[awalGarg]

Sorry I am pretty much a noob (specially at containerization) compared to most of the peeps here, but I can't resist asking - what is the need for this? What exactly does the author mean by "services (like databases) without needing to setup and maintain these environments manually"? Don't we just install node/mongo once ever followed by only updates?

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?

[ksafranski]

While purely local development works, the problem you often face in a fast-paced development environment is differences between your local system and production. That's not only dealing with things like file paths and service versions and configuration; it's also very stateful.

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.

[bobfromhuddle]

I don't see much use for this project over, say, Compose, but I use dockerised services for development in my current day-job.

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.

[Natanael_L]

As far as I can tell, the entire point is to ensure your code is platform independent from the start instead of making subtle assumptions derived from how your development boxes are configured. Making the test environment as similar to the deployment environment as you can.

[awalGarg]

Begging for an example ;)

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).

[drumdance]

Here's an example:

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.

[maccard]

We just have the particular versions of tools we rely on checked into perforce, and when there's an upgrade, you don't have to do anything different and everyone gets the upgrade.

[IanCal]

I have several projects that use different versions of elasticsearch and different versions of postgres.

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.