Performance Benchmarks

Parse-stage performance of the GoSQLX parser, re-measured with the in-repo benchmarks.

Measured 2026-10-22 · Apple M4 · Go 1.27 · parse stage only

2.87M ops/sec

Simple SELECT, parse stage (Apple M4)

349 ns/op

Simple SELECT parse latency

Clean

Race detector in CI

Parse Benchmarks

Parse-stage throughput (Apple M4, Go 1.27)

Simple SELECTGoSQLX2.87M ops/sec

INSERTGoSQLX2.46M ops/sec

CTEGoSQLX1.55M ops/sec

Window FunctionGoSQLX1.58M ops/sec

Complex QueryGoSQLX810K ops/sec

Measured on Apple M4 · Go 1.27 · parse stage · median of 10 runs · 2026-10-22 · source BenchmarkPerformanceBaseline. Reproduce

GoSQLX Parse Benchmarks
Benchmark	Query Type	Apple M4 (median)	CI tolerance baseline
Simple SELECT	SELECT id, name FROM users	349 ns/op (2.87M ops/sec)	≤ 800 ns/op
INSERT	Simple INSERT statement	406 ns/op (2.46M ops/sec)	≤ 950 ns/op
CTE	WITH clause	644 ns/op (1.55M ops/sec)	≤ 1400 ns/op
Window Function	ROW_NUMBER() OVER (PARTITION BY ... ORDER BY ...)	631 ns/op (1.58M ops/sec)	≤ 1500 ns/op
Complex Query	JOIN, WHERE, ORDER BY, LIMIT	1234 ns/op (810K ops/sec)	≤ 2900 ns/op

← Swipe to see all columns →

Competitor Comparison

End-to-end stage (SQL text to AST) · AMD EPYC 9V74 80-Core Processor (linux/amd64) · go1.26.9 · median of 10 runs · measured 2026-10-22 · commit e9c5db9

Median nanoseconds per statement, end-to-end, lower is faster
Query	GoSQLX	pg_query_go	Vitess sqlparser	TiDB parser	xwb1989/sqlparser
Simple SELECT	2,138 ns	10,811 ns	1,587 ns	2,801 ns	7,897 ns
SELECT with WHERE, ORDER BY, LIMIT	4,439 ns	32,990 ns	3,626 ns	7,218 ns	11,826 ns
Two-table JOIN	6,919 ns	43,528 ns	4,700 ns	8,535 ns	25,931 ns
GROUP BY with HAVING	4,818 ns	25,832 ns	3,090 ns	6,572 ns	12,192 ns
IN subquery	4,076 ns	23,527 ns	3,977 ns	6,279 ns	21,828 ns
Common table expression	5,829 ns	33,530 ns	4,415 ns	9,158 ns	not supported
Window function	4,062 ns	21,010 ns	3,903 ns	6,022 ns	not supported
Multi-JOIN analytical query	18,293 ns	116,213 ns	12,305 ns	21,303 ns	42,654 ns
INSERT VALUES	3,185 ns	13,373 ns	2,202 ns	3,914 ns	9,105 ns
UPDATE with WHERE	2,641 ns	14,108 ns	2,044 ns	3,922 ns	9,033 ns
DELETE with WHERE	1,772 ns	8,876 ns	1,343 ns	2,312 ns	6,736 ns

Median ns per statement; lower is faster and the fastest library per query is in bold. Hover a query name to see its SQL.

← Swipe to see all columns →

Read with care. The libraries implement different SQL dialects and build different ASTs (pg_query_go wraps the real PostgreSQL parser through CGo and converts its tree to Go structs; Vitess, TiDB and xwb1989 implement MySQL grammars). Each figure is the end-to-end cost of turning the same SQL text into that library's AST on the stated hardware, not identical work.

GoSQLX: github.com/ajitpratap0/GoSQLX commit e9c5db91a8ed · multi-dialect (default settings)
pg_query_go: github.com/pganalyze/pg_query_go/v6 v6.2.5 · PostgreSQL (the real PostgreSQL grammar via CGo)
Vitess sqlparser: vitess.io/vitess v0.24.5 · MySQL
TiDB parser: github.com/pingcap/tidb/pkg/parser v0.0.0-20261015143638-7193401b6746 · MySQL (TiDB)
xwb1989/sqlparser: github.com/xwb1989/sqlparser v0.0.0-20180606152119-120387863bf2 · MySQL (2018 Vitess fork, unmaintained)

Reproduce with task bench:publish using the benchmark harness, which pins every library version and keeps the raw output for benchstat. See also the feature comparison.

Reproduce these numbers

Clone the repository, then run the commands below. Absolute numbers depend on your hardware; compare ratios, not raw ns/op.

Parse stage (the charts above)

go test -run='^$' -bench=BenchmarkPerformanceBaseline -benchmem -count=20 ./pkg/sql/parser/

Summarise the 10 runs with benchstat to get the median.

Competitor comparison (end-to-end)
```
cd benchmarks/compare && go test -run=^$ -bench=^BenchmarkParse$ -benchmem -count=10 .
```
Needs CGO for pg_query_go. Summarise with benchstat; `task bench:publish` runs this and writes results/latest.json and latest.txt.
CI regression check (not a measurement)
```
go test -run TestPerformanceRegression ./pkg/sql/parser/
```
Compares against the loose ceilings in the baselines file; skipped under the race detector.

Parse benchmark source: pkg/sql/parser/performance_regression_test.go
CI baselines: performance_baselines.json
Competitor harness: benchmarks/compare
CI runs: Competitor Benchmarks workflow

Methodology

Go 1.27 on Apple M4 (arm64); every figure is for the parse stage only (pre-tokenized input to AST), not tokenize or end-to-end
Source: BenchmarkPerformanceBaseline, median of 20 runs, re-measured 2026-10-22, run with -benchmem

Reproduce:

go test -run='^$' -bench=BenchmarkPerformanceBaseline -benchmem -count=20 ./pkg/sql/parser/

Race detector disabled during benchmarks (adds 3–5x overhead)
Object pooling enabled for all runs (production configuration)
The CI regression baseline is a deliberately loose ceiling used with a tolerance on slower shared runners; it is not a measurement

Ready to use GoSQLX in your project?

Get Started Try Playground