Skupper v2
16 October 2024
Skupper v2 is a major change and a major improvement over v1. Here's why we're doing it and what it means for our users and contributors.
Why a new major version?
Skupper first became available four years ago. Since then, we've learned a lot about what users need from Skupper. We've also learned about the pain points for both users and Skupper developers in our existing design and implementation. The changes we are making for v2 will result in a version of Skupper that is easier to operate, easier to extend, and easier to maintain.
The move to custom resources
Skupper v1 uses a combination of ConfigMaps and resource annotations as its declarative interface. Skupper v2 instead uses Kubernetes custom resources.
Custom resources have two key advantages. First, they are subject to standard Kubernetes role-based access control, so cluster admins can use standard tooling to control use of Skupper if they choose. Second, they provide a standard mechanism for reporting resource status.
Choosing custom resources comes with a trade off: installing custom resource definitions (CRDs) requires cluster admin privileges, something v1 did not require. This is an advantage for some of our users, but a disadvantage for others. We believe that custom resources are, on balance, the right choice.
A uniform declarative API
V2 has a new, uniform API for site configuration, site linking, and service exposure. In v2, all of Skupper's interfaces and platforms use this common API.
The following are the core API resources in v2, organized by function:
| Site configuration | Site |
| Site linking | Link, AccessGrant, AccessToken |
| Service exposure | Connector, Listener |
The new API is designed to enable automation with GitOps and other tools and to provide a foundation for third-party integrations.
Service exposure in particular sees a change in v2. In v1, service exposure is implicit: exposing a service in one site by default exposed it in all the linked sites. In v2, service exposure is instead explicit. A connector binds a local workload to a routing key. In another site, a listener with a matching routing key makes it available for application connections. Only those sites with a matching listener can connect to the service.
A new controller and CLI
The new controller combines the previous service and site controllers into one that can be deployed at cluster or namespace scope. The improved implementation is easier to maintain and understand.
The new controller also addresses a v1 pain point: it allows site configuration changes without requiring re-creation of the site. Notably, you can reconfigure your site without losing existing site-to-site links.
The new CLI closely follows the API. Indeed, in v2 the CLI is really just a thin layer on top of the API. To simplify its use, the CLI blocks until operations are complete.
Router improvements
The router in v2 has a new, faster TCP adaptor with improved buffer handling and reduced threading overhead. The new TCP adaptor incorporates lightweight protocol observers for capturing HTTP traffic metrics. Together these reduce application latency and router CPU utilization.
Non-Kube sites
Skupper is not just for Kubernetes. Skupper sites can run on Docker, Podman, VMs, or bare metal. In v2, we've made the support for non-Kube sites simpler and more uniform. They use the same YAML resources as Kube sites. One codebase implements support for all of the non-Kube platforms.
Upgrades
V2 is not backward compatible with v1. We are developing tooling to convert v1 configuration to v2 custom resources. Those upgrading from v1 to v2 will have a small period of downtime while the v1 components shut down and the v2 components start up.
When will v2 be available?
The preview 1 release is available now. Preview 2 is set for the end of October, and we are aiming to release 2.0 at the end of November.
Note that things are still changing in the preview releases as we review interfaces and make improvements.
Try it today
The best way to start exploring v2 is with our examples. With preview 1, we have converted our CLI and YAML Hello World examples:
We would love to get your feedback!