What happened?

The clientConnection configuration (QPS and burst) defined in the Configuration API
(configuration_types.go#L189-L202)
is validated and defaulted but never applied to the main controller manager's Kubernetes client.

Additionally:

• The Helm chart doesn't emit clientConnection in the ConfigMap template at all, while
  kustomize does — creating a parity gap
• The status server correctly wires its own QPS/burst in pkg/statusserver/setup.go,
  making the main manager inconsistency more visible

What did you expect to happen?

The clientConnection values should be applied to the manager's rest.Config so operators can
tune API server request rates.

Impact

• Documented defaults (QPS=50, burst=100) are silently ignored
• Manager runs with client-go defaults (QPS=5, burst=10), which can throttle under moderate load
• Operators configuring clientConnection believe they're tuning performance, but values have
  no effect

Environment

Kubernetes version:
N/A — this is a code-level configuration bug, not runtime-specific.

Kubeflow Trainer version:
Affects all Trainer v2 versions (v2.0 through current master).
The clientConnection config in configuration_types.go is validated and defaulted but never wired in main.go or config.go.

Kubeflow Python SDK version:
N/A

Impacted by this bug?

Give it a 👍 We prioritize the issues with most 👍