About Tracker, and being perfect

How hard can it be to just make a simple service to track a phone? Not that hard. Unless you want to be a perfectionist and overengineer the whole thing, several times over. And that is why I never seem to get anywhere close to finishing Tracker.

In the last five or so years I've learned a lot about web services and would of course like to apply ALL that knowledge somewhere. A hobby project seems like the perfect place!

  • Code generation and custom tooling: avoiding repeative tasks by writing ten times as much code to maintain. And then there is this one thing that was not in the original specification and you'd basicly have to rewrite the whole thing.
  • High availability and fault tolerance: considerable additional design and implementation complexity. Good luck getting anything ready when working on the project just some occasional weekends.
  • Metrics: a whole universe in itself to set up, integrate and maintain.
  • Devops, service orchestration and distributed configuration management: yet another whole universe, with yet more design and implementation complexity. As if you even get to this point. Sure you can devops your stub services, but that if anything is wasted time.
  • Distributed tracing: better reinvent the wheel again! That way it works just the way I want.
  • Extremely high performance: each small part is tuned for the best achiveable performance before integrating it to the larger system, and never finishing anything.
And of course everything needs to be P E R F E C T!

What good is a if it doesn't function as well as theoretically possible? This intra-process message queue can only pass 300 000 messages per second?? It should be able to do at least 500 000!! Better spend a week trying to optimize it before moving on to the next thing! A decryption routine that can only decrypt 20 000 packets per second? Needs to be able to do at least twenty times as much! Better to just give up or spend ages looking for a faster implementation.

How about this config file? Why it is on the file system? How to make it seamlessly work with multiple environments with different configs. What if those configs change dynamically? Better get to work already, there is much to do!

Hey, see that single point of failure? You'd better engineer it away! What failure point? I need to first map the whole system and get realtime metrics from every single thing it does, or doesn't.

And hey, see that middleware solution for solving that one problem you have? Better not use it and write my own. That way I learn so much more about the problem domain and can leave out unnecessary parts for blazing fast performance.

I'd like to have public test server for this? I wonder how I can set that up automatically. And better make sure that it does database migrations and all that automatically.

And did you know that you could run your code inside a custom operating system to save on context switching time with networking?

Oh, and there is this web application, too. What if I used modern JavaScript this time? And then the mobile app needs to be rewritten, too.

And what about...!


No comments: