use a non-cryptographic hash to generate unique statement names

[drakkan]

you can benchmark like this

func hashSha256(sql string) string {
	digest := sha256.Sum256([]byte(sql))
	return "stmtcache_" + hex.EncodeToString(digest[0:24])
}

func hashFNV1(sql string) string {
	h := fnv.New64a()
	h.Write([]byte(sql))
	return "stmtcache_" + strconv.FormatUint(h.Sum64(), 10)
}

func hashXXHash(sql string) string {
	h := xxhash.New()
	h.Write([]byte(sql))
	return "stmtcache_" + strconv.FormatUint(h.Sum64(), 10)
}

func BenchmarkSha256(b *testing.B) {
	sql := `SELECT u.id,u.username,u.password,u.public_keys,u.home_dir,u.uid,u.gid,u.max_sessions,u.quota_size,u.quota_files,u.permissions,u.used_quota_size,u.used_quota_files,u.last_quota_update,u.upload_bandwidth,u.download_bandwidth,u.expiration_date,u.last_login,u.status,u.filters,u.filesystem,u.additional_info,u.description,u.email,u.created_at,u.updated_at,u.upload_data_transfer,u.download_data_transfer,u.total_data_transfer,u.used_upload_data_transfer,u.used_download_data_transfer,u.deleted_at,u.first_download,u.first_upload,r.name,u.last_password_change FROM users u LEFT JOIN roles r on r.id = u.role_id WHERE u.deleted_at = 0 ORDER BY u.username ASC LIMIT $1 OFFSET $2`
	for i := 0; i < b.N; i++ {
		hashSha256(sql)
	}
}

func BenchmarkFNV1(b *testing.B) {
	sql := `SELECT u.id,u.username,u.password,u.public_keys,u.home_dir,u.uid,u.gid,u.max_sessions,u.quota_size,u.quota_files,u.permissions,u.used_quota_size,u.used_quota_files,u.last_quota_update,u.upload_bandwidth,u.download_bandwidth,u.expiration_date,u.last_login,u.status,u.filters,u.filesystem,u.additional_info,u.description,u.email,u.created_at,u.updated_at,u.upload_data_transfer,u.download_data_transfer,u.total_data_transfer,u.used_upload_data_transfer,u.used_download_data_transfer,u.deleted_at,u.first_download,u.first_upload,r.name,u.last_password_change FROM users u LEFT JOIN roles r on r.id = u.role_id WHERE u.deleted_at = 0 ORDER BY u.username ASC LIMIT $1 OFFSET $2`
	for i := 0; i < b.N; i++ {
		hashFNV1(sql)
	}
}

func BenchmarkXXhash(b *testing.B) {
	sql := `SELECT u.id,u.username,u.password,u.public_keys,u.home_dir,u.uid,u.gid,u.max_sessions,u.quota_size,u.quota_files,u.permissions,u.used_quota_size,u.used_quota_files,u.last_quota_update,u.upload_bandwidth,u.download_bandwidth,u.expiration_date,u.last_login,u.status,u.filters,u.filesystem,u.additional_info,u.description,u.email,u.created_at,u.updated_at,u.upload_data_transfer,u.download_data_transfer,u.total_data_transfer,u.used_upload_data_transfer,u.used_download_data_transfer,u.deleted_at,u.first_download,u.first_upload,r.name,u.last_password_change FROM users u LEFT JOIN roles r on r.id = u.role_id WHERE u.deleted_at = 0 ORDER BY u.username ASC LIMIT $1 OFFSET $2`
	for i := 0; i < b.N; i++ {
		hashXXHash(sql)
	}
}

xxhash is faster but requires an external dependency. FNV-1 should be enough.

Results on my laptop:

BenchmarkFNV1-12    	  807478	      1304 ns/op	     760 B/op	       3 allocs/op

BenchmarkSha256-12    	  365073	      2831 ns/op	     864 B/op	       4 allocs/op

This is just a suggestion, even sha256 should be ok for general use. Thank you

[jackc]

The unique statement name generation should only happen once per SQL text per connection. Given that it is not a per query cost, I think the additional collision resistance is worth the extra time.

[drakkan]

No problem, it was just a suggestion 😄

I'm not sure why sha256 is truncated

hex.EncodeToString(digest[0:24])

this should be

hex.EncodeToString(digest[0:32])

or

hex.EncodeToString(digest[:])

am I missing something? Thank you

[jackc]

The reason it is truncated is to stay under the identifier length limit of 63 bytes. https://www.postgresql.org/docs/current/limits.html