liuyu/pgvector
1
0
Fork 0
mirror of https://github.com/pgvector/pgvector.git synced 2026-07-02 18:50:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

Reduced memory usage for small tables for IVFFlat index builds - resolves #995 and resolves #996

Browse Source 
Co-authored-by: Itai Spiegel <itai@mave.com>
This commit is contained in:
Andrew Kane
2026-07-01 11:54:21 -07:00
parent 1d458ad5d7
commit b383e4d191
4 changed files with 54 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
CHANGELOG.md
View File

@@ -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

7
src/ivfbuild.c
View File

@@ -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);

24
test/expected/ivfflat_vector.out
View File

@@ -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;

19
test/sql/ivfflat_vector.sql
View File

@@ -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;