Skip to content

Let the apiserver reach argocd and longhorn-replica-affinity - #80

Merged
yama6a merged 1 commit into
mainfrom
cilium-apiserver-ingress
Sep 11, 2026
Merged

Let the apiserver reach argocd and longhorn-replica-affinity#80
yama6a merged 1 commit into
mainfrom
cilium-apiserver-ingress

Conversation

@yama6a

@yama6a yama6a commented Sep 11, 2026

Copy link
Copy Markdown
Owner

Both gaps showed up as AUDIT verdicts in Hubble while policyAuditMode is on, so this is traffic that flows today and would stop the moment enforcement goes on.

The real bug is the longhorn webhook's 8443 rule: it never matched its own traffic. The admission call arrives SNATed to the node hosting the apiserver, so the identity is remote-node, not kube-apiserver, and the SYN was auditing. With failurePolicy: Ignore that fails silently, which is exactly what the existing comment warns about.

The rest is ingress from kube-apiserver on argocd and both longhorn-replica-affinity pods, for watch replies that arrive after the conntrack entry ages out mid-stream. Egress to the apiserver was already allowed; only the late retransmit lacks a rule.

@yama6a
yama6a merged commit 29ebdab into main Sep 11, 2026
4 checks passed
@yama6a
yama6a deleted the cilium-apiserver-ingress branch September 11, 2026 20:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant