Use pairing heap [skip ci]

CHANGELOG.md

@@ -1,7 +1,6 @@
-## 0.4.2 (2023-05-13)
+## 0.4.2 (unreleased)
 
 - Added notice when index created with little data
-- Fixed dimensions check for some direct function calls
 - Fixed installation error with Postgres 12.0-12.2
 
 ## 0.4.1 (2023-03-21)

META.json

@@ -2,7 +2,7 @@
 	"name": "vector",
 	"abstract": "Open-source vector similarity search for Postgres",
 	"description": "Supports L2 distance, inner product, and cosine distance",
-	"version": "0.4.2",
+	"version": "0.4.1",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.4.2",
+			"version": "0.4.1",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.4.2
+EXTVERSION = 0.4.1
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)
@@ -63,9 +63,3 @@ dist:
 
 docker:
 	docker build --pull --no-cache --platform linux/amd64 -t ankane/pgvector:latest .
-
-.PHONY: docker-release
-
-docker-release:
-	docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 -t ankane/pgvector:latest .
-	docker buildx build --push --platform linux/amd64,linux/arm64 -t ankane/pgvector:v$(EXTVERSION) .

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.4.2
+EXTVERSION = 0.4.1
 
 OBJS = src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
 

README.md

@@ -16,7 +16,7 @@ Compile and install the extension (supports Postgres 11+)
 
 ```sh
 cd /tmp
-git clone --branch v0.4.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -307,11 +307,10 @@ Yes, pgvector uses the write-ahead log (WAL), which allows for replication and p
 
 #### What if I want to index vectors with more than 2,000 dimensions?
 
-You’ll need to use [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction) at the moment.
+Two things you can try are:
 
-#### Why am I seeing less results after adding an index?
+1. use dimensionality reduction
-
+2. compile Postgres with a larger block size (`./configure --with-blocksize=32`) and edit the limit in `src/ivfflat.h`
-The index was likely created with too little data for the number of lists. Drop the index until the table has more data.
 
 ## Reference
 
@@ -355,11 +354,7 @@ If your machine has multiple Postgres installations, specify the path to [pg_con
 export PG_CONFIG=/Applications/Postgres.app/Contents/Versions/latest/bin/pg_config
 ```
 
-Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
+Then re-run the installation instructions (run `make clean` before `make` if needed)
-
-```sh
-sudo --preserve-env=PG_CONFIG make install
-```
 
 ### Missing Header
 
@@ -379,7 +374,7 @@ Support for Windows is currently experimental. Use `nmake` to build:
 
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\15"
-git clone --branch v0.4.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -400,7 +395,7 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.4.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 docker build -t pgvector .
 ```
@@ -413,8 +408,6 @@ With Homebrew Postgres, you can use:
 brew install pgvector
 ```
 
-Note: This only adds it to the `postgresql@14` formula
-
 ### PGXN
 
 Install from the [PostgreSQL Extension Network](https://pgxn.org/dist/vector) with:
@@ -451,9 +444,10 @@ pgvector is available on [these providers](https://github.com/pgvector/pgvector/
 
 To request a new extension on other providers:
 
+- Amazon RDS - follow the instructions on [this page](https://aws.amazon.com/rds/postgresql/faqs/)
 - Google Cloud SQL - vote or comment on [this page](https://issuetracker.google.com/issues/265172065)
 - Azure Database - vote or comment on [this page](https://feedback.azure.com/d365community/idea/7b423322-6189-ed11-a81b-000d3ae49307)
-- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/managed-database/p/pgvector-extension-for-postgresql)
+- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/app-framework-services/p/pgvector-extension-for-postgresql)
 - Heroku Postgres - vote or comment on [this page](https://github.com/heroku/roadmap/issues/156)
 
 ## Upgrading

sql/vector--0.4.1--0.4.2.sql

@@ -1,2 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.2'" to load this file. \quit

src/ivfbuild.c

@@ -438,8 +438,8 @@ ComputeCenters(IvfflatBuildState * buildstate)
 		{
 			ereport(NOTICE,
 					(errmsg("ivfflat index created with little data"),
-					 errdetail("This will cause low recall."),
+					 errdetail("this will cause poor recall"),
-					 errhint("Drop the index until the table has more data.")));
+					 errhint("drop the index until the table has more data")));
 		}
 	}
 

src/ivfflat.h

@@ -187,6 +187,14 @@ typedef struct IvfflatScanList
 	double		distance;
 }			IvfflatScanList;
 
+typedef struct IvfflatScanItem
+{
+	pairingheap_node ph_node;
+	BlockNumber searchPage;
+	double		distance;
+	ItemPointerData tid;
+}			IvfflatScanItem;
+
 typedef struct IvfflatScanOpaqueData
 {
 	int			probes;
@@ -204,6 +212,13 @@ typedef struct IvfflatScanOpaqueData
 	FmgrInfo   *normprocinfo;
 	Oid			collation;
 
+	/* Items */
+	int			maxItems;
+	int			itemCount;
+	pairingheap *itemQueue;
+	IvfflatScanItem *items;
+	IvfflatScanItem **sortedItems;
+
 	/* Lists */
 	pairingheap *listQueue;
 	IvfflatScanList lists[FLEXIBLE_ARRAY_MEMBER];	/* must come last */

src/ivfscan.c

@@ -26,6 +26,21 @@ CompareLists(const pairingheap_node *a, const pairingheap_node *b, void *arg)
 	return 0;
 }
 
+/*
+ * Compare item distances
+ */
+static int
+CompareItems(const pairingheap_node *a, const pairingheap_node *b, void *arg)
+{
+	if (((const IvfflatScanItem *) a)->distance > ((const IvfflatScanItem *) b)->distance)
+		return 1;
+
+	if (((const IvfflatScanItem *) a)->distance < ((const IvfflatScanItem *) b)->distance)
+		return -1;
+
+	return 0;
+}
+
 /*
  * Get lists and sort by distance
  */
@@ -111,13 +126,10 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	Datum		datum;
 	bool		isnull;
 	TupleDesc	tupdesc = RelationGetDescr(scan->indexRelation);
-	double		tuples = 0;
+	int			i;
-
+	double		distance;
-#if PG_VERSION_NUM >= 120000
+	IvfflatScanItem *scanitem;
-	TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
+	double		maxDistance = DBL_MAX;
-#else
-	TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc);
-#endif
 
 	/*
 	 * Reuse same set of shared buffers for scan
@@ -143,25 +155,40 @@ GetScanItems(IndexScanDesc scan, Datum value)
 			{
 				itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offno));
 				datum = index_getattr(itup, 1, tupdesc, &isnull);
+				distance = DatumGetFloat8(FunctionCall2Coll(so->procinfo, so->collation, datum, value));
 
-				/*
+				if (so->itemCount < so->maxItems)
-				 * Add virtual tuple
+				{
-				 *
+					scanitem = &so->items[so->itemCount];
-				 * Use procinfo from the index instead of scan key for
+					scanitem->searchPage = searchPage;
-				 * performance
+					scanitem->tid = itup->t_tid;
-				 */
+					scanitem->distance = distance;
-				ExecClearTuple(slot);
+					so->itemCount++;
-				slot->tts_values[0] = FunctionCall2Coll(so->procinfo, so->collation, datum, value);
-				slot->tts_isnull[0] = false;
-				slot->tts_values[1] = PointerGetDatum(&itup->t_tid);
-				slot->tts_isnull[1] = false;
-				slot->tts_values[2] = Int32GetDatum((int) searchPage);
-				slot->tts_isnull[2] = false;
-				ExecStoreVirtualTuple(slot);
 
-				tuplesort_puttupleslot(so->sortstate, slot);
+					/* Add to heap */
+					pairingheap_add(so->itemQueue, &scanitem->ph_node);
 
-				tuples++;
+					/* Calculate max distance */
+					if (so->itemCount == so->maxItems)
+					{
+						maxDistance = ((IvfflatScanItem *) pairingheap_first(so->itemQueue))->distance;
+						scanitem = &so->items[so->itemCount];
+					}
+				}
+				else if (distance < maxDistance)
+				{
+					/* Reuse */
+					scanitem->searchPage = searchPage;
+					scanitem->tid = itup->t_tid;
+					scanitem->distance = distance;
+					pairingheap_add(so->itemQueue, &scanitem->ph_node);
+
+					/* Remove */
+					scanitem = (IvfflatScanItem *) pairingheap_remove_first(so->itemQueue);
+
+					/* Update max distance */
+					maxDistance = ((IvfflatScanItem *) pairingheap_first(so->itemQueue))->distance;
+				}
 			}
 
 			searchPage = IvfflatPageGetOpaque(page)->nextblkno;
@@ -170,14 +197,10 @@ GetScanItems(IndexScanDesc scan, Datum value)
 		}
 	}
 
-	/* TODO Scan more lists */
+	for (i = 0; i < so->itemCount; i++)
-	if (tuples < 100)
+		so->sortedItems[i] = (IvfflatScanItem *) pairingheap_remove_first(so->itemQueue);
-		ereport(DEBUG1,
-				(errmsg("index scan found few tuples"),
-				 errdetail("Index may have been created with little data."),
-				 errhint("Recreate the index and possibly decrease lists.")));
 
-	tuplesort_performsort(so->sortstate);
+	Assert(pairingheap_is_empty(so->itemQueue));
 }
 
 /*
@@ -189,10 +212,6 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	IndexScanDesc scan;
 	IvfflatScanOpaque so;
 	int			lists;
-	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Float8LessOperator};
-	Oid			sortCollations[] = {InvalidOid};
-	bool		nullsFirstFlags[] = {false};
 	int			probes = ivfflat_probes;
 
 	scan = RelationGetIndexScan(index, nkeys, norderbys);
@@ -211,27 +230,14 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	so->normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
 	so->collation = index->rd_indcollation[0];
 
