Teleport

Teleport is Lumie's operator access plane. The repo manages it as two ArgoCD applications in the same namespace: the teleport cluster itself and teleport-agent, which registers browser apps behind the Teleport proxy.

Responsibility

• Terminate operator access at lumie-infra.com and the wildcard subdomains backed by the same certificate.
• Store Teleport cluster state in PostgreSQL instead of local disk.
• Publish browser apps such as Coder, Headlamp, OpenClaw, Umami, Vault, and Grafana.
• Expose Kubernetes and app access roles through Teleport custom resources.

Source paths

Path	Role
lumie-infra/security/teleport/argocd.yaml	Two ArgoCD Application objects: teleport and teleport-agent
lumie-infra/security/teleport/helm-values.yaml	Teleport cluster settings, auth storage, TLS, and HA proxy settings
lumie-infra/security/teleport/kustomization.yaml	Helm chart entry plus NodePort patch
lumie-infra/security/teleport/agent/helm-values.yaml	App registrations and rewrite rules
lumie-infra/security/teleport/agent/kustomization.yaml	Init-container patch that substitutes secret-backed rewrite headers
lumie-infra/security/teleport/manifests/certificate.yaml	Wildcard certificate request for lumie-infra.com
lumie-infra/security/teleport/manifests/provision-token.yaml	Kubernetes join token for the app agent
lumie-infra/security/teleport/manifests/role-admin.yaml	Teleport role kube-admin with full Kubernetes and app labels
lumie-infra/security/teleport/manifests/vault-static-secret.yaml	Vault secret for Teleport's PostgreSQL credentials
lumie-infra/security/teleport/manifests/teleport-agent-credentials-vss.yaml	Vault secret for Zot basic auth rewrite

Ownership boundaries

• security/teleport/** owns the Teleport cluster and app agent only.
• Audit-event export is a separate application under applications/teleport-event-handler/**; it is not part of the Teleport cluster overlay even though it shares the teleport namespace.
• App workloads such as Coder and OpenClaw do not define their own public ingress; Teleport owns that access contract from the security slice.

Public surface and contracts

Surface	Contract
Cluster name	lumie-infra.com
Public addresses	lumie-infra.com:443 and teleport.teleport.svc.cluster.local:443
Service type	NodePort patched to 30443 on port 443
TLS	teleport-tls secret from cert-manager, covering apex and wildcard names
Storage backend	PostgreSQL database teleport on shared infra-db
Join method	Kubernetes in_cluster via provision token teleport-agent

The kube agent connects to teleport.teleport.svc.cluster.local:443 with insecureSkipProxyTLSVerify: true, because it joins through the internal service rather than the public certificate name.

Runtime flow

App access contract

The app agent registers internal services as Teleport apps, including:

• coder at coder.lumie-infra.com
• openclaw
• headlamp
• umami
• vault
• grafana
• argocd
• keycloak
• rabbitmq
• gitea
• tilt
• code

Some apps require request rewriting. Two important examples are checked in explicitly:

• coder rewrites the Host header to coder.lumie-infra.com so same-origin and WebSocket checks pass.
• openclaw injects the gateway bearer token and rewrites Origin.

Because Teleport does not expand environment variables in rewrite.headers, the agent overlay patches the StatefulSet with an init container that pre-processes the config before startup.

Failure behavior and operational risks

• Wildcard certificate failure breaks the whole access plane, not just a single app.
• A bad infra-db-credentials secret prevents auth-state reads and writes, which affects login and certificate-backed session flow.
• If the agent's init-container substitution fails, Zot and OpenClaw can register but still be unusable because the required auth headers were never rendered.
• The repo defines the kube-admin role, but role assignment to users is outside the checked-in contract here.
• The NodePort 30443 is part of the desired state; any load balancer or DNS layer in front of it must stay aligned with that patch.

Observability

• kubectl logs on the Teleport cluster and agent workloads are the primary operational signal.
• The separate teleport-event-handler application exports audit events; if audit-event forwarding is broken, debug it there rather than in this overlay.
• Service readiness, certificate status, and ArgoCD app health cover the standing control-plane contract.

Verification

kubectl get applications.argoproj.io -n argocd teleport teleport-agent
kubectl get deploy -n teleport -l app.kubernetes.io/instance=teleport
kubectl get statefulset teleport-agent -n teleport
kubectl get svc teleport -n teleport
kubectl get secret,certificate teleport-tls -n teleport
kubectl get teleportprovisiontokens.resources.teleport.dev teleport-agent -n teleport
kubectl get teleportrolev7.resources.teleport.dev kube-admin -n teleport
kubectl logs -n teleport -l app.kubernetes.io/instance=teleport --all-containers --tail=200
kubectl logs -n teleport statefulset/teleport-agent --all-containers --tail=200

Success signals:

• Both Argo CD applications, teleport and teleport-agent, are Healthy and Synced.
• The teleport service is NodePort with node port 30443, matching the Kustomize patch and Terraform NLB contract.
• Certificate teleport-tls is ready, and the backing secret exists.
• TeleportProvisionToken teleport-agent and TeleportRoleV7 kube-admin are present in namespace teleport.
• The Teleport cluster logs do not show PostgreSQL storage connection failures, and the teleport-agent StatefulSet logs do not show config-substitution or app-registration errors.

Related pages

• Security overview
• Keycloak
• Applications overview