tests: replace benchmarks #477
Conversation
oleg-jukovec
force-pushed
the
oleg-jukovec/perf-tests
branch
3 times, most recently
from
6008328 to
a95a5cd
Compare
tarantool_test.go
Outdated
| 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.
There was a problem hiding this comment.
Updated, thank you.
tarantool_test.go
Outdated
| } | ||
| } | ||
| }() | ||
| _, 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.
There was a problem hiding this comment.
Added t.Skip().
| // 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.
There was a problem hiding this comment.
I'd prefer consistency across the project. All integration tests requiring Tarantool are located in tarantool_test.go , let's benchmarks will lead the logic too.
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
a95a5cd to
334beab
Compare
It should help to fix flaky tests on CI.
It seems that a connection's file descriptor is not always fully ready after the connection is created. The hack helps to avoid the "bad file descriptor" flaky errors for the test.
oleg-jukovec
force-pushed
the
oleg-jukovec/perf-tests
branch
from
f6fd089 to
683d1e4
Compare
Existing benchmarks are no longer relevant - no one knows what they're testing. Instead, benchmarks have been added for the following metrics:
The pull request also attempts to fix flaky tests found.