pgvector.git / summary / log / commit / refs

commit    b383e4d191e7946ad8ecfee6834ec847cc5ebf2d
Author:   Andrew Kane <andrew@ankane.org>
Date:     Wed Jul 01 18:54:21 2026 +0000

    Reduced memory usage for small tables for IVFFlat index builds - resolves #995 and resolves #996
    
    Co-authored-by: Itai Spiegel <itai@mave.com>


 CHANGELOG.md                     |  4 ++++
 src/ivfbuild.c                   |  7 +++++++
 test/expected/ivfflat_vector.out | 24 ++++++++++++++++++++++++
 test/sql/ivfflat_vector.sql      | 19 +++++++++++++++++++
 4 files changed, 54 insertions(+)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 44bca3fa..e62cd9ee 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,3 +1,7 @@
+## 0.8.5 (unreleased)
+
+- Reduced memory usage for small tables for IVFFlat index builds
+
 ## 0.8.4 (2026-06-30)
 
 - Fixed `hnsw graph not repaired` error with HNSW vacuuming
diff --git a/src/ivfbuild.c b/src/ivfbuild.c
index e35c3521..7515c437 100644
--- a/src/ivfbuild.c
+++ b/src/ivfbuild.c
@@ -447,6 +447,13 @@ ComputeCenters(IvfflatBuildState * buildstate)
 	/* Skip samples for unlogged table */
 	if (buildstate->heap == NULL)
 		numSamples = 1;
+	else
+	{
+		int64		maxTuples = (int64) RelationGetNumberOfBlocks(buildstate->heap) * MaxHeapTuplesPerPage;
+
+		/* Save memory since will not have more than max tuples */
+		numSamples = Max(Min(numSamples, maxTuples), 1);
+	}
 
 	/* Sample rows */
 	buildstate->memoryUsed += VECTOR_ARRAY_SIZE(numSamples, buildstate->itemsize);
diff --git a/test/expected/ivfflat_vector.out b/test/expected/ivfflat_vector.out
index aad8226c..23e1c014 100644
--- a/test/expected/ivfflat_vector.out
+++ b/test/expected/ivfflat_vector.out
@@ -173,3 +173,27 @@ ERROR:  0 is outside the valid range for parameter "ivfflat.max_probes" (1 .. 32
 SET ivfflat.max_probes = 32769;
 ERROR:  32769 is outside the valid range for parameter "ivfflat.max_probes" (1 .. 32768)
 DROP TABLE t;
+-- memory
+CREATE TABLE t (val vector(2000));
+CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 4096);
+NOTICE:  ivfflat index created with little data
+DETAIL:  This will cause low recall.
+HINT:  Drop the index until the table has more data.
+DROP TABLE t;
+SET maintenance_work_mem = '1MB';
+CREATE TABLE t (val vector(2000));
+CREATE INDEX ON t USING ivfflat (val vector_l2_ops);
+NOTICE:  ivfflat index created with little data
+DETAIL:  This will cause low recall.
+HINT:  Drop the index until the table has more data.
+DROP TABLE t;
+RESET maintenance_work_mem;
+SET maintenance_work_mem = '5MB';
+CREATE TABLE t (val vector(2000));
+INSERT INTO t (val) VALUES (array_fill(0, ARRAY[2000]));
+CREATE INDEX ON t USING ivfflat (val vector_l2_ops);
+NOTICE:  ivfflat index created with little data
+DETAIL:  This will cause low recall.
+HINT:  Drop the index until the table has more data.
+DROP TABLE t;
+RESET maintenance_work_mem;
diff --git a/test/sql/ivfflat_vector.sql b/test/sql/ivfflat_vector.sql
index 4c20d2fa..826f6f8b 100644
--- a/test/sql/ivfflat_vector.sql
+++ b/test/sql/ivfflat_vector.sql
@@ -97,3 +97,22 @@ SET ivfflat.max_probes = 0;
 SET ivfflat.max_probes = 32769;
 
 DROP TABLE t;
+
+-- memory
+
+CREATE TABLE t (val vector(2000));
+CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 4096);
+DROP TABLE t;
+
+SET maintenance_work_mem = '1MB';
+CREATE TABLE t (val vector(2000));
+CREATE INDEX ON t USING ivfflat (val vector_l2_ops);
+DROP TABLE t;
+RESET maintenance_work_mem;
+
+SET maintenance_work_mem = '5MB';
+CREATE TABLE t (val vector(2000));
+INSERT INTO t (val) VALUES (array_fill(0, ARRAY[2000]));
+CREATE INDEX ON t USING ivfflat (val vector_l2_ops);
+DROP TABLE t;
+RESET maintenance_work_mem;


[parent: 1d458ad5d737]