Go语言已经为开发者内置配套了很多性能调优监控的好工具和方法，这大大提升了我们profile分析的效率。此外本文还将重点介绍和推荐uber开源的go-torch，其生成的火焰图更方便更直观的帮我们进行性能调优。我也是在实际一次的性能调优中，接触到go-torch，非常棒。
go tool pprof简介
Golang内置cpu, mem, block profiler
Go强大之处是它已经在语言层面集成了profile采样工具,并且允许我们在程序的运行时使用它们，使用Go的profiler我们能获取以下的样本信息：
- cpu profiles
- mem profiles
- block profile
Golang常见的profiling使用场景
基准测试文件：例如使用命令go test . -bench . -cpuprofile prof.cpu生成采样文件后，再通过命令 go tool pprof [binary] prof.cpu 来进行分析。
import _ net/http/pprof：如果我们的应用是一个web服务，我们可以在http服务启动的代码文件(eg: main.go)添加 import _ net/http/pprof，这样我们的服务 便能自动开启profile功能，有助于我们直接分析采样结果。
通过在代码里面调用runtime.StartCPUProfile或者runtime.WriteHeapProfile等内置方法，即可方便的进行数据采样。
更多Golang Profiling的使用，推荐https://blog.golang.org/profiling-go-programs。
go tool pprof的使用方法
go tool pprof的参数很多，不做详细介绍，自己help看看。在这里，我主要用到的命令为：
go tool pprof --seconds 25 http://localhost:9090/debug/pprof/profile
命令中，设置了25s的采样时间，当25s采样结束后，就生成了我们想要的profile文件，然后在pprof交互命令行中输入web，从浏览器中打开，就能看到对应的整个调用链的性能树形图。
root@garnett:~/# go tool pprof -h
usage: pprof [options] [binary] <profile source> ...
Output format (only set one):
-callgrind Outputs a graph in callgrind format
-disasm=p Output annotated assembly for functions matching regexp or address
-dot Outputs a graph in DOT format
-eog Visualize graph through eog
-evince Visualize graph through evince
-gif Outputs a graph image in GIF format
-gv Visualize graph through gv
-list=p Output annotated source for functions matching regexp
-pdf Outputs a graph in PDF format
-peek=p Output callers/callees of functions matching regexp
-png Outputs a graph image in PNG format
-proto Outputs the profile in compressed protobuf format
-ps Outputs a graph in PS format
-raw Outputs a text representation of the raw profile
-svg Outputs a graph in SVG format
-tags Outputs all tags in the profile
-text Outputs top entries in text form
-top Outputs top entries in text form
-tree Outputs a text rendering of call graph
-web Visualize graph through web browser
-weblist=p Output annotated source in HTML for functions matching regexp or address
Output file parameters (for file-based output formats):
-output=f Generate output on file f (stdout by default)
Output granularity (only set one):
-functions Report at function level [default]
-files Report at source file level
-lines Report at source line level
-addresses Report at address level
Comparison options:
-base <profile> Show delta from this profile
-drop_negative Ignore negative differences
Sorting options:
-cum Sort by cumulative data
Dynamic profile options:
-seconds=N Length of time for dynamic profiles
Profile trimming options:
-nodecount=N Max number of nodes to show
-nodefraction=f Hide nodes below <f>*total
-edgefraction=f Hide edges below <f>*total
Sample value selection option (by index):
-sample_index Index of sample value to display
-mean Average sample value over first value
Sample value selection option (for heap profiles):
-inuse_space Display in-use memory size
-inuse_objects Display in-use object counts
-alloc_space Display allocated memory size
-alloc_objects Display allocated object counts
Sample value selection option (for contention profiles):
-total_delay Display total delay at each region
-contentions Display number of delays at each region
-mean_delay Display mean delay at each region
Filtering options:
-runtime Show runtime call frames in memory profiles
-focus=r Restricts to paths going through a node matching regexp
-ignore=r Skips paths going through any nodes matching regexp
-tagfocus=r Restrict to samples tagged with key:value matching regexp
Restrict to samples with numeric tags in range (eg "32kb:1mb")
-tagignore=r Discard samples tagged with key:value matching regexp
Avoid samples with numeric tags in range (eg "1mb:")
Miscellaneous:
-call_tree Generate a context-sensitive call tree
-unit=u Convert all samples to unit u for display
-divide_by=f Scale all samples by dividing them by f
-buildid=id Override build id for main binary in profile
-tools=path Search path for object-level tools
-help This message
Environment Variables:
PPROF_TMPDIR Location for saved profiles (default $HOME/pprof)
PPROF_TOOLS Search path for object-level tools
PPROF_BINARY_PATH Search path for local binary files
default: $HOME/pprof/binaries
finds binaries by $name and $buildid/$namego-torch简介
go-torch是Uber公司开源的一款针对Golang程序的火焰图生成工具，能收集 stack traces,并把它们整理成火焰图，直观地程序给开发人员。go-torch是基于使用BrendanGregg创建的火焰图工具生成直观的图像，很方便地分析Go的各个方法所占用的CPU的时间。
