Replies: 1 comment
|
💬 Your Product Feedback Has Been Submitted 🎉 Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users. Here's what you can expect moving forward ⏩
Where to look to see what's shipping 👀
What you can do in the meantime 💻
As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities. Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐ |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
🏷️ Discussion Type
Product Feedback
💬 Feature/Topic Area
ARC (Actions Runner Controller)
Discussion Details
Exploring Multi-Cluster Runner Scheduling with ARC and Karmada
The Problem
ARC's documented multi-cluster HA model uses independent ARC deployments. This provides control-plane isolation, but no unified, resource-aware decision about which cluster should execute a job:
maxRunnerslimits runner count, not workload feasibility.Result: A Runner Pod can remain pending while another cluster has compatible idle resources. A similar problem is described in ARC #4515.
We built an ARC + Karmada PoC to explore another approach. We'd welcome community feedback on whether external multi-cluster schedulers could become a useful integration direction for ARC.
Examples
Consider a CI platform with three Kubernetes clusters (illustrative values):
Each runner specification retains its
runs-onlabel; physical cluster selection remains transparent to users.Proposal
Separate runner lifecycle management from cross-cluster placement:
One logical Runner Scale Set uses the Karmada API to run Runner Pods across member clusters. Karmada offers
PropagationPolicy, Scheduler Estimator, and Scheduler Framework extension points for evolving placement policies independently of ARC.Pros
Cons
This complements ARC's existing HA model; it does not replace its control-plane isolation benefits.
PoC State
Tested with unmodified ARC 0.15.0, Karmada 1.19.0, and two member clusters.
Validated: One logical Runner Scale Set and
runs-onlabel; jobs executed in both clusters; Runner Pod and JIT Secret propagation; Pod status synchronization; cleanup in verified runs.Not validated: Resource-aware placement, guaranteed load balancing, global fairness, automatic failover, or production-grade HA.
Thanks to ARC's maintainers and contributors! We'd appreciate your thoughts on this integration model, especially the lifecycle, status, and resource-ownership contracts needed for external schedulers. We're happy to help with documentation, compatibility tests, or focused upstream improvements.
References:
All reactions