Lectures

Let’s break down why you should run tests with -race, what exactly counts as a data race in Go, how to read race detector reports, and why the absence of warnings is not a correctness guarantee. Using an in-memory store with a map and a nextID counter as an example, we’ll show how to reproduce a race, strengthen the test, then fix the issue using sync.Mutex and re-verify the result.

We break down how to speed up and “harden” tests in Go using parallel execution: where and why to call t.Parallel(), how it works with subtests t.Run(), how to isolate the environment via t.TempDir(), and how to build an anti-flaky approach (without “treating” instability with time.Sleep()). We also discuss common pitfalls: shared state, env variables, ports, loop variables in table-driven tests, and time dependence.

We break down why concurrency tests often “hang”, and how to write them so they wait for completion but don’t wait forever: the select pattern with a timeout via time.After, done channels ( chan struct{}), waiting for sync.WaitGroup with a time limit, and using context.WithTimeout as a “single kill switch” for cancellation. At the end: common mistakes— time.Sleep instead of waiting for an event, wg.Wait() without a timeout, and unbuffered result channels.

We cover practical microbenchmarks in Go: how BenchmarkXxx functions work with * testing.B, how to run iterations correctly using b.N or b.Loop(), why you need the timer ( b.ResetTimer(), b.StopTimer(), b.StartTimer()), how to run measurements with flags -bench, -run, -benchtime, and why, without a sink variable, you can end up “measuring the compiler’s optimizations” rather than your code.

We cover the practical minimum of profiling in Go with pprof: how to capture CPU and heap profiles using go test and the -cpuprofile/ -memprofile flags, how to open them with go tool pprof, how to read top and top -cum, how to find hotspots and jump to lines with list. With examples, we show why fmt.Sprintf and string concatenation are dangerous in a “hot” loop, and how to interpret profiles without confusing CPU and memory.