DEV Community

Cover image for How to Docker Compose a developer environment: an open source example
Dan Lebrero
Dan Lebrero

Posted on • Originally published at danlebrero.com

How to Docker Compose a developer environment: an open source example

This article originally appeared on my blog

As we mentioned in a previous blog post, you should strive to have a simple and repeatable way of setting up a dev environment for your project.

In this blog post we are going to go into details of an example from one of the open source projects at Akvo.

IMHO, having a painless way of setting up a dev environment is one of the key aspects to remove some of the friction for contributors to open source projects.

Akvo Lumen Architecture

The project that we are going to look at is called Akvo Lumen, which is an “easy to use data mashup, analysis and publishing platform”.

Akvo Lumen is a Javascript single page application (SPA) with a Clojure backend, as follows:

Akvo Lumen architecture

  • Keycloak is an open source single sign on application. It is shared with other Akvo products.
  • Nginx serves the SPA and proxies the request to the Backend.
  • The Backend is the backend ...

Before docker compose

The instructions to setup the dev environment were what we can call the “classic” ones: install Postgres, run this init script, download this build tool, run this command ...

But in this particular case they were split into three files: keycloak, backend and SPA.

Personally, while following the instructions, I did not realized that I had to run KeyCloak locally and one of the npm dependencies would not compile in my developer box. I still do not know why. I don't really want to know.

Of course, there is a better way.

Instructions after docker compose

The new instructions after docker compose are here, which can be summarized as:

sudo sh -c 'echo "127.0.0.1 t1.lumen.localhost t2.lumen.localhost auth.lumen.localhost" >> /etc/hosts'
docker-compose up -d && docker-compose logs -f --tail=10
Enter fullscreen mode Exit fullscreen mode

The first step is required because Lumen is a multi tenant product and the tenant is based on the host.

The second step is basically the same as docker-compose up, but without docker-compose holding your console hostage.

Interestingly, note the absence of any npm install or mvn install from the instructions.

After that, we will be running:

Akvo Lumen developer docker compose setup

The docker compose file

The whole docker compose file can be found at here.

Postgres and Keycloak

Let’s start by looking at the Postgres image:

postgres:
  build: postgres
  ports:
   - "5432:5432"
Enter fullscreen mode Exit fullscreen mode

Strictly speaking, we do not need to expose the Postgres port, but it is useful during development to be able to inspect the DB tables with some UI tool.

The Dockerfile is extremely simple:

FROM postgres:9.5
ADD ./provision  /docker-entrypoint-initdb.d/
Enter fullscreen mode Exit fullscreen mode

Following the instructions of the Postgres official image, we copy our initial setup scripts and they will be run the first time the container starts.

The scripts just creates a bunch of empty databases. It will be the Lumen Backend the one that creates the required tables and reference data as part of the DB migration logic.

The Keycloak image is very similar, but setups the initial set of users, passwords and credentials in a Keycloaky way.

Lumen Backend

The Lumen Backend is a Clojure service. Its docker compose configuration looks like:

backend:
 build:
   context: ./backend
   dockerfile: Dockerfile-dev
 volumes:
    - ./backend:/app
    - ~/.m2:/root/.m2
    - ~/.lein:/root/.lein
 links:
    - keycloak:auth.lumen.localhost
 ports:
    - "47480:47480"
Enter fullscreen mode Exit fullscreen mode

The first interesting point is that it uses a different Dockerfile than the production one.

During development, we need our build tools, Lein in our case, plus we want the fast feedback cycle that a good REPL provides, while in production we just want a fast start up time.

Note that because our build tools come as part of the Docker image, everybody in the team will be running exactly the same version of Lein, on exactly the same JVM and OS. Other projects could use different Lein versions, or different tools, but containers isolate one project from the others.

We do not want to be rebuilding and restarting our Backend Docker image every time we make a change in our source files, so the first line of “volumes” (- ./backend:/app) makes the source code available to the Docker container: any change in the source files will be immediately visible inside the container.

The second volume that we mount is the local maven repository (- ~/.m2:/root/.m2). This somehow pollutes your developer box, as deleting the Docker container will not get rid of the downloaded dependencies, but in theory your local maven repository is just a cache, so you can delete it without repercussions whenever it gets too big.

If you don’t want to pollute your developer box at all, you can make use of layers and download the dependencies just when there is a change in the project file.

The last volume (- ~/.lein:/root/.lein) makes the Lein global profiles available to the container. Use with care as you want to avoid any “it works on my machine” issues.

Even if the Keycloak container is accessible by the Backend using the hostname “keycloak”, we need the link (- keycloak:auth.lumen.localhost) due to JWT validation requiring the single sign-on host to be the same one for the client (the browser) and the backend.

Last, we make the REPL port available so you can connect to it with your favourite IDE. You will need to explicitly configure the lein :repl-options to listen to that port and to allow connections from any host.

Lumen Client

The Lumen Client image is a Nginx that servers the SPA and proxies other requests to the backend.

client:
 build:
   context: ./client
   dockerfile: Dockerfile-dev
 volumes:
    - ./client:/lumen
 ports:
    - "3030:3030"
Enter fullscreen mode Exit fullscreen mode

Again, for development we prioritize a fast feedback cycle, so the Docker images between production and development are different.

