Benchmarks

Grove is designed for near-raw performance. These benchmarks compare Grove against raw database/sql, Bun, and GORM using SQLite in-memory databases.

Running Benchmarks

# Run all benchmarks
make bench

# Generate markdown report
make bench-report

# Update README.md and docs with latest results
make bench-update

Results

Benchmarks generated on 2026-02-22 with go1.25.7 on darwin/arm64. Each benchmark ran 5 times; values are averages.

Insert

Library	ns/op	B/op	allocs/op	vs Raw SQL
Raw SQL	4,015	880	20	baseline
Grove	4,381	1,283	28	+9.1%
Bun	8,459	5,470	27	+110.7%
GORM	10,265	4,954	66	+155.7%

SelectOne

Library	ns/op	B/op	allocs/op	vs Raw SQL
Raw SQL	4,747	1,096	39	baseline
Grove	5,575	1,458	34	+17.4%
Bun	6,695	5,944	43	+41.0%
GORM	8,070	4,383	81	+70.0%

SelectMulti

Library	ns/op	B/op	allocs/op	vs Raw SQL
Raw SQL	58,079	20,984	388	baseline
Grove	45,686	23,413	584	-21.3%
Bun	67,184	23,192	395	+15.7%
GORM	83,217	28,570	765	+43.3%

Update

Library	ns/op	B/op	allocs/op	vs Raw SQL
Raw SQL	2,821	520	13	baseline
Grove	3,617	1,010	21	+28.2%
Bun	4,074	5,205	19	+44.4%
GORM	5,468	4,020	49	+93.8%

Delete

Library	ns/op	B/op	allocs/op	vs Raw SQL
Raw SQL	3,479	232	8	baseline
Grove	5,083	553	13	+46.1%
Bun	5,147	4,880	12	+47.9%
GORM	6,660	2,856	36	+91.4%

BulkInsert100

Library	ns/op	B/op	allocs/op	vs Raw SQL
Raw SQL	151,006	60,141	1524	baseline
Grove	105,572	42,111	1421	-30.1%
Bun	143,532	30,389	224	-4.9%
GORM	187,914	93,706	1251	+24.4%

BulkInsert1000

Library	ns/op	B/op	allocs/op	vs Raw SQL
Raw SQL	1,401,957	605,384	16513	baseline
Grove	967,190	409,753	14021	-31.0%
Bun	1,359,612	408,694	2033	-3.0%
GORM	1,782,186	894,166	12051	+27.1%

BuildSelect

Variant	ns/op	B/op	allocs/op
Grove	453	864	11
Bun	825	1,744	17

BuildInsert

Variant	ns/op	B/op	allocs/op
Grove	842	930	19
Bun	1,012	1,425	17

BuildUpdate

Variant	ns/op	B/op	allocs/op
Grove	380	736	11
Bun	597	1,184	14

SchemaCache

Variant	ns/op	B/op	allocs/op
CacheHit	11	0	0
ColdStart	3,123	3,728	75

TagResolution

Variant	ns/op	B/op	allocs/op
GroveTags	3,101	3,728	75
BunFallback	3,398	3,736	74

Methodology

Database: SQLite in-memory (:memory:) -- isolates ORM overhead from network/disk
Runs: 5 iterations per benchmark (-count=5) for statistical stability
Model: User struct with 6 fields (ID, Name, Email, Age, Active, CreatedAt)
Metrics: ns/op, B/op (bytes allocated), allocs/op (heap allocations)
Overhead: Computed as (ORM - RawSQL) / RawSQL * 100%

What's Measured

Category	Description
Insert	Single row INSERT
SelectOne	SELECT single row by PK
SelectMulti	SELECT 50 rows with WHERE + LIMIT
Update	UPDATE 2 columns by PK
Delete	DELETE single row by PK
BulkInsert100	INSERT 100 rows in one operation
BulkInsert1000	INSERT 1000 rows in one operation
BuildSelect	SQL generation only (no I/O)
BuildInsert	SQL generation only (no I/O)
BuildUpdate	SQL generation only (no I/O)
SchemaCache	Schema registry cache hit vs cold start
TagResolution	Grove tags vs bun tag fallback resolution

Benchmarks

On this page