Wednesday, September 21, 2016

concept of docker-compose-cauldron tool

The docker-compose.yml  format is very stiff (and Dockerfile as well ) :
  • environment variables (.env) don't allow to have default values, lists etc 
  • extend is very limited - multi-value sets are merged, links don't inherited etc
  • Dockerfile can have conditions for RUN, but not for other commands
  • there is no repository for compose.yml files and collections 
as the end every project ends with very specialized config which hard to support (updates, new features, fixes)

So there will be great to have tool which generates all needed files with use of some templating engine (if/for/while support inside docker-compose.yml, Dockerfile and other files as well)

use cases:

1. docker-compose-cauldron init {git url or name at repository}
- downloads source of template (into local subdir or somewhere under /home/.cache/
- creates docker-couldron.yml.example and docker-couldron.yml (project wide settings, i.e. will be stored in git) and .env.example (sample config for local environment (i.e. development/dev settings etc), stored in git but copied/edited to .env for local environment )

2. docker-compose-cauldron build
Builds docker-compose.yml and related "build" files for local environment (does not stored in git) based on settings from docker-couldron.yml and template

3. docker-compose-cauldron update
Updates source of template, as well docker-couldron.yml.example and .env.example , user should manually update docker-couldron.yml and .env files



Monday, September 5, 2016

SOLVED: docker-compose. Error starting userland proxy: listen tcp 0.0.0.0:80: bind: address already in use

I configured new set of containers (via docker-compose),  and tried them to run

>docker-compose up -d
...
Error starting userland proxy: listen tcp 0.0.0.0:80: bind: address already in use 


googling returns me pretty inconvenient solutions:
  • just stop local web/mysql server or use different port for container. 
  • don't publish ports at all, i.e. remove '-ports'/'-p'  
  • use reverse-proxy before containers for web servers 
But I need something less troublesome, I need different/independent sets of containers running, i.e. different projects with different needs, with easy way to setup/remove, so I ended with next solution:
  • tie each set of containers (from single docker-compose.yml ) to docker custom networks i.e. all containers from set are bound to some unique IP range
  • use /etc/hosts to bind IPs to names  and always have regular ports 80, 443, 3306 etc accessible from local host
  • use of environment file (.env) for local config (so team can use the same config, while each member can customize important values) 

so, this is how it looks:
>cat .env
GATEWAY_IP=172.16.240.1
SUBNET=172.16.240.0/24
COMPOSE_PROJECT_NAME=myproject1

>cat docker-compose.yml
version: '2'

services:
  nginx:
    image: nginx:latest
    ports:
      - "${GATEWAY_IP}:80:80"
    networks:
      - export

networks:
  export:
    driver: bridge
    ipam:
      driver: default
      config:
      - subnet: "${SUBNET}"
        gateway: "${GATEWAY_IP}"

and docker marges both files and see config as:
>docker-compose config                  
networks:
  export:
    driver: bridge
    ipam:
      config:
      - gateway: 172.16.240.1
        subnet: 172.16.240.0/24
      driver: default
services:
  nginx:
    image: nginx:latest
    networks:
      export: null
    ports:
    - 172.16.240.1:80:80
version: '2.0'
volumes: {}


>docker-compose up -d                   
Creating network "myproject1_export" with driver "bridge"
Creating myproject1_nginx_1


>docker-compose ps                    
       Name                Command          State                Ports              
------------------------------------------------------------------------------------
myproject1_nginx_1   nginx -g daemon off;   Up      443/tcp, 172.16.240.1:80->80/tcp



Now, just add into your /etc/hosts file some like
172.16.240.1 myprojecthost

and go to http://myprojecthost to check if it works (MySQL will be on myprojecthost:3306 etc)

Some additional notes:
  • you will need to link every service in docker-compose.yml  to same network, i.e.

        networks:
          - export


    Otherwise, docker-compose will create two networks - default and custom one, and if services don't share common network, they won't communicate.
    Of course, you can use default network too for services you want to publish:

        networks:
          - export
         
          - default

    then there no need to update every service
     
  • do not add .env in git , just create .env.sample , so every developer can copy it to .env and use customized config 
  • docker-compose names containers based on directory name, this is inconvenient if you work with different project with similar directory structure, because docker tools checks only name of container (i.e. down/restart etc will catch containers from other directories).
    To make things manageable COMPOSE_PROJECT_NAME environment variable is used. Still, it should be unique per docker host/machine, or docker tools will work incorrectly