Skip to content

Promote GALACTIC_CNI_EBPF_INTERFACES to a per-node CR field, not a hand-written env override #458

Description

@privateip

Problem

Multi-homed nodes where eBPF ingress-hook auto-detection is ambiguous need
GALACTIC_CNI_EBPF_INTERFACES set explicitly (#456, #457). The only way to
set it today is a raw env var on the galactic-cni DaemonSet.

That var is cluster-wide. A per-node override means a hand-written
Kustomize patch per affected cluster in datum-cloud/infra.

Proposal

Mirror how srv6Locator/nodeID already template into BGPRouter.spec.

Add an interfaces field to BGPRouter.spec. Have galactic-cni's
install-cni and credential-refresh read it from the node's own BGPRouter CR
by node name, instead of requiring the env var. Let infra template it from
a per-host value instead of a raw patch.

Why not now

No node in infra's hostvars model can express more than one real uplink
today. The override has no live use case after #456 and #457's auto-detect
and bond-slave-expansion fixes.

This is a should-have for the day a genuinely multi-homed node shows up,
not a should-have now.

Context

Follow-up from a stopgap in datum-cloud/infra, dropped once #456 and #457
landed. See those PRs for the original cross-region debugging writeup.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Fields

    Priority

    None yet

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions