Version bump to 0.8.3 [skip ci]

Co-authored-by: Bhagyesh Chaturvedi <bhagyeshc@google.com>

.github/workflows/build.yml

@@ -97,7 +97,7 @@ jobs:
         with:
           postgres-version: ${{ matrix.postgres }}
       - run: |
-          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
+          call "C:\Program Files\Microsoft Visual Studio\${{ matrix.os == 'windows-2025' && 18 || 2022 }}\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
           nmake /NOLOGO /F Makefile.win && ^
           nmake /NOLOGO /F Makefile.win install && ^
           nmake /NOLOGO /F Makefile.win installcheck ${{ matrix.postgres != 17 && 'PG_REGRESS=$(PGROOT)\bin\pg_regress' || '' }} && ^

CHANGELOG.md

@@ -1,6 +1,11 @@
+## 0.8.3 (2026-06-17)
+
+- Fixed possible index corruption with HNSW vacuuming
+- Fixed performance regression with Hamming distance and Jaccard distance with Postgres 18
+
 ## 0.8.2 (2026-02-25)
 
-- Fixed buffer overflow with parallel HNSW index build
+- Fixed buffer overflow with parallel HNSW index build - [more info](https://github.com/pgvector/pgvector/issues/959)
 - Improved `install` target on Windows
 - Fixed `Index Searches` in `EXPLAIN` output for Postgres 18
 

Dockerfile

@@ -5,7 +5,7 @@ ARG DEBIAN_CODENAME=bookworm
 FROM postgres:$PG_MAJOR-$DEBIAN_CODENAME
 ARG PG_MAJOR
 
-ADD https://github.com/pgvector/pgvector.git#v0.8.2 /tmp/pgvector
+ADD https://github.com/pgvector/pgvector.git#v0.8.3 /tmp/pgvector
 
 RUN apt-get update && \
 		apt-mark hold locales && \

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": "0.8.2",
+	"version": "0.8.3",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.8.2",
+			"version": "0.8.3",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.8.2
+EXTVERSION = 0.8.3
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*--*.sql)

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.8.2
+EXTVERSION = 0.8.3
 
 DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql
 OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj

README.md

@@ -11,6 +11,8 @@ Store your vectors with the rest of your data. Supports:
 
 Plus [ACID](https://en.wikipedia.org/wiki/ACID) compliance, point-in-time recovery, JOINs, and all of the other [great features](https://www.postgresql.org/about/) of Postgres
 
+Have a lot of vectors? Use [quantization](#scaling) to scale
+
 [![Build Status](https://github.com/pgvector/pgvector/actions/workflows/build.yml/badge.svg)](https://github.com/pgvector/pgvector/actions)
 
 ## Installation
@@ -21,7 +23,7 @@ Compile and install the extension (supports Postgres 13+)
 
 ```sh
 cd /tmp
-git clone --branch v0.8.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.3 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -38,7 +40,7 @@ Ensure [C++ support in Visual Studio](https://learn.microsoft.com/en-us/cpp/buil
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\18"
 cd %TEMP%
-git clone --branch v0.8.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.3 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -314,6 +316,8 @@ For a large number of workers, you may need to increase `max_parallel_workers` (
 
 The [index options](#index-options) also have a significant impact on build time (use the defaults unless seeing low recall)
 
+Use [binary quantization](#binary-quantization) for faster build times at scale
+
 ### Indexing Progress
 
 Check [indexing progress](https://www.postgresql.org/docs/current/progress-reporting.html#CREATE-INDEX-PROGRESS-REPORTING)
@@ -443,13 +447,7 @@ Exact indexes work well for conditions that match a low percentage of rows. Othe
 CREATE INDEX ON items USING hnsw (embedding vector_l2_ops);
 ```
 
-With approximate indexes, filtering is applied *after* the index is scanned. If a condition matches 10% of rows, with HNSW and the default `hnsw.ef_search` of 40, only 4 rows will match on average. For more rows, increase `hnsw.ef_search`.
-
-```sql
-SET hnsw.ef_search = 200;
-```
-
-Starting with 0.8.0, you can enable [iterative index scans](#iterative-index-scans), which will automatically scan more of the index when needed.
+With approximate indexes, filtering is applied *after* the index is scanned. If a condition matches 10% of rows, with HNSW and the default `hnsw.ef_search` of 40, only 4 rows will match on average. For more rows, enable [iterative index scans](#iterative-index-scans), which will automatically scan more of the index when needed.
 
 ```sql
 SET hnsw.iterative_scan = strict_order;
@@ -673,6 +671,10 @@ SHOW shared_buffers;
 
 Be sure to restart Postgres for changes to take effect.
 
+### Storing
+
+Use the `halfvec` type instead of `vector` for a smaller working set.
+
 ### Loading
 
 Use `COPY` for bulk loading data ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/loading/example.py)).
@@ -687,6 +689,8 @@ Add any indexes *after* loading the initial data for best performance.
 
 See index build time for [HNSW](#index-build-time) and [IVFFlat](#index-build-time-1).
 
+Use [binary quantization](#binary-quantization) for smaller indexes and faster build times at scale.
+
 In production environments, create indexes concurrently to avoid blocking writes.
 
 ```sql
@@ -717,6 +721,8 @@ SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
 
 #### Approximate Search
 
+Use [binary quantization](#binary-quantization) with re-ranking to keep indexes in-memory at scale.
+
 To speed up queries with an IVFFlat index, increase the number of inverted lists (at the expense of recall).
 
 ```sql
@@ -732,21 +738,20 @@ REINDEX INDEX CONCURRENTLY index_name;
 VACUUM table_name;
 ```
 
+## Scaling
+
+For a smaller working set:
+
+1. Use the `halfvec` type instead of `vector` for tables
+2. Use [binary quantization](#binary-quantization) for indexes (with re-ranking for search)
+
+Scale vertically by increasing memory, CPU, and storage on a single instance. Use existing tools to [tune parameters](#tuning) and [monitor performance](#monitoring).
+
+Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus), [PgDog](https://github.com/pgdogdev/pgdog), or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus/example.py)).
+
 ## Monitoring
 
-Monitor performance with [pg_stat_statements](https://www.postgresql.org/docs/current/pgstatstatements.html) (be sure to add it to `shared_preload_libraries`).
-
-```sql
-CREATE EXTENSION pg_stat_statements;
-```
-
-Get the most time-consuming queries with:
-
-```sql
-SELECT query, calls, ROUND((total_plan_time + total_exec_time) / calls) AS avg_time_ms,
-    ROUND((total_plan_time + total_exec_time) / 60000) AS total_time_min
-    FROM pg_stat_statements ORDER BY total_plan_time + total_exec_time DESC LIMIT 20;
-```
+Use existing tools like [pg_stat_statements](https://www.postgresql.org/docs/current/pgstatstatements.html) or [PgHero](https://github.com/ankane/pghero) to monitor performance.
 
 Monitor recall by comparing results from approximate search with exact search.
 
@@ -757,14 +762,6 @@ SELECT ...
 COMMIT;
 ```
 
-## Scaling
-
-Scale pgvector the same way you scale Postgres.
-
-Scale vertically by increasing memory, CPU, and storage on a single instance. Use existing tools to [tune parameters](#tuning) and [monitor performance](#monitoring).
-
-Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus) or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus/example.py)).
-
 ## Languages
 
 Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
@@ -878,6 +875,8 @@ No, but like other index types, you’ll likely see better performance if they d
 SELECT pg_size_pretty(pg_relation_size('index_name'));
 ```
 
+Use [half-precision indexing](#half-precision-indexing) or [binary quantization](#binary-quantization) for smaller indexes.
+
 ## Troubleshooting
 
 #### Why isn’t a query using an index?
@@ -1152,23 +1151,23 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 
 Supported tags are:
 
-- `pg18-trixie`, `0.8.2-pg18-trixie`
-- `pg18-bookworm`, `0.8.2-pg18-bookworm`, `pg18`, `0.8.2-pg18`
-- `pg17-trixie`, `0.8.2-pg17-trixie`
-- `pg17-bookworm`, `0.8.2-pg17-bookworm`, `pg17`, `0.8.2-pg17`
-- `pg16-trixie`, `0.8.2-pg16-trixie`
-- `pg16-bookworm`, `0.8.2-pg16-bookworm`, `pg16`, `0.8.2-pg16`
-- `pg15-trixie`, `0.8.2-pg15-trixie`
-- `pg15-bookworm`, `0.8.2-pg15-bookworm`, `pg15`, `0.8.2-pg15`
-- `pg14-trixie`, `0.8.2-pg14-trixie`
-- `pg14-bookworm`, `0.8.2-pg14-bookworm`, `pg14`, `0.8.2-pg14`
-- `pg13-trixie`, `0.8.2-pg13-trixie`
-- `pg13-bookworm`, `0.8.2-pg13-bookworm`, `pg13`, `0.8.2-pg13`
+- `pg18-trixie`, `0.8.3-pg18-trixie`
+- `pg18-bookworm`, `0.8.3-pg18-bookworm`, `pg18`, `0.8.3-pg18`
+- `pg17-trixie`, `0.8.3-pg17-trixie`
+- `pg17-bookworm`, `0.8.3-pg17-bookworm`, `pg17`, `0.8.3-pg17`
+- `pg16-trixie`, `0.8.3-pg16-trixie`
+- `pg16-bookworm`, `0.8.3-pg16-bookworm`, `pg16`, `0.8.3-pg16`
+- `pg15-trixie`, `0.8.3-pg15-trixie`
+- `pg15-bookworm`, `0.8.3-pg15-bookworm`, `pg15`, `0.8.3-pg15`
+- `pg14-trixie`, `0.8.3-pg14-trixie`
+- `pg14-bookworm`, `0.8.3-pg14-bookworm`, `pg14`, `0.8.3-pg14`
+- `pg13-trixie`, `0.8.3-pg13-trixie`
+- `pg13-bookworm`, `0.8.3-pg13-bookworm`, `pg13`, `0.8.3-pg13`
 
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.8.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.3 https://github.com/pgvector/pgvector.git
 cd pgvector
 docker build --pull --build-arg PG_MAJOR=18 -t myuser/pgvector .
 ```
@@ -1330,7 +1329,7 @@ make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
 To enable benchmarking:
 
