wordpress3

Part 3: Using nginx as a load balancer

The next part of our journey will involve creating a load balancer configuration in nginx, and using it to spread the incoming load to our WordPress site across three identical WordPress instances serving the same content. To do this, we create an upstream cluster in our nginx configuration file. For this phase of the project, we will use the files in the wordpress3 folder of our repository (this directory).

First, we will modify docker-compose.yml to start three WordPress containers, wordpress1, wordpress2, and (you guessed it) wordpress3. These containers will have identical settings – in particular, all three will share the same volume for the WordPress distribution files – this guarantees that all three will have the same settings and plug-in setup. Our docker-compose.yml now looks like this:

version: "3"

services:
  proxy:
    restart: always
    image: nginx:latest
    networks: 
      - external_network
    ports:
      - 80:80
    volumes:
      - ./nginx/tmp:/var/run/nginx
      - ./nginx/conf.d:/etc/nginx/conf.d
    command: [ nginx-debug, '-g', 'daemon off;' ]

  wordpress1:
    restart: always
    image: wordpress:6.1-apache
    depends_on:
      db:
        condition: service_healthy
    environment:
      WORDPRESS_DB_HOST: 'db'
      WORDPRESS_DB_USER: 'wordpress'
      WORDPRESS_DB_PASSWORD: 'password1234'
      WORDPRESS_DB_NAME: 'wordpress'
    networks:
      - external_network
    volumes:
      - ./wordpress:/var/www/html

  wordpress2:
    exact copy of wordpress1

  wordpress3:
    exact copy of wordpress1

  db:
    unchanged
networks:
  external_network:

In nginx/conf.d/wordpress.conf we now use an upstream group to define the services we will load balance across. The default load balancing method is round-robin (1, 2, 3, 1, 2, 3, …), but there are many options documented in the nginx docs. Among these: we can send requests to the node that has the least active connections, or to a random node in the cluster. For session-aware applications, we can always send requests from the same client to the same server. We can also weight certain containers more or less (enabling, for example, A-B testing using the load balancer). Incidentally, the nginx docs mention a http directive which is unnecessary for us, as it is already included by the configuration in the container - our wordpress.conf file is being included by that configuration. Our nginx configuration now looks like this:

upstream wpcluster {
        server wordpress1;
        server wordpress2;
        server wordpress3;
}

server {
        listen 80 default_server;
        listen   [::]:80 default_server;

        server_name example.com;
        location / {
            proxy_pass http://wpcluster;

        }
}

This time, when we start our service in wordpress3, we see five containers starting. No container images are downloaded, since we are just starting three copies of the same image for WordPress that we downloaded in the first step.

[opc@cloud-native-wordpress wordpress3]$ docker-compose up -d
[+] Running 6/6
 ⠿ Network wordpress3_external_network  Created                                                                  0.2s
 ⠿ Container wordpress3-proxy-1         Started                                                                  0.6s
 ⠿ Container wordpress3-db-1            Healthy                                                                 21.1s
 ⠿ Container wordpress3-wordpress3-1    Started                                                                 21.6s
 ⠿ Container wordpress3-wordpress1-1    Started                                                                 21.5s
 ⠿ Container wordpress3-wordpress2-1    Started                                                                 21.7s

If you run docker-compose logs -f now, you will see requests for resources are being directed to all three WordPress nodes by the proxy service. As all three share the same volume for WordPress files, and connect to the same back-end database, from the user’s perspective, they are identical.

Now that we have our service load balancing, we should figure out how to do less work, not more. That means enabling some caching. In our next step, we will add a Redis object cache for WordPress, and add a caching plug-in to WordPress to provide page caching.