Facing the problem.
Reusable and maintainable development environments
A common challenge we face while building applications, is replicating the production environment in our development machine.
This would be easier if we were the only developer in a project or we had just a one single application. In such case, the easiest way to prepare the development environment is to install all the needed packages and dependencies into our local machine.
However, within a team or dealing with multiple services and applications, replicating the same environment than production is way more difficult. Different applications have different runtime dependencies. Replicating the same configuration manually, over every single development machine could lead to inconsistencies among developers. This could be a maintainability nightmare.
In order to make this task easier and build reusable and maintainable configurations across applications and team members Vagrant comes to the rescue. Creating virtual machines for development environments has never been easier.
This article is not about how to get started with Vagrant. It assumes a basic knowledge of how Vagrant works. If you don’t have ever get in touch with Vagrant, I suggest to read the Getting Started Vagrant guide before continue reading the rest of the entry.
The solution to the problem could have been to use Vagrant to handle the creation of development environments.
1 2 3 4 5 6 7 8
I think this could be the minimum
Vagrantfile that a project could have. Generally applications that build its
development environment using Vagrant have slightly more complex provisioning setups that could use Ansible,
Puppet, Chef or some other IT automation system. Currently,
Vagrant supports all of the mentioned and many many more.
Tradeoffs, tradeoffs, tradeoffs …
This is a good approach to maintain the creation of development environments. So every time we need to bootstrap any development environment it’s as easy as
And vagrant brings a virtual machine with all of the application’s runtime set up properly to start (or continue) working in the project. And yes, you’ve read it well. Every time we need a new development environment, vagrant creates one or more new virtual machines if not have been yet created.
So if we have a slightly more complex architecture in production, it means probably that we will need several virtual machines to be able to handle this. For example, a typical web scenario could be to have some sort of CDN (Akamai, Amazon CloudFront, etc.) to handle all the static content of the site. If it doesn’t have the static asset, it asks for it to our web infrastructure. Generally, this infrastructure consists of an nginx or cluster of nginx that talks with the CDN.
This is just an example, but I would not catalog it as complex. It’s a common example of a web scenario. Another example, could be if we have some sort of intermediate caching layer. I think of a Varnish for example. Or if we have some operations in the business logic that performs something in memory with some Redis instance.
How do we face the creation of all those services? We could create a virtual machine for every service for example. Having a dedicated virtual machine for every single service just to mimic our production infrastructure, would be the ideal case but I think it’s not possible. Just because, unless we have truly monsters as local host machines, the resources of them are limited.
So one possible solution to this problem, is to build a single vagrant machine to provide all the services that the application needs. Although this is a perfectly fine solution to the problem, we can come with another lightweight solution by using Docker.
Unless you live secluded in a dark cave, you probably have heard about Docker. In short, Docker is a client/server application that is capable to run processes in an isolated way, by running each one in an isolated container. That is, Docker is just another virtualization platform that make use of LXC to provide virtualization at process level rather than in entire machine level.
It’s out of the scope of the entry to provide a Getting started with Docker. In fact, It assumes that the reader has a basic knowledge of how Docker works. So if you don’t have tried Docker, I suggest you to stop here and take the tutorial that Docker’s website provides.
(NOTE: If you are not using OS X you can skip ahead this section and go directly to Docker to the rescue)
The primarily limitation I’ve found when using Docker for building development environments, is that OS X (I’m currently using a Mac Book Pro) does not have native support for LXC. So Docker cannot run natively in OS X.
But this is not a major real problem, because the Docker guys have already thought of this and the docker client can
be configured to run the Docker commands to the Docker server we want. They have created a project called
boot2docker is a command-line utility that spawns a lightweight
virtual machine through VirtualBox, that supports natively LXC and can run Docker commands.
But, although boot2docker is an excellent project that allow us to spawn Docker containers in hosts which don’t support LXC, it has from my point of view a major limitation that makes it unable to be used in the development of large applications: It only supports file sharing through VirtualBox Shared Folders. That is, it does not have support for NFS and VirtualBox SharedFolder are terrible slow with a large number of files.
