The host is treated as unavailable because of one failed check.

It is possible that Zabbix will check the same failed item several times in row during the unreachable period and mark a host as unavailable. In other words, one incorrect or slow item can disable monitoring for the host.

Zabbix should put such item in the end of "queue" and use other items during the unreachable period.