Enrico La Cava

A CI/CD pipeline built with CircleCI and Spinnaker to deploy ephemeral development, staging, and production environments.

I used AWS Lambda functions and an API gateway to interface Spinnaker with the outside world (CircleCI).

The development environment data was filled by a batch procedure that anonymizes and extracts a subset of data from the production database.

That anonymized subset was loaded in a temporary RDS database and managed with a tiny in-house CRD. This way, when a developer opened a pull request (PR) and triggered such an environment, we could deploy every resource he needed in Kubernetes and AWS and then clean it up when the PR was closed.

Right after the first strike of the pandemic, I was tasked to save as much as possible and deliver a plan to optimize the existing AWS infrastructure and third-party tools.

My work started with a heavy refactoring of the existing Terraform code to add meaningful and standardized tags to each piece of the infrastructure.
Then I cleaned all the unused and abandoned resources and added the newly discovered ones to terraform.

With tags in place, I assessed the cost of each piece of infrastructure of each product and made informed proposals on how to optimize the cost.

Finally, with some CI and dashboards, I provided transparency on costs to all the concerned stakeholders. I build the CI, creating excel reports for upper management based on AWS cost data. Developers were informed via dashboards of the cost estimates on the Terraform PR.

The team struggled to work and deploy with a compiled multi-language (C++/Python/Bash) 3D-asset conversion pipeline due to bad containerization and lack of tools.

I had to re-architect the building stack from scratch, leveraging a lot on the Docker multistage and rebuilding both CI/CD and local development tools to release them from the pain of compiling their application.

To ease some tricky compiler dependencies, I built helper tools leveraging docker container layers and volumes.

Before my work, it took up to three hours to test most of the code changes on that stack, and after it became a matter of minutes, if not seconds.