As a small engineering team of 40 at Clever, we aim to focus all of our efforts on building feature depth and improve resiliency. As a company focussed on K-12 education, we want to maximize time working with our customers and not on building orchestration infrastructure. However, we also know that well designed infrastructure and developer tooling allows us to move faster safely.

Our infrastructure team mirrors our product teams’ extreme extreme focus on user experience, and we constantly evaluate our options. Over time we have moved our container orchestration system from a internally built prototype in 2014 to Mesos Marathon and finally Amazon Elastic Container Service. We build infrastructure when required, but move to an off-the-shelf solution when it satisfies our requirements to minimize ongoing maintenance. This has allowed our small team to build reliable products that support education in over 60% of K-12 schools in the US.

In this session I want to share our learnings on how to build developer control planes to allow your infrastructure team to make changes without disrupting engineers. Specifically I will talk about

• Lessons learnt about building control planes using snapshots of our own service deployment orchestration tooling over the last four years. A lot of our building blocks are available as public repositories on Github
• Designing infrastructure tooling for rapid evolution and change using examples from the rollout of our batch processing system over the last year.
• Evaluation and decision making frameworks for choosing between using cloud-managed, open source and build-your-own options through our own move from self-hosting containers to using a containers-as-a-service platform.