[jlamillan]

Summary

Add an optional --max-node-backends-per-backend-set flag for Flannel node-port backend sets. The default of 0 keeps the current all-node behavior.

Why

On large clusters that use NodePort service based backend set can grow an OCI load balancer's total backend count enough to block updates once it hits backends-per-lb-limit. A per-set cap provides a controlled way to cap that count while retaining NodePort routing through the selected nodes.

What changed

• Select a stable, backend-set-specific subset of nodes when a positive cap is configured.
• Apply the same selection during initial backend-set creation and later reconciliation.
• Expose the flag through Helm and the deployment manifest, with 0 as the default.
• Add focused tests for the cap, stable selection, and initial backends.

This PR does not include the backend repair modes. The cap applies only to Cluster traffic policy node backends; it does not cap the total across an LB or change Local traffic policy and default_ingress behavior.

Special notes for your reviewer:

• These proposed changes have been running in our production clusters for a number of weeks.

See also

#163 which prevents a deadlock keeping old deleted node backends once backends-per-lb-limit is hit.