在线时间:8:00-16:00
迪恩网络APP
随时随地掌握行业动态
扫描二维码
关注迪恩网络微信公众号
pprof是go语言内置 的分析性能,分析数据的工具,pprof开启后,每隔一段时间(10ms)就会收集下当前的堆栈信息,获取格格函数占用的CPU以及内存资源;最后通过对这些采样数据进行分析,形成一个性能分析报告。pprof用 profile.proto 读取分析样本的集合,并声称可视化报告,以帮助分析数据 (支持文本和图形报告)
profile.proto 是一个 Protobuf v3的描述文件,它描述了一组 callstack和 symbolization 信息,它的作用是统计分析一组采样的调用栈,配置文件格式是 stacktrace.
采样方式
使用模式
可以做什么
简单例子package main import ( "log" "net/http" _ "net/http/pprof" "time" ) var datas []string func main() { go func() { for { log.Printf("len:%d",Add("gwyy")) time.Sleep(time.Microsecond + 10) } }() _ = http.ListenAndServe(":6060",nil) } func Add(str string) int { data := []byte(str) datas = append(datas,string(data)) return len(datas) }
为什么要初始化 net/http/pprof在前面例子中,在引用上对 net/http/pprof 包进行了默认的初始化,(也就是 _) ,如果没在对该包进行初始化,则无法调用pprof的相关接口,这是为什么呢,我们可以一起看看该包初始化方法:
func init() { http.HandleFunc("/debug/pprof/", Index) http.HandleFunc("/debug/pprof/cmdline", Cmdline) http.HandleFunc("/debug/pprof/profile", Profile) http.HandleFunc("/debug/pprof/symbol", Symbol) http.HandleFunc("/debug/pprof/trace", Trace) }
实际上,在初始化函数中, net/http/pprof 会对标准库中的 net/http 默认提供 DefaultServeMux 进行路由注册,源码如下
func HandleFunc(pattern string, handler func(ResponseWriter, *Request)) { DefaultServeMux.HandleFunc(pattern, handler) } 我们在例子中使用的 HTTP Server,也是标准库默认提供的,因此便可以注册进去。
在实际项目中 我们有独立的 ServeMux的,这时候只需要将PProf对应的路由注册进去即可
mux := http.NewServeMux() mux.HandleFunc("/debug/pprof",pprof.Index) mux.HandleFunc("/debug/pprof/cmdline",pprof.cmdline) mux.HandleFunc("/debug/pprof/profile", pprof.Profile) mux.HandleFunc("/debug/pprof/symbol", pprof.Symbol) mux.HandleFunc("/debug/pprof/trace", pprof.Trace)
好了 我们运行了上面的程序后,浏览器直接访问:如果在访问路径上加上 ?debug=1 则可以直接在浏览器中访问。
这个页面中有许多子页面,咱们继续深究下去,看看可以得到什么?
一般在线上环境,为了网络安全,通常不会这么做。另外debug的访问方式是具有时效性的,在实际场景中,我们常常需要及时将当前状态下的 profile文件给存储下来,便于二次分析。
通过终端访问第二种是通过命令行完整对正在运行的程序 pprof进行抓取和分析 //在执行命令后,需要等待 60s (可以调整seconds的值) PProf 会进行 cpu Profiling , 结束后将默认进入PProf 的命令行交互模式,查看或者导出分析结果。 // 如果是以TLS方式启动的HTTP Server 那么在调用需要改成 go tool pprof https+insecure://localhost:6060/debug/pprof/profile?seconds=60 go tool pprof http://localhost:6060/debug/pprof/profile?seconds=60 Fetching profile over HTTP from http://localhost:6060/debug/pprof/profile?seconds=60 Saved profile in C:\Users\admin\pprof\pprof.samples.cpu.001.pb.gz Type: cpu Time: Oct 12, 2020 at 11:40pm (CST) Duration: 1mins, Total samples = 13.53s (22.50%) Entering interactive mode (type "help" for commands, "o" for options) (pprof)
top命令输入命令 top 10 查看对应资源开销 (例如 cpu就是执行耗时/开销 Memory 就是内存占用大小)排名前十的函数
(pprof) top 10 Showing nodes accounting for 10.82s, 79.97% of 13.53s total Dropped 64 nodes (cum <= 0.07s) Showing top 10 nodes out of 61 flat flat% sum% cum cum% 7.50s 55.43% 55.43% 7.56s 55.88% runtime.stdcall1 1.88s 13.90% 69.33% 1.90s 14.04% runtime.cgocall 0.31s 2.29% 71.62% 2.25s 16.63% runtime.timerproc 0.26s 1.92% 73.54% 0.26s 1.92% runtime.(*mcache).prepareForSweep 0.20s 1.48% 75.02% 0.53s 3.92% runtime.acquirep 0.18s 1.33% 76.35% 0.18s 1.33% runtime.casgstatus 0.15s 1.11% 77.46% 1.18s 8.72% runtime.exitsyscall 0.12s 0.89% 78.34% 0.52s 3.84% runtime.goroutineReady 0.11s 0.81% 79.16% 8.79s 64.97% runtime.systemstack 0.11s 0.81% 79.97% 0.11s 0.81% runtime.unlock
大多数情况下,我们可以得出一个应用程序的运行情况,知道当前是什么函数,正在做什么事情,占用了多少资源等。