-	/* Create tuple description for sorting */
-#if PG_VERSION_NUM >= 120000
-	so->tupdesc = CreateTemplateTupleDesc(3);
-#else
-	so->tupdesc = CreateTemplateTupleDesc(3, false);
-#endif
-	TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
-	TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
-	TupleDescInitEntry(so->tupdesc, (AttrNumber) 3, "indexblkno", INT4OID, -1, 0);
-
-	/* Prep sort */
-	so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
-
-#if PG_VERSION_NUM >= 120000
-	so->slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsMinimalTuple);
-#else
-	so->slot = MakeSingleTupleTableSlot(so->tupdesc);
-#endif
-
 	so->listQueue = pairingheap_allocate(CompareLists, scan);
 
+	so->maxItems = 1024;
+	so->itemCount = 0;
+	so->itemQueue = pairingheap_allocate(CompareItems, scan);
+	so->items = palloc(sizeof(IvfflatScanItem) * (so->maxItems + 1));
+	so->sortedItems = palloc(sizeof(IvfflatScanItem *) * so->maxItems);
+
 	scan->opaque = so;
 
 	return scan;
@@ -245,13 +251,10 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
 {
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
 
-#if PG_VERSION_NUM >= 130000
-	if (!so->first)
-		tuplesort_reset(so->sortstate);
-#endif
-
 	so->first = true;
 	pairingheap_reset(so->listQueue);
+	pairingheap_reset(so->itemQueue);
+	so->itemCount = 0;
 
 	if (keys && scan->numberOfKeys > 0)
 		memmove(scan->keyData, keys, scan->numberOfKeys * sizeof(ScanKeyData));
@@ -311,15 +314,18 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 			pfree(DatumGetPointer(value));
 	}
 
-	if (tuplesort_gettupleslot(so->sortstate, true, false, so->slot, NULL))
+	if (so->itemCount > 0)
 	{
-		ItemPointer tid = (ItemPointer) DatumGetPointer(slot_getattr(so->slot, 2, &so->isnull));
+		IvfflatScanItem *scanitem;
-		BlockNumber indexblkno = DatumGetInt32(slot_getattr(so->slot, 3, &so->isnull));
+
+		so->itemCount--;
+
+		scanitem = so->sortedItems[so->itemCount];
 
 #if PG_VERSION_NUM >= 120000
-		scan->xs_heaptid = *tid;
+		scan->xs_heaptid = scanitem->tid;
 #else
-		scan->xs_ctup.t_self = *tid;
+		scan->xs_ctup.t_self = scanitem->tid;
 #endif
 
 		if (BufferIsValid(so->buf))
@@ -331,7 +337,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		 *
 		 * https://www.postgresql.org/docs/current/index-locking.html
 		 */
-		so->buf = ReadBuffer(scan->indexRelation, indexblkno);
+		so->buf = ReadBuffer(scan->indexRelation, scanitem->searchPage);
 
 		scan->xs_recheckorderby = false;
 		return true;
@@ -353,7 +359,10 @@ ivfflatendscan(IndexScanDesc scan)
 		ReleaseBuffer(so->buf);
 
 	pairingheap_free(so->listQueue);
-	tuplesort_end(so->sortstate);
+
+	pairingheap_free(so->itemQueue);
+	pfree(so->items);
+	pfree(so->sortedItems);
 
 	pfree(so);
 	scan->opaque = NULL;

src/vector.c

@@ -396,8 +396,10 @@ array_to_vector(PG_FUNCTION_ARGS)
 	get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
 	deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, &nullsp, &nelemsp);
 
-	CheckDim(nelemsp);
+	if (typmod == -1)
-	CheckExpectedDim(typmod, nelemsp);
+		CheckDim(nelemsp);
+	else
+		CheckExpectedDim(typmod, nelemsp);
 
 	result = InitVector(nelemsp);
 	for (i = 0; i < nelemsp; i++)
@@ -950,7 +952,6 @@ vector_avg(PG_FUNCTION_ARGS)
 
 	/* Create vector */
 	dim = STATE_DIMS(statearray);
-	CheckDim(dim);
 	result = InitVector(dim);
 	for (int i = 0; i < dim; i++)
 	{

test/expected/cast.out

@@ -46,8 +46,6 @@ SELECT '[1,2,3]'::vector::real[];
 
 SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
 ERROR:  vector cannot have more than 16000 dimensions
-SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;
-ERROR:  vector cannot have more than 16000 dimensions
 -- ensure no error
 SELECT ARRAY[1,2,3] = ARRAY[1,2,3];
  ?column? 

test/expected/functions.out

@@ -102,5 +102,3 @@ SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
 
 SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
 ERROR:  expected 2 dimensions, not 1
-SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
-ERROR:  vector cannot have more than 16000 dimensions

test/sql/cast.sql

@@ -10,7 +10,6 @@ SELECT '{-Infinity}'::real[]::vector;
 SELECT '{}'::real[]::vector;
 SELECT '[1,2,3]'::vector::real[];
 SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
-SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;
 
 -- ensure no error
 SELECT ARRAY[1,2,3] = ARRAY[1,2,3];

test/sql/functions.sql

@@ -24,4 +24,3 @@ SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
 SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
 SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
 SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
-SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat access method'
-default_version = '0.4.2'
+default_version = '0.4.1'
 module_pathname = '$libdir/vector'
 relocatable = true