laravel octane 基于docker安装实践
一、安装
- RoadRunner
- 安装扩展
composer require laravel/octane spiral/roadrunner - 安装 RoadRunner 二进制包
./vendor/bin/rr get-binary- 不行的话 加代理，比如我在这里直接加前缀
vendor/spiral/roadrunner-cli/src/Repository/Asset.php:33代码为$this->uri = 'https://dl.ghpig.top/'. $uri;
- 不行的话 加代理，比如我在这里直接加前缀
- 安装扩展
- Swoole
前提安装 swoole: pecl install swoole
- 安装扩展:
composer require laravel/octane
- 安装扩展:
- OpenSwoole
其实跟Swoole一样，名字不一样而已
- 安装到项目中
- 生成配置文件等:
php artisan octane:install- 选择你的安装的情况 roadrunner or swoole
- 开发环境 推荐 ，不然你的watch会报错
apk add nodejs npmnpm config set registry https://registry.npm.taobao.orgnpm install --save-dev chokidar
- 生成配置文件等:
二、nginx 配置
# 默认
map $http_upgrade $connection_upgrade {
default upgrade;
'' close;
}
# 容器内 只能通过upstream方式转发
upstream backend {
server php8:8000;
}
server {
listen 80;
server_name backendstudy.local.com;
#server_tokens off;
root /www/backend/public;
add_header X-Frame-Options "SAMEORIGIN";
add_header X-Content-Type-Options "nosniff";
index index.php;
charset utf-8;
# 因为不是全栈 所以只是部分path转发 与官方给的默认配置有出入
location ~* ^/(api|web|horizon|telescope)/ {
try_files $uri $uri/ @octane;
}
# horizon vendor
location /vendor {
alias /www/backend/public/vendor/;
}
# fronend
location / {
alias /www/backend/public/dist/;
}
#location /index.php {
# try_files /not_exists @octane;
#}
#location / {
# try_files $uri $uri/ @octane;
#}
location = /favicon.ico { access_log off; log_not_found off; }
location = /robots.txt { access_log off; log_not_found off; }
error_page 404 /index.php;
location @octane {
set $suffix "";
if ($uri = /index.php) {
set $suffix ?$query_string;
}
proxy_http_version 1.1;
proxy_set_header Host $http_host;
proxy_set_header Scheme $scheme;
proxy_set_header SERVER_PORT $server_port;
proxy_set_header REMOTE_ADDR $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection $connection_upgrade;
proxy_pass http://backend$suffix;
}
location ~ /\.(?!well-known).* {
deny all;
}
}
三、使用
-
添加Makefile
fixed 在docker里，host必须绑定到 0.0.0.0
# 小心缩进，是Tab，不是空格 help: ## Display this help screen @grep -E '^[a-zA-Z_-]+:.*?## .*$$' $(MAKEFILE_LIST) | awk 'BEGIN {FS = ":.*?## "}; {printf "\033[36m%-20s\033[0m %s\n", $$1, $$2}' dev: ## octane start in dev php artisan octane:start --watch --workers=4 --task-workers=2 --host=0.0.0.0 --port=8000 start: ## octane start in dev php artisan octane:start --workers=4 --task-workers=2 --host=0.0.0.0 --port=8000 reload: ## octane reload php artisan octane:reload stop: ## octane stop php artisan octane:stop run: ## run start by octane on production php artisan octane:start --host=0.0.0.0 status: ## show octane status php artisan octane:status optimize: ## optimize cache and autoload composer dump-autoload php artisan optimize
四、生产使用
- 建议参照文档，将命令打包到docker中去
五、不靠谱的性能测试
- 测试并发，使用wrk
当前测试不是很严格，部分硬件性能没有完全压榨完全，另外测试是同一台宿主机的两台虚拟机 这次测试主要想看下php阻塞对并发影响，虽然加上了内存模式+非阻塞异步事件IO，业务那块其实还是阻塞的 本来使用ab，不过试了之后，发现ab测试会飘 每个都是测试至少3遍，选其中表现最好的
- 环境
- 硬件 4核8G
- php版本
- php-fpm: PHP 8.1.23
- octane: PHP 8.1.23
- hyperf PHP 8.1.25
- webman: PHP 8.1.23，框架的确简陋了点，连个环境变量配置都没有
-
hello world 组
-
php-fpm static 200
wrk -t8 -c400 -d30s http://fpm.local.com/api/hello Running 30s test @ http://fpm.local.com/api/hello 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 1.26s 190.76ms 1.80s 80.33% Req/Sec 40.61 24.20 333.00 77.16% 9315 requests in 30.10s, 3.15MB read Requests/sec: 309.47 Transfer/sec: 107.29KB -
