AWS
Most of the production workloads I've run, I've run here. Solutions Architect Professional. I care about account structure, networking and cost models that hold up once real traffic arrives.
I'm Andrea Sosso. I have spent my career on the receiving end of the pager: AWS, Kubernetes and OpenTofu, currently as Service Manager at Criticalcase, in Turin. I measure a system by how rarely it needs me — the best week is the one nobody had to think about.
Four things I keep coming back to
Most of the production workloads I've run, I've run here. Solutions Architect Professional. I care about account structure, networking and cost models that hold up once real traffic arrives.
Clusters as a means, not an end. The interesting parts are the boring ones: upgrade paths, resource limits that match reality, and GitOps so the state of a cluster is a file someone can review.
Infrastructure as code, reviewed like code. Modules with narrow interfaces, plans attached to pull requests, state nobody edits by hand.
The perimeter is gone; identity is what's left. Access in front of internal apps, Tunnel instead of open inbound ports. Nothing to babysit.
The short version
tennisteen.it — The most followed Italian tennis forum of those years, and the first server I ever had to keep online. Everything I know about capacity started as somebody else's traffic problem.
An art gallery — Invitations, posters, catalogues for the artists. The website and the newsletter too — Joomla and phpList.
Politecnico di Torino — Freelance: a Joomla platform for the universities collaborating on the Sofcom project.
Bitonda — Small-business servers consolidated onto virtual machines, high-traffic sites kept up on somebody else's cloud, and two REST APIs in Symfony: one extracted individual receipts from a cash register's electronic journal, one was single sign-on for an energy company's portals.
Ansel Technology — A Symfony2 client, front and back, over a WSDL API for telephone bookings.
COLT Engine — Software Architect, and the first job with a datacentre to look after: KVM in high availability, Foreman and Ansible to install bare metal without walking to the rack, Drone and Docker Swarm to build and ship what ran on top.
Criticalcase — Automate what was still being done by hand; git was the first tool introduced. VMware, then OpenStack, then Kubernetes — all on premise, where a provider would have handed us these services ready to use and here you build them and keep them running yourself: Ansible, Packer, Terraform, HAProxy.
Criticalcase Digital Solutions — The cloud side of the same company. Service Manager; AWS Solutions Architect, Associate and then Professional. Terraform stops being a tool I run and becomes the interface everyone changes things through. Terraform on GitHub Actions, through Digger — now OpenTofu, through OpenTACO. Plans attached to pull requests; nobody applies from a laptop.
RKE2 on premise, where the nodes are the ones we bought; EKS and AKS in the cloud, where Karpenter decides how many we actually need. In 2005 keeping a server up meant being awake when it went down. Now a node dies and the cluster replaces it on its own: the job has moved from fixing what broke to building the thing that fixes itself.
I'm building something new. Ask me again when it stays up on its own.
Verifiable on Credly
The full list
Nothing runs on deploy, no cookies
There is a build, it just does not run in the cloud: the scripts that generate the design tokens, the favicons and the fonts run on my machine, and their output is committed. So nothing runs when this page is deployed, and what is in the repository is what the browser receives — three HTML files, two stylesheets, three small modules. One request goes to a domain that is not this one: the analytics script, in the table below.