer heap_tid, bool building);
void HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building);
bool HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPointer heaptid, bool building);
void HnswUpdateNeighborsOnDisk(Relation index, HnswSupport * support, HnswElement e, int m, bool checkExisting, bool building);
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, float *maxDistance);
void HnswLoadElement(HnswElement element, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance);
bool HnswFormIndexValue(Datum *out, Datum *values, bool *isnull, const HnswTypeInfo * typeInfo, HnswSupport * support);
void HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element);
void HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
void HnswLoadNeighbors(HnswElement element, Relation index, int m);
void HnswUpdateConnection(char *base, HnswNeighborArray * neighbors, HnswElement newElement, float distance, int lm, int *updateIdx, Relation index, HnswSupport * support);
bool HnswLoadNeighborTids(HnswElement element, ItemPointerData *indextids, Relation index, int m, int lm, int lc);
void HnswInitLockTranche(void);
const HnswTypeInfo *HnswGetTypeInfo(Relation index);
PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);

View File

@@ -366,7 +366,7 @@ AddElementInMemory(char *base, HnswGraph * graph, HnswElement element)
* Update neighbors
*/
static void
UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswElement e, int m)
UpdateNeighborsInMemory(char *base, HnswSupport * support, HnswElement e, int m)
{
for (int lc = e->level; lc >= 0; lc--)
{
@@ -388,7 +388,7 @@ UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswEleme
Assert(neighborElement);
LWLockAcquire(&neighborElement->lock, LW_EXCLUSIVE);
HnswUpdateConnection(base, e, hc, lm, lc, NULL, NULL, procinfo, collation);
HnswUpdateConnection(base, HnswGetNeighbors(base, neighborElement, lc), e, hc->distance, lm, NULL, NULL, support);
LWLockRelease(&neighborElement->lock);
}
}
@@ -398,7 +398,7 @@ UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswEleme
* Update graph in memory
*/
static void
UpdateGraphInMemory(FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, HnswBuildState * buildstate)
UpdateGraphInMemory(HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, HnswBuildState * buildstate)
{
HnswGraph *graph = buildstate->graph;
char *base = buildstate->hnswarea;
@@ -411,7 +411,7 @@ UpdateGraphInMemory(FmgrInfo *procinfo, Oid collation, HnswElement element, int
AddElementInMemory(base, graph, element);
/* Update neighbors */
UpdateNeighborsInMemory(base, procinfo, collation, element, m);
UpdateNeighborsInMemory(base, support, element, m);
/* Update entry point if needed (already have lock) */
if (entryPoint == NULL || element->level > entryPoint->level)
@@ -424,9 +424,8 @@ UpdateGraphInMemory(FmgrInfo *procinfo, Oid collation, HnswElement element, int
static void
InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
{
FmgrInfo *procinfo = buildstate->procinfo;
Oid collation = buildstate->collation;
HnswGraph *graph = buildstate->graph;
HnswSupport *support = &buildstate->support;
HnswElement entryPoint;
LWLock *entryLock = &graph->entryLock;
LWLock *entryWaitLock = &graph->entryWaitLock;
@@ -458,10 +457,10 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
}
/* Find neighbors for element */
HnswFindElementNeighbors(base, element, entryPoint, NULL, procinfo, collation, m, efConstruction, false);
HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false);
/* Update graph in memory */
UpdateGraphInMemory(procinfo, collation, element, m, efConstruction, entryPoint, buildstate);
UpdateGraphInMemory(support, element, m, efConstruction, entryPoint, buildstate);
/* Release entry lock */
LWLockRelease(entryLock);
@@ -473,30 +472,19 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
static bool
InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, HnswBuildState * buildstate)
{
const HnswTypeInfo *typeInfo = buildstate->typeInfo;
HnswGraph *graph = buildstate->graph;
HnswElement element;
HnswAllocator *allocator = &buildstate->allocator;
HnswSupport *support = &buildstate->support;
Size valueSize;
Pointer valuePtr;
LWLock *flushLock = &graph->flushLock;
char *base = buildstate->hnswarea;
Datum value;
/* Detoast once for all calls */
Datum value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
/* Check value */
if (typeInfo->checkValue != NULL)
typeInfo->checkValue(DatumGetPointer(value));
/* Normalize if needed */
if (buildstate->normprocinfo != NULL)
{
if (!HnswCheckNorm(buildstate->normprocinfo, buildstate->collation, value))
return false;
value = HnswNormValue(typeInfo, buildstate->collation, value);
}
/* Form index value */
if (!HnswFormIndexValue(&value, values, isnull, buildstate->typeInfo, support))
return false;
/* Get datum size */
valueSize = VARSIZE_ANY(DatumGetPointer(value));
@@ -509,7 +497,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
{
LWLockRelease(flushLock);
return HnswInsertTupleOnDisk(index, value, values, isnull, heaptid, true);
return HnswInsertTupleOnDisk(index, support, value, heaptid, true);
}
/*
@@ -541,7 +529,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
LWLockRelease(flushLock);
return HnswInsertTupleOnDisk(index, value, values, isnull, heaptid, true);
return HnswInsertTupleOnDisk(index, support, value, heaptid, true);
}
/* Ok, we can proceed to allocate the element */
@@ -607,7 +595,7 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
* Initialize the graph
*/
static void
InitGraph(HnswGraph * graph, char *base, long memoryTotal)
InitGraph(HnswGraph * graph, char *base, Size memoryTotal)
{
/* Initialize the lock tranche if needed */
HnswInitLockTranche();
@@ -704,11 +692,9 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
buildstate->indtuples = 0;
/* Get support functions */
buildstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
buildstate->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
buildstate->collation = index->rd_indcollation[0];
HnswInitSupport(&buildstate->support, index);
InitGraph(&buildstate->graphData, NULL, maintenance_work_mem * 1024L);
InitGraph(&buildstate->graphData, NULL, (Size) maintenance_work_mem * 1024L);
buildstate->graph = &buildstate->graphData;
buildstate->ml = HnswGetMl(buildstate->m);
buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);