 ```sh
-make clean && PG_CFLAGS="-DIVFFLAT_BENCH" make && make install
+make clean && PG_CFLAGS="-DHNSW_BENCH -DIVFFLAT_BENCH" make && make install
 ```
 
 To show memory usage:

sql/vector--0.8.2--0.8.3.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.3'" to load this file. \quit

src/bitutils.c

@@ -31,10 +31,12 @@
 #define BIT_TARGET_CLONES
 #endif
 
-/* Use built-ins when possible for inlining */
-#if defined(HAVE__BUILTIN_POPCOUNT) && defined(HAVE_LONG_INT_64)
+/* Use built-ins when possible for Postgres < 19 for inlining */
+#if PG_VERSION_NUM >= 190000
+#define popcount64(x) pg_popcount64(x)
+#elif defined(HAVE__BUILTIN_POPCOUNT) && (defined(HAVE_LONG_INT_64) || SIZEOF_LONG == 8)
 #define popcount64(x) __builtin_popcountl(x)
-#elif defined(HAVE__BUILTIN_POPCOUNT) && defined(HAVE_LONG_LONG_INT_64)
+#elif defined(HAVE__BUILTIN_POPCOUNT) && (defined(HAVE_LONG_LONG_INT_64) || SIZEOF_LONG_LONG == 8)
 #define popcount64(x) __builtin_popcountll(x)
 #elif !defined(_MSC_VER)
 /* Fails to resolve with MSVC */
@@ -169,7 +171,7 @@ BitJaccardDistanceAvx512Popcount(uint32 bytes, unsigned char *ax, unsigned char
 #endif
 
 TARGET_XSAVE static bool
-SupportsAvx512Popcount()
+SupportsAvx512Popcount(void)
 {
 	unsigned int exx[4] = {0, 0, 0, 0};
 

src/hnsw.c

@@ -13,6 +13,7 @@
 #include "hnsw.h"
 #include "miscadmin.h"
 #include "nodes/pg_list.h"
+#include "storage/lwlock.h"
 #include "utils/float.h"
 #include "utils/guc.h"
 #include "utils/relcache.h"

src/hnsw.h

@@ -10,10 +10,18 @@
 #include "lib/pairingheap.h"
 #include "nodes/execnodes.h"
 #include "port.h"				/* for random() */
+#include "storage/bufpage.h"
+#include "storage/condition_variable.h"
+#include "storage/lwlock.h"
+#include "storage/s_lock.h"
 #include "utils/relptr.h"
 #include "utils/sampling.h"
 #include "vector.h"
 
+#ifdef HNSW_BENCH
+#include "portability/instr_time.h"
+#endif
+
 #if PG_VERSION_NUM >= 190000
 typedef Pointer Item;
 #endif
@@ -74,6 +82,21 @@ typedef Pointer Item;
 #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)
@@ -408,6 +431,7 @@ typedef struct HnswVacuumState
 	BufferAccessStrategy bas;
 	HnswNeighborTuple ntup;
 	HnswElementData highestPoint;
+	HnswElementData fallbackPoint;
 
 	/* Memory */
 	MemoryContext tmpCtx;

src/hnswbuild.c

@@ -36,6 +36,8 @@
  */
 #include "postgres.h"
 
+#include <limits.h>
+
 #include "access/genam.h"
 #include "access/parallel.h"
 #include "access/relscan.h"
@@ -52,6 +54,7 @@
 #include "nodes/execnodes.h"
 #include "optimizer/optimizer.h"
 #include "storage/bufmgr.h"
+#include "storage/condition_variable.h"
 #include "tcop/tcopprot.h"
 #include "utils/datum.h"
 #include "utils/memutils.h"
@@ -800,7 +803,11 @@ HnswParallelScanAndInsert(Relation heapRel, Relation indexRel, HnswShared * hnsw
 	buildstate.hnswarea = hnswarea;
 	InitAllocator(&buildstate.allocator, &HnswSharedMemoryAlloc, &buildstate);
 	scan = table_beginscan_parallel(heapRel,
-									ParallelTableScanFromHnswShared(hnswshared));
+									ParallelTableScanFromHnswShared(hnswshared)
+#if PG_VERSION_NUM >= 190000
+									,SO_NONE
+#endif
+		);
 	reltuples = table_index_build_scan(heapRel, indexRel, indexInfo,
 									   true, progress, BuildCallback,
 									   (void *) &buildstate, scan);

src/hnswinsert.c

@@ -6,6 +6,7 @@
 #include "nodes/execnodes.h"
 #include "storage/bufmgr.h"
 #include "storage/lmgr.h"
+#include "storage/lwlock.h"
