// the find
apache/streampark
The Open Control Plane for Real-Time Data
StreamPark is an Apache top-level project that wraps Flink and Spark with a development framework and a web control plane for deploying, versioning, and monitoring streaming jobs. It is aimed at platform and data engineering teams that run more than one Flink or Spark version across Standalone, YARN, or Kubernetes and want one place to manage them.
- It runs multiple Flink and Spark versions side by side from one control plane. Teams usually end up building this themselves once they have two cluster generations in production.
- It supports Standalone, YARN (Hadoop 2.x and 3.x), and Kubernetes, which covers most of the places Flink jobs actually end up running.
- The project has Apache TLP governance and an Apache 2.0 license. The CI surface is visible in .github/workflows, with separate backend, frontend, unit-test, and e2e workflows plus CodeQL.
- Shared code is split into modules such as streampark-common rather than kept in one monolith, which makes the core configuration and version handling easier to find and reuse.
- The README is mostly a feature list. It has no architecture section, does not say how job state is stored or recovered when the StreamPark server restarts, and gives no scale limits. You learn the failure model by reading code or running it.
- The 'zero to running jobs in minutes' claim describes a single-service demo. The README does not show what a YARN or Kubernetes setup involves, and that is where most of the operational work sits.
- The quickstart pipes a remote shell script from the project website into sh, and the Docker example pins nothing and uses the mutable latest tag. That is fine for a trial, but a shared or production setup should download the script, read it, and pin the image.
- Support runs through the dev mailing list and GitHub issues, and the README sets no response expectations. A team adopting this for production should plan to debug problems itself.