notes.md

pprof notes

1. pprof

• start a server exposing pprof endpoint
• go tool pprof -seconds=10 localhost:8080/debug/pprof/profile
  --10 pprof -seconds=10 localhost:8080/debug/pprof/profile ** profile the application for 10 seconds
• start load testing

2. load test benchmark tool go get -u github.com/tsliwowicz/go-wrk ** 5 seconds of load testing with 10 goroutines go-wrk -d 5 http://localhost:8080/[email protected]

3. flame graphs go-torch go-torch -t 5 ** runs the same as go tool pprof but produces flame graphs (torch.svg)

• horizontal axis is the time spent while profiling the application
• vertical axis the the stack trace of each function call
• order and color does not irrelevant

4. Optimize notes:

   • move regex must compile out

5. Benchmark go test -bench . goos: linux goarch: amd64 pkg: github.com/dnguy078/pprofit BenchmarkHandler-4 50000 27855 ns/op PASS ok github.com/dnguy078/pprofit 1.694s

means that the loop ran for 50000 times at a speed of 27855 ns/op ** If a benchmark needs some expensive setup before running, the benchmark time can be reset ie) func BenchmarkBigLen(b *testing.B) { big := NewBig() b.ResetTimer() for i := 0; i < b.N; i++ { big.Len() } }

• Creates a cpu profile go test -bench . -cpuprofile prof.cpu outputs:
  • prof.cpu - cpu profile data
  • pprofit.test benchmark test holds symbolic links to variables?

go get github.com/google/pprof pprof -http [host]:[port] [options] [binary] ... host and port - to view web interface ie) show cpu data on localhost:1234 pprof -http "localhost:1234" prof.cpu

alternatively you can do pprof pprofit.test prof.cpu