 #include "utils/datum.h"
 #include "utils/memutils.h"
 #include "utils/rel.h"

src/hnswscan.c

@@ -1,5 +1,7 @@
 #include "postgres.h"
 
+#include <limits.h>
+
 #include "access/genam.h"
 #include "access/relscan.h"
 #include "hnsw.h"
@@ -151,7 +153,7 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
 	/* Calculate max memory */
 	/* Add 256 extra bytes to fill last block when close */
 	maxMemory = (double) work_mem * hnsw_scan_mem_multiplier * 1024.0 + 256;
-	so->maxMemory = Min(maxMemory, (double) SIZE_MAX);
+	so->maxMemory = Min(maxMemory, (double) (SIZE_MAX / 2));
 
 	scan->opaque = so;
 

src/hnswutils.c

@@ -546,6 +546,9 @@ HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, Hns
 
 	Assert(HnswIsElementTuple(etup));
 
+	if (unlikely(etup->deleted))
+		elog(ERROR, "cannot load deleted element");
+
 	/* Calculate distance */
 	if (distance != NULL)
 	{

src/hnswvacuum.c

@@ -37,17 +37,20 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 {
 	BlockNumber blkno = HNSW_HEAD_BLKNO;
 	HnswElement highestPoint = &vacuumstate->highestPoint;
+	HnswElement fallbackPoint = &vacuumstate->fallbackPoint;
 	Relation	index = vacuumstate->index;
 	BufferAccessStrategy bas = vacuumstate->bas;
-	HnswElement entryPoint = HnswGetEntryPoint(vacuumstate->index);
 	IndexBulkDeleteResult *stats = vacuumstate->stats;
 
-	/* Store separately since highestPoint.level is uint8 */
+	/* Store separately since HnswElement level is uint8 */
 	int			highestLevel = -1;
+	int			fallbackLevel = -1;
 
-	/* Initialize highest point */
+	/* Initialize highest point and fallback point */
 	highestPoint->blkno = InvalidBlockNumber;
 	highestPoint->offno = InvalidOffsetNumber;
+	fallbackPoint->blkno = InvalidBlockNumber;
+	fallbackPoint->offno = InvalidOffsetNumber;
 
 	while (BlockNumberIsValid(blkno))
 	{
@@ -119,14 +122,31 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 				tidhash_insert(vacuumstate->deleted, ip, &found);
 				Assert(!found);
 			}
-			else if (etup->level > highestLevel && !(entryPoint != NULL && blkno == entryPoint->blkno && offno == entryPoint->offno))
+			else if (etup->level > highestLevel)
 			{
-				/* Keep track of highest non-entry point */
+				if (BlockNumberIsValid(highestPoint->blkno))
+				{
+					/* Current highest point becomes fallback */
+					fallbackPoint->blkno = highestPoint->blkno;
+					fallbackPoint->offno = highestPoint->offno;
+					fallbackPoint->level = highestPoint->level;
+					fallbackLevel = highestLevel;
+				}
+
+				/* Keep track of highest point */
 				highestPoint->blkno = blkno;
 				highestPoint->offno = offno;
 				highestPoint->level = etup->level;
 				highestLevel = etup->level;
 			}
+			else if (etup->level > fallbackLevel)
+			{
+				/* Keep track of second highest point */
+				fallbackPoint->blkno = blkno;
+				fallbackPoint->offno = offno;
+				fallbackPoint->level = etup->level;
+				fallbackLevel = etup->level;
+			}
 		}
 
 		blkno = HnswPageGetOpaque(page)->nextblkno;
@@ -138,6 +158,10 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
 
 		UnlockReleaseBuffer(buf);
 	}
+
+#ifdef HNSW_MEMORY
+	elog(INFO, "memory: %zu KB", MemoryContextMemAllocated(CurrentMemoryContext, true) / 1024);
+#endif
 }
 
 /*
@@ -269,12 +293,27 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 		/* Get a shared lock */
 		LockPage(index, HNSW_UPDATE_LOCK, ShareLock);
 
-		/* Load element */
-		HnswLoadElement(highestPoint, NULL, NULL, index, support, true, NULL);
+		/* Get latest entry point */
+		entryPoint = HnswGetEntryPoint(index);
 
-		/* Repair if needed */
-		if (NeedsUpdated(vacuumstate, highestPoint))
-			RepairGraphElement(vacuumstate, highestPoint, HnswGetEntryPoint(index));
+		/* Use fallback point if highest point is entry point */
+		if (entryPoint != NULL && entryPoint->blkno == highestPoint->blkno && entryPoint->offno == highestPoint->offno)
+		{
+			highestPoint = &vacuumstate->fallbackPoint;
+
+			if (!BlockNumberIsValid(highestPoint->blkno))
+				highestPoint = NULL;
+		}
+
+		if (highestPoint != NULL)
+		{
+			/* Load element */
+			HnswLoadElement(highestPoint, NULL, NULL, index, support, true, NULL);
+
+			/* Repair if needed */
+			if (NeedsUpdated(vacuumstate, highestPoint))
+				RepairGraphElement(vacuumstate, highestPoint, entryPoint);
+		}
 
 		/* Release lock */
 		UnlockPage(index, HNSW_UPDATE_LOCK, ShareLock);
@@ -441,6 +480,99 @@ RepairGraph(HnswVacuumState * vacuumstate)
 		/* Reset memory context */
 		MemoryContextSwitchTo(oldCtx);
 		MemoryContextReset(vacuumstate->tmpCtx);
+
+#ifdef HNSW_VACUUM_PROGRESS
+		if (!BlockNumberIsValid(blkno) || (blkno - HNSW_HEAD_BLKNO) % 1000 == 0)
+		{
+			BlockNumber totalBlocks = RelationGetNumberOfBlocks(index);
+			BlockNumber currentBlocks = BlockNumberIsValid(blkno) ? blkno : totalBlocks;
+
+			elog(INFO, "hnsw vacuum progress: %.1f%%", 100.0 * currentBlocks / totalBlocks);
+		}
+#endif
+	}
+}
+
+/*
+ * Confirm graph was repaired
+ */
+static void
+ConfirmRepaired(HnswVacuumState * vacuumstate)
+{
+	BlockNumber blkno = HNSW_HEAD_BLKNO;
+	Relation	index = vacuumstate->index;
+	BufferAccessStrategy bas = vacuumstate->bas;
+
+	while (BlockNumberIsValid(blkno))
+	{
+		Buffer		buf;
+		Page		page;
+		OffsetNumber offno;
+		OffsetNumber maxoffno;
+
+		vacuum_delay_point();
+
+		buf = ReadBufferExtended(index, MAIN_FORKNUM, blkno, RBM_NORMAL, bas);
+		LockBuffer(buf, BUFFER_LOCK_SHARE);
+		page = BufferGetPage(buf);
+		maxoffno = PageGetMaxOffsetNumber(page);
+
+		/* Iterate over nodes */
+		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
+		{
+			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
+			HnswNeighborTuple ntup;
+			Buffer		nbuf;
+			Page		npage;
+			BlockNumber neighborPage;
+			OffsetNumber neighborOffno;
+
+			/* Skip neighbor tuples */
+			if (!HnswIsElementTuple(etup))
+				continue;
+
+			/* Skip if being deleted */
+			if (!ItemPointerIsValid(&etup->heaptids[0]))
+				continue;
+
+			/* Get neighbor page */
+			neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
+			neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
+
+			if (neighborPage == blkno)
+			{
+				nbuf = buf;
+				npage = page;
+			}
+			else
+			{
+				nbuf = ReadBufferExtended(index, MAIN_FORKNUM, neighborPage, RBM_NORMAL, bas);
+				LockBuffer(nbuf, BUFFER_LOCK_SHARE);
+				npage = BufferGetPage(nbuf);
+			}
+
+			ntup = (HnswNeighborTuple) PageGetItem(npage, PageGetItemId(npage, neighborOffno));
+
+			/* Check neighbors */
+			for (int i = 0; i < ntup->count; i++)
+			{
+				ItemPointer indextid = &ntup->indextids[i];
+
+				if (!ItemPointerIsValid(indextid))
+					continue;
+
+				/* Check if in deleted list */
+				if (DeletedContains(vacuumstate->deleted, indextid))
+					elog(ERROR, "hnsw graph not repaired");
+			}
+
+			if (nbuf != buf)
+				UnlockReleaseBuffer(nbuf);
+		}
+
