Skip to content

cert-manager

Issues and renews TLS certificates for every public hostname in the cluster. Every TLS Secret (argocd-server-tls, grafana-tls, auth-tls, the app's tomoda-app-tls) is generated by cert-manager from Let's Encrypt.

Installed by k8s/envs/platform/cert-manager/application.yaml. The ClusterIssuer is defined in k8s/envs/platform/cert-manager/cluster-issuer.yaml.

Chart and source

Field Value
Helm chart cert-manager
Repository https://charts.jetstack.io
Version v1.21.2
crds.enabled true (passed as a Helm parameter)
Destination namespace cert-manager (created by Argo CD)
Argo CD Application cert-manager

crds.enabled: true ships the Certificate, Issuer, ClusterIssuer, CertificateRequest, Order, and Challenge CRDs that the per-app Ingresses depend on. The ClusterIssuer carries SkipDryRunOnMissingResource=true so a fresh-cluster platform sync doesn't abort before the CRDs exist.

ClusterIssuer

One issuer, letsencrypt-prod, against the production ACME endpoint (acme-v02.api.letsencrypt.org, account email admin@tomoda.life), with two solvers:

Solver Hosts Why
DNS-01 (Cloudflare) tomoda.life, app.tomoda.life, www.tomoda.life, api.tomoda.life Proxied prod hosts sit behind the Cloudflare edge, which can redirect an HTTP-01 request on port 80 before it reaches Traefik
HTTP-01 (Ingress class traefik) everything else (dev, Argo CD, Grafana, admin UIs, ...) Default

The DNS-01 token is the cloudflare-dns01-token Secret in cert-manager, an ExternalSecret (k8s/envs/platform/manifests/external-secrets-config.yaml) over the same GSM tomoda-cloudflare-api-token External-DNS uses.

  • Production endpoint. Certificates and Let's Encrypt rate limits are real (50 certs per registered domain per week, 5 duplicate certs per week). There is no staging issuer, so don't create and delete certs in a loop.
  • HTTP-01 depends on Traefik. cert-manager writes a temporary challenge Ingress with class: traefik; if Traefik is down, HTTP-01 hosts can't issue or renew.
  • No wildcards. Every Ingress that wants TLS lists its exact host(s).

How a workload requests a cert

A per-app Ingress declares two things:

metadata:
  annotations:
    cert-manager.io/cluster-issuer: letsencrypt-prod
spec:
  tls:
    - hosts:
        - argo-app.tomoda.life
      secretName: argocd-server-tls

cert-manager's Ingress shim notices the annotation, creates a Certificate resource targeting argocd-server-tls, runs the challenge, and writes the resulting cert+key into argocd-server-tls in the same namespace. Traefik picks the Secret up automatically and serves it for that host.

Concrete examples in this repo:

  • argocd-server-tls for argo-app.tomoda.life (k8s/envs/platform/argocd-ingress/ingress.yaml)
  • grafana-tls for grafana.tomoda.life (k8s/envs/platform/monitoring/values.yaml)
  • auth-tls for auth.tomoda.life (k8s/envs/platform/oauth2-proxy/application.yaml)

Auto-renewal

cert-manager renews certs roughly 30 days before expiry (Let's Encrypt issues 90-day certs). Renewal uses the same solver as issuance. No human action is required; if a cert shows Ready: False, check the Order and Challenge resources in that namespace.

kubectl get certificate -A
kubectl describe certificate <name> -n <namespace>
kubectl get order,challenge -A

Operational notes

  • Removing a TLS Secret manually triggers re-issuance. Useful for forcing a refresh, but counts against the rate limit.
  • CRDs are installed via the chart, not separately. Uninstalling the cert-manager Application removes the CRDs, every Certificate and then every TLS Secret in the same sync. Don't do this without a plan.

For the broader TLS story (Cloudflare edge TLS, certificate inventory, what hostnames terminate where), see Security → TLS.