使用 Go 对 Nginx 进行性能测试

目前有很多提供Go语言HTTP应用服务的方法，但其中最好的选择取决于每个应用的实际情况。目前，Nginx看起来是每个新项目的标准Web服务器，即使在有其他许多不错Web服务器的情况下。然而，在Nginx上提供Go应用服务的开销是多少呢？我们需要一些nginx的特性参数（vhosts，负载均衡，缓存，等等）或者直接使用Go提供服务？如果你需要nginx，最快的连接机制是什么？这就是在这我试图回答的问题。该基准测试的目的不是要验证Go比nginx的快或慢。那将会很愚蠢。

下面是我们要比较不同的设置：

• Go HTTP standalone (as the control group)
• Nginx proxy to Go HTTP
• Nginx fastcgi to Go TCP FastCGI
• Nginx fastcgi to Go Unix Socket FastCGI

硬件

因为我们将在相同的硬件下比较所有设置，硬件选择的是廉价的一个。这不应该是一个大问题。

• Samsung 笔记本 NP550P5C-AD1BR
• Intel Core i7 3630QM @2.4GHz (quad core, 8 threads)
• CPU caches: (L1: 256KiB, L2: 1MiB, L3: 6MiB)
• RAM 8GiB DDR3 1600MHz

软件

• Ubuntu 13.10 amd64 Saucy Salamander (updated)
• Nginx 1.4.4 (1.4.4-1~saucy0 amd64)
• Go 1.2 (linux/amd64)
• wrk 3.0.4

设置

内核

只需很小的一点调整，将内核的limits调高。如果你对这一变量有更好的想法，请在写在下面评论处：

fs.file-max                    9999999 fs.nr_open                    9999999 net.core.netdev_max_backlog    4096 net.core.rmem_max              16777216 net.core.somaxconn            65535 net.core.wmem_max              16777216 net.ipv4.ip_forward            0 net.ipv4.ip_local_port_range  1025      65535 net.ipv4.tcp_fin_timeout      30 net.ipv4.tcp_keepalive_time    30 net.ipv4.tcp_max_syn_backlog  20480 net.ipv4.tcp_max_tw_buckets    400000 net.ipv4.tcp_no_metrics_save  1 net.ipv4.tcp_syn_retries      2 net.ipv4.tcp_synack_retries    2 net.ipv4.tcp_tw_recycle        1 net.ipv4.tcp_tw_reuse          1 vm.min_free_kbytes            65536 vm.overcommit_memory          1

Limits

供root和www-data打开的最大文件数限制被配置为200000。

Nginx

有几个必需得Nginx调整。有人跟我说过，我禁用了gzip以保证比较公平。下面是它的配置文件/etc/nginx/nginx.conf：

user www-data; worker_processes auto; worker_rlimit_nofile 200000; pid /var/run/nginx.pid;

events {     worker_connections 10000;     use epoll;     multi_accept on; }

http {     sendfile on;     tcp_nopush on;     tcp_nodelay on;     keepalive_timeout 300;     keepalive_requests 10000;     types_hash_max_size 2048;

    open_file_cache max=200000 inactive=300s;     open_file_cache_valid 300s;     open_file_cache_min_uses 2;     open_file_cache_errors on;

    server_tokens off;     dav_methods off;

    include /etc/nginx/mime.types;     default_type application/octet-stream;

    access_log /var/log/nginx/access.log combined;     error_log /var/log/nginx/error.log warn;

    gzip off;     gzip_vary off;

    include /etc/nginx/conf.d/*.conf;     include /etc/nginx/sites-enabled/*.conf; }

Nginx vhosts

upstream go_http {     server 127.0.0.1:8080;     keepalive 300; }

server {     listen 80;     server_name go.http;     access_log off;     error_log /dev/null crit;

    location / {         proxy_pass http://go_http;         proxy_http_version 1.1;         proxy_set_header Connection "";     } }

upstream go_fcgi_tcp {     server 127.0.0.1:9001;     keepalive 300; }

server {     listen 80;     server_name go.fcgi.tcp;     access_log off;     error_log /dev/null crit;

    location / {         include fastcgi_params;         fastcgi_keep_conn on;         fastcgi_pass go_fcgi_tcp;     } }

upstream go_fcgi_unix {     server unix:/tmp/go.sock;     keepalive 300; }

server {     listen 80;     server_name go.fcgi.unix;     access_log off;     error_log /dev/null crit;

    location / {         include fastcgi_params;         fastcgi_keep_conn on;         fastcgi_pass go_fcgi_unix;     } }

