Skip to content

CFP: Rollout of cilium-agent with bgpControlPlane should wait for bgp established state #39468

@pasteley

Description

@pasteley

Cilium Feature Proposal

Is your proposed feature related to a problem?
When running cilium-agent with load balancer control plane inside a cluster (e.g., on 3 nodes), performing a rollout can lead to a temporary loss of all BGP announcements while the pods are being restarted sequentially.

Describe the feature you'd like
A readiness probe for the cilium-agent that checks the state of BGP session(s). This would prevent pod transitions during rollouts until BGP sessions are properly established, ensuring continuous service availability.

Describe your proposed solution
Implement BGP session state monitoring in the cilium-agent's readiness probe when there are matching ciliumbgppeeringpolicies. This should:

  1. Monitor the BGP control plane state
  2. Be configurable via Helm when bgpControlPlane is enabled
  3. Include configuration options to specify whether the probe should:
    • Wait for all configured BGP sessions to reach "Established" state
    • OR require at least one session to be "Established"

Metadata

Metadata

Assignees

Labels

area/bgpImpacts the Border Gateway Protocol feature.kind/cfpCilium Feature Proposalkind/featureThis introduces new functionality.

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions