01 / 06
QuantaRun
- Type
- Distributed systems
- Year
- 2026
- Disciplines
- Backend, architecture and testing
- Stack
- Java 25
- Spring Boot 4
- PostgreSQL
- React
- TypeScript
- OpenTelemetry
- Docker
- Links
A control plane for scheduling, executing, recovering and replaying distributed AI workloads. It is built around the cases where such systems fail: a worker that dies mid-task, a result that arrives late, two schedulers racing for the last free slot, or one tenant flooding the queue.
PostgreSQL is the only coordinator (SKIP LOCKED, row locks and constraints), with no message broker. Workers hold their jobs under leases: when one dies, its job moves to another worker and resumes from the last checkpoint. It has six scheduling policies that explain every decision, a deterministic simulator that runs the scheduler's own code, fault injection and OpenTelemetry tracing.
It has 417 tests against real PostgreSQL and Playwright end-to-end tests. Performance is measured: median time to start a job fell from 526 ms to 11 ms.