Heap ProfilingPS C:\Users\admin> go tool pprof http://localhost:6060/debug/pprof/heap Fetching profile over HTTP from http://localhost:6060/debug/pprof/heap Saved profile in C:\Users\admin\pprof\pprof.alloc_objects.alloc_space.inuse_objects.inuse_space.001.pb.gz Type: inuse_space Time: Oct 12, 2020 at 11:50pm (CST) No samples were found with the default sample value type. Try "sample_index" command to analyze different sample values. Entering interactive mode (type "help" for commands, "o" for options) (pprof)
这个命令能够很快的拉取到结果。不需要采样等待, 需要注意的是 Type 这个选项默认是 inuse_space,实际上,它可以对多种内存情况进行分析。常见的有:
inuse_space 分析应用常驻内存的占用情况
PS C:\Users\admin> go tool pprof -inuse_space http://localhost:6060/debug/pprof/heap Fetching profile over HTTP from http://localhost:6060/debug/pprof/heap Saved profile in C:\Users\admin\pprof\pprof.alloc_objects.alloc_space.inuse_objects.inuse_space.002.pb.gz Type: inuse_space Time: Oct 12, 2020 at 11:52pm (CST) Entering interactive mode (type "help" for commands, "o" for options) (pprof) top Showing nodes accounting for 3.38MB, 100% of 3.38MB total flat flat% sum% cum cum% 2.38MB 70.40% 70.40% 2.38MB 70.40% main.Add 1MB 29.60% 100% 3.38MB 100% main.main.func1
alloc_objects 分析应用程序的内存临时分配情况
PS C:\Users\admin> go tool pprof -alloc_objects http://localhost:6060/debug/pprof/heap Fetching profile over HTTP from http://localhost:6060/debug/pprof/heap Saved profile in C:\Users\admin\pprof\pprof.alloc_objects.alloc_space.inuse_objects.inuse_space.003.pb.gz Type: alloc_objects Time: Oct 12, 2020 at 11:53pm (CST) Entering interactive mode (type "help" for commands, "o" for options) (pprof) top Showing nodes accounting for 98305, 100% of 98312 total Dropped 14 nodes (cum <= 491) flat flat% sum% cum cum% 98305 100% 100% 98311 100% main.main.func1 (pprof)
另外还有 inuse_objects 和 alloc_space 类别,分别对应查看每个函数的对象数量和分配的内存空间大小。
Goroutine ProfilingPS C:\Users\admin> go tool pprof http://localhost:6060/debug/pprof/goroutine Fetching profile over HTTP from http://localhost:6060/debug/pprof/goroutine Saved profile in C:\Users\admin\pprof\pprof.goroutine.001.pb.gz Type: goroutine Time: Oct 13, 2020 at 12:15am (CST) Entering interactive mode (type "help" for commands, "o" for options) (pprof) 在查看goroutine 时可以使用traces命令,这个命令会打印出对应的所有调用栈,以及指标信息,通过这个命令我们可以很方便的查看整改调用链路有什么,分别在哪里使用了多岁个goroutine,并且通过分析可以知道谁才是真正的调用方,输出结果如下:
(pprof) traces Type: goroutine Time: Oct 13, 2020 at 12:15am (CST) -----------+------------------------------------------------------- 1 runtime.cgocall syscall.Syscall9 syscall.WSARecv internal/poll.(*FD).Read.func1 internal/poll.(*ioSrv).ExecIO internal/poll.(*FD).Read net.(*netFD).Read net.(*conn).Read net/http.(*connReader).backgroundRead -----------+------------------------------------------------------- 1 runtime.gopark runtime.netpollblock internal/poll.runtime_pollWait internal/poll.(*pollDesc).wait internal/poll.(*ioSrv).ExecIO internal/poll.(*FD).acceptOne internal/poll.(*FD).Accept net.(*netFD).accept net.(*TCPListener).accept net.(*TCPListener).Accept net/http.(*Server).Serve net/http.(*Server).ListenAndServe net/http.ListenAndServe main.main runtime.main -----------+------------------------------------------------------- 1 runtime.gopark runtime.goparkunlock time.Sleep main.main.func1 -----------+------------------------------------------------------- 1 runtime/pprof.writeRuntimeProfile runtime/pprof.writeGoroutine runtime/pprof.(*Profile).WriteTo net/http/pprof.handler.ServeHTTP net/http/pprof.Index net/http.HandlerFunc.ServeHTTP net/http.(*ServeMux).ServeHTTP net/http.serverHandler.ServeHTTP net/http.(*conn).serve -----------+------------------------------------------------------- (pprof)
调用栈上的展示是自下而上的,也就是说 runtime.main方法调用了 main.main方法,而main.main方法又调用了 net/http.ListenAndServe 方法,排查起来比较方便。
