profiling

README

Profiling Code

We can use the go tooling in conjunction with the Graph Visualization Tools and Ghostscript. These tools will allow us to graph the profiles we create.

Note: Unless you are running OS X El Capitan, profiling on the Mac is broken. This post talks about how to hack to OS X Kernel to make it work.
http://research.swtch.com/macpprof
https://godoc.org/rsc.io/pprof_mac_fix

Installing Tools

download files

Graph Visualization Tools

Download the package for your target OS/Arch: http://www.graphviz.org/Download.php

Ghostscript

This is not an easy step for Mac users since there is no prebuilt distribution.

Download and uncompress the source code: http://ghostscript.com/download/gsdnld.html

./configure
make
sudo make install

Static Profiling

We can have our program produce profiling data and once the program quits, write the profiling data to disk so it can be visualized.

Code Changes

go get Dave Cheney's profiling package. He has done a nice job abstracting all the boilerplate code required.

go get github.com/davecheney/profile

We need to make some changes to main to get the profiling data we need.

import "github.com/davecheney/profile"

// main is the entry point for the application.
func main() {
	cfg := profile.Config{
		MemProfile:     true,
		CPUProfile:     true,
		ProfilePath:    ".",  // store profiles in current directory
		NoShutdownHook: true, // do not hook SIGINT
	}

	// p.Stop() must be called before the program exits to
	// ensure profiling information is written to disk.
	p := profile.Start(&cfg)
	defer p.Stop()

	. . .
}

Running and Creating Profile Graphs

Build and run the service:
	go build
	./profiling

In a separate terminal generate requests:
	while true && do curl http://localhost:6060/english && done
	<control> C

Generate the call graphs:
	go tool pprof --pdf ./profiling cpu.pprof > cpugraph.pdf
	go tool pprof --pdf ./profiling mem.pprof > memgraph.pdf

Dynamic Profiling

We can have our program produce profiling data and during runtime, write the profiling data to disk so it can be visualized.

Code Changes

Reset the example back to the original code. Then add this import and this will add a route to the existing service we can leverage to get the profile data.

_ "net/http/pprof"

Running and Creating Profile Graphs

Build and run the service:
	go build
	./profiling

In a separate terminal generate requests:
	while true && do curl http://localhost:6060/english && done

In a separate terminal run a pprof tool:
	go tool pprof http://localhost:6060/debug/pprof/heap		(Memory Profile)
	go tool pprof http://localhost:6060/debug/pprof/profile   	(CPU Profile)

pprof Commands

We will use the CPU profile data:

go tool pprof http://localhost:6060/debug/pprof/profile

The top command will show top functions and methods based on CPU Profile

(pprof) top
810ms of 810ms total (  100%)
Showing top 10 nodes out of 98 (cum >= 10ms)
      flat  flat%   sum%        cum   cum%
     320ms 39.51% 39.51%      330ms 40.74%  syscall.Syscall
     140ms 17.28% 56.79%      140ms 17.28%  runtime.kevent
     120ms 14.81% 71.60%      120ms 14.81%  runtime.mach_semaphore_signal
      70ms  8.64% 80.25%       70ms  8.64%  runtime.usleep
      60ms  7.41% 87.65%       60ms  7.41%  runtime.mach_semaphore_wait
      30ms  3.70% 91.36%       30ms  3.70%  runtime.mach_semaphore_timedwait
      30ms  3.70% 95.06%       30ms  3.70%  runtime.memmove
      20ms  2.47% 97.53%       20ms  2.47%  syscall.Syscall6
      10ms  1.23% 98.77%       10ms  1.23%  runtime.cas64
      10ms  1.23%   100%       10ms  1.23%  syscall.RawSyscall

The web command will generate a call graph.

Links

http://golang.org/blog/profiling-go-programs

http://golang.org/pkg/runtime/pprof/

https://golang.org/pkg/net/http/pprof/

https://godoc.org/rsc.io/pprof_mac_fix

Code Review

HTTP Service (Go Playground)

---

All material is licensed under the Apache License Version 2.0, January 2004.

Documentation

Overview

Sample program to show how profiling works. This is the base code. Use the readme for changes needed to be made to the code.