Composing containers

As Docker compose is the same as Docker in many aspects, it's more efficient to understand how to write docker-compose.yml with examples than start from docker-compose syntax. Let's now go back to the kiosk-example we looked at earlier and start with a version definition and four services:

version: '3'
services:
  kiosk-example:
  recorder-example:
  lcredis:
  lmysql:

The docker run arguments for kiosk-example are pretty simple. They include a publishing port and an environment variable. On the Docker compose side, we fill the source image, the publishing port, and environment variables accordingly. Because Docker compose is able to handle docker build, it can build images if those images can't be found locally. We want to use this to decrease the effort of image management:

kiosk-example:
  image: kiosk-example
  build: ./kiosk
  ports:
  - "5000:5000"
  environment:
    REDIS_HOST: lcredis

Converting the Docker run of the recorder-example and redis in the same manner, we have a template that looks as follows:

version: '3'
services:
  kiosk-example:
    image: kiosk-example
    build: ./kiosk
    ports:
    - "5000:5000"
    environment:
      REDIS_HOST: lcredis
  recorder-example:
    image: recorder-example
    build: ./recorder
    environment:
      REDIS_HOST: lcredis
      MYSQL_HOST: lmysql
      MYSQL_USER: root
      MYSQL_ROOT_PASSWORD: mysqlpass
  lcredis:
    image: redis
    ports:
    - "6379"

For the MySQL part, MySQL requires a data volume to keep its data as well as its configurations. In addition to the lmysql section, we add volumes at the level of services and an empty map called mysql-vol to claim a data volume:

  lmysql:
    image: mysql:5.7
    environment:
      MYSQL_ROOT_PASSWORD: mysqlpass
      MYSQL_DATABASE: db
      MYSQL_USER: user
      MYSQL_PASSWORD: pass
    volumes:
    - mysql-vol:/var/lib/mysql
    ports:
    - "3306"
volumes:
  mysql-vol: {}

One of the benefits of this is that we can manage the launching order between the components with depends_on. What this does is maintain the order; it can't detect whether the components that it will use are ready. This means our application could still connect and write to the database before the database is ready. All in all, as our program is a part of a distributed system with many moving parts, it's a good idea to make it resilient to the changes of its dependencies.

Combining all of the preceding configurations, including depends_on, we have the final template, as follows:

version: '3'
services:
  kiosk-example:
    image: kiosk-example
    build: ./kiosk
    ports:
    - "5000:5000"
    environment:
      REDIS_HOST: lcredis
    depends_on:
    - lcredis
  recorder-example:
    image: recorder-example
    build: ./recorder
    environment:
      REDIS_HOST: lcredis
      MYSQL_HOST: lmysql
      MYSQL_USER: root
      MYSQL_ROOT_PASSWORD: mysqlpass
    depends_on:
    - lmysql
    - lcredis
  lcredis:
    image: redis
    ports:
    - "6379"
  lmysql:
    image: mysql:5.7
    environment:
      MYSQL_ROOT_PASSWORD: mysqlpass
      MYSQL_DATABASE: db
      MYSQL_USER: user
      MYSQL_PASSWORD: pass
    volumes:
    - mysql-vol:/var/lib/mysql
    ports:
    - "3306"
volumes:
  mysql-vol: {}

This file is put in the root folder of a project. The corresponding file tree is shown here:

├── docker-compose.yml
├── kiosk
│   ├── Dockerfile
│   ├── app.py
│   └── requirements.txt
└── recorder
    ├── Dockerfile
    ├── process.py
    └── requirements.txt

Finally, run docker-compose up to check everything is fine. We can check every component is linked nicely using kiosk:

$ curl localhost:5000
Gooday!
 Import tickets with "curl -XPOST -F 'value=<int>' /tickets"
 Purchase a ticket with "curl -XPOST /buy
 Get current tickets with "curl -XGET /tickets"
$ curl -XGET localhost:5000/tickets
0
$ curl -XPOST -F 'value=10' localhost:5000/tickets
SUCCESS
$ curl -XGET localhost:5000/tickets
10
$ curl -XPOST localhost:5000/buy
SUCCESS
$ docker exec chapter2_lmysql_1  mysql -u root -pmysqlpass \
              -e "select * from kiosk.sellinglog;"
id  ts
1   1536704902204

Writing a template for Docker compose is as simple as this. We're now able to run an application in the stack with ease.