Ceph Services (OSDs) seemingly binding to cluster_network instead of public_network

Issue

• Slow operations or failed operations on RBD volumes in Open Stack or Openshift.
• socket closed errors seen in dmesg on client nodes, the IP is an IP from the cluster_network instead of the public_network.
• The osdmap from the MON shows some OSDs with the wrong network in the public IP. See Diagnostic section for details.