Skip to content

Latest commit

 

History

History
415 lines (311 loc) · 14.4 KB

README.md

File metadata and controls

415 lines (311 loc) · 14.4 KB
Raw

Python Docker Compose Control

This library can be used to control multiple docker compose-based projects in a coordinated way, allowing you to start, stop, and developing on separate projects as necessary.

This project was built from requirements at Adobe that arose when splitting a monolith project into microservices. It simplifies this transition and makes it easy to run multiple projects at the same time whether pulling from built/deployed upstream docker images or developing using local code.

Features

  • Coordinates the execution of multiple docker compose projects in a single network/namespace
  • Start or stop an entire application, no matter the number of projects or docker-compose files, with a single command
  • Extremely configurable, add your own commands and projects
  • Integrates directly with docker compose, as long as you can use docker compose files, this project can control them
  • Allows easy development of a single project or multiple projects at the same time
  • Generate dynamic docker-compose files using Jinja templates for any number of projects
  • Manages pre-commit config file(s) by copying and installing in each configured project

Setup

Application filesystem layout

This project assumes a flat directory layout for projects, with the control project located in the same directory as all other projects. For example:

# Arbitrary directory containing all projects for an application,
# may also contain other directories/applications as well
my-app/
  # Control project, uses pydc_control
  app-control/
    # This is arbitrary, but recommended to be able to add to the path
    bin/
      # Control script - executable python script that uses pydc_control
      appctl
    # Configuration file used to build base docker-compose template
    config.yml
    # Environment variable configuration, symlinked to all projects automatically
    docker-compose.env
    # Optional pre-commit config file(s)
    # This needs to be configured in each project to be copied and installed
    my-pre-commit-config.yaml
  # Arbitrary projects
  project1/
    # The only required file for a project is a docker-compose file
    docker-compose.yml
  project2/
    # A jinja-templated file may be used instead (see advanced features below)
    docker-compose-template.yml
  ...

Determine configuration variables

Determine these variables up front:

  • Docker network name (mynetwork in the examples below)
  • Docker compose service and container prefix (mynamespace_ in the examples below)
  • Core service prefix (core_ in the examples below)
  • Docker compose project name (myproject in the examples below)
  • Docker registry and tags that will be used for deployed containers

Create a control project

Create a new project for your control project. It can be configured however you'd like, but it should either have a script available on the path for easy execution or install it into a local environment for easy execution.

Control script

Add a single python script to the project. Using naming such as <app>ctl is recommended to make it easy to execute.

#!/usr/bin/env python3

import os
import sys
import pydc_control


if __name__ == '__main__':
    # The base path is the location of your control project
    base_path = os.path.realpath(os.path.join(os.path.dirname(__file__), '..'))
    # Run pydc_control, returns an exit code
    sys.exit(pydc_control.run(base_path))

Configuration file

Copy the config.yml.example file (or create your own) and add it to the control project. Every project in your application should have an entry in the projects list in this file, with one or more services (containers) attached to it. The syntax is similar to a docker-compose file, but has a layer of abstraction to handle projects with multiple services. This file will be used to generate the base docker compose file during control script runs.

For services that have no associated project, for example, a MySQL or message bus container, you can create an arbitrarily named project that has null values for the directory and repository properties.

For projects that have no associated services, for example a project containing static configuration, you can create a project with the services property set to an empty list ([]).

Note that several settings near the top of the file are required.

Environment variables file

Copy the docker-compose.env.example file (or create your own) and add it to the control project. This file should be ignored by your VCS. It may be helpful to create a docker-compose.env.example file of your own in the project with placeholders that may be filled in by individual developers when setting up their own environment.

Any environment variables defined here will be added automatically to any running container (unless they define env_file: [] in the configuration file).

Setup projects

Each project that pydc control can own should contain at least a docker-compose.yml file. The docker compose should be quite standard, but typically must contain at least one service that is prefixed with the service/container prefix determined above and the network as determined above. The service should also be attached to that network.

For example:

version: '3'

services:
  # This prefix MUST match the service prefix determined above
  mynamespace_ping:
    image: containous/whoami:latest
    ports:
    - "80"
    networks:
    - mynetwork

networks:
  mynetwork:
    external: true

Setup template projects

Projects with docker compose templates can add some variables that are pulled automatically from the config file:

version: '3'

services:
  {{ service_prefix }}_ping:
    image: containous/whoami:latest
    ports:
    - "80"
    networks:
    - {{ network }}

networks:
  {{ network }}:
    external: true

Add README

Copy the README-example.rst to the README.rst file in your control project. Follow the instructions at the top of the file to customize it for your project.

Usage

NOTE: All examples using appctl as the command is just an example. This script may be named whatever you would like.

To use your control script, simply add it to the PATH and call it from any directory. The behavior changes based on the directory you are in:

  • If called from a project directory (listed in config.yml), that project is assumed to be currently developed, meaning it is automatically added to the -p parameters in the control script. This will use the development docker-compose file from the project and all other containers from the generated base docker-compose in the control project (built from config.yml).
  • If called from any other project directory (included the control project), it will only include the generated docker compose from the control project (built from config.yml).

