yacron2 + Kubernetes Lease leadership backend
=============================================

This example runs several yacron2 replicas that elect a single leader through a
coordination.k8s.io/v1 Lease. Only the lease holder runs the `Leader` jobs, so
you can scale for availability without double-running scheduled work -- with a
fenced, exactly-once guarantee while the apiserver is reachable (unlike the
best-effort gossip backend).

Files
-----

  yacron2tab.yaml   the yacron2 config: `cluster.backend: kubernetes` plus one
                    job of each clusterPolicy (Leader / PreferLeader / EveryNode)
  deployment.yaml   a ServiceAccount, a Role granting get/create/update on the
                    Lease, a RoleBinding, and a 3-replica Deployment

How the backend reaches the apiserver
-------------------------------------

In-cluster it uses the pod's service-account token and CA automatically. By
default it uses the official `kubernetes` Python client if that is installed in
the image, otherwise it falls back to a hand-rolled apiserver REST transport
over the core aiohttp dependency -- so no extra dependency is required, and it
works on every architecture. To require the native client install the extra
(`pip install yacron2[kubernetes]`) and set `clientLibrary: library`; to force
the hand-rolled path set `clientLibrary: http`.

Try it
------

  # 1. create the config ConfigMap from the file in this directory
  kubectl create configmap yacron2tab --from-file=yacron2tab.yaml

  # 2. apply the RBAC + Deployment
  kubectl apply -f deployment.yaml

  # 3. see who holds the lease. holderIdentity is the pod name plus a
  #    per-process token (e.g. yacron2-0#a1b2c3d4e5f6) so a duplicate identity
  #    cannot make two pods both hold it; the dashboard and GET /cluster show
  #    the readable pod name without the token.
  kubectl get lease yacron2-leader \
    -o jsonpath='{.spec.holderIdentity}{"\n"}'

  # 4. watch leadership fail over: delete the holder pod and re-check the lease
  kubectl delete pod <holder-pod-name>
  #    a survivor acquires it within ~leaseDurationSeconds (15s here)

Each pod also serves the dashboard on :8080 (port-forward to inspect the
cluster panel, which shows the backend, the current holder, and the lease
expiry).

Out-of-cluster / local testing
------------------------------

Point the backend at a kubeconfig (e.g. a local k3d/kind cluster) instead of
the in-cluster credentials:

  cluster:
    backend: kubernetes
    kubernetes:
      kubeconfig: /path/to/kubeconfig
      leaseNamespace: default
