Problem

In dual-stack or IPv6-enabled clusters, the agent can crash when it attempts to announce or withdraw a BGP path for an IPv6 address, but the node does not have a corresponding IPv6 address configured in HostMetadata.

level=error msg="error making path to announce: no ip6 address for node" component=routing
level=warning msg="Tomb function errored with error making path to announce: no ip6 address for node"
level=error msg="tomb Dying error making path to announce: no ip6 address for node"
level=fatal msg="GrpcServer Server returned grpc: the server has been stopped" component=cni

RCA

• common.MakePath() is called to construct a BGP path for announcement/withdrawal.
• When the required node IP is missing (nodeIPv6 == nil), MakePath() returns an error.
• The error propagates through routing_server.announceLocalAddress() or prefix_watcher.WatchPrefix().
• The Go() wrapper passes the error to tomb, which enters the Dying state, signaling all goroutines to stop.
• The agent crashes, disrupting all network connectivity for pods on the node.

Impact

• agent crash - all networking functionality is lost.
• pod connectivity disruption - other pods on the node lose network access.
• cluster instability - affected node becomes unavailable for other workloads.

Solution

Implement graceful degradation by treating missing node IP as a non-fatal, recoverable condition.
The missing node IP is a transient or configuration condition, it does not mean the routing component is broken. Returning an error would stop the routing/prefix watchers and trigger the tomb to kill the whole agent.
By returning early, the routing/prefix watchers continue, and missing paths can be restored later. The logic keeps state in memory (localAddressMap), so once the node IP becomes available, RestoreLocalAddresses() will re-announce.