In 2020, Foyer switched off its mainframe in production. Twenty months, 12,000+ COBOL/EGL programs and batch, moved from z/OS to Linux and Java, with a three-day go-live and a single month of code freeze. That was not an open-source project, but it is where open source became a real option.

This is a consumer's story, told by two of the people who lived it. Since 2017, we have replaced proprietary layers one at a time: z/OS to Linux, IBM WebSphere to Tomcat, Tanzu to vanilla Kubernetes. Our platforms now run on Kubernetes, offering services such as MongoDB, Kafka, Elasticsearch, MinIO,... and supporting technologies like Angular, Go, Scala/Play and Java.

Our most ambitious bet yet is a fully open-source Data Platform based on Apache Spark, Iceberg, Polaris, airflow, openmetadata, dbt, Trino and Superset.

We will walk through that timeline honestly, including what did not work: tools we adopted and dropped, and licence changes that forced us to decide all over again, sometimes to stay, sometimes to leave.

One outcome we did not plan for: open source changed who wanted to work for us. Several of our infrastructure and platform engineers joined Foyer because of this stack. Open source has raised our overall skill level, and it helps us attract talent.

The last part is about the debt: nine years of consuming open source and giving too little back. We will share what we are committing to now, contributing upstream, opening internal code, and building the internal organisation that makes it repeatable.