Litterboxing

Sandboxing is a great idea. Isolate your processes. Isolate your data. Chinese Walls everywhere to ensure everything is kept separate, independent and secure. Whether that’s containers or SELinux

Litterbox – it’s like a sandbox but when you look closer it’s full of shit

“There’s the inkling of a good idea in there. He’s almost describing a sandbox (without, of course, any of the isolation required) and front-end microservices (without a pipeline, or components). Can we call this pattern litterboxing? It’s a sandbox until you look closer and see it’s full of catnip.”

https://thedailywtf.com/articles/comments/classic-wtf-i-am-right-and-the-entire-industry-is-wrong/1#comment-506416

Cloud thinking : Executable documentation.

Documentation is just comments in a separate file. At least developers can see the comments when they change code. Tests are better comments. Tests know when they don’t match the code.

Infrastructure is the same. I can write a checklist to set up an environment, and write an architecture diagram to define it, but as soon as I change something in production, it’s out of date, unless it’s very high level, and therefore only useful to provide an outline, not detail.

Unit tests document code, acceptance tests document requirements, code analysis documents style and readability, and desired-state-configuration documents infrastructure. All of them can be checked automatically every time you commit.

Documentation as code means the documentation is executable. It doesn’t always mean it’s human readable; ARM templates in particular can be impenetrable at times. If machines understand it, the documentation can be tested continuously, repeated endlessly across multiple environments, reconfigured and redeployed at the stroke of a keyboard.

The more human you have in a process, the more opportunities for human error. It doesn’t remove mistakes, but it’s much easier to stop the same mistake happening twice.

Modular doesn’t mean interchangeable

Microservices are great. Containers are great. Packages are great. Objects are great. Functions are great. At each level they encapsulate functionality into re-usable modules. One place to make changes. One place to optimise. One abstraction to help you move to the best layer.

For the sake of argument though, I’m going to focus on microservices, specifically those accessed via an API, as that’s what I know best.

Interfaces and APIs are good for test harnesses, but isolating systems for your own benefit (to make modules open for extension but closed for change) bears no relation to making that module useful to other systems.

Sure, I could rip out the modules that contain business logic, and package it as a framework, but that doesn’t mean it’s useful for anyone else. It could well be an inner system . It may still be coupled to domain/architectural knowledge that make it too valuable to swap out. There could be an implicit dependency for just that one version of that one tool.

It is of course possible to make modules re-usable, but they have to be designed that way, whether up front, or evolved. And anything interchangeable has to be tested, in at least 2 conditions, to make sure it’s suitably robust.

Design an interface. Evolve the design. But don’t mistake solving one problem in one place for a general solution. And don’t think that adding an interface is all it takes to turn a local SQLite data store into a cloud file system repository.