Infrastructure and reliability

At Mobilauto I set up and kept the Kubernetes cluster the products ran on, and the storage under it. That's also where my habit around dependencies comes from. Of all the industries I've worked in, taxi forgives the least. Outages happen to everybody, of course. The problem in taxi is who gets to the failed machine, because when the service goes down together with the infrastructure under it, the engineer on call can't get a taxi to reach it. So I keep outside dependencies as few as possible and keep my own "hosting access and execution control" over everything important. The cost is mine to carry: every vulnerability, patch, upgrade and reliability problem, often at 3 a.m. with nobody else to call.

The rider gateway there has one guarded call, and it's sign-up, because sign-up sends a one-time code by text message to a phone. The gateway watches sign-ups by IP address and with hCaptcha, and each taxi firm picks one of 3 modes: off, a captcha on every sign-up, or a middle mode. In the middle mode a rider inside the firm's own country passes with no check, a rider from anywhere else gets a captcha, and any address with more than 10 sign-up requests in a day gets a captcha wherever it is. Each firm's mode lives in the operator's database, and the gateway reloads it on a short cycle. So a firm that changes its mode in the middle of an attack is covered within minutes.

I joined The Parker Avery Group in 2022, and what I learned there is discipline around a change: a proper test suite, a pipeline that builds and deploys every change the same way each time, and tracing on the AI, so the team can look back at exactly what it did with a given question. Still, a passing change isn't automatically a safe one. I assume the one failure the tests missed is there and keep a plan B behind plan A, and I don't treat plan A as fully certain either.

My first real company job was at Qberry, an outsourcing and outstaffing company, in 2019-2020. I was a full stack developer who also carried the DevOps and SRE side. The server work ran through every client project next to the features, mostly on machines that belonged to the clients: Apache, and the tooling to migrate sites.