From Concept to Implementation: Building Fast Stack

Just six months ago, a vision was formed after years of frustration with the current solutions available on the market. A vision was born, a blueprint defined, and work began on transferring that vision into a reality. Endless nights, weekends, and just about any time in between resulted in what might be called an what might be considered an 24 hours a day effort, with a few small brakes between development cycles.

It turns out, digiesting metrics is relatively easy, scaling that effort is hard, and providing effective visualizations into those metrics is even harder. An initial server side architecture was developed in under a week. Rough, functional, an effective draft. This provided a great foundation for establishing API contracts. It was well understood that the effort would be an 'disposable' one. What was not expected was that two additional separate efforts were made, battle tested, and untilately dismanteled. Load testing quickly exposed design flaws. This has been an intersting excersize in that we are not ingesting metrics for a single application, however creating a solution for ingesting metrics for hundreds if not thousands of applications.

Beyond scalability concearns, uptime is another important topic. Configuration changes, deployments, etc need to be performed in realtime. Interrupting service to customers for these types of changes is not acceptable. I becomes neccesary to establish a solution where changes may be introduced in an asynchronous (hot reload) manner, where inbound requests support backpressure. Essntially we need a system that scales infinately, and aims to provide a no downtime experience. It's hard than it might initially appear.