CPU Profiling

In this lesson, explore cpu profiling through one small Go program, and connect its output to the language rules behind it.

CPU profiling samples which functions consume processor time. Run benchmarks with -cpuprofile or call runtime/pprof.StartCPUProfile in a suitable environment.

Use go tool pprof to inspect measured hotspots; a hotspot identifies where time went, not automatically what change is correct.

Follow this diagram before executing the example. Check where a value or decision changes what runs next.

Try go test -run ^$ -bench . -cpuprofile cpu.out in a package with benchmarks, then go tool pprof cpu.out. For pprof help, see Profiling Go Programs.

This example is a complete program. Save it as main.go or paste it into the Go Playground.

Go code

package main

import "fmt"

func compute() int {
	total := 0
	for i := 1; i <= 1000; i++ {
		total += i * i
	}
	return total
}
func main() { fmt.Println("Computed:", compute()) }

Output

Computed: 333833500

compute performs deterministic CPU work and prints its result. A CPU profile of a suitably repeated workload can identify where sampled CPU time accumulates.

• Too-short runs can yield uninformative profiles.
• A profile from a synthetic loop is not evidence about production load.

Benchmark compute and write a CPU profile, then locate the most expensive function in pprof.

Tip: Edit the example above, predict the output, and run the modified code to test your reasoning.

You have practiced CPU Profiling in a small, self-contained program. The important habit is to predict what runs, inspect the actual output and distinguish the language rule from the particular example.

Before continuing, make sure you can explain the program to someone who has not seen it, including the line that changes its result.