Steps to reproduce:

1. Using Zabbix server/agent 5.0.6 (same problem occurs in earlier versions)
2. Agents are installed on CentOS 7.3.1611
3. Hosts are configured with template "Template OS Linux by Zabbix agent"
4. Hosts have 36 physical disks.
5. Host's disks are configured to use LVM resulting in 279 device mapper devices (/dev/dm-*)

Result:
After adding the host to zabbix server the number of applications/items/triggers grows exceptionally large.  In my case one of the hosts had over 90000 items.  Nearly all of these were no longer discovered and were set to be deleted.  They are all disks with names that are seemingly random numbers.  These disk names are anywhere between 0 and 10 (or more?) digits long but contain only numbers (i.e. NOT dm-0 or dm-254 but rather 0 or 32432322 for example)
**

Expected:
Normal physical disks should be detected but these random number disks should not be.  I'm not sure if the device mapper devices should be included or not - but if they were,  that would be acceptable.

NOTE:
I was able to workaround this by appending '|[0-9]' to the {$VFS.DEV.DEVNAME.NOT_MATCHES} macro in "Template Module Linux block devices by Zabbix agent".  Initially I added '|dm-[0-9]' which did NOT work.  This leads me to believe there is some parsing error when parsing /proc/diskstats - particularly with regard to the hyphen in the name of the device mapper devices (i.e. dm-0 or dm-245).  I think that this is instead getting the disk name from the frequently changing values that follow.

Attached is an example snapshot of the contents of /proc/diskstats from one of the machines.