Discovered VMware datastore items become not supported then supported again multiple times per day. Datastores are shared between numerous hosts. VMware collector may be storing the performance data in shared memory with identical names. When the poller retrieves one set of data from memory it is no longer available the next time a different host attempts to retrieve the data for the its datastore with the same name.

Current Zabbix 3.3 implementation monitoring 60 vmware virtual hosts with a total of 2000 vmware item checked with intervals between 60 and 120.

StartVMwareCollectors=75
VMwareFrequency=60
VMwarePerfFrequency=60
VMwareCacheSize=256M
VMwareTimeout=60

VMware collector self check is on average 6.65% busy.
WMware cache is on average 76% free.

Current zabbix server log attached.