View File

@@ -340,6 +340,107 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
*updatedInsertPage = newInsertPage;
}
/*
* Load neighbors
*/
static HnswNeighborArray *
HnswLoadNeighbors(HnswElement element, Relation index, int m, int lm, int lc)
{
char *base = NULL;
HnswNeighborArray *neighbors = HnswInitNeighborArray(lm, NULL);
ItemPointerData indextids[HNSW_MAX_M * 2];
if (!HnswLoadNeighborTids(element, indextids, index, m, lm, lc))
return neighbors;
for (int i = 0; i < lm; i++)
{
ItemPointer indextid = &indextids[i];
HnswElement e;
HnswCandidate *hc;
if (!ItemPointerIsValid(indextid))
break;
e = HnswInitElementFromBlock(ItemPointerGetBlockNumber(indextid), ItemPointerGetOffsetNumber(indextid));
hc = &neighbors->items[neighbors->length++];
HnswPtrStore(base, hc->element, e);
}
return neighbors;
}
/*
* Load elements for insert
*/
static void
LoadElementsForInsert(HnswNeighborArray * neighbors, HnswQuery * q, int *idx, Relation index, HnswSupport * support)
{
char *base = NULL;
for (int i = 0; i < neighbors->length; i++)
{
HnswCandidate *hc = &neighbors->items[i];
HnswElement element = HnswPtrAccess(base, hc->element);
double distance;
HnswLoadElement(element, &distance, q, index, support, true, NULL);
hc->distance = distance;
/* Prune element if being deleted */
if (element->heaptidsLength == 0)
{
*idx = i;
break;
}
}
}
/*
* Get update index
*/
static int
GetUpdateIndex(HnswElement element, HnswElement newElement, float distance, int m, int lm, int lc, Relation index, HnswSupport * support, MemoryContext updateCtx)
{
char *base = NULL;
int idx = -1;
HnswNeighborArray *neighbors;
MemoryContext oldCtx = MemoryContextSwitchTo(updateCtx);
/*
* Get latest neighbors since they may have changed. Do not lock yet since
* selecting neighbors can take time. Could use optimistic locking to
* retry if another update occurs before getting exclusive lock.
*/
neighbors = HnswLoadNeighbors(element, index, m, lm, lc);
/*
* Could improve performance for vacuuming by checking neighbors against
* list of elements being deleted to find index. It's important to exclude
* already deleted elements for this since they can be replaced at any
* time.
*/
if (neighbors->length < lm)
idx = -2;
else
{
HnswQuery q;
q.value = HnswGetValue(base, element);
LoadElementsForInsert(neighbors, &q, &idx, index, support);
if (idx == -1)
HnswUpdateConnection(base, neighbors, newElement, distance, lm, &idx, index, support);
}
MemoryContextSwitchTo(oldCtx);
MemoryContextReset(updateCtx);
return idx;
}
/*
* Check if connection already exists
*/
@@ -360,14 +461,94 @@ ConnectionExists(HnswElement e, HnswNeighborTuple ntup, int startIdx, int lm)
return false;
}
/*
* Update neighbor
*/
static void
UpdateNeighborOnDisk(HnswElement element, HnswElement newElement, int idx, int m, int lm, int lc, Relation index, bool checkExisting, bool building)
{
Buffer buf;
Page page;
GenericXLogState *state;
HnswNeighborTuple ntup;
int startIdx;
OffsetNumber offno = element->neighborOffno;
/* Register page */
buf = ReadBuffer(index, element->neighborPage);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
if (building)
{
state = NULL;
page = BufferGetPage(buf);
}
else
{
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
}
/* Get tuple */
ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, offno));
/* Calculate index for update */
startIdx = (element->level - lc) * m;
/* Check for existing connection */
if (checkExisting && ConnectionExists(newElement, ntup, startIdx, lm))
idx = -1;
else if (idx == -2)
{
/* Find free offset if still exists */
/* TODO Retry updating connections if not */
for (int j = 0; j < lm; j++)
{
if (!ItemPointerIsValid(&ntup->indextids[startIdx + j]))
{
idx = startIdx + j;
break;
}
}
}
else
idx += startIdx;
/* Make robust to issues */
if (idx >= 0 && idx < ntup->count)
{
ItemPointer indextid = &ntup->indextids[idx];
/* Update neighbor on the buffer */
ItemPointerSet(indextid, newElement->blkno, newElement->offno);
/* Commit */
if (building)
MarkBufferDirty(buf);
else
GenericXLogFinish(state);
}
else if (!building)
GenericXLogAbort(state);
UnlockReleaseBuffer(buf);
}
/*
* Update neighbors
*/
void
HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building)
HnswUpdateNeighborsOnDisk(Relation index, HnswSupport * support, HnswElement e, int m, bool checkExisting, bool building)
{
char *base = NULL;
/* Use separate memory context to improve performance for larger vectors */
MemoryContext updateCtx = GenerationContextCreate(CurrentMemoryContext,
"Hnsw insert update context",
#if PG_VERSION_NUM >= 150000
128 * 1024, 128 * 1024,
#endif
128 * 1024);
for (int lc = e->level; lc >= 0; lc--)
{
int lm = HnswGetLayerM(m, lc);
@@ -376,96 +557,20 @@ HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, Hns
for (int i = 0; i < neighbors->length; i++)
{
HnswCandidate *hc = &neighbors->items[i];
Buffer buf;
Page page;
GenericXLogState *state;
HnswNeighborTuple ntup;
int idx = -1;
int startIdx;
HnswElement neighborElement = HnswPtrAccess(base, hc->element);
OffsetNumber offno = neighborElement->neighborOffno;
int idx;
/*
* Get latest neighbors since they may have changed. Do not lock
* yet since selecting neighbors can take time. Could use