Develop projects

To add more development projects besides the current project directory, simply use the -p flag:

appctl -p project2 config

Run docker compose commands

Most common docker compose commands have shortcuts added to the control script, with an additional command that may be used to pass arbitrary commands to docker compose.

appctl config
appctl up
appctl up-detach
appctl down
appctl stop
appctl build
appctl pull
appctl docker-compose -- <additional docker compose args>
# Alias for docker-compose command
appctl dc -- <additional docker compose args>
# see appctl --help for all commands available

Container startup order

Services are started up using the following method:

  • All services (containers) marked with core: true are started detached

  • All open ports defined on the core services are checked to make sure they are open

  • Any wait-for-ports (see below) are waited for via requests

  • All services belonging to projects that are not being developed are started detached

    • Note: If no projects are being developed (see above), all services are started and are not detached.

  • All developed project services are started and logs are displayed for only these services

The reason for the detached behavior for many of the containers is that they can run in the background and are often not changed. Additionally, this prevents logs from showing up for them in the console, which could cause extra noise when developing on only one or two projects.

If the control script process is interrupted (via Ctrl+C for example), the developed project services are attempted to be stopped (and only these) so that they may be restarted again. All of these operations and choices are to ensure the smoothest experience with docker compose logs and interactions so that you can focus only on your developed projects.

Perform VCS checkout, pull, status (git only)

To clone and/or update all projects listed in the config.yml automatically, use the following command:

appctl checkout
# Alias
appctl co

To check the repository status for every project:

appctl repo-status
# Alias
appctl rs

Advanced features

Dynamically generating image references

While many image references can be hardcoded, it may be desirable to generate image references based on specific tags or with specific registries based on those tags. This can be done by using the image_path key in the service definition in config.yml instead of image. The full image reference is generated from 3 pieces of information:

  • The image path defined in the image_path property for the service in config.yml

  • The tag defined on the command line of your control script (-t)

    • The full list of tags is defined in the docker-compose.tags property of the config.yml file.

  • The docker-compose.registry property defined in config.yml

    • The docker-compose.registries-by-tag property may be used to override the registry based on the tag value.

The full image reference is of the form <registry><path>:<tag>.

Templating docker-compose.yml

In order to use a template instead of a direct docker-compose.yml file in a project, simply create a file named docker-compose-template.yml and then make sure that docker-compose.yml file is ignored by your VCS since it will be regenerated on each run of the control script. The template file is processed via Jinja and has several variables available to it such as:

  • dev_project_names - The names of the project currently under development
  • enabled_services - The services that are marked as enabled and are currently enabled
  • tag - The docker tag selected
  • registry - The docker registry from the config file
  • network - The docker network from the config file
  • core_prefix - The prefix to use for core containers
  • service_prefix - The prefix to use for service containers

Making services enabled/disabled

Sometimes it is desirable to not start all services when starting up the application with the control script, but be able to start these services when needed. For example, a service that takes a lot of resources to run and is rarely used may not need to be started every time, but when testing functionality involving the service, it may be enabled explicitly. Similarly, it may be desirable to be able to disable specific services.

Simply add the enable: true or disable: true flags to your service definition in config.yml to make the service disabled by default or enabled by default respectively. Flags are added to the control script automatically based on the service name to enable/disable the service.

For example, if a service needs to be disabled by default, in your config.yml:

projects:
  project1:
    ...
    service:
    - name: service1
      enable: true
      ...

This adds the --enable-service1 flag to the control script automatically and prevents the service from being included in docker compose otherwise.

If you would like to use the same behavior in development projects, use the enabled_services dictionary in your docker-compose-template.yml file to tell if a service should be included or not:

services:
  {%- if enabled_services.get('service1') %}
  mynamespace_service1:
    ...
  {%- endif %}

Using enable flags from other services

Sometimes it is desired to use a single enable flag to enable multiple services across projects. This may be done by setting the enable or disable flags to the name of the service to use for enabling/disabling. This requires that the other service also has the same flag set to true. Multiple levels of enable/disable redirection are not possible (e.g. s1 is enabled with s2 which is enabled with s3).

Create your own control script commands

It is easy to create additional commands using pydc_control.

#!/usr/bin/env python3

import argparse
import os
import sys

import pydc_control


def run_db_connect(args: argparse.Namespace):
    """
    Connects to the mongo my_db database running on "my_mongo_container"
    """
    pydc_control.call_commands(
        ['docker', 'exec', '-it', 'my_mongo_container', 'mongo', 'my_db']
    )
    return os.EX_OK


def configure_parsers(parser: argparse.ArgumentParser, commands_parser: argparse._SubParsersAction):
    # DB
    db_parser = commands_parser.add_parser(
        'db',
        help='Connects to the mongo database in an interactive shell',
    )
    db_parser.set_defaults(
        func=run_db_connect,
    )


if __name__ == '__main__':
    base_path = os.path.realpath(os.path.join(os.path.dirname(__file__), '..'))
    sys.exit(pydc_control.run(base_path, configure_parsers))

Frequently Asked Questions

Why do we duplicate docker compose configuration between projects and the control config?

While it may seem like duplicated code/configuration, the docker compose files in each project and the configuration file in the control project serve different purposes and often look different:

  • The configuration file in the control project should use a deployed (production or stage-like) docker image and configuration for the project. The code for each project is not dynamically modified and is usually promoted/deployed to environments outside of pydc-control.
  • The docker compose file in the individual (developed) project should use a development build of the docker image with, ideally, application files and configuration mounted dynamically into the container so that development is seamless and immediate. Live-reloading should be used when available to speed development of a project.