每个调用栈信息都是用 ------- 分割,函数方法前面的是指标数据,例如,Gorutine Profiling 展示的是该方法占用的 goroutine的数量,而Heap Profiling 展示 的是占用内存大小。如图
PS C:\Users\admin> go tool pprof http://localhost:6060/debug/pprof/heap Fetching profile over HTTP from http://localhost:6060/debug/pprof/heap Saved profile in C:\Users\admin\pprof\pprof.alloc_objects.alloc_space.inuse_objects.inuse_space.005.pb.gz Type: inuse_space Time: Oct 13, 2020 at 12:24am (CST) No samples were found with the default sample value type. Try "sample_index" command to analyze different sample values. Entering interactive mode (type "help" for commands, "o" for options) (pprof) traces Type: inuse_space Time: Oct 13, 2020 at 12:24am (CST) -----------+------------------------------------------------------- bytes: 304kB 678.95kB main.Add main.main.func1 -----------+------------------------------------------------------- bytes: 192kB 0 main.Add main.main.func1 -----------+------------------------------------------------------- (pprof)
实际上,PProf中的所有功能都会根据 Profile的不同类型展示不同的对应结果
Mutex Profiling一般来说,在调用 chan (通道)、sync.Mutex (同步锁)或者 time.Sleep() 时会造成阻塞,下面看个例子:
func init() { runtime.SetMutexProfileFraction(1) } func main() { var m sync.Mutex var datas = make(map[int]struct{}) for i:=0;i<999;i++ { go func(i int) { m.Lock() defer m.Unlock() datas[i] = struct{}{} }(i) } _ = http.ListenAndServe(":6060",nil) }
需要特别注意的是 runtime.SetMutexProfileFraction(1) 语句,如果未来希望对互斥锁进行采集,则需要调用该方法设置采集频率,如果没有设置,或者设置数值小于0,则不进行采集
接下里调用 go tool pprof 进行分析:
PS C:\Users\admin> go tool pprof http://localhost:6060/debug/pprof/mutex Fetching profile over HTTP from http://localhost:6060/debug/pprof/mutex Saved profile in C:\Users\admin\pprof\pprof.contentions.delay.001.pb.gz Type: delay Time: Oct 13, 2020 at 12:31am (CST) Entering interactive mode (type "help" for commands, "o" for options) (pprof)
调用 top 命令,查看互斥锁排名
(pprof) top Showing nodes accounting for 149.51us, 100% of 149.51us total flat flat% sum% cum cum% 149.51us 100% 100% 149.51us 100% sync.(*Mutex).Unlock 0 0% 100% 149.51us 100% main.main.func1 (pprof)
调用 list命令 查看指定函数的代码情况 (包含特定的指标信息,如耗时)若函数名不明确,则默认对该函数名进行模糊匹配:
(pprof) list main Total: 149.51us ROUTINE ======================== main.main.func1 in D:\goroot\projects\_practice\go-learn\pprof\main.go 0 149.51us (flat, cum) 100% of Total . . 16: for i:=0;i<999;i++ { . . 17: go func(i int) { . . 18: m.Lock() . . 19: defer m.Unlock() . . 20: datas[i] = struct{}{} . 149.51us 21: }(i) . . 22: } . . 23: _ = http.ListenAndServe(":6060",nil) . . 24:} (pprof)
从输出分析中可以看到引起互斥锁函数,以及锁开销的位置。
Block Profiling与 Mutex 的 runtime.SetMutexProfileFraction 语句类似,Block也需要调用 runtime.SetBlockProfileRate 语句进行设置,如果没有设置,或者设置数值小于0,则不进行采集 func init() { runtime.SetMutexProfileFraction(1) runtime.SetBlockProfileRate(1) }
调用 top 命令,查看阻塞情况排名:
PS C:\Users\admin> go tool pprof http://localhost:6060/debug/pprof/block Fetching profile over HTTP from http://localhost:6060/debug/pprof/block Saved profile in C:\Users\admin\pprof\pprof.contentions.delay.004.pb.gz Type: delay Time: Oct 13, 2020 at 12:35am (CST) Entering interactive mode (type "help" for commands, "o" for options) (pprof) top Showing nodes accounting for 83.75us, 100% of 83.75us total flat flat% sum% cum cum% 83.75us 100% 100% 83.75us 100% sync.(*Cond).Wait 0 0% 100% 83.75us 100% net/http.(*conn).serve 0 0% 100% 83.75us 100% net/http.(*connReader).abortPendingRead 0 0% 100% 83.75us 100% net/http.(*response).finishRequest (pprof)
也可以用 list 命令查看具体阻塞情况。调用方式和 Mutex Profiling 一样
|
请发表评论