For some reasons (let it be not synced time on specific host) we have numerical data coming with time stamp from the past (let it be >2 hours from now)
These values are kept in the cache till end of current hour. At the beginning of next hour trends should be synced and if we have data from the past probably recalculated.
Zabbix checks if trends where already flushed in the DB:
now = > 25 Jun 19:16

 
"select distinct itemid from trends_uint where clock>=<25 Jun 15:00> and (itemid in (12345678,12345679,....

Zabbix is also checking for the trends values if they exist

"select value_avg,num from trends_uint where itemid=234567890 and clock>=<25 Jun 15:00> and clock<=<25 Jun 16:00>"

These queries can be heavy and lead to overall performance degradation

We can keep latest calculated trends in the cache. In this case if we have latest calculated trends for 25 Jun 16:00 - there is no need to check this itemid with first query or if there are for 25 Jun 15:00 we don't need to run both

Regards, Elina