When discovering hosts they are checked for duplicates by iterating through all hosts for each host, this is very inefficient and takes long time, hashes should be used instead.

LLD for 100k hosts.

Before (takes unreasonably long 715 seconds):
 1410257 85.1  1.1 328816 187932 pts/0   S+   13:42  11:39 ./sbin/zabbix_server: lld worker #1 [processed 1 LLD rules, idle 8.372470 sec during 715.515829 sec]

After patch it is finished in reasonable 53 seconds:
68.1  1.1 328800 187700 pts/0   S+   13:40   0:36 ./sbin/zabbix_server: lld worker #1 [processed 1 LLD rules, idle 9.348398 sec during 53.101660 sec]