octane swoole
wrk -t8 -c400 -d30s http://octane.local.com/api/hello Running 30s test @ http://octane.local.com/api/hello 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 537.92ms 112.21ms 958.60ms 70.75% Req/Sec 95.31 46.90 380.00 70.21% 22144 requests in 30.10s, 6.06MB read Requests/sec: 735.71 Transfer/sec: 206.20KB # octane 的worker设置到8个,结果不是很理想 wrk -t8 -c400 -d30s http://octane.local.com/api/hello Running 30s test @ http://octane.local.com/api/hello 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 581.83ms 167.72ms 1.34s 74.55% Req/Sec 92.64 48.08 410.00 64.35% 20476 requests in 30.10s, 5.60MB read Requests/sec: 680.28 Transfer/sec: 190.66KB -
hyperf
wrk -t8 -c400 -d30s http://hyperf.local.com/hello Running 30s test @ http://hyperf.local.com/hello 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 59.98ms 30.18ms 223.07ms 62.56% Req/Sec 842.18 407.92 3.13k 78.64% 201302 requests in 30.09s, 43.95MB read Requests/sec: 6689.38 Transfer/sec: 1.46MB -
webman
# process 8 wrk -t8 -c400 -d30s http://webman.local.com/test/hello Running 30s test @ http://webman.local.com/test/hello 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 9.89ms 8.71ms 128.72ms 90.04% Req/Sec 5.72k 1.30k 13.28k 58.71% 1367121 requests in 30.09s, 218.97MB read Requests/sec: 45427.13 Transfer/sec: 7.28MB # process 12 wrk -t8 -c400 -d30s http://webman.local.com/test/hello Running 30s test @ http://webman.local.com/test/hello 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 10.27ms 8.21ms 120.76ms 85.99% Req/Sec 5.41k 1.45k 13.83k 62.25% 1295066 requests in 30.10s, 207.43MB read Requests/sec: 43031.45 Transfer/sec: 6.89MB
-
-
本地数据库，通过User Id查询用户信息
-
php-fpm static 200
wrk -t6 -c400 -d30s http://fpm.local.com/api/test Running 30s test @ http://fpm.local.com/api/test 6 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 1.18s 134.97ms 1.38s 93.06% Req/Sec 115.95 108.88 370.00 67.72% 9909 requests in 30.10s, 5.60MB read Requests/sec: 329.24 Transfer/sec: 190.54KB -
octane swoole
wrk -t6 -c400 -d30s http://octane.local.com/api/test Running 30s test @ http://octane.local.com/api/test 6 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 764.12ms 146.47ms 1.24s 73.90% Req/Sec 86.13 35.96 376.00 70.27% 15351 requests in 30.08s, 8.02MB read Requests/sec: 510.35 Transfer/sec: 272.95KB # octane 的worker设置到8个,结果不是很理想 wrk -t8 -c400 -d30s http://octane.local.com/api/test Running 30s test @ http://octane.local.com/api/test 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 720.49ms 188.45ms 1.58s 81.37% Req/Sec 69.49 31.71 404.00 78.24% 16474 requests in 30.10s, 8.60MB read Requests/sec: 547.30 Transfer/sec: 292.72KB -
hyperf
wrk -t6 -c400 -d30s http://hyperf.local.com/test Running 30s test @ http://hyperf.local.com/test 6 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 211.14ms 86.29ms 610.61ms 67.81% Req/Sec 314.08 85.60 540.00 66.22% 56308 requests in 30.07s, 30.12MB read Requests/sec: 1872.36 Transfer/sec: 1.00MB -
webman
