| name | karmada-k8s-switch-skill |
| description | Use when the user wants to prepare a local Karmada lab, install and verify the Karmada control plane, deploy the tested member1/member2 stateless failover PoC in /root/karmada, switch traffic between Kubernetes clusters, validate cutover behavior, or tear the environment down. This skill favors the repo's proven scripts and manifests over ad hoc setup so the PoC is faster and more accurate. |
Karmada K8s Switch Skill
Overview
This skill drives the verified local Karmada failover PoC that was already implemented in /root/karmada. Default to the repo's bootstrap scripts, manifests, and demo helpers instead of inventing a new deployment flow.
Use This Skill When
- The user wants a working Karmada environment on a single host.
- The user wants a stateless Kubernetes cluster switch demo with
member1 and member2.
- The user wants to prove failover or failback through a stable host entrypoint.
- The user wants cleanup of the local Karmada and kind environment.
Default Workflow
- Read references/environment-prep.md before changing the host.
- Read references/karmada-bootstrap.md before bringing Karmada up or down.
- Read references/failover-demo-poc.md before deploying, switching, or validating the PoC.
- Reuse the existing assets in
/root/karmada/samples/failover-demo unless the user explicitly asks for a redesign.
Operating Rules
- Prefer
hack/local-up-karmada.sh for the first working environment. It is the fastest verified bootstrap in this repo.
- Accept that the standard bootstrap creates
member3. For this PoC, use member1 and member2 and ignore member3 unless the user explicitly asks for a two-member custom bootstrap.
- Prefer the host-side proxy in
/root/karmada/samples/failover-demo/scripts/start-proxy.sh so the browser entrypoint stays fixed during cutover.
- Treat the demo as stateless. Do not describe it as state migration. The proof comes from successful probes, heartbeat continuity, and backend switch events.
- Do not claim "no interruption" unless the probe timeline and endpoint checks stayed healthy during the switch.
- When cleanup is requested, prefer
hack/local-down-karmada.sh; if the environment is partially broken, fall back to explicit kind delete cluster commands.
Fast Path
Use this order unless the user asks for a custom flow:
- Prepare the host and verify Docker,
kind, kubectl, and Go.
- Run
hack/local-up-karmada.sh.
- Build and deploy
/root/karmada/samples/failover-demo.
- Start the host proxy and verify
http://127.0.0.1:8088/.
- Trigger failover to
member2 and validate the switch.
- Converge Karmada placement to
member2 if the user wants migration completion.
- Recover
member1 and fail back only if the user asks for that second step.
Validation Gates
- Karmada control plane is reachable via
/root/.kube/karmada.config.
kubectl --kubeconfig=/root/.kube/karmada.config --context=karmada-apiserver get clusters shows member1 and member2 as ready.- The demo app rolls out on both member clusters before traffic switching starts.
http://127.0.0.1:8088/status clearly reports cluster 1 or cluster 2.- Direct member endpoints and the proxied endpoint agree with the intended traffic target.
Deliverables
When reporting back to the user, include:
- what was installed or changed on the host
- which commands were run
- current cluster and proxy status
- whether traffic is on
cluster 1 or cluster 2
- any deviations from the verified PoC path