+		blkno = HnswPageGetOpaque(page)->nextblkno;
+
+		UnlockReleaseBuffer(buf);
 	}
 }
 
@@ -634,13 +766,16 @@ hnswbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 	InitVacuumState(&vacuumstate, info, stats, callback, callback_state);
 
 	/* Pass 1: Remove heap TIDs */
-	RemoveHeapTids(&vacuumstate);
+	HnswBench("RemoveHeapTids", RemoveHeapTids(&vacuumstate));
 
 	/* Pass 2: Repair graph */
-	RepairGraph(&vacuumstate);
+	HnswBench("RepairGraph", RepairGraph(&vacuumstate));
 
-	/* Pass 3: Mark as deleted */
-	MarkDeleted(&vacuumstate);
+	/* Pass 3: Confirm repaired */
+	HnswBench("ConfirmRepaired", ConfirmRepaired(&vacuumstate));
+
+	/* Pass 4: Mark as deleted */
+	HnswBench("MarkDeleted", MarkDeleted(&vacuumstate));
 
 	FreeVacuumState(&vacuumstate);
 

src/ivfbuild.c

@@ -22,6 +22,7 @@
 #include "nodes/execnodes.h"
 #include "optimizer/optimizer.h"
 #include "storage/bufmgr.h"
+#include "storage/condition_variable.h"
 #include "tcop/tcopprot.h"
 #include "utils/memutils.h"
 #include "utils/rel.h"
@@ -62,15 +63,13 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
 
 	/*
-	 * Normalize with KMEANS_NORM_PROC since spherical distance function
-	 * expects unit vectors
+	 * Check with KMEANS_NORM_PROC that the value can be normalized since
+	 * spherical distance function expects unit vectors
 	 */
 	if (buildstate->kmeansnormprocinfo != NULL)
 	{
 		if (!IvfflatCheckNorm(buildstate->kmeansnormprocinfo, buildstate->collation, value))
 			return;
-
-		value = IvfflatNormValue(buildstate->typeInfo, buildstate->collation, value);
 	}
 
 	if (samples->length < targsamples)
@@ -81,7 +80,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 	else
 	{
 		if (buildstate->rowstoskip < 0)
-			buildstate->rowstoskip = reservoir_get_next_S(&buildstate->rstate, samples->length, targsamples);
+			buildstate->rowstoskip = reservoir_get_next_S(&buildstate->rstate, buildstate->samplerows, targsamples);
 
 		if (buildstate->rowstoskip <= 0)
 		{
@@ -97,6 +96,9 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 
 		buildstate->rowstoskip -= 1;
 	}
+
+	/* Increment after reservoir_get_next_S */
+	buildstate->samplerows += 1;
 }
 
 /*
@@ -133,6 +135,7 @@ SampleRows(IvfflatBuildState * buildstate)
 	int			targsamples = buildstate->samples->maxlen;
 	BlockNumber totalblocks = RelationGetNumberOfBlocks(buildstate->heap);
 
+	buildstate->samplerows = 0;
 	buildstate->rowstoskip = -1;
 
 	BlockSampler_Init(&buildstate->bs, totalblocks, targsamples, RandomInt());
@@ -142,8 +145,24 @@ SampleRows(IvfflatBuildState * buildstate)
 	{
 		BlockNumber targblock = BlockSampler_Next(&buildstate->bs);
 
+		/* Set anyvisible to false like table_index_build_scan */
 		table_index_build_range_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
-									 false, true, false, targblock, 1, SampleCallback, (void *) buildstate, NULL);
+									 false, false, false, targblock, 1, SampleCallback, (void *) buildstate, NULL);
+	}
+
+	/* Normalize if needed */
+	if (buildstate->kmeansnormprocinfo != NULL)
+	{
+		VectorArray samples = buildstate->samples;
+
+		for (int i = 0; i < samples->length; i++)
+		{
+			Datum		value = PointerGetDatum(VectorArrayGet(samples, i));
+			Datum		normValue = IvfflatNormValue(buildstate->typeInfo, buildstate->collation, value);
+
+			VectorArraySet(samples, i, DatumGetPointer(normValue));
+			pfree(DatumGetPointer(normValue));
+		}
 	}
 }
 