# process 8 wrk -t8 -c400 -d30s http://webman.local.com/test/test Running 30s test @ http://webman.local.com/test/test 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 86.96ms 60.02ms 1.03s 84.61% Req/Sec 610.23 84.88 1.23k 76.46% 146078 requests in 30.09s, 72.15MB read Requests/sec: 4854.74 Transfer/sec: 2.40MB # process 12 wrk -t8 -c400 -d30s http://webman.local.com/test/test Running 30s test @ http://webman.local.com/test/test 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 91.43ms 69.79ms 1.35s 91.84% Req/Sec 584.84 78.12 1.03k 74.83% 139951 requests in 30.07s, 69.13MB read Requests/sec: 4654.85 Transfer/sec: 2.30MB
-
-
本地数据库，循环查执行一百条查询sql，我观测到cpu hyperf 干到100%，但是 octane_swoole在70%-80%之间
-
php-fpm static 200
wrk -t8 -c400 -d30s http://fpm.local.com/api/batch Running 30s test @ http://fpm.local.com/api/batch 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 0.00us 0.00us 0.00us -nan% Req/Sec 13.03 11.23 80.00 78.76% 1144 requests in 30.04s, 662.05KB read Socket errors: connect 0, read 0, write 0, timeout 1144 Requests/sec: 38.08 Transfer/sec: 22.04KB -
octane swoole
wrk -t8 -c400 -d30s http://octane.local.com/api/batch Running 30s test @ http://octane.local.com/api/batch 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 961.31ms 565.22ms 1.99s 57.73% Req/Sec 17.43 12.34 70.00 61.07% 1297 requests in 30.04s, 693.53KB read Socket errors: connect 0, read 0, write 0, timeout 1200 Requests/sec: 43.17 Transfer/sec: 23.08KB # octane 的worker设置到8个,结果不是很理想 wrk -t8 -c400 -d30s http://octane.local.com/api/batch Running 30s test @ http://octane.local.com/api/batch 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 1.07s 559.04ms 1.98s 58.16% Req/Sec 10.78 8.83 69.00 71.53% 1496 requests in 30.05s, 799.83KB read Socket errors: connect 0, read 0, write 0, timeout 1398 Requests/sec: 49.78 Transfer/sec: 26.61KB -
hyperf
wrk -t8 -c400 -d30s http://hyperf.local.com/batch Running 30s test @ http://hyperf.local.com/batch 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 1.29s 434.40ms 1.97s 54.43% Req/Sec 22.56 34.25 420.00 91.90% 3474 requests in 30.04s, 1.08MB read Socket errors: connect 0, read 0, write 0, timeout 3395 Non-2xx or 3xx responses: 2176 Requests/sec: 115.66 Transfer/sec: 36.81KB -
webman
# process 8 wrk -t8 -c400 -d30s http://webman.local.com/test/batch Running 30s test @ http://webman.local.com/test/batch 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 1.37s 494.45ms 2.00s 64.28% Req/Sec 23.68 15.70 110.00 73.67% 3453 requests in 30.07s, 1.71MB read Socket errors: connect 0, read 0, write 0, timeout 1734 Requests/sec: 114.82 Transfer/sec: 58.07KB # process 12 wrk -t8 -c400 -d30s http://webman.local.com/test/batch Running 30s test @ http://webman.local.com/test/batch 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 1.33s 453.25ms 2.00s 64.50% Req/Sec 17.77 11.77 80.00 68.64% 3612 requests in 30.08s, 1.78MB read Socket errors: connect 0, read 0, write 0, timeout 1705 Requests/sec: 120.08 Transfer/sec: 60.74KB
-
- 环境
- 结果：
- webman的并发效果还是挺不错的，应该是框架很精简的原因吧，不过随着阻塞io的增多，协程模式优势会慢慢体现出来
- hyperf的性能的确比octane.swoole的好
- 按道理两边都是用Swoole Server 基于非阻塞事件IO，在单查询时，两边响应时长差别不大，但当并发上来后，octane.swoole就有点力不从心了
- 原因：
- octane.swoole应该时控制器里的逻辑遇到阻塞IO时将整个进程都阻塞导致，因为CPU利用率没打满
- Swoole 跟 Roadrunner 区别
六、总结与解决
- 解决：
- 跟踪CPU负载，如果CPU消耗低于85-95%（持续），再增加worker数。
- 建议将不同请求拆分到不同实例上，然后再根据不同响应情况，配置worker数
- 将耗时的api拎出来优化，拆分，异步
- 总结：
- 非阻塞异步事件IO+内存模式在有限的硬件资源里的确可以加快php响应
- 传统大部分php组件扩展都是阻塞型的，就算用上了非阻塞事件IO也是会拖慢整个进程或线程响应，没法做到纯协程那样worker数等于cpu核心数，可以适当提高些
- 如果使用协程式的框架，后续使用其他扩展时，也需要了解里面是否存在阻塞IO操作