Switched to runtime check [skip ci]

CHANGELOG.md

@@ -1,11 +1,11 @@
-## 0.6.0 (2024-01-29)
+## 0.6.0 (unreleased)
 
-If upgrading with Postgres 12 or Docker, see [these notes](https://github.com/pgvector/pgvector#060).
+If upgrading with Postgres 12 or Docker, see [these notes](https://github.com/pgvector/pgvector#060-unreleased).
 
 - Changed storage for vector from `extended` to `external`
+- Improved performance of HNSW
 - Added support for parallel index builds for HNSW
 - Added validation for GUC parameters
-- Improved performance of HNSW
 - Reduced memory usage for HNSW index builds
 - Reduced WAL generation for HNSW index builds
 - Fixed error with logical replication

Dockerfile

@@ -1,4 +1,4 @@
-ARG PG_MAJOR=16
+ARG PG_MAJOR=15
 FROM postgres:$PG_MAJOR
 ARG PG_MAJOR
 

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": "0.6.0",
+	"version": "0.5.1",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.6.0",
+			"version": "0.5.1",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.6.0
+EXTVERSION = 0.5.1
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.6.0
+EXTVERSION = 0.5.1
 
 OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
 HEADERS = src\vector.h

README.md

@@ -20,7 +20,7 @@ Compile and install the extension (supports Postgres 11+)
 
 ```sh
 cd /tmp
-git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -44,7 +44,7 @@ Then use `nmake` to build:
 
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\16"
-git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -264,7 +264,7 @@ HINT:  Increase maintenance_work_mem to speed up builds.
 
 Note: Do not set `maintenance_work_mem` so high that it exhausts the memory on the server
 
-Starting with 0.6.0, you can also speed up index creation by increasing the number of parallel workers (2 by default)
+Starting with 0.6.0 [unreleased], you can also speed up index creation by increasing the number of parallel workers (2 by default)
 
 ```sql
 SET max_parallel_maintenance_workers = 7; -- plus leader
@@ -651,10 +651,10 @@ If compilation fails and the output includes `warning: no such sysroot directory
 
 ### Docker
 
-Get the [Docker image](https://hub.docker.com/r/pgvector/pgvector) with:
+Get the [Docker image](https://hub.docker.com/r/ankane/pgvector) with:
 
 ```sh
-docker pull pgvector/pgvector:pg16
+docker pull ankane/pgvector
 ```
 
 This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (run it the same way).
@@ -662,9 +662,9 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
 cd pgvector
-docker build --build-arg PG_MAJOR=16 -t myuser/pgvector .
+docker build --build-arg PG_MAJOR=15 -t myuser/pgvector .
 ```
 
 ### Homebrew
@@ -741,7 +741,7 @@ SELECT extversion FROM pg_extension WHERE extname = 'vector';
 
 ## Upgrade Notes
 
-### 0.6.0
+### 0.6.0 [unreleased]
 
 #### Postgres 12
 

src/hnsw.h

@@ -60,7 +60,6 @@
 #define PROGRESS_HNSW_PHASE_LOAD		2
 
 #define HNSW_MAX_SIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - sizeof(ItemIdData))
-#define HNSW_TUPLE_ALLOC_SIZE BLCKSZ
 
 #define HNSW_ELEMENT_TUPLE_SIZE(size)	MAXALIGN(offsetof(HnswElementTupleData, data) + (size))
 #define HNSW_NEIGHBOR_TUPLE_SIZE(level, m)	MAXALIGN(offsetof(HnswNeighborTupleData, indextids) + ((level) + 2) * (m) * sizeof(ItemPointerData))
@@ -355,7 +354,6 @@ typedef struct HnswVacuumState
 	/* Variables */
 	struct tidhash_hash *deleted;
 	BufferAccessStrategy bas;
-	HnswNeighborTuple ntup;
 	HnswElementData highestPoint;
 
 	/* Memory */

src/hnswbuild.c

@@ -148,6 +148,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 {
 	Relation	index = buildstate->index;
 	ForkNumber	forkNum = buildstate->forkNum;
+	Size		etupAllocSize;
 	Size		maxSize;
 	HnswElementTuple etup;
 	HnswNeighborTuple ntup;
@@ -159,11 +160,12 @@ CreateGraphPages(HnswBuildState * buildstate)
 	char	   *base = buildstate->hnswarea;
 
 	/* Calculate sizes */
+	etupAllocSize = BLCKSZ;
 	maxSize = HNSW_MAX_SIZE;
 
 	/* Allocate once */
-	etup = palloc0(HNSW_TUPLE_ALLOC_SIZE);
-	ntup = palloc0(HNSW_TUPLE_ALLOC_SIZE);
+	etup = palloc0(etupAllocSize);
+	ntup = palloc0(BLCKSZ);
 
 	/* Prepare first page */
 	buf = HnswNewBuffer(index, forkNum);
@@ -182,7 +184,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 		iter = element->next;
 
 		/* Zero memory for each element */
-		MemSet(etup, 0, HNSW_TUPLE_ALLOC_SIZE);
+		MemSet(etup, 0, etupAllocSize);
 
 		/* Calculate sizes */
 		etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(valuePtr));
@@ -190,7 +192,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 		combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
 
 		/* Initial size check */
-		if (etupSize > HNSW_TUPLE_ALLOC_SIZE)
+		if (etupSize > etupAllocSize)
 			elog(ERROR, "index tuple too large");
 
 		HnswSetElementTuple(base, etup, element);
@@ -252,43 +254,38 @@ WriteNeighborTuples(HnswBuildState * buildstate)
 	int			m = buildstate->m;
 	HnswElementPtr iter = buildstate->graph->head;
 	char	   *base = buildstate->hnswarea;
-	HnswNeighborTuple ntup;
-
-	/* Allocate once */
-	ntup = palloc0(HNSW_TUPLE_ALLOC_SIZE);
 
 	while (!HnswPtrIsNull(base, iter))
 	{
-		HnswElement element = HnswPtrAccess(base, iter);
+		HnswElement e = HnswPtrAccess(base, iter);
 		Buffer		buf;
 		Page		page;
-		Size		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, m);
+		ItemId		itemid;
+		HnswNeighborTuple ntup;
 
 		/* Update iterator */
-		iter = element->next;
-
-		/* Zero memory for each element */
-		MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);
+		iter = e->next;
 
 		/* Can take a while, so ensure we can interrupt */
 		/* Needs to be called when no buffer locks are held */
 		CHECK_FOR_INTERRUPTS();
 
-		buf = ReadBufferExtended(index, forkNum, element->neighborPage, RBM_NORMAL, NULL);
+		buf = ReadBufferExtended(index, forkNum, e->neighborPage, RBM_NORMAL, NULL);
 		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
 		page = BufferGetPage(buf);
+		itemid = PageGetItemId(page, e->neighborOffno);
+		ntup = (HnswNeighborTuple) PageGetItem(page, itemid);
 
-		HnswSetNeighborTuple(base, ntup, element, m);
+		/* Check expected size */
+		Assert(ItemIdGetLength(itemid) == HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m));
 
-		if (!PageIndexTupleOverwrite(page, element->neighborOffno, (Item) ntup, ntupSize))
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
+		/* Update page in-place */
+		HnswSetNeighborTuple(base, ntup, e, m);
 
 		/* Commit */
 		MarkBufferDirty(buf);
 		UnlockReleaseBuffer(buf);
 	}
-
-	pfree(ntup);
 }
 
 /*

src/hnswinsert.c

@@ -355,7 +355,9 @@ HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, Hns
 			Buffer		buf;
 			Page		page;
 			GenericXLogState *state;
+			ItemId		itemid;
 			HnswNeighborTuple ntup;
+			Size		ntupSize;
 			int			idx = -1;
 			int			startIdx;
 			HnswElement neighborElement = HnswPtrAccess(base, hc->element);
@@ -394,7 +396,9 @@ HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, Hns
 			}
 
 			/* Get tuple */
-			ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, offno));
+			itemid = PageGetItemId(page, offno);
+			ntup = (HnswNeighborTuple) PageGetItem(page, itemid);
+			ntupSize = ItemIdGetLength(itemid);
 
 			/* Calculate index for update */
 			startIdx = (neighborElement->level - lc) * m;
@@ -423,9 +427,13 @@ HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, Hns
 			{
 				ItemPointer indextid = &ntup->indextids[idx];
 
-				/* Update neighbor on the buffer */
+				/* Update neighbor */
 				ItemPointerSet(indextid, e->blkno, e->offno);
 
+				/* Overwrite tuple */
+				if (!PageIndexTupleOverwrite(page, offno, (Item) ntup, ntupSize))
+					elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
+
 				/* Commit */
 				if (building)
 					MarkBufferDirty(buf);
@@ -449,7 +457,9 @@ AddDuplicateOnDisk(Relation index, HnswElement element, HnswElement dup, bool bu
