Passenger information at a large transport operator
Release testing a data platform: from GitLab CI jobs to automated end-to-end tests
- Industry
- Transport, passenger information
- Frybad's role
- Solution architecture of the test orchestration, development of test tooling, removal of legacy; part of an interdisciplinary test team
- Stack
- Kafka, Avro, Spring Boot, Kotlin, Node.js, Argo Workflows, Crossplane, Kubernetes, GitLab CI
Starting point
The platform receives timetable and real-time data from many providers, matches it and consolidates it into passenger information that reaches passengers across several channels. The platform is event-driven, with Kafka and Avro as its backbone; most teams build with Spring Boot.
An interdisciplinary team tests the platform end to end: data in, information out, across all teams. Its tooling had grown over time: the Node.js test runner had been written by a tester and lacked a sustainable architecture; release tests ran as a chain of plain GitLab CI jobs; and the company had DevSecOps requirements the team was behind on.
What we did
- Test runner refactored. The Node.js test runner was brought onto a sustainable structure step by step, without interrupting the running tests.
- Test tooling as services. REST APIs in Spring Boot and Kotlin were built for test data, checks and evaluation, so tests call them like any other service of the platform.
- Orchestration with Argo Workflows. Release tests were rebuilt from a chain of plain CI jobs into workflows: steps with dependencies, parallel where possible, repeatable, with artefacts and evaluation per run.
- Release environments on demand. With Crossplane CRDs on Kubernetes, release environments are requested declaratively and torn down again. A release gets its own environment instead of waiting for a shared staging one.
- DevSecOps requirements. The legacy that stood in the way of the company’s requirements was removed and the tooling brought up towards them.
What Crossplane does here: a release environment is described as a Kubernetes object; Crossplane creates the resources it needs and removes them when the object is deleted. The environment becomes part of the test run, not a ticket to another team.
What this means for your project
If your release tests depend on a shared environment or hand-maintained CI jobs, start with Cloud-native DevOps & SRE. If it is about the test tooling itself, APIs, runners and architecture, it is Software development & architecture.
Result
Example from an ongoing client project.
Release tests run as automated end-to-end workflows instead of a chain of plain CI jobs.
Release environments are created on demand via Crossplane and disappear after the run.
The test runner has a sustainable architecture; test tooling consists of dedicated services.
Not included: numbers.
What we would do differently today
Refactor the test runner earlier. A tool everyone uses daily gets more expensive to rebuild with every month. The rebuild should have come before the orchestration work, not alongside it.
Environments as code from the start. Shared staging environments are the most common reason releases wait. Crossplane environments per release solve that; introduced earlier, they would have saved waiting earlier.
DevSecOps requirements into the pipeline, not onto a list. What a pipeline checks stays met. What sits on a list goes stale.
Similar situation?
Tell us how a release is tested today and where it waits. We will say what can be automated first.