Overview / AutoTrader
Caching as a stability problem, not a speed problem
The platform
Five years of this was AutoSync — the platform thousands of Canadian dealerships run their inventory on. A large codebase I did not design, shared with other engineers, with Vue on the front end and Node.js, PHP, MySQL and REST APIs behind it. The PHP layer was a custom object-oriented architecture built in-house rather than on Laravel or Symfony, which meant the load-bearing questions were about our own architecture rather than about a framework’s conventions.
The caching work
The work I would point at first is the caching. I implemented and tuned Redis Object Cache Pro against AWS-hosted Redis. Caching is usually sold as a speed improvement, and it was one — but the result that actually mattered was stability. After rollout, network-level downtime dropped to near zero. The platform stopped falling over, which is a different and better outcome than pages loading faster.
What I cannot tell you
I want to be precise about what I can and cannot tell you there. I can count what I wrote, because the repository counts it for me. I cannot tell you what it did to page speed or to conversion: I did not own those dashboards and I did not take the figures with me. So there is no percentage here and I am not going to invent one. What I can describe exactly is the failure mode before and its absence afterwards. On a page that ends with a table of my own unfixed gaps, a number I cannot source is worth less than a mechanism I can explain.
The rest of the five years
The rest of the five years: technical SEO on a marketplace where dealer inventory discoverability is the product rather than a marketing concern, and mentoring engineers through code review and pair programming. That last part is worth naming given the year I have spent as the only engineer on my own projects — the review habits on this page came from somewhere.
How the work was run
Five years of this was team work, not solo work. Agile delivery, code review on every change, and pair programming when something was load-bearing enough to be worth two people’s time. I reviewed other engineers’ work and had mine reviewed, on a codebase I did not design and could not unilaterally change. That is a different discipline from the studio year, and the more common one.
What I would do differently
I would have kept my own record. I can describe the caching work precisely and I cannot quantify it, because the dashboards belonged to the platform team and I took no figures with me. The mechanism survives; the evidence did not. Every project on this site since has a measurement written into it before the feature ships, and that habit started with not being able to answer this question.