It started, as these things often do, with a single process running on a single machine in a server room nobody liked visiting. The system took in shipment requests, enriched them with a few heuristics, and spat out routing hints. Nothing fancy, just enough logic to save operations teams a few hours a day, and Marta built it on her own as a side project.
Marta wasn’t new to Java, she had spent a few months working in a microservices project in Spring and liked the serverless functional-style thinking: small transformations composed together. She had enough scar-tissue by now having experienced the inevitable issues with Spring and wanted to experiment with Quarkus anyway.
She picked TPF because, on top of featuring a canvas UI to design the pipeline, she got the complete app's scaffolding as a pure Quarkus project. She could keep developing on her IDE as usual, only with the addition of the Quarkus environment plugin. And, she got to use Dev Services which she also wanted to try.
Her choices for the script were deliberately boring:
- a JVM runtime (a single process) with
- no high availability
- a traditional direct method invocation between the orchestrator and the pipeline steps
- no DB, just basic HTML responses to human queries
Each step in the pipeline was just a function. Input came in as loosely structured key/value data. No rigid schema, no upfront modeling, no SQL persistence.
She didn’t yet have to think about
- execution tracking or audits
- fan-out/fan-in logic
- retry semantics
- error handling
- availability
- high throughput
- latency
- integrations with 3rd parties
- monitoring
- security
And because she had started using AI-assisted coding tools, even the tests came almost for free, as TPF’s step isolation made it trivial to generate meaningful unit tests for each transformation.
It was fast. It was understandable. It worked and, crucially, it already had shape—even if nobody called it that yet.
But during one summer's day, a sandstorm from the Sahara impacted Valencia and caused temperatures to rise (a phenomenon called kalima). The A/C unit in the server room failed and, by the next morning morning, alerts were firing, CPUs were throttling, and someone mentioned, only half jokingly, the possibility of actual fire.
Top comments (0)