Go源码

package main

import (     "fmt"     "log"     "net"     "net/http"     "net/http/fcgi"     "os"     "os/signal"     "syscall" )

var (     abort bool )

const (     SOCK = "/tmp/go.sock" )

type Server struct { }

func (s Server) ServeHTTP(w http.ResponseWriter, r *http.Request) {     body := "Hello World\n"     // Try to keep the same amount of headers     w.Header().Set("Server", "gophr")     w.Header().Set("Connection", "keep-alive")     w.Header().Set("Content-Type", "text/plain")     w.Header().Set("Content-Length", fmt.Sprint(len(body)))     fmt.Fprint(w, body) }

func main() {     sigchan := make(chan os.Signal, 1)     signal.Notify(sigchan, os.Interrupt)     signal.Notify(sigchan, syscall.SIGTERM)

    server := Server{}

    go func() {         http.Handle("/", server)         if err := http.ListenAndServe(":8080", nil); err != nil {             log.Fatal(err)         }     }()

    go func() {         tcp, err := net.Listen("tcp", ":9001")         if err != nil {             log.Fatal(err)         }         fcgi.Serve(tcp, server)     }()

    go func() {         unix, err := net.Listen("unix", SOCK)         if err != nil {             log.Fatal(err)         }         fcgi.Serve(unix, server)     }()

    <-sigchan

    if err := os.Remove(SOCK); err != nil {         log.Fatal(err)     } }

检查HTTP header

为公平起见，所有的请求必需大小相同。

$ curl -sI http://127.0.0.1:8080/ HTTP/1.1 200 OK Connection: keep-alive Content-Length: 12 Content-Type: text/plain Server: gophr Date: Sun, 15 Dec 2013 14:59:14 GMT

$ curl -sI http://127.0.0.1:8080/ | wc -c 141

$ curl -sI http://go.http/ HTTP/1.1 200 OK Server: nginx Date: Sun, 15 Dec 2013 14:59:31 GMT Content-Type: text/plain Content-Length: 12 Connection: keep-alive

$ curl -sI http://go.http/ | wc -c 141

$ curl -sI http://go.fcgi.tcp/ HTTP/1.1 200 OK Content-Type: text/plain Content-Length: 12 Connection: keep-alive Date: Sun, 15 Dec 2013 14:59:40 GMT Server: gophr

$ curl -sI http://go.fcgi.tcp/ | wc -c 141

$ curl -sI http://go.fcgi.unix/ HTTP/1.1 200 OK Content-Type: text/plain Content-Length: 12 Connection: keep-alive Date: Sun, 15 Dec 2013 15:00:15 GMT Server: gophr

$ curl -sI http://go.fcgi.unix/ | wc -c 141

启动引擎

• 使用sysctl配置内核
• 配置Nginx
• 配置Nginx vhosts
• 用www-data启动服务
• 运行基准测试

基准测试

GOMAXPROCS = 1

Go standalone

# wrk -t100 -c5000 -d30s http://127.0.0.1:8080/ Running 30s test @ http://127.0.0.1:8080/   100 threads and 5000 connections   Thread Stats  Avg      Stdev    Max  +/- Stdev     Latency  116.96ms  17.76ms 173.96ms  85.31%     Req/Sec  429.16    49.20  589.00    69.44%   1281567 requests in 29.98s, 215.11MB read Requests/sec:  42745.15 Transfer/sec:      7.17MB

Nginx + Go through HTTP

# wrk -t100 -c5000 -d30s http://go.http/ Running 30s test @ http://go.http/   100 threads and 5000 connections   Thread Stats  Avg      Stdev    Max  +/- Stdev     Latency  124.57ms  18.26ms 209.70ms  80.17%     Req/Sec  406.29    56.94    0.87k    89.41%   1198450 requests in 29.97s, 201.16MB read Requests/sec:  39991.57 Transfer/sec:      6.71MB

Nginx + Go through FastCGI TCP

# wrk -t100 -c5000 -d30s http://go.fcgi.tcp/ Running 30s test @ http://go.fcgi.tcp/   100 threads and 5000 connections   Thread Stats  Avg      Stdev    Max  +/- Stdev     Latency  514.57ms  119.80ms  1.21s    71.85%     Req/Sec    97.18    22.56  263.00    79.59%   287416 requests in 30.00s, 48.24MB read   Socket errors: connect 0, read 0, write 0, timeout 661 Requests/sec:  9580.75 Transfer/sec:      1.61MB

Nginx + Go through FastCGI Unix Socket

# wrk -t100 -c5000 -d30s http://go.fcgi.unix/ Running 30s test @ http://go.fcgi.unix/   100 threads and 5000 connections   Thread Stats  Avg      Stdev    Max  +/- Stdev     Latency  425.64ms  80.53ms 925.03ms  76.88%     Req/Sec  117.03    22.13  255.00    81.30%   350162 requests in 30.00s, 58.77MB read   Socket errors: connect 0, read 0, write 0, timeout 210 Requests/sec:  11670.72 Transfer/sec:      1.96MB

GOMAXPROCS = 8

Go standalone

# wrk -t100 -c5000 -d30s http://127.0.0.1:8080/ Running 30s test @ http://127.0.0.1:8080/   100 threads and 5000 connections   Thread Stats  Avg      Stdev    Max  +/- Stdev     Latency    39.25ms    8.49ms  86.45ms  81.39%     Req/Sec    1.29k  129.27    1.79k    69.23%   3837995 requests in 29.89s, 644.19MB read Requests/sec: 128402.88 Transfer/sec:    21.55MB

Nginx + Go through HTTP

# wrk -t100 -c5000 -d30s http://go.http/ Running 30s test @ http://go.http/   100 threads and 5000 connections   Thread Stats  Avg      Stdev    Max  +/- Stdev     Latency  336.77ms  297.88ms 632.52ms  60.16%     Req/Sec    2.36k    2.99k  19.11k    84.83%   2232068 requests in 29.98s, 374.64MB read Requests/sec:  74442.91 Transfer/sec:    12.49MB

Nginx + Go through FastCGI TCP

# wrk -t100 -c5000 -d30s http://go.fcgi.tcp/ Running 30s test @ http://go.fcgi.tcp/   100 threads and 5000 connections   Thread Stats  Avg      Stdev    Max  +/- Stdev     Latency  217.69ms  121.22ms  1.80s    75.14%     Req/Sec  263.09    102.78  629.00    62.54%   721027 requests in 30.01s, 121.02MB read   Socket errors: connect 0, read 0, write 176, timeout 1343 Requests/sec:  24026.50 Transfer/sec:      4.03MB

Nginx + Go through FastCGI Unix Socket

# wrk -t100 -c5000 -d30s http://go.fcgi.unix/ Running 30s test @ http://go.fcgi.unix/   100 threads and 5000 connections   Thread Stats  Avg      Stdev    Max  +/- Stdev     Latency  694.32ms  332.27ms  1.79s    62.13%     Req/Sec  646.86    669.65    6.11k    87.80%   909836 requests in 30.00s, 152.71MB read Requests/sec:  30324.77 Transfer/sec:      5.09MB

结论

第一组基准测试时一些Nginx的设置还没有很好的优化（启用gzip，Go的后端没有使用keep-alive连接）。当改为wrk以及按建议优化Nginx后结果有较大差异。

当GOMAXPROCS=1时，Nginx的开销不是那么大，但当OMAXPROCS=8时差异就很大了。以后可能会再试一下其他设置。如果你需要使用Nginx像虚拟主机，负载均衡，缓存等特性，使用HTTP proxy，别使用FastCGI。有些人说Go的FastCGI还没有被很好优化，这也许就是测试结果中巨大差异的原因。