tests: replace benchmarks #477
Conversation
oleg-jukovec
force-pushed
the
oleg-jukovec/perf-tests
branch
2 times, most recently
from
7da7935 to
6008328
Compare
Existing benchmarks are no longer relevant - no one knows what they're testing. Instead, benchmarks have been added for the following metrics: * Number of allocations. * RPS on synchronous and asynchronous modes.
oleg-jukovec
force-pushed
the
oleg-jukovec/perf-tests
branch
from
6008328 to
a95a5cd
Compare
| b.Errorf("request error: %s", err) | ||
| } | ||
|
|
||
| tuple := data[0].([]any) |
There was a problem hiding this comment.
Maybe GetTyped([][]uint16) will perform better here?
There was a problem hiding this comment.
It won't work since it is an array of []any actually.
| b.Errorf("Failed to replace: %s", err) | ||
| } | ||
| rps := float64(requests) / duration.Seconds() | ||
| fmt.Println("requests :", requests) // nolint |
There was a problem hiding this comment.
Let's use t.Log/t.Logf instead of fmt.Println here.
| } | ||
| } | ||
| }() | ||
| _, err := conns[0].Replace(spaceNo, []interface{}{uint(1111), "hello", "world"}) |
There was a problem hiding this comment.
Let's not use deprecated APIs in new/modified benches
| // TestPerformance is a benchmark for the async API that is unable to implement | ||
| // with a Go-benchmark. It can be used to test performance with different | ||
| // numbers of connections and processing goroutines. | ||
| func TestBenchmarkAsync(t *testing.T) { |
There was a problem hiding this comment.
I'm not strongly agains that kind of "bench", but i'd suggest it to be t.Skip()ped by default, only run it in case of CLI argument is provided.
There was a problem hiding this comment.
nitpicking: we can configure this bench using CLI flags too.
| // TestPerformance is a benchmark for the async API that is unable to implement | ||
| // with a Go-benchmark. It can be used to test performance with different | ||
| // numbers of connections and processing goroutines. | ||
| func TestBenchmarkAsync(t *testing.T) { |
There was a problem hiding this comment.
I'd prefer this test to be separated from usual integration tests (for example - in other file). The same thing applies to usual benches, but again - no prio.
Existing benchmarks are no longer relevant - no one knows what they're testing. Instead, benchmarks have been added for the following metrics: