The landscape of applications is an ever-expanding one from desktop to web to mobile. With increasing popularity, a torrent of new development and automation tools are popping in which are making the lives of developers blissful. It has also introduced a new perplexities of choice between a new (flashy) tool because one is excited and a battle-tested tool that you’ve got experience with (nil for a newbie). Among all the automations and increasing levels of abstraction, a lot of fundamentals are being obscured from the developer. What is left for a developer is just one click of a button and automagically, everything is setup.
This article is a collection of highly opinionated principles and philosophies around the lifecycle of a system and the thought process behind the engineering decisions needed to make in today’s world of highly complex systems. It is a refresher for the principles we’ve always known, mixed with new age requirements and philosophies. So stop being the lazy developer you’ve become and be the pragmatic engineer.
WAIT! If you are looking for a tutorial on how to orchestrate things in a real production environment, this is not the right place. Those will follow.
That’s right! Don’t try to achieve scale. While it may seem to be a contrarian thought among all the noise about cloud systems but this is one single fact that can help you achieve scale. To understand in terms of mathematical induction,
Build for the base case, iterate for scale, restart for greater scale.
More concretely, if you are building a CRUD system, build the best on a single machine and scale one machine at a time. When building a chat system, build the best for two users and scale each user at a time. When building a distributed database system, build the most efficient distributed database for two systems, and scale them into a ring.
Given the minimal set of resources and context, does your application survive the litmus test? That is what stateless is all about. An application should come with it’s own batteries and make the least number of assumptions.
Applications should be context free, configuration based.
This helps achieve two things — minimized side effects and maximized flexibility. The former helps maximize development and debugging (in case you were still lazy enough to leave that bug) velocity. The latter helps achieve a varied application behaviors with a minimal configuration set.
Considering today’s highly distributed systems, it is very easy to loose consistency across systems. Inconsistency could show itself in the form of inconsistent code design patterns, inconsistent multi-machine behavior due to state mismatch across machines or database inconsistencies due to network latencies. Such inconsistencies compound over time if not taken care of, from day one.
All systems must listen to one true master.
In more practical terms, a master machine (which itself could have redundant deployments) should be the orchestrator of your infrastructure. It should always persist the configuration for each moving part of your behemoth. On any changes to the master, each service must obediently conform to the new configuration (ideally in real time).
Predictability ensures replicability. I am an advocate of great code quality. If you don’t write code conforming to standards (industry or even in house!), you have failed as a developer.
Don’t take your system to the grave.
If you can’t write code that your fellow developers and future contributors cannot understand, you are the bug in the application’s lifetime which will never be fixed. Code obfuscation is the root of all bugs. Be as verbose and as simplistic as possible. Maintainability of the code is just as important as any other code characteristic because tomorrow when somebody builds on top of your shaky efforts, you don’t want everybody crashing down with you. High velocity teams cannot survive without high code quality.
Building predictable systems demands vigorous compliance to certain protocols and a protocol is only worth if everyone follows. Even though we always have a myriad number of choices to work with, figuring out the best set of rules and standards early on can help maximize productivity.
Channel in only the sufficient and necessary direction.
The pipeline starts from the smallest idea to the complete product being served to the end user. If the pipeline does not have a stringent set of guidelines, you will soon be shipping needless features and consequently products. For instance, consider the Medium editor — it is very strict in terms of its philosophy to allow minimal distractions during writing, no unnecessary features but just plain simple text and images. It is forcing the writer to focus on what was originally planned — WRITING, rather than tinkering around with options.
This statement is generally perceived as an overkill by early stage projects. But one very important thing we always forget is automation starts at the heart of the project not at Google-scale infrastructures.
Be the programmer, not the robot.
Strict workflow will bring in strict conformity requirements. Automation might seem a heavy investment during the early stages of a system, but you can reap tremendous benefits after the one time investment you made. Where you could be debugging where things went wrong during your server deploy, automating with a state management tool will bring predictability to the system.
If you now work backwards, automation is possible when we adhere to a strict-ruled pipeline, helping achieve predictable system lifecycle. Single source of truth separates context from systems, giving a constraint free environment to scale.
These philosophies help achieve system clarity and work at any level of operation, whether it be just local development or deploying mammoth clusters of heavy duty applications.
What’s next? If we are on the same page, then following in line with these principles, we’ll discover a more detailed and precise philosophies related to various moving parts of a distributed system. Here is a complete list of high level parts that we will explore in order of execution (ideally!).
If you see mistakes or want to suggest changes, please create issues or revisions against source.