During cluster bootstrap, the podman-etcd agent fails to properly clean up standalone and learner Pacemaker properties. This failure occurs because the agent's code is hard-coded to expect exactly two nodes for the cleanup operation, even though the cluster temporarily presents three nodes during the bootstrap phase.

The "Learner" property is intended to temporarily restrict a joining node from accessing the etcd member list until a voting node formally adds it and clears this property. This signals to the joining node that it's now a recognized member and can proceed to access the member list. However, because the podman-etcd agent fails to clear the "Learner" property (due to the incorrect node count expectation), the joining node is never signaled that it belongs to the member list. Consequently, it remains blocked from accessing the etcd member list, preventing its full integration into the cluster.

While it should not be expected to have more than 2 nodes in a Two node with Fencing cluster, the podman-etcd agent should be allowed to clear the Standalone and Learner properties up if the node is not alone anymore, which happens when 2 *or more* members are in the list.

*Rare*. It happened only once.

    1.
    2.
    3.