* optimistic locking to retry if another update occurs before
* getting exclusive lock.
*/
HnswLoadNeighbors(neighborElement, index, m);
/*
* Could improve performance for vacuuming by checking neighbors
* against list of elements being deleted to find index. It's
* important to exclude already deleted elements for this since
* they can be replaced at any time.
*/
/* Select neighbors */
HnswUpdateConnection(NULL, e, hc, lm, lc, &idx, index, procinfo, collation);
idx = GetUpdateIndex(neighborElement, e, hc->distance, m, lm, lc, index, support, updateCtx);
/* New element was not selected as a neighbor */
if (idx == -1)
continue;
/* Register page */
buf = ReadBuffer(index, neighborElement->neighborPage);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
if (building)
{
state = NULL;
page = BufferGetPage(buf);
}
else
{
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
}
/* Get tuple */
ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, offno));
/* Calculate index for update */
startIdx = (neighborElement->level - lc) * m;
/* Check for existing connection */
if (checkExisting && ConnectionExists(e, ntup, startIdx, lm))
idx = -1;
else if (idx == -2)
{
/* Find free offset if still exists */
/* TODO Retry updating connections if not */
for (int j = 0; j < lm; j++)
{
if (!ItemPointerIsValid(&ntup->indextids[startIdx + j]))
{
idx = startIdx + j;
break;
}
}
}
else
idx += startIdx;
/* Make robust to issues */
if (idx >= 0 && idx < ntup->count)
{
ItemPointer indextid = &ntup->indextids[idx];
/* Update neighbor on the buffer */
ItemPointerSet(indextid, e->blkno, e->offno);
/* Commit */
if (building)
MarkBufferDirty(buf);
else
GenericXLogFinish(state);
}
else if (!building)
GenericXLogAbort(state);
UnlockReleaseBuffer(buf);
UpdateNeighborOnDisk(neighborElement, e, idx, m, lm, lc, index, checkExisting, building);
}
}
MemoryContextDelete(updateCtx);
}
/*
@@ -555,7 +660,7 @@ FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
* Update graph on disk
*/
static void
UpdateGraphOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
{
BlockNumber newInsertPage = InvalidBlockNumber;
@@ -571,7 +676,7 @@ UpdateGraphOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement
HnswUpdateMetaPage(index, 0, NULL, newInsertPage, MAIN_FORKNUM, building);
/* Update neighbors */
HnswUpdateNeighborsOnDisk(index, procinfo, collation, element, m, false, building);
HnswUpdateNeighborsOnDisk(index, support, element, m, false, building);
/* Update entry point if needed */
if (entryPoint == NULL || element->level > entryPoint->level)
@@ -582,14 +687,12 @@ UpdateGraphOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement
* Insert a tuple into the index
*/
bool
HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull, ItemPointer heap_tid, bool building)
HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPointer heaptid, bool building)
{
HnswElement entryPoint;
HnswElement element;
int m;
int efConstruction = HnswGetEfConstruction(index);
FmgrInfo *procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
Oid collation = index->rd_indcollation[0];
LOCKMODE lockmode = ShareLock;
char *base = NULL;
@@ -604,7 +707,7 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
HnswGetMetaPageInfo(index, &m, &entryPoint);
/* Create an element */
element = HnswInitElement(base, heap_tid, m, HnswGetMl(m), HnswGetMaxLevel(m), NULL);
element = HnswInitElement(base, heaptid, m, HnswGetMl(m), HnswGetMaxLevel(m), NULL);
HnswPtrStore(base, element->value, DatumGetPointer(value));
/* Prevent concurrent inserts when likely updating entry point */
@@ -622,10 +725,10 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
}
/* Find neighbors for element */
HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, false);
HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false);
/* Update graph on disk */
UpdateGraphOnDisk(index, procinfo, collation, element, m, efConstruction, entryPoint, building);
UpdateGraphOnDisk(index, support, element, m, efConstruction, entryPoint, building);
/* Release lock */
UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);
@@ -637,31 +740,19 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
* Insert a tuple into the index
*/
static void
HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid)
HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid)
{
Datum value;
const HnswTypeInfo *typeInfo = HnswGetTypeInfo(index);
FmgrInfo *normprocinfo;
Oid collation = index->rd_indcollation[0];
HnswSupport support;
/* Detoast once for all calls */
value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
HnswInitSupport(&support, index);
/* Check value */
if (typeInfo->checkValue != NULL)
typeInfo->checkValue(DatumGetPointer(value));
/* Form index value */
if (!HnswFormIndexValue(&value, values, isnull, typeInfo, &support))
return;
/* Normalize if needed */
normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
if (normprocinfo != NULL)
{
if (!HnswCheckNorm(normprocinfo, collation, value))
return;
value = HnswNormValue(typeInfo, collation, value);
}
HnswInsertTupleOnDisk(index, value, values, isnull, heap_tid, false);
HnswInsertTupleOnDisk(index, &support, value, heaptid, false);
}
/*

