In a large (>2,000+ servers), heterogenous environment grown organically over 8-10 years, we've got issues with broken reverse DNS entries and multi-homed machines which greatly complicate discovery. Since zabbix is finding bad or broken reverse entries and using them as an authoritative source for the hostname of the discovered box this makes auto-discovery useless to us. We do have entirely reliably produced FQDNs as the second argument returned from system.uname(), however, which can be used to uniquely identify hosts. While its arguable that our environment is horribly broken and we should fix it (and we will, but it is a multi-month project to clean it all up), I think you'll also run into this issue with multi-homed machines if zabbix is capable of reaching more than one interface on the server. Using system.uname()'s return makes sense since the PTR records from different interfaces on a machine typically have different DNS names which would result in multiple entries for the same host on discovery, while system.uname() can uniquely identify the server no matter which interface the zabbix server connects to.

I'm working on a patch to the discovery thread in zabbix to trust the return of system.uname() over doing DNS reverse lookups.