 	Buffer		buf;
 	Page		page;
 	GenericXLogState *state;
+	ItemId		itemid;
 	HnswElementTuple etup;
+	Size		etupSize;
 	int			i;
 
 	/* Read page */
@@ -467,7 +477,9 @@ AddDuplicateOnDisk(Relation index, HnswElement element, HnswElement dup, bool bu
 	}
 
 	/* Find space */
-	etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, dup->offno));
+	itemid = PageGetItemId(page, dup->offno);
+	etup = (HnswElementTuple) PageGetItem(page, itemid);
+	etupSize = ItemIdGetLength(itemid);
 	for (i = 0; i < HNSW_HEAPTIDS; i++)
 	{
 		if (!ItemPointerIsValid(&etup->heaptids[i]))
@@ -483,9 +495,13 @@ AddDuplicateOnDisk(Relation index, HnswElement element, HnswElement dup, bool bu
 		return false;
 	}
 
-	/* Add heap TID, modifying the tuple on the page directly */
+	/* Add heap TID */
 	etup->heaptids[i] = element->heaptids[0];
 
+	/* Overwrite tuple */
+	if (!PageIndexTupleOverwrite(page, dup->offno, (Item) etup, etupSize))
+		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
+
 	/* Commit */
 	if (building)
 		MarkBufferDirty(buf);

src/hnswvacuum.c

@@ -60,7 +60,8 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 		/* Iterate over nodes */
 		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
 		{
-			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
+			ItemId		itemid = PageGetItemId(page, offno);
+			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, itemid);
 			int			idx = 0;
 			bool		itemUpdated = false;
 
@@ -91,10 +92,15 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 
 				if (itemUpdated)
 				{
+					Size		etupSize = ItemIdGetLength(itemid);
+
 					/* Mark rest as invalid */
 					for (int i = idx; i < HNSW_HEAPTIDS; i++)
 						ItemPointerSetInvalid(&etup->heaptids[i]);
 
+					if (!PageIndexTupleOverwrite(page, offno, (Item) etup, etupSize))
+						elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
+
 					updated = true;
 				}
 			}
@@ -192,8 +198,8 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 	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);
+	ItemId		itemid;
+	HnswNeighborTuple ntup;
 	char	   *base = NULL;
 
 	/* Skip if element is entry point */
@@ -207,23 +213,23 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 	/* Find neighbors for element, skipping itself */
 	HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, true);
 
-	/* Zero memory for each element */
-	MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);
-
-	/* Update neighbor tuple */
-	/* Do this before getting page to minimize locking */
-	HnswSetNeighborTuple(base, ntup, element, m);
-
 	/* Get neighbor page */
 	buf = ReadBufferExtended(index, MAIN_FORKNUM, element->neighborPage, RBM_NORMAL, bas);
 	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
 	state = GenericXLogStart(index);
 	page = GenericXLogRegisterBuffer(state, buf, 0);
 
-	/* Overwrite tuple */
-	if (!PageIndexTupleOverwrite(page, element->neighborOffno, (Item) ntup, ntupSize))
+	/* Get neighbor tuple */
+	itemid = PageGetItemId(page, element->neighborOffno);
+	ntup = (HnswNeighborTuple) PageGetItem(page, itemid);
+
+	/* Check expected size */
+	if (ItemIdGetLength(itemid) != HNSW_NEIGHBOR_TUPLE_SIZE(element->level, m))
 		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
 
+	/* Update page in-place */
+	HnswSetNeighborTuple(base, ntup, element, m);
+
 	/* Commit */
 	GenericXLogFinish(state);
 	UnlockReleaseBuffer(buf);
@@ -476,8 +482,11 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 		/* Update element and neighbors together */
 		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
 		{
-			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
+			ItemId		itemid = PageGetItemId(page, offno);
+			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, itemid);
 			HnswNeighborTuple ntup;
+			Size		etupSize;
+			Size		ntupSize;
 			Buffer		nbuf;
 			Page		npage;
 			BlockNumber neighborPage;
@@ -501,6 +510,10 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 			if (ItemPointerIsValid(&etup->heaptids[0]))
 				continue;
 
+			/* Calculate sizes */
+			etupSize = ItemIdGetLength(itemid);
+			ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(etup->level, vacuumstate->m);
+
 			/* Get neighbor page */
 			neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
 			neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
@@ -527,10 +540,13 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 			for (int i = 0; i < ntup->count; i++)
 				ItemPointerSetInvalid(&ntup->indextids[i]);
 
-			/*
-			 * We modified the tuples in place, no need to call
-			 * PageIndexTupleOverwrite
-			 */
+			/* Overwrite element tuple */
+			if (!PageIndexTupleOverwrite(page, offno, (Item) etup, etupSize))
+				elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
+
+			/* Overwrite neighbor tuple */
+			if (!PageIndexTupleOverwrite(npage, neighborOffno, (Item) ntup, ntupSize))
+				elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
 
 			/* Commit */
 			GenericXLogFinish(state);
@@ -575,7 +591,6 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
 	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);
@@ -595,7 +610,6 @@ FreeVacuumState(HnswVacuumState * vacuumstate)
 {
 	tidhash_destroy(vacuumstate->deleted);
 	FreeAccessStrategy(vacuumstate->bas);
-	pfree(vacuumstate->ntup);
 	MemoryContextDelete(vacuumstate->tmpCtx);
 }
 

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.6.0'
+default_version = '0.5.1'
 module_pathname = '$libdir/vector'
 relocatable = true