So if you have an application with a large number of dynamic includes (that is applications written in an interpreted
language: PHP, Ruby, Python, etc.) I don’t recommend the usage of the virtual machine that provides
Maybe in the future they will address this major problem. We, all of the web developers that work with large
applications, hope so :)
Fortunately, the usage of the virtual machine spawned by
boot2docker is not mandatory. We can use another
custom virtual machine that could operate directly with
With Docker the schema changes a bit. Instead of having a virtual machine with all the services running inside or having several virtual machines one per service to allow them to run in an isolated way, we can bootstrap a single virtual machine and have all the processes that the application need to run, in an isolated way.
Setting up a VM to operate with Docker (OS X)
Again this only applies if you’re under OS X. If you are under some distribution with native support for LXC you will probably don’t have to worry about all of this.
We can use Vagrant to spawn a virtual machine that acts as if it was provided by boot2docker, but without all the
limitations of boot2docker. That means, we can spawn a new virtual machine with NFS support that will be able to
create Docker containers. In fact, someone has already created a Vagrant box for that:
https://vagrantcloud.com/yungsang/boxes/boot2docker. This is an
improved version of
boot2docker with some extra benfits but the primary, at least for me, is the support for
NFS shared folders.
To make the point clear this does not mean that docker volumes and data-only volume containers are mounted through NFS, this means that we can mount NFS shared folders on that machine and then make use of that folders to build volumes and data-only volume containers. So we still have to manually mount the shared folders on the machine, but this drawback with Vagrant is totally painless and can be assumed with almost no cost.
So we could have a Vagrantfile similar to
1 2 3 4 5 6 7 8 9 10 11 12
This is a very basic version of the Vagrantfile. With this we only have to bring up the VM and tell
it must operate with this instance.
And to check that everything works properly, the following commands should not fail
1 2 3
With this you are ready to start working with Docker. You’re able at this point to start building images and creating isolated containers that could use the synced folders with an acceptable performance.
Development environments with Vagrant + Docker
To be fair and if you don’t like Vagrant, there are other alternatives: the fig project. Again if you’re under OS X, you will probably have to spawn a VM to save the performance issues with boot2docker, so in the end you’ll have to use an intermediate custom VM either with Vagrant or not.
So now, you’re truly ready to start building the development environment with Vagrant and Docker. First of all we should prepare all the Docker files in order to build the Docker images. For the example we will use
- A data-only volume container that will contain all the source code.
- A container that will run nginx.
- A container that will run php-fpm.
So a possible application layout could be one as follows
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Next we should restructure the Vagrantfile in order to make it more maintainable, and also share the application code as a NFS mount to achieve an acceptable performance
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Things to note here:
- With the instruction
default.vm.synced_folder ".", "/var/www", type: "nfs"we have mounted through NFS the application source code.
- Then we specified a
dockerprovisioner for the
defaultmachine that will build all the needed Docker images (This has been done that way, because I was not able to make Vagrant to build the images automatically) with the names:
To end with the
Vagrantfile, the only needed thing is to specify the Docker containers as if they were common
Vagrant VMs. First, in the same Vagrantfile where we defined the
default machine, let’s define the
Docker container, a data-only volume container that will host the application source code
1 2 3 4 5 6 7 8 9 10 11
Since the only purpose of this container is to host the application source code, we don’t need a
build an image for it. So we only need to configure the base image through
docker.image = "debian".
Next we give the name
app to the image, add a data volume with
docker.volumes = ["/var/www:/var/www"] and
tell Vagrant that this container won’t be permanently running with
docker.remains_running = false. Note that the
data-volumes, have a direct correspondence with the synced folders of the
Additionally, Vagrant needs to know which VM should spawn through
docker.vagrant_vagrantfile = __FILE__. With
this we are telling Vagrant that this container will be running on the
default machine in the current
Vagrantfile. If you’re in the case that you don’t need an intermediate VM to create Docker container you should add the
docker.force_host_vm = false. Now we only need to define the container configuration for the
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Note that with this two containers we are specifying the images
nginx, that correspond to the
image names we used in the
default machine into the
docker provisioner. This is the full Vagrantfile
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
With this we are ready to run the full development environment with Docker and through Vagrant. It’s important that the
containers should be created in the order specified: first the
app that will hold the application source code,
phpfpm container that will be used by the
nginx container to provide the application entry point
through the port 80. To tell Vagrant that it should create the containers in that order, we should use the
I think Docker is an awesome tool. It’s pretty easy to build lightweight development environments either with Vagrant or fig. In the worst case, we only need to spun up a tiny virtual machine before. Compared in the case where we have n virtual machines one for each service or one for each application that agglutinates all the services that the application needs. With Docker development environments will bootstrap a lot faster once they are created.
To conclude with the entry I have created a Github repo with an example application that runs its development environment with Vagrant and Docker. Check it out at https://github.com/theUniC/vagrant-docker-example
In fact. Annotations are evil. If you’re using it, you’re doing it wrong. But why? Why are they so evil? Next, I’m going to present a real-world use-case in which the use of annotations lead to a serious coupling situation.
Supose you have an application in which you want to integarate an ORM with annotated support to define your domain objects, in order to keep under control the growing complex business logic. You look at the resulting code, and … Wow! Have you ever seen such a clean code?
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Coupling, coupling and more coupling
This piece of code flaws of a serious coupling problem. YOU ARE ACTUALLY COUPLING YOUR DOMAIN OBJECT TO ITS OWN CONFIGURATION. A statement of domain-driven design that I really like is that “Separating the domain layer from the infrastructure and user interface layers allows much cleaner design of each layer. Isolated layers are much less expensive to mantain, because they tend to evolve at different rates and respond to different needs.”
But, how could it be? It’s only a plain PHP class with no defined behaviour and a few comments, it can be changed whenever it’s needed.
OK! Supose that there’s a need to reuse this entity in another persistence engine. How could it be done with annotations? It cannot be done, because I cannot define the same annotation twice without overriding the previous one!
The simplest example could be different table names between engines that share common entities. How could it be mapped without breaking each other? Or the use of sequences in Oracle/Postgres … So, indeed, this entity is coupled with the persistence engine.
The solution would be to move all the persistence mapping (aka entity metadata) to another format like YAML, XML or PHP (no annotated code). Doing it this way several mapping formats could be used and share common entities between each of them.
And with couping comes the rest …
In case we need to perform unit testing or to do some kind of runtime substitution, the entity is absolutely contextualized and the dependencies cannot be changed at runtime. It’s actually tied to an implementation. So there is no chance to substitute any of the dependencies.
At a first glance annotations could seem a big chance to keep your code simple and clean. But It actually leads to a coupling problem and serious design flaws. If you don’t mind about that coupling level, then probably annotations are for you.
Today I’m reopening my personal blog. And it comes with a big change. I’ve migrated from Wordpress to Octopress. Mainly for the content-orientation aspect of Octopress. I’m very tired to fight against the so cool web interface of Wordpress for managing posts and all the stuff. I think Wordpress it’s a great product besides the internal crappy code, but I like the simplicity of Octopress & Jekyll.
Mainly the reasons for the change have been
- Octopress is dead simple. You don’t need complicated and bloated web interfaces. Instead is a matter of write content. As simple as that.
- It’s fast than Wordpress. Remember Octopress is based on Jekyll, a static blog-aware web site generator, that dumps all your posts to plain HTML.
- Levarages git. Git is used to keep track of all the changes in your content :)
- Plays nice with Github. Jekyll is intended to be deployed to github pages, so to publish the blog had been matter of minutes.
Must admit. Octopress / Jekyll isn’t the silver bullet.
- It’s hard to get started. I’m a ruby newby! :(
- Don’t use Octopress if your main concern it’s not only to write content.