In this case, we have replaced Nginx with a webpack Dev Server which will recompile the SPA and do a hot code reload on your browser whenever we make a change on our source code. The mounted volume is to make the source code accessible inside the container.

The exposed port is just the main application entrypoint.

Running tests and other build tasks

As you don't need to install any npm, lein or maven in your local box, to run any tasks provided by those tools, you just need to run them from within the Docker container.

For example, to run the Backend tests:

docker-compose exec backend lein test
Enter fullscreen mode Exit fullscreen mode

Or if you are going to run several commands, you can always start a bash shell:

docker-compose exec backend bash
Enter fullscreen mode Exit fullscreen mode

Tip: if you want to preserve the bash history, just add another volume to the Docker Compose file that mounts the home directory.

A note on startup dependencies

Docker Compose provides very little help to ensure the startup order of the containers.

It is up to you to make sure that the dependant container waits long enough for the dependency to be ready, usually by polling with some maximum time limit.

For this project, the Backend depends both on Keycloak and Postgres, but the Backend consistently takes longer than both to startup, so we are ignoring the issue for now.

Examples in other projects on how to deal with the startup dependency issue:

  1. Checking that a DB is ready by querying the last table created for some data
  2. Checking that Kafka is ready by listing the topics and finding the last created one

An environment upgrade

It happens that one of the new features in Akvo Lumen is to provide some interactive maps.

This means that the project now needs:

  1. PostGis extension installed on Postgres.
  2. A Windshaft node.js server.
  3. A Redis DB.

Which now that we have Docker Compose can be done with:

--- a/postgres/Dockerfile
+++ b/postgres/Dockerfile
-FROM postgres:9.5
+FROM mdillon/postgis:9.6 

--- a/postgres/provision/helpers/create-extensions.sql
+++ b/postgres/provision/helpers/create-extensions.sql
+CREATE EXTENSION IF NOT EXISTS postgis WITH SCHEMA public;

--- a/docker-compose.yml
+++ b/docker-compose.yml
+ redis:
+   image: redis:3.2.9
+ windshaft:
+   image: akvo/akvo-maps:2469ae0cb95ba090412f042fdfa8c7038273fe0e
+   environment:
+     - NODE_ENV=development
+   volumes:
+      - ./windshaft/config/dev:/config
Enter fullscreen mode Exit fullscreen mode

And one docker-compose down; docker-compose up --build later, the whole team is enjoying the new setup.

Isn’t that beautiful?

Top comments (13)

Collapse
 
prees1 profile image
Philip Rees • Edited

Instead of adding entries to your local /etc/hosts file. You could use readme.localtest.me/

and use t1.lumen.localtest.me and admin.lumen.localtest.me during development.

Collapse
 
danlebrero profile image
Dan Lebrero

Love it! What a nifty idea. I will give it a try.

Collapse
 
jaapio profile image
Jaap van Otterdijk

Or try to use a dsnmasq solution polderknowledge.nl/2017/02/16/dock...

Collapse
 
danfromisrael profile image
Dan Mordechay

Very nice article!
I was trying to do a similar setup and also ran into a problem with dependencies startup.
I still didn't have time but its worth looking into dockers health check feature
Looks like exactly what we need.
Maybe thats the next blog post?
😉

Collapse
 
danlebrero profile image
Dan Lebrero

In version 2.1 of docker compose, there was a condition option on the depends_on which was removed on version 3.

The 3.0 version states:

“Version 3 no longer supports the condition form of depends_on.“

And points again to docs.docker.com/compose/startup-or...

So I think we are out of luck right now, but let me if you find a sensible way!

Collapse
 
danfromisrael profile image
Dan Mordechay
Thread Thread
 
danlebrero profile image
Dan Lebrero

That is what Docker Compose v2.1 was using on the condition option on the depends_on, and that is what Docker Compose v3 removed.

Of course, I may be wrong, so by all means please investigate and let me know!

Thread Thread
 
danfromisrael profile image
Dan Mordechay • Edited

i think you're right about depends_on but we might be lucky with healthcheck:
docs.docker.com/compose/compose-fi...
(this will only tell us an indication if the container is healthy or not).
here's an example to try (docker not docker-compose):
github.com/tomwillfixit/healthcheck

also, i was trying to find depends_on with v3 out there in the wild and found this example:
github.com/Eficode/Docker-Selenium...

so i must say i'm confused.
i hope i'll have the time to play with it myself soon to figure it out

Thread Thread
 
danlebrero profile image
Dan Lebrero

Please let me know if you find anything!

Collapse
 
jrock2004 profile image
John Costanzo • Edited

The other issue you can run into is the more files and folders that go into the m2 the more it can slow down your container. The reason being its more files for it to monitor on both the container and the host side

Also ~/.m2:/root/.m2 will that tilda work on windows hosts?

Collapse
 
danlebrero profile image
Dan Lebrero • Edited

Great tip! Thanks a lot!

Not sure about the tilda. Nobody is using Windows over here :)

Collapse
 
terceranexus6 profile image
Paula

Very useful thanks!

Collapse
 
danlebrero profile image
Dan Lebrero

You are welcome!