Sorry I can't help with your problem, but I just wanted to let you know that I'm seeing something similar with another system that's been upgraded from 9.11p3 -> 9.20.
I haven't quite gotten to the bottom of it yet, but it looks like it's pretty much the same thing you're seeing - some nodes that had been loaded from a seed IP previously (thus forcing a specific IP to be used for name resolution) are now using a hostname based on the reverse DNS of an interface IP.
My name selection criteria is still short DNS Name, short sysName, IP. I could change to use short sysName everywhere, but that might change other nodes.
It's a bit frustrating, because now several devices are not in the right node groups, as my node group filters were based on nodeName, not sysName. Looks like I'm going to have to try removing/re-adding some devices, using the seed IP.
# Lookup Management IP Address for the Node Host Name # # This parameter determines if disco should use management IP for the host name. Before 9.0, disco will lookup # all available IP addresses to find a valid host name. The default value is true. # If management IP does not have corresponding DNS name, IP address will be set to host name directly. # It is highly recommended that customers only change this parameter at initial time to make the node host name # stable. Late time change could break other configurations, such as node group settings and SNMP specific node settings, # because node name could be changed. #com.hp.ov.nms.disco.HostNameMatchManagementIP = false