View File

@@ -1,48 +1,47 @@
#include "postgres.h"
#include <float.h>
#include "access/relscan.h"
#include "hnsw.h"
#include "pgstat.h"
#include "storage/bufmgr.h"
#include "storage/lmgr.h"
#include "utils/float.h"
#include "utils/memutils.h"
/*
* Algorithm 5 from paper
*/
static List *
GetScanItems(IndexScanDesc scan, Datum q)
GetScanItems(IndexScanDesc scan, Datum value)
{
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
Relation index = scan->indexRelation;
FmgrInfo *procinfo = so->procinfo;
Oid collation = so->collation;
HnswSupport *support = &so->support;
List *ep;
List *w;
int m;
HnswElement entryPoint;
char *base = NULL;
HnswQuery *q = &so->q;
/* Get m and entry point */
HnswGetMetaPageInfo(index, &m, &entryPoint);
so->q = q;
q->value = value;
so->m = m;
if (entryPoint == NULL)
return NIL;
ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, procinfo, collation, false));
ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, support, false));
for (int lc = entryPoint->level; lc >= 1; lc--)
{
w = HnswSearchLayer(base, q, ep, 1, lc, index, procinfo, collation, m, false, NULL, NULL, NULL, true, NULL);
w = HnswSearchLayer(base, q, ep, 1, lc, index, support, m, false, NULL, NULL, NULL, true, NULL);
ep = w;
}
return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL, &so->v, hnsw_streaming ? &so->discarded : NULL, true, &so->tuples);
return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, &so->v, hnsw_iterative_search != HNSW_ITERATIVE_SEARCH_OFF ? &so->discarded : NULL, true, &so->tuples);
}
/*
@@ -53,8 +52,6 @@ ResumeScanItems(IndexScanDesc scan)
{
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
Relation index = scan->indexRelation;
FmgrInfo *procinfo = so->procinfo;
Oid collation = so->collation;
List *ep = NIL;
char *base = NULL;
int batch_size = hnsw_ef_search;
@@ -65,17 +62,17 @@ ResumeScanItems(IndexScanDesc scan)
/* Get next batch of candidates */
for (int i = 0; i < batch_size; i++)
{
HnswSearchCandidate *hc;
HnswSearchCandidate *sc;
if (pairingheap_is_empty(so->discarded))
break;
hc = HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded));
sc = HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded));
ep = lappend(ep, hc);
ep = lappend(ep, sc);
}
return HnswSearchLayer(base, so->q, ep, batch_size, 0, index, procinfo, collation, so->m, false, NULL, &so->v, &so->discarded, false, &so->tuples);
return HnswSearchLayer(base, &so->q, ep, batch_size, 0, index, &so->support, so->m, false, NULL, &so->v, &so->discarded, false, &so->tuples);
}
/*
@@ -98,8 +95,8 @@ GetScanValue(IndexScanDesc scan)
Assert(!VARATT_IS_EXTENDED(DatumGetPointer(value)));
/* Normalize if needed */
if (so->normprocinfo != NULL)
value = HnswNormValue(so->typeInfo, so->collation, value);
if (so->support.normprocinfo != NULL)
value = HnswNormValue(so->typeInfo, so->support.collation, value);
}
return value;
@@ -126,9 +123,7 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
ALLOCSET_DEFAULT_SIZES);
/* Set support functions */
so->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
so->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
so->collation = index->rd_indcollation[0];
HnswInitSupport(&so->support, index);
scan->opaque = so;
@@ -151,7 +146,7 @@ hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int no
so->first = true;
so->tuples = 0;
so->previousDistance = -INFINITY;
so->previousDistance = -get_float8_infinity();
MemoryContextReset(so->tmpCtx);
if (keys && scan->numberOfKeys > 0)
@@ -201,7 +196,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
*/
LockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
HnswBench("scan iteration", so->w = GetScanItems(scan, value));
so->w = GetScanItems(scan, value);
/* Release shared lock */
UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
@@ -216,13 +211,13 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
for (;;)
{
char *base = NULL;
HnswSearchCandidate *hc;
HnswSearchCandidate *sc;
HnswElement element;
ItemPointer heaptid;
if (list_length(so->w) == 0)
{
if (!hnsw_streaming)
if (hnsw_iterative_search == HNSW_ITERATIVE_SEARCH_OFF)
break;
/* Empty index */
@@ -230,7 +225,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
break;
/* Reached max number of additional tuples */
if (hnsw_ef_stream != -1 && so->tuples >= hnsw_ef_search + hnsw_ef_stream)
if (hnsw_max_iterative_tuples != -1 && so->tuples >= hnsw_ef_search + hnsw_max_iterative_tuples)
{
if (pairingheap_is_empty(so->discarded))
break;
@@ -243,7 +238,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
{
if (pairingheap_is_empty(so->discarded))
{
ereport(NOTICE,
ereport(DEBUG1,
(errmsg("hnsw index scan exceeded work_mem after " INT64_FORMAT " tuples", so->tuples),
errhint("Increase work_mem to scan more tuples.")));
@@ -266,7 +261,7 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
*/
LockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
HnswBench("scan iteration", so->w = ResumeScanItems(scan));
so->w = ResumeScanItems(scan);
UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
@@ -279,8 +274,8 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
break;
}
hc = llast(so->w);
element = HnswPtrAccess(base, hc->element);
sc = llast(so->w);
element = HnswPtrAccess(base, sc->element);
/* Move to next element if no valid heap TIDs */
if (element->heaptidsLength == 0)
@@ -288,10 +283,10 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
so->w = list_delete_last(so->w);
/* Mark memory as free for next iteration */
if (hnsw_streaming)
if (hnsw_iterative_search != HNSW_ITERATIVE_SEARCH_OFF)
{
pfree(element);
pfree(hc);
pfree(sc);
}
continue;
@@ -299,10 +294,13 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
heaptid = &element->heaptids[--element->heaptidsLength];
if (hc->distance < so->previousDistance)
continue;
if (hnsw_iterative_search == HNSW_ITERATIVE_SEARCH_STRICT)
{
if (sc->distance < so->previousDistance)
continue;
so->previousDistance = hc->distance;
so->previousDistance = sc->distance;
}
MemoryContextSwitchTo(oldCtx);

