Benchmarks

:::note Local benchmarks only These results were generated against local Docker containers and in-memory SQLite — not a substitute for production profiling. Results will vary significantly with real network latency, connection pool sizing, and concurrent load. Reproduce locally with:

go test -bench=. -benchmem ./...

See benchmark_test.go in the main repository for the full benchmark suite. :::

The current benchmark report was generated on 2026-05-02 using local Docker instances for server databases and in-memory SQLite.

Average operation time

Engine	Operation	Average time	Total run
SQLite in-memory	Insert	6.12 microseconds	61.19 ms
SQLite in-memory	Update	3.24 microseconds	16.18 ms
SQLite in-memory	Delete	1.92 microseconds	9.59 ms
PostgreSQL	Insert	198.55 microseconds	1.99 s
PostgreSQL	Update	129.76 microseconds	648.82 ms
PostgreSQL	Delete	128.53 microseconds	642.63 ms
MySQL	Insert	979.43 microseconds	9.79 s
MySQL	Update	275.57 microseconds	1.38 s
MySQL	Delete	266.18 microseconds	1.33 s
MSSQL	Insert	651.50 microseconds	6.52 s
MSSQL	Update	266.88 microseconds	1.33 s
MSSQL	Delete	265.41 microseconds	1.33 s
Oracle	Insert	431.73 microseconds	4.32 s
Oracle	Update	271.94 microseconds	1.36 s
Oracle	Delete	269.34 microseconds	1.35 s

Benchmark notes

• SQLite used in-memory mode.
• PostgreSQL, MySQL, MSSQL, and Oracle ran against local Docker services.
• Insert tests used 10,000 operations.
• Update and delete tests used 5,000 operations each.
• Progress logs were emitted every 1,000 operations to verify sustained progress.

Design takeaway

QUARK's reflection cache keeps the ORM overhead low after model metadata is parsed. The query builder keeps application ergonomics close to dynamic ORMs while using generics for typed results.