Show HN: DevLab – Docker Containerization for Local Development
github.com
github.com
How is DevLab different than Docker Compose?
Compose is actually part of the reason we wrote this tool. We tried so hard to make it make sense for a TDD workflow, but it was always cumbersome.
Compose wants you to build your application into a container, and build and run that container every time you have a task to run. This takes time and a fair amount of cleanup, especially when you want a clean environment to run your tests that doesn't persist to the next run. DevLab just wants you to specify the environment to plug your application into, and doesn't build a container at all.
The result is:
- No manual cleanup
- No pile of images or processes that stack up
- Your project doesn't have to be a Docker project. It doesn't need a Dockerfile. You can use this for something you plan to clone on an EC2 node and run from an upstart script.
- No Vagrant/Ansible/Chef/Puppet, server config, anything. You pick out an image that matches your environment's needs (node:4.2, wordpress, go, etc., hub.docker.com is a great place to start) and DevLab plugs your app into it.
- More development-oriented features outside of the pure Docker ecosystem that Compose stays confined to. For example, soon, Mac users running docker-machine will get to enjoy ports bound to localhost much like Linux docker users do [1].
To get this functionality outside of Compose, we had a monolithic makefile to maintain all of this for us. It wasn't DRY and it wasn't smooth. I hope you enjoy DevLab!
You don't have to rebuild each time, just use docker-compose rm -f and then run again. The old containers will be nuked, built docker images will be kept.
But where cleanup really comes into play is when you have multiple tasks. You can specify multiple command paths in Compose, but each one will be a newly built image. With DevLab, you could `lab install` a node project, followed by a `lab lint`, or a `lab test`. You could `lab build` to compile your app at any time. And when you do this, whether it's the first time or the twenty-first time, overhead versus what you'd experience locally is just a second -- and the only image you have is your environment container. No build process, ever.
'lab test -f node:4'
There's no additional overhead or cleanup, or modification of files.
DevLab allowed us to get everyone on the team using containerization and now (after a few weeks rolling on DevLab) dev's are feeling comfortable with the core concepts and know how to work with Docker when they need to get more "bare metal" with it.
"... Between Makefiles that spun up commands and Docker-Compose we were tired of our tooling getting in our way, or creating extra tasks for us to manage. The goal of DevLab is to have a tool with a small footprint, both in application and configuration requirements, which allows for local containerization using Docker. ..."
My question is not just for this particular thing, but all "lets dev in a container" thing like python's virtual env, npm's "local hard copy of all the modules" etc.? Does it really happen that your node binary or $language interpreter/compiler gets infected/changed somehow? Or one of your crucial modules gets borked? And if it does, shouldn't we be looking at investigating why such a thing really happened in the first place instead of duplicating everything?
As for maintenance, what kind of maintenance does node (or any $compiler or $interpreter) require? Don't they just sit waiting for invocation with some code to return the output/compiler binary? The only thing I have ever needed is stuff like nvm/rvm for using multiple versions of the same package/interpreter/compiler. Can anyone educate me?
With DevLab, you not only get an environment that very closely matches (or can be made to closely match) production, you also get a clean, blank-slate installation any time you run tests or any other task against your app. If you're developing locally, you're prone to database cruft, evolving configurations, pre-existing tempfiles, and other complications that can be easily avoided with the use of good development tooling.
Now extend that idea to multiple applications: chances are, your team isn't developing just one thing. You probably have multiple development concerns, and if you have an architecture like an SOA, you may have a multitude of microservices. When you update the language they're based on, be it node, go, python, or anything else, you need the ability to test and update your applications individually. You need your entire team to be using the same versions for the same applications. And more than anything, you need your potentially large team to get all these updates without collective hours of manual work to orchestrate those configurations on their own local machines.
My colleague addressed the question of DevLab vs. Compose elsewhere in this topic, but even if you don't use DevLab, I do heavily encourage virtualizing your environment to avoid these pitfalls in the way that makes the most sense for your team.
I'm an architect helping to move a a cluster of monoliths to a proper service-oriented setup; some of the teams work in PHP Symfony with MySQL, some work in Python 2.7 with Postgres, some in Python 3.4 with [insert hipster technology here]. We use Redis and Couchbase at various versions for different systems, Eventstore for messaging, and Elasticsearch.
That's a lot of moving parts to keep updated, but because we're deploying docker containers to production anyway, it's really simple to get up and running by just running `docker-compose up` from the root of a project.
tl;dr you're correct, unless you have to work in multiple systems that can version their infrastructure separately.
I do understand this. And I do agree. But the more I think about it, the more I find that those arguments don't matter in real life. Code written in high level languages/platforms like node/ruby/python/etc. is meant to be platform independent since the underlying VM abstracts the low-level details. Can you give me an example of the "configuration" differences that you (can?) encounter on different systems while developing on such platforms? (Except for the obvious and documented ones which containers can't solve either).
A few weeks ago we had an issue where a db migration would not work on some of our developer machines (it worked fine on mine and in staging and in production). After a lot of sleuthing, we found that it was because their version of the database was slightly different because they had installed it months after I did. Furthermore, it was a real pain to roll back to the appropriate version.
A container that locked down the db version would have avoided all that.
Elasicsearch has differences through versions which make some of the code incompatible (I recall changes with security around custom scoring functions). Some projects depended on an older version which allowed this code, other projects depended on newer features. Being able to have a different configuration per project can make things much simpler.
Even with the same versions, I'd like my ES process to be used by only one project, otherwise there's a risk of them stamping over each other causing issues which wouldn't be seen in production.