go-torch的具体使用参加如下help信息，在这里，我们主要使用到-u和-t参数:
go-torch -u http://localhost:9090 -t 30
root@garnett:~/# go-torch -h
Usage:
go-torch [options] [binary] <profile source>
pprof Options:
-u, --url= Base URL of your Go program (default: http://localhost:8080)
-s, --suffix= URL path of pprof profile (default: /debug/pprof/profile)
-b, --binaryinput= File path of previously saved binary profile. (binary profile is anything accepted by https://golang.org/cmd/pprof)
--binaryname= File path of the binary that the binaryinput is for, used for pprof inputs
-t, --seconds= Number of seconds to profile for (default: 30)
--pprofArgs= Extra arguments for pprof
Output Options:
-f, --file= Output file name (must be .svg) (default: torch.svg)
-p, --print Print the generated svg to stdout instead of writing to file
-r, --raw Print the raw call graph output to stdout instead of creating a flame graph; use with Brendan Gregg's flame graph perl script (see
https://github.com/brendangregg/FlameGraph)
--title= Graph title to display in the output file (default: Flame Graph)
--width= Generated graph width (default: 1200)
--hash Colors are keyed by function name hash
--colors= set color palette. choices are: hot (default), mem, io, wakeup, chain, java, js, perl, red, green, blue, aqua, yellow, purple, orange
--cp Use consistent palette (palette.map)
--reverse Generate stack-reversed flame graph
--inverted icicle graph
Help Options:
-h, --help Show this help message环境准备
安装FlameGraph脚本
git clone https://github.com/brendangregg/FlameGraph.git
cp flamegraph.pl /usr/local/bin在终端输入 flamegraph.pl -h 是否安装FlameGraph成功:
$ flamegraph.pl -h
Option h is ambiguous (hash, height, help)
USAGE: /usr/local/bin/flamegraph.pl [options] infile > outfile.svg
--title # change title text
--width # width of image (default 1200)
--height # height of each frame (default 16)
--minwidth # omit smaller functions (default 0.1 pixels)
--fonttype # font type (default "Verdana")
--fontsize # font size (default 12)
--countname # count type label (default "samples")
--nametype # name type label (default "Function:")
--colors # set color palette. choices are: hot (default), mem, io,
# wakeup, chain, java, js, perl, red, green, blue, aqua,
# yellow, purple, orange
--hash # colors are keyed by function name hash
--cp # use consistent palette (palette.map)
--reverse # generate stack-reversed flame graph
--inverted # icicle graph
--negate # switch differential hues (blue<->red)
--help # this message
eg,
/usr/local/bin/flamegraph.pl --title="Flame Graph: malloc()" trace.txt > graph.svg安装go-torch
有了flamegraph的支持，我们接下来要使用go-torch展示profile的输出:
go get -v github.com/uber/go-torchDemo
启动待调优的程序
在我的实例中，是一个简单的web Demo，go run main.go -printStats启动之后，浏览器能正常访问待调优的接口: http://localhost:9090/demo。每次该接口的访问，都会打印访问信息，如下所示：
root@garnett:/# go run main.go -printStats
Starting Server on :9090
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 67.984µs
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 339.656µs
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 55.749µs
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 89.34µs
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 59.606µs
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 47.917µs
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 42.768µs
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 1.270416ms
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 34.518µs
IncCounter: handler.received.garnett.advance.no-os.no-browser = 1
RecordTimer: handler.latency.garnett.advance.no-os.no-browser = 281.014µs启动压力测试
接下来，我们对该接口进行压力测试，看看它在大并发情况下的性能表现。
我们使用go-wrk工具进行试压，go-wrk的安装请前往github官网https://github.com/adjust/go-wrk，只要把代码clone下来go build一下即可。
执行如下命令，进行35s 1W次高并发场景模拟：
go-wrk -d 35 -n 10000 http://localhost:9090/demo
使用go tool pprof
在上面的压测过程中，我们再新建一个终端窗口输入以下命令，生成我们的profile文件：
go tool pprof --seconds 25 http://localhost:9090/debug/pprof/profile命令中，我们设置了25秒的采样时间，当看到(pprof)的时候，我们输入 web, 表示从浏览器打开,可见下图：
看到这个图，你可能已经懵逼了。在我这个简单的Demo中，已经这么难看了，更何况在实际的性能调优中呢！
使用go-torch
在上面的压测过程中，这次我们使用go-torch来生成采样报告：
go-torch -u http://localhost:9090 -t 3030s后，go-torch完成采样，输出以下信息：
Writing svg to torch.svgtorch.svg是go-torch采样结束后自动生成的profile文件，我们也用浏览器打开,可见下图：
这就是go-torch生成的火焰图，看起来是不是舒服多了。
火焰图的y轴表示cpu调用方法的先后，x轴表示在每个采样调用时间内，方法所占的时间百分比，越宽代表占据cpu时间越多
有了火焰图，我们就可以更清楚的看到哪个方法调用耗时长了，然后不断的修正代码，重新采样，不断优化。
好了，本文只有一个目的，就是希望让你对golang程序的性能调优更有兴趣。接下来，你可以在自己的golang项目中对那些耗时太长的接口进行调优了。
有疑问加站长微信联系（非本文作者）