View File

@@ -100,12 +100,6 @@ hash_offset(Size offset)
#define SH_DEFINE
#include "lib/simplehash.h"
typedef union
{
HnswElement element;
ItemPointerData indextid;
} HnswUnvisited;
/*
* Get the max number of connections in an upper layer for each element in the index
*/
@@ -146,6 +140,17 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
return index_getprocinfo(index, 1, procnum);
}
/*
* Init support functions
*/
void
HnswInitSupport(HnswSupport * support, Relation index)
{
support->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
support->collation = index->rd_indcollation[0];
support->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
}
/*
* Normalize value
*/
@@ -159,9 +164,9 @@ HnswNormValue(const HnswTypeInfo * typeInfo, Oid collation, Datum value)
* Check if non-zero norm
*/
bool
HnswCheckNorm(FmgrInfo *procinfo, Oid collation, Datum value)
HnswCheckNorm(HnswSupport * support, Datum value)
{
return DatumGetFloat8(FunctionCall1Coll(procinfo, collation, value)) > 0;
return DatumGetFloat8(FunctionCall1Coll(support->normprocinfo, support->collation, value)) > 0;
}
/*
@@ -190,7 +195,7 @@ HnswInitPage(Buffer buf, Page page)
/*
* Allocate a neighbor array
*/
static HnswNeighborArray *
HnswNeighborArray *
HnswInitNeighborArray(int lm, HnswAllocator * allocator)
{
HnswNeighborArray *a = HnswAlloc(allocator, HNSW_NEIGHBOR_ARRAY_SIZE(lm));
@@ -389,6 +394,33 @@ HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, Bloc
UnlockReleaseBuffer(buf);
}
/*
* Form index value
*/
bool
HnswFormIndexValue(Datum *out, Datum *values, bool *isnull, const HnswTypeInfo * typeInfo, HnswSupport * support)
{
/* Detoast once for all calls */
Datum value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
/* Check value */
if (typeInfo->checkValue != NULL)
typeInfo->checkValue(DatumGetPointer(value));
/* Normalize if needed */
if (support->normprocinfo != NULL)
{
if (!HnswCheckNorm(support, value))
return false;
value = HnswNormValue(typeInfo, support->collation, value);
}
*out = value;
return true;
}
/*
* Set element tuple, except for neighbor info
*/
@@ -446,69 +478,6 @@ HnswSetNeighborTuple(char *base, HnswNeighborTuple ntup, HnswElement e, int m)
ntup->version = e->version;
}
/*
* Load neighbors from page
*/
static void
LoadNeighborsFromPage(HnswElement element, Relation index, Page page, int m)
{
char *base = NULL;
HnswNeighborTuple ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, element->neighborOffno));
int neighborCount = (element->level + 2) * m;
Assert(HnswIsNeighborTuple(ntup));
HnswInitNeighbors(base, element, m, NULL);
/* Ensure expected neighbors */
if (ntup->count != neighborCount)
return;
for (int i = 0; i < neighborCount; i++)
{
HnswElement e;
int level;
HnswCandidate *hc;
ItemPointer indextid;
HnswNeighborArray *neighbors;
indextid = &ntup->indextids[i];
if (!ItemPointerIsValid(indextid))
continue;
e = HnswInitElementFromBlock(ItemPointerGetBlockNumber(indextid), ItemPointerGetOffsetNumber(indextid));
/* Calculate level based on offset */
level = element->level - i / m;
if (level < 0)
level = 0;
neighbors = HnswGetNeighbors(base, element, level);
hc = &neighbors->items[neighbors->length++];
HnswPtrStore(base, hc->element, e);
}
}
/*
* Load neighbors
*/
void
HnswLoadNeighbors(HnswElement element, Relation index, int m)
{
Buffer buf;
Page page;
buf = ReadBuffer(index, element->neighborPage);
LockBuffer(buf, BUFFER_LOCK_SHARE);
page = BufferGetPage(buf);
LoadNeighborsFromPage(element, index, page, m);
UnlockReleaseBuffer(buf);
}
/*
* Load an element from a tuple
*/
@@ -543,11 +512,20 @@ HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHe
}
}
/*
* Calculate the distance between values
*/
static inline double
HnswGetDistance(Datum a, Datum b, HnswSupport * support)
{
return DatumGetFloat8(FunctionCall2Coll(support->procinfo, support->collation, a, b));
}
/*
* Load an element and optionally get its distance from q
*/
static void
HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec, float *maxDistance, HnswElement * element)
HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance, HnswElement * element)
{
Buffer buf;
Page page;
@@ -565,10 +543,10 @@ HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, float *distance, Datu
/* Calculate distance */
if (distance != NULL)
{
if (DatumGetPointer(*q) == NULL)
if (DatumGetPointer(q->value) == NULL)
*distance = 0;
else
*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->data)));
*distance = HnswGetDistance(q->value, PointerGetDatum(&etup->data), support);
}
/* Load element */
@@ -587,36 +565,37 @@ HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, float *distance, Datu
* 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, float *maxDistance)
HnswLoadElement(HnswElement element, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance)
{
HnswLoadElementImpl(element->blkno, element->offno, distance, q, index, procinfo, collation, loadVec, maxDistance, &element);
HnswLoadElementImpl(element->blkno, element->offno, distance, q, index, support, loadVec, maxDistance, &element);
}
/*
* Get the distance for an element
*/
static float
GetElementDistance(char *base, HnswElement element, Datum q, FmgrInfo *procinfo, Oid collation)
static double
GetElementDistance(char *base, HnswElement element, HnswQuery * q, HnswSupport * support)
{
Datum value = HnswGetValue(base, element);
return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q, value));
return HnswGetDistance(q->value, value, support);
}
/*
* Create a candidate for the entry point
*/
HnswSearchCandidate *
HnswEntryCandidate(char *base, HnswElement entryPoint, Datum q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
HnswEntryCandidate(char *base, HnswElement entryPoint, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec)
{
HnswSearchCandidate *hc = palloc(sizeof(HnswSearchCandidate));
HnswSearchCandidate *sc = palloc(sizeof(HnswSearchCandidate));
bool inMemory = index == NULL;
HnswPtrStore(base, hc->element, entryPoint);
if (index == NULL)
hc->distance = GetElementDistance(base, entryPoint, q, procinfo, collation);
HnswPtrStore(base, sc->element, entryPoint);
if (inMemory)
sc->distance = GetElementDistance(base, entryPoint, q, support);
else
HnswLoadElement(entryPoint, &hc->distance, &q, index, procinfo, collation, loadVec, NULL);
return hc;
HnswLoadElement(entryPoint, &sc->distance, q, index, support, loadVec, NULL);
return sc;
}
/*
@@ -668,9 +647,9 @@ CompareFurthestCandidates(const pairingheap_node *a, const pairingheap_node *b,
* Init visited
*/
static inline void
InitVisited(char *base, visited_hash * v, Relation index, int ef, int m)
InitVisited(char *base, visited_hash * v, bool inMemory, int ef, int m)
{
if (index != NULL)
if (!inMemory)
v->tids = tidhash_create(CurrentMemoryContext, ef * m * 2, NULL);
else if (base != NULL)
v->offsets = offsethash_create(CurrentMemoryContext, ef * m * 2, NULL);
@@ -682,9 +661,9 @@ InitVisited(char *base, visited_hash * v, Relation index, int ef, int m)
