the demo keeps nothing, so it claims nothing

A 20Gi RWO claim takes local-path on k3s, which puts the volume on one
node and pins the pod there — the image is a multi-arch index and the
claim was the only reason it never ran on the other node.

Co-authored-by: Hanzo Dev <[email protected]>

deploy/lux-bank/bankd.yaml

@@ -78,29 +78,22 @@ ports:
 - containerPort: 8070
   name: http
   servicePort: 80
-# The volume bankd keeps its ledger on. It was the one object live2values.py
-# could not see — it reads the Deployment, and a claim a Deployment merely MOUNTS
-# is not part of it — so the values described a pod that names a volume nobody
-# declares, and the scheduler refuses such a pod outright: no node is available,
-# forever, with no image ever pulled. 20Gi RWO is what k8s/lux-k8s/bankd carried
-# before that directory was superseded.
+# The demo keeps nothing, so it claims nothing. A 20Gi RWO claim on the
+# cluster's default provisioner is local-path on k3s, which puts the volume on
+# ONE node and pins the pod there for the life of the claim — the image is a
+# proper multi-arch index and could run on either node, and the volume was the
+# only reason it never did.
 #
-# No storageClass on purpose. Naming one binds this file to one cluster's
-# provisioner; omitting it takes the cluster's default, which is the block class
-# in the cloud and local-path on k3s. Same text, both places. The spec is
-# immutable once bound, so this is a statement made at creation and never edited.
-claims:
-- name: bankd-data
-  size: 20Gi
-  accessModes:
-  - ReadWriteOnce
+# Ephemeral, not shared: a second replica would get a second emptyDir and serve
+# a different database from the first, so this stays at one replica. Duplicating
+# it across both nodes is a shared-store change (per-org Postgres through the
+# DBConnect seam base already exposes), not a replica count.
 volumes:
 - name: data
-  persistentVolumeClaim:
-    claimName: bankd-data
+  emptyDir: {}
 imagePullSecrets:
 - name: ghcr-secret
-strategy: Recreate
+strategy: RollingUpdate
 topologySpread:
   enabled: false
 lifecycle: null