@@ -374,9 +393,13 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
 	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
 	TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 3, "vector", TupleDescAttr(buildstate->tupdesc, 0)->atttypid, -1, 0);
+#if PG_VERSION_NUM >= 190000
+	TupleDescFinalize(buildstate->sortdesc);
+#endif
 
 	buildstate->slot = MakeSingleTupleTableSlot(buildstate->sortdesc, &TTSOpsVirtual);
 
+	/* TODO Ensure within maintenance_work_mem */
 	buildstate->centers = VectorArrayInit(buildstate->lists, buildstate->dimensions, buildstate->typeInfo->itemSize(buildstate->dimensions));
 	buildstate->listInfo = palloc(sizeof(ListInfo) * buildstate->lists);
 
@@ -435,7 +458,7 @@ ComputeCenters(IvfflatBuildState * buildstate)
 	buildstate->samples = VectorArrayInit(numSamples, buildstate->dimensions, buildstate->centers->itemsize);
 	if (buildstate->heap != NULL)
 	{
-		SampleRows(buildstate);
+		IvfflatBench("sample rows", SampleRows(buildstate));
 
 		if (buildstate->samples->length < buildstate->lists)
 		{
@@ -650,7 +673,11 @@ IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, S
 	ivfspool->sortstate = InitBuildSortState(buildstate.sortdesc, sortmem, coordinate);
 	buildstate.sortstate = ivfspool->sortstate;
 	scan = table_beginscan_parallel(ivfspool->heap,
-									ParallelTableScanFromIvfflatShared(ivfshared));
+									ParallelTableScanFromIvfflatShared(ivfshared)
+#if PG_VERSION_NUM >= 190000
+									,SO_NONE
+#endif
+		);
 	reltuples = table_index_build_scan(ivfspool->heap, ivfspool->index, indexInfo,
 									   true, progress, BuildCallback,
 									   (void *) &buildstate, scan);

src/ivfflat.h

@@ -9,6 +9,7 @@
 #include "lib/pairingheap.h"
 #include "nodes/execnodes.h"
 #include "port.h"				/* for random() */
+#include "storage/condition_variable.h"
 #include "utils/sampling.h"
 #include "utils/tuplesort.h"
 #include "vector.h"
@@ -213,7 +214,8 @@ typedef struct IvfflatBuildState
 	/* Sampling */
 	BlockSamplerData bs;
 	ReservoirStateData rstate;
-	int			rowstoskip;
+	double		samplerows;
+	double		rowstoskip;
 
 	/* Sorting */
 	Tuplesortstate *sortstate;

src/ivfscan.c

@@ -298,6 +298,9 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	so->tupdesc = CreateTemplateTupleDesc(2);
 	TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
 	TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "heaptid", TIDOID, -1, 0);
+#if PG_VERSION_NUM >= 190000
+	TupleDescFinalize(so->tupdesc);
+#endif
 
 	/* Prep sort */
 	so->sortstate = InitScanSortState(so->tupdesc);

src/sparsevec.c

@@ -182,10 +182,10 @@ sparsevec_isspace(char ch)
 static int
 CompareIndices(const void *a, const void *b)
 {
-	if (((SparseInputElement *) a)->index < ((SparseInputElement *) b)->index)
+	if (((const SparseInputElement *) a)->index < ((const SparseInputElement *) b)->index)
 		return -1;
 
-	if (((SparseInputElement *) a)->index > ((SparseInputElement *) b)->index)
+	if (((const SparseInputElement *) a)->index > ((const SparseInputElement *) b)->index)
 		return 1;
 
 	return 0;

src/vector.c

@@ -40,7 +40,7 @@
 #endif
 
 #if PG_VERSION_NUM >= 180000
-PG_MODULE_MAGIC_EXT(.name = "vector",.version = "0.8.2");
+PG_MODULE_MAGIC_EXT(.name = "vector", .version = "0.8.3");
 #else
 PG_MODULE_MAGIC;
 #endif

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.8.2'
+default_version = '0.8.3'
 module_pathname = '$libdir/vector'
 relocatable = true