When host become unavailable, zabbix server marks some triggers (not time based) as unknown.
The problem is that a select query does not limit triggers by normal and lld-created ones, so trigger prototypes being updated too, which is bad itself and may cause other problems.
For example it caused a crash which is reported in
Another problem that internal events and escalation (if internal action for triggers does exist) being generated and alerts being send, although for users such alert looks like ones for already deleted trigger, which may mislead too.
Later, such trigger prototype values will never be cleaned out, which is reported as
ZBX-10761 to be somehow fixed.
See debug sections below.
Attached patch fixes the issue.