The chart can deploy the Publisher as either a DaemonSet or a
StatefulSet. Pick via workload.type.

DaemonSet (workload.type: daemonset) — default

1
2
3
workload:
  type: daemonset
  replicas: 1   # ignored for DaemonSet
  • One Publisher pod per matching node.
  • Use nodeSelector / tolerations to constrain which nodes get
    Publishers. Otherwise it lands on every schedulable node.
  • Each pod registers with the same enrollment.commonName. In
    mode: api the chart appends the node hostname to disambiguate.
  • Good for: dedicated “publisher nodes”, host-network deployments,
    k3s single-node setups.

StatefulSet (workload.type: statefulset) — API mode only

1
2
3
workload:
  type: statefulset
  replicas: 3
  • N pods with stable identities: <release>-0, <release>-1, …
  • Each pod’s commonName becomes <configured-commonName>-<ordinal> —
    the chart appends only the StatefulSet ordinal (0, 1, 2, …), so
    with commonName: npa-publisher the replicas register as
    npa-publisher-0, npa-publisher-1, npa-publisher-2 and appear as
    distinct Publishers in the tenant.
  • Works only with enrollment.mode: api — token enrollment cannot
    scale to multiple replicas without manual token-per-pod handling.
  • Good for: HA pairs, replicated regional deployments inside one
    cluster, autoscaling by replica count.

Which should I use?

Scenario Recommendation
Single host, k3s, kind DaemonSet, replicas 1
Production cluster, HA, want explicit replica count StatefulSet, replicas 2–3
One Publisher per worker node by design DaemonSet + node selector
Token enrollment (air-gapped tenant) DaemonSet, replicas 1