worker_connections 还不够——Nginx、docker

rav*_*ven 6 nginx reverse-proxy wss docker mqtt

在我的生产服务器中,我们有几个上游,它们是在带有 nginx 的反向代理后面运行的 docker 容器。其中一个容器是 mqtt 代理(mosquitto),我们用来通过 websockets 进行连接。这是我们的 nginx.conf 文件:

worker_processes 1;

events {
  worker_connections  1024;
}

http {
    upstream br-frontend {
        server br-frontend:3000;
    }

    upstream br-backend {
        server br-backend:5000;
    }

    upstream mosquitto {
        server mosquitto:9001;
    }

    map $http_upgrade $connection_upgrade {
        default upgrade;
        '' close;
    }

    ssl_certificate /etc/nginx/certs/server.crt;
    ssl_certificate_key /etc/nginx/certs/server.key;

    server {
        listen 443 ssl default_server;
        server_name _;

        location / {
            proxy_pass http://br-frontend/;
        }

        location /api {
            proxy_pass http://br-backend;
        }

        location /swagger.json {
            proxy_pass http://br-backend/swagger.json;
        }

        location /swaggerui {
            proxy_pass http://br-backend/swaggerui;
        }

        location /mosquitto-ws {
            proxy_pass http://mosquitto;
            proxy_http_version 1.1;
            proxy_set_header Upgrade $http_upgrade;
            proxy_set_header Connection $connection_upgrade;
        }
    }

    server {
        listen 80 default_server;
        server_name _;
        return 301 https://$host$request_uri;
    }
}
Run Code Online (Sandbox Code Playgroud)

昨天,我们的生产服务器因为以下错误而崩溃。我读过我可以增加 worker_connections 的数量,但我认为这不是最好的解决方案。我还在其他问题中读到我的 nginx.conf 文件中可能有一个无限循环,但我看不到它。

2018/12/14 00:23:12 [alert] 6#6: 1024 worker_connections are not enough
2018/12/14 00:23:13 [alert] 6#6: *14666 1024 worker_connections are not enough while connecting to upstream, client: *.*.*.*, server: _, request: "GET /mosquitto-ws HTTP/1.1", upstream: "http://172.21.0.5:9001/mosquitto-ws", host: "****"
Run Code Online (Sandbox Code Playgroud)

更新: docker-compose.yml

version: '3'
services:
    mongodb:
      image: mongo:latest
      volumes:
        - './data/db:/data/db'
        - './data/configdb:/data/configdb'
      ports:
        - 27017:27017
    br-backend-express:
      working_dir: /app
      command: npm run execute-prod
      image: ${ACR}/br-backend-express:${tag}
      ports:
        - "5000:5000"
      depends_on:
        - mongodb
    mosquitto:
      image: ${ACR}/mosquitto:${tag}
      depends_on:
        - br-backend-express
    br-bridge:
      working_dir: /app
      image: ${ACR}/br-bridge:${tag}
      command: npm run execute-prod
      depends_on:
        - mosquitto
        - mongodb

    br-frontend:
      image: ${ACR}/br-frontend:${tag}

    nginx:
      image: ${ACR}/nginx:${tag}
      ports:
        - 443:443
        - 80:80
      depends_on:
        - br-frontend
        - br-backend-express
Run Code Online (Sandbox Code Playgroud)

任何帮助,将不胜感激。谢谢。

poi*_*ige 1

根据Nginx 文档worker_connections“\xe2\x80\xa6 设置工作进程可以打开的最大并发连接数。应记住,该数字包括所有连接(例如与代理服务器的连接等) ,不仅仅是与客户端的连接。\xe2\x80\xa6"

\n\n

鉴于您的配置使用代理连接,在足够密集的流量下,Nginx 会开始缺乏可用连接,这并不奇怪。

\n\n

里面也提到了worker_rlimit_nofile要调整的地方,所以大家也要注意这一点。

\n\n

最后,虽然它与问题没有直接关系,但我还是建议使用它worker_processes auto来在可用核心之间更好地分配 Nginx 负载。

\n