Building TCloud:
self-service container platform at KPN retail

During a period of three years we led the team responsible for building the TCloud container platform, which is used by over 750 engineers within the business-to-consumer (B2C) department. TCloud is a first choice for applications within B2C. It is a true internal development platform that offers an on-demand, multi-tenant, self-service runtime to teams, allowing them to bring their ideas to market in record time. It cuts down teams’ cognitive load by a significant margin; teams can focus on development of new features without worrying about infrastructure. The platform exposes well-known APIs (Kubernetes, etc) and does not lock teams into using limited abstractions. The platform team ensures that the platform is secure, cost-effective, transparant, highly-available, scalable, maintainable and self-service. The platform team tries to find the right balance between meeting all these requirements.

The platform hosts hundreds of applications, including critical workloads like:

  • Order/shop APIs. KPN was the only ISP whose shop was available during the iPhone-launch in 2021 and 2022;
  • TV-backend applications. Migration of the TV applications to cloud was possible with minimal changes to the platform; allowing teams to migrate their applications without help from the platform team;
  • Training of GPU-enabled machine-learning models;
  • Applications for sales, finance, etc;
  • Applications of which the platform team is not aware, because they did not have to be involved in their deployment

The platform team practices continuous deployment of the platform itself; the team can deploy up to 20 changes per week to live clusters. This is made possible by a testing pipeline that ensures that changes are made without impact. Within 30 minutes a platform-engineer can spin up his own cluster with all features enabled, allowing him or her to test his changes before touching a live cluster. This process allows for rapid development, reducing the time-to-market for the platform itself.