* Add to visited
*/
static inline void
AddToVisited(char *base, visited_hash * v, HnswElementPtr elementPtr, Relation index, bool *found)
AddToVisited(char *base, visited_hash * v, HnswElementPtr elementPtr, bool inMemory, bool *found)
{
if (index != NULL)
if (!inMemory)
{
HnswElement element = HnswPtrAccess(base, elementPtr);
ItemPointerData indextid;
@@ -745,7 +724,7 @@ HnswLoadUnvisitedFromMemory(char *base, HnswElement element, HnswUnvisited * unv
HnswCandidate *hc = &localNeighborhood->items[i];
bool found;
AddToVisited(base, v, hc->element, NULL, &found);
AddToVisited(base, v, hc->element, true, &found);
if (!found)
unvisited[(*unvisitedLength)++].element = HnswPtrAccess(base, hc->element);
@@ -753,18 +732,15 @@ HnswLoadUnvisitedFromMemory(char *base, HnswElement element, HnswUnvisited * unv
}
/*
* Load unvisited neighbors from disk
* Load neighbor index TIDs
*/
static void
HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *unvisitedLength, visited_hash * v, Relation index, int m, int lm, int lc)
bool
HnswLoadNeighborTids(HnswElement element, ItemPointerData *indextids, Relation index, int m, int lm, int lc)
{
Buffer buf;
Page page;
HnswNeighborTuple ntup;
int start;
ItemPointerData indextids[HNSW_MAX_M * 2];
*unvisitedLength = 0;
buf = ReadBuffer(index, element->neighborPage);
LockBuffer(buf, BUFFER_LOCK_SHARE);
@@ -779,14 +755,29 @@ HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *u
if (ntup->version != element->version || ntup->count != (element->level + 2) * m)
{
UnlockReleaseBuffer(buf);
return;
return false;
}
/* Copy to minimize lock time */
start = (element->level - lc) * m;
memcpy(&indextids, ntup->indextids + start, lm * sizeof(ItemPointerData));
memcpy(indextids, ntup->indextids + start, lm * sizeof(ItemPointerData));
UnlockReleaseBuffer(buf);
return true;
}
/*
* Load unvisited neighbors from disk
*/
static void
HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *unvisitedLength, visited_hash * v, Relation index, int m, int lm, int lc)
{
ItemPointerData indextids[HNSW_MAX_M * 2];
*unvisitedLength = 0;
if (!HnswLoadNeighborTids(element, indextids, index, m, lm, lc))
return;
for (int i = 0; i < lm; i++)
{
@@ -807,7 +798,7 @@ HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *u
* Algorithm 2 from paper
*/
List *
HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples)
HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples)
{
List *w = NIL;
pairingheap *C = pairingheap_allocate(CompareNearestCandidates, NULL);
@@ -820,6 +811,7 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
int lm = HnswGetLayerM(m, lc);
HnswUnvisited *unvisited = palloc(lm * sizeof(HnswUnvisited));
int unvisitedLength;
bool inMemory = index == NULL;
if (v == NULL)
{
@@ -829,14 +821,14 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
if (initVisited)
{
InitVisited(base, v, index, ef, m);
InitVisited(base, v, inMemory, ef, m);
if (discarded != NULL)
*discarded = pairingheap_allocate(CompareNearestDiscardedCandidates, NULL);
}
/* Create local memory for neighborhood if needed */
if (index == NULL)
if (inMemory)
{
neighborhoodSize = HNSW_NEIGHBOR_ARRAY_SIZE(lm);
localNeighborhood = palloc(neighborhoodSize);
@@ -845,26 +837,26 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
/* Add entry points to v, C, and W */
foreach(lc2, ep)
{
HnswSearchCandidate *hc = (HnswSearchCandidate *) lfirst(lc2);
HnswSearchCandidate *sc = (HnswSearchCandidate *) lfirst(lc2);
bool found;
if (initVisited)
{
AddToVisited(base, v, hc->element, index, &found);
AddToVisited(base, v, sc->element, inMemory, &found);
if (tuples != NULL)
(*tuples)++;
}
pairingheap_add(C, &hc->c_node);
pairingheap_add(W, &hc->w_node);
pairingheap_add(C, &sc->c_node);
pairingheap_add(W, &sc->w_node);
/*
* Do not count elements being deleted towards ef when vacuuming. It
* would be ideal to do this for inserts as well, but this could
* affect insert performance.
*/
if (CountElement(skipElement, HnswPtrAccess(base, hc->element)))
if (CountElement(skipElement, HnswPtrAccess(base, sc->element)))
wlen++;
}
@@ -879,7 +871,7 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
cElement = HnswPtrAccess(base, c->element);
if (index == NULL)
if (inMemory)
HnswLoadUnvisitedFromMemory(base, cElement, unvisited, &unvisitedLength, v, lc, localNeighborhood, neighborhoodSize);
else
HnswLoadUnvisitedFromDisk(cElement, unvisited, &unvisitedLength, v, index, m, lm, lc);
@@ -891,15 +883,15 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
{
HnswElement eElement;
HnswSearchCandidate *e;
float eDistance;
double eDistance;
bool alwaysAdd = wlen < ef;
f = HnswGetSearchCandidate(w_node, pairingheap_first(W));
if (index == NULL)
if (inMemory)
{
eElement = unvisited[i].element;
eDistance = GetElementDistance(base, eElement, q, procinfo, collation);
eDistance = GetElementDistance(base, eElement, q, support);
}
else
{
@@ -909,7 +901,10 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
/* Avoid any allocations if not adding */
eElement = NULL;
HnswLoadElementImpl(blkno, offno, &eDistance, &q, index, procinfo, collation, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
HnswLoadElementImpl(blkno, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
if (eElement == NULL)
continue;
}
if (eElement == NULL || !(eDistance < f->distance || alwaysAdd))
@@ -961,9 +956,9 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
/* Add each element of W to w */
while (!pairingheap_is_empty(W))
{
HnswSearchCandidate *hc = HnswGetSearchCandidate(w_node, pairingheap_remove_first(W));
HnswSearchCandidate *sc = HnswGetSearchCandidate(w_node, pairingheap_remove_first(W));
w = lappend(w, hc);
w = lappend(w, sc);
}
return w;
@@ -1017,32 +1012,22 @@ CompareCandidateDistancesOffset(const ListCell *a, const ListCell *b)
return 0;
}
/*
* Calculate the distance between elements
*/
static float
HnswGetDistance(char *base, HnswElement a, HnswElement b, FmgrInfo *procinfo, Oid collation)
{
Datum aValue = HnswGetValue(base, a);
Datum bValue = HnswGetValue(base, b);
return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, aValue, bValue));
}
/*
* Check if an element is closer to q than any element from R
*/
static bool
CheckElementCloser(char *base, HnswCandidate * e, List *r, FmgrInfo *procinfo, Oid collation)
CheckElementCloser(char *base, HnswCandidate * e, List *r, HnswSupport * support)
{
HnswElement eElement = HnswPtrAccess(base, e->element);
Datum eValue = HnswGetValue(base, eElement);
ListCell *lc2;
foreach(lc2, r)
{
HnswCandidate *ri = lfirst(lc2);
HnswElement riElement = HnswPtrAccess(base, ri->element);
float distance = HnswGetDistance(base, eElement, riElement, procinfo, collation);
Datum riValue = HnswGetValue(base, riElement);
float distance = HnswGetDistance(eValue, riValue, support);
if (distance <= e->distance)
return false;
@@ -1055,15 +1040,14 @@ CheckElementCloser(char *base, HnswCandidate * e, List *r, FmgrInfo *procinfo, O
* Algorithm 4 from paper
*/
static List *
SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid collation, HnswElement e2, HnswCandidate * newCandidate, HnswCandidate * *pruned, bool sortCandidates)
SelectNeighbors(char *base, List *c, int lm, HnswSupport * support, bool *closerSet, HnswCandidate * newCandidate, HnswCandidate * *pruned, bool sortCandidates)
{
List *r = NIL;
List *w = list_copy(c);
HnswCandidate **wd;
int wdlen = 0;
int wdoff = 0;
HnswNeighborArray *neighbors = HnswGetNeighbors(base, e2, lc);
bool mustCalculate = !neighbors->closerSet;
bool mustCalculate = !(*closerSet);
List *added = NIL;
bool removedAny = false;
@@ -1090,7 +1074,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
/* Use previous state of r and wd to skip work when possible */
if (mustCalculate)
e->closer = CheckElementCloser(base, e, r, procinfo, collation);
e->closer = CheckElementCloser(base, e, r, support);
else if (list_length(added) > 0)
{
/* Keep Valgrind happy for in-memory, parallel builds */
@@ -1103,7 +1087,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
*/
if (e->closer)
{
e->closer = CheckElementCloser(base, e, added, procinfo, collation);
e->closer = CheckElementCloser(base, e, added, support);
if (!e->closer)
removedAny = true;
@@ -1116,7 +1100,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
*/
if (removedAny)
{
e->closer = CheckElementCloser(base, e, r, procinfo, collation);
e->closer = CheckElementCloser(base, e, r, support);
if (e->closer)
added = lappend(added, e);
}
@@ -1124,7 +1108,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
}
else if (e == newCandidate)
{
e->closer = CheckElementCloser(base, e, r, procinfo, collation);
e->closer = CheckElementCloser(base, e, r, support);
if (e->closer)
added = lappend(added, e);
}
@@ -1140,7 +1124,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
}
/* Cached value can only be used in future if sorted deterministically */
neighbors->closerSet = sortCandidates;
*closerSet = sortCandidates;
/* Keep pruned connections */
while (wdoff < wdlen && list_length(r) < lm)
@@ -1175,18 +1159,16 @@ AddConnections(char *base, HnswElement element, List *neighbors, int lc)
* Update connections
*/
void
HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation)
HnswUpdateConnection(char *base, HnswNeighborArray * neighbors, HnswElement newElement, float distance, int lm, int *updateIdx, Relation index, HnswSupport * support)
{
HnswElement hce = HnswPtrAccess(base, hc->element);
HnswNeighborArray *currentNeighbors = HnswGetNeighbors(base, hce, lc);
HnswCandidate hc2;
HnswCandidate newHc;
HnswPtrStore(base, hc2.element, element);
hc2.distance = hc->distance;
HnswPtrStore(base, newHc.element, newElement);
newHc.distance = distance;
if (currentNeighbors->length < lm)
if (neighbors->length < lm)
{
currentNeighbors->items[currentNeighbors->length++] = hc2;
neighbors->items[neighbors->length++] = newHc;
/* Track update */
if (updateIdx != NULL)
@@ -1195,54 +1177,26 @@ HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm
else
{
/* Shrink connections */
List *c = NIL;
HnswCandidate *pruned = NULL;
/* Load elements on insert */
if (index != NULL)
{
Datum q = HnswGetValue(base, hce);
/* Add candidates */
for (int i = 0; i < neighbors->length; i++)
c = lappend(c, &neighbors->items[i]);
c = lappend(c, &newHc);
for (int i = 0; i < currentNeighbors->length; i++)
{
HnswCandidate *hc3 = &currentNeighbors->items[i];
HnswElement hc3Element = HnswPtrAccess(base, hc3->element);
if (HnswPtrIsNull(base, hc3Element->value))
HnswLoadElement(hc3Element, &hc3->distance, &q, index, procinfo, collation, true, NULL);
else
hc3->distance = GetElementDistance(base, hc3Element, q, procinfo, collation);
/* Prune element if being deleted */
if (hc3Element->heaptidsLength == 0)
{
pruned = &currentNeighbors->items[i];
break;
}
}
}
SelectNeighbors(base, c, lm, support, &neighbors->closerSet, &newHc, &pruned, true);
/* Should not happen */
if (pruned == NULL)
{
List *c = NIL;
/* Add candidates */
for (int i = 0; i < currentNeighbors->length; i++)
c = lappend(c, &currentNeighbors->items[i]);
c = lappend(c, &hc2);
SelectNeighbors(base, c, lm, lc, procinfo, collation, hce, &hc2, &pruned, true);
/* Should not happen */
if (pruned == NULL)
return;
}
return;
/* Find and replace the pruned element */
for (int i = 0; i < currentNeighbors->length; i++)
for (int i = 0; i < neighbors->length; i++)
{
if (HnswPtrEqual(base, currentNeighbors->items[i].element, pruned->element))
if (HnswPtrEqual(base, neighbors->items[i].element, pruned->element))
{
currentNeighbors->items[i] = hc2;
neighbors->items[i] = newHc;
/* Track update */
if (updateIdx != NULL)
@@ -1302,17 +1256,20 @@ PrecomputeHash(char *base, HnswElement element)
* Algorithm 1 from paper
*/
void
HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing)
HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing)
{
List *ep;
List *w;
int level = element->level;
int entryLevel;
Datum q = HnswGetValue(base, element);
HnswQuery q;
HnswElement skipElement = existing ? element : NULL;
bool inMemory = index == NULL;
q.value = HnswGetValue(base, element);
/* Precompute hash */
if (index == NULL)
if (inMemory)
PrecomputeHash(base, element);
/* No neighbors if no entry point */
@@ -1320,13 +1277,13 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
return;
/* Get entry point and level */
ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, procinfo, collation, true));
ep = list_make1(HnswEntryCandidate(base, entryPoint, &q, index, support, true));
entryLevel = entryPoint->level;
/* 1st phase: greedy search to insert level */
for (int lc = entryLevel; lc >= level + 1; lc--)
{
w = HnswSearchLayer(base, q, ep, 1, lc, index, procinfo, collation, m, true, skipElement, NULL, NULL, true, NULL);
w = HnswSearchLayer(base, &q, ep, 1, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL);
ep = w;
}
@@ -1345,7 +1302,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
List *lw = NIL;
ListCell *lc2;
w = HnswSearchLayer(base, q, ep, efConstruction, lc, index, procinfo, collation, m, true, skipElement, NULL, NULL, true, NULL);
w = HnswSearchLayer(base, &q, ep, efConstruction, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL);
/* Convert search candidates to candidates */
foreach(lc2, w)
@@ -1361,7 +1318,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
/* Elements being deleted or skipped can help with search */
/* but should be removed before selecting neighbors */
if (index != NULL)
if (!inMemory)
lw = RemoveElements(base, lw, skipElement);
/*
@@ -1369,7 +1326,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
* sortCandidates to true for in-memory builds to enable closer
* caching, but there does not seem to be a difference in performance.
*/
neighbors = SelectNeighbors(base, lw, lm, lc, procinfo, collation, element, NULL, NULL, false);
neighbors = SelectNeighbors(base, lw, lm, support, &HnswGetNeighbors(base, element, lc)->closerSet, NULL, NULL, false);
AddConnections(base, element, neighbors, lc);

View File

@@ -184,13 +184,12 @@ static void
RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswElement entryPoint)
{
Relation index = vacuumstate->index;
HnswSupport *support = &vacuumstate->support;
Buffer buf;
Page page;
GenericXLogState *state;
int m = vacuumstate->m;
int efConstruction = vacuumstate->efConstruction;
FmgrInfo *procinfo = vacuumstate->procinfo;
Oid collation = vacuumstate->collation;
BufferAccessStrategy bas = vacuumstate->bas;
HnswNeighborTuple ntup = vacuumstate->ntup;
Size ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, m);
@@ -205,7 +204,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
element->heaptidsLength = 0;
/* Find neighbors for element, skipping itself */
HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, true);
HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, true);
/* Zero memory for each element */
MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);
@@ -229,7 +228,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
UnlockReleaseBuffer(buf);
/* Update neighbors */
HnswUpdateNeighborsOnDisk(index, procinfo, collation, element, m, true, false);
HnswUpdateNeighborsOnDisk(index, support, element, m, true, false);
}
/*
@@ -239,6 +238,7 @@ static void
RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
{
Relation index = vacuumstate->index;
HnswSupport *support = &vacuumstate->support;
HnswElement highestPoint = &vacuumstate->highestPoint;
HnswElement entryPoint;
MemoryContext oldCtx = MemoryContextSwitchTo(vacuumstate->tmpCtx);
@@ -256,7 +256,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
LockPage(index, HNSW_UPDATE_LOCK, ShareLock);
/* Load element */
HnswLoadElement(highestPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL);
HnswLoadElement(highestPoint, NULL, NULL, index, support, true, NULL);
/* Repair if needed */
if (NeedsUpdated(vacuumstate, highestPoint))
@@ -294,7 +294,7 @@ 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, NULL);
HnswLoadElement(entryPoint, NULL, NULL, index, support, true, NULL);
if (NeedsUpdated(vacuumstate, entryPoint))
{
@@ -581,13 +581,13 @@ 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->ntup = palloc0(HNSW_TUPLE_ALLOC_SIZE);
vacuumstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
"Hnsw vacuum temporary context",
ALLOCSET_DEFAULT_SIZES);
HnswInitSupport(&vacuumstate->support, index);
/* Get m from metapage */
HnswGetMetaPageInfo(index, &vacuumstate->m, NULL);

View File

@@ -69,6 +69,8 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
GenericCosts costs;
int lists;
double ratio;
double sequentialRatio = 0.5;
double startupPages;
double spc_seq_page_cost;
Relation index;
@@ -85,6 +87,8 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
MemSet(&costs, 0, sizeof(costs));
genericcostestimate(root, path, loop_count, &costs);
index = index_open(path->indexinfo->indexoid, NoLock);
IvfflatGetMetaPageInfo(index, &lists, NULL);
index_close(index, NoLock);
@@ -94,34 +98,26 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
if (ratio > 1.0)
ratio = 1.0;
/*
* This gives us the subset of tuples to visit. This value is passed into
* the generic cost estimator to determine the number of pages to visit
* during the index scan.
*/
costs.numIndexTuples = path->indexinfo->tuples * ratio;
genericcostestimate(root, path, loop_count, &costs);
get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
/* Change some page cost from random to sequential */
costs.indexTotalCost -= sequentialRatio * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Startup cost is cost before returning the first row */
costs.indexStartupCost = costs.indexTotalCost * ratio;
/* Adjust cost if needed since TOAST not included in seq scan cost */
if (costs.numIndexPages > path->indexinfo->rel->pages && ratio < 0.5)
startupPages = costs.numIndexPages * ratio;
if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
{
/* Change all page cost from random to sequential */
costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Change rest of page cost from random to sequential */
costs.indexStartupCost -= (1 - sequentialRatio) * startupPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Remove cost of extra pages */
costs.indexTotalCost -= (costs.numIndexPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
}
else
{
/* Change some page cost from random to sequential */
costs.indexTotalCost -= 0.5 * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
costs.indexStartupCost -= (startupPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
}
/* Use total cost since most work happens before first tuple is returned */
*indexStartupCost = costs.indexTotalCost;
*indexStartupCost = costs.indexStartupCost;
*indexTotalCost = costs.indexTotalCost;
*indexSelectivity = costs.indexSelectivity;
*indexCorrelation = costs.indexCorrelation;

View File

@@ -41,8 +41,7 @@ my $c = int(rand() * $nc);
my $explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
));
# TODO Do not use index
like($explain, qr/Index Scan using idx/);
like($explain, qr/Seq Scan/);
# Test attribute filtering with few rows removed
$explain = $node->safe_psql("postgres", qq(
@@ -60,8 +59,7 @@ like($explain, qr/Index Scan using idx/);
$explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE c < 1 ORDER BY v <-> '$query' LIMIT $limit;
));
# TODO Do not use index
like($explain, qr/Index Scan using idx/);
like($explain, qr/Seq Scan/);
# Test attribute filtering with few rows removed like
$explain = $node->safe_psql("postgres", qq(

View File

@@ -17,12 +17,11 @@ $node->safe_psql("postgres", "CREATE EXTENSION vector;");
for my $dim (@dims)
{
my $array_sql = join(",", ('random()') x $dim);
my $n = $dim == 384 ? 2000 : 1000;
# Create table and index
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, $n) i;"
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 2000) i;"
);
$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
$node->safe_psql("postgres", "ANALYZE tst;");
@@ -40,6 +39,21 @@ for my $dim (@dims)
));
like($explain, qr/Index Scan using idx/);
# 3x the rows are needed for distance filters
# since the planner uses DEFAULT_INEQ_SEL for the selectivity (should be 1)
# Recreate index for performance
$node->safe_psql("postgres", "DROP INDEX idx;");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(2001, 6000) i;"
);
$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
$node->safe_psql("postgres", "ANALYZE tst;");
$explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query' LIMIT $limit;
));
like($explain, qr/Index Scan using idx/);
$node->safe_psql("postgres", "DROP TABLE tst;");
}

View File

@@ -39,6 +39,11 @@ for my $dim (@dims)
));
like($explain, qr/Index Scan using idx/);
$explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query' LIMIT $limit;
));
like($explain, qr/Index Scan using idx/);
$node->safe_psql("postgres", "DROP TABLE tst;");
}

View File

@@ -58,6 +58,7 @@ foreach ((30000, 50000, 70000))
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
SET enable_seqscan = off;
SET hnsw.streaming = on;
SET client_min_messages = debug1;
SET work_mem = '2MB';
SELECT COUNT(*) FROM (SELECT v FROM tst WHERE i % 10000 = 0 ORDER BY v <-> (SELECT v FROM tst LIMIT 1) LIMIT 11) t;
));