// the find
apache/dubbo-admin
The ops and reference implementation for Apache Dubbo.
Dubbo Admin is the management console and control plane for Apache Dubbo, the Java/Go RPC framework widely used in Chinese enterprise stacks. It handles service discovery visualization, traffic rule management (condition routes, tag routes, dynamic config), and acts as a bridge between Dubbo services and Kubernetes or traditional registries like Zookeeper and Nacos. Aimed squarely at teams already running Dubbo in production — there is no reason to look at this otherwise.
The architecture is cleanly separated: discovery (Zookeeper, Nacos), engine (Kubernetes), governor, and store are all pluggable via factory interfaces, so swapping a registry backend doesn't ripple through the whole codebase. The resource model mirrors Kubernetes CRD conventions with typed helpers per resource kind, which makes the code navigable if you know how to read k8s controller code. Traffic rules (condition routes, tag routes, affinity routes) are first-class typed objects with proto definitions and helper methods — not stringly-typed config blobs. The presence of `.agents/skills/` with structured eval files suggests someone is actively thinking about AI-assisted development workflows in a non-trivial way.
The README is nearly empty — 'refer to official website' and 'refer to roadmap' does not help anyone get started locally, and the official docs site is in Chinese first. The Go codebase uses GORM for distributed locking (`pkg/lock/gorm`), which is a fragile choice — database-backed locks under network partition or high contention will cause cascading failures in the control plane. Test coverage appears sparse outside of a handful of unit tests on proto utilities and the dependency graph; the lock implementation has a test but the core discovery and routing logic does not seem to. If you're not already in the Dubbo ecosystem, the learning curve is steep and the payoff is zero.