Zabbix server generates slow "select clock,ns,value from history_uint..." queries in case of missing data for the items

It takes a lot of time (hours!) to get values with select clock,ns,value from history_uint where itemid=<itemid> and clock<=<timestamp> and clock><timestamp> order by clock desc limit 1 query from a big partitioned DB (e.g. ~40-50GB per each partition). These queries lead to 100% disk IO on the DB server.

A possible solution is to replace the slow SQL query:

select clock,ns,value from history_uint where itemid=123 and clock<=1460364250 and clock>1459759450 order by clock desc limit 1

with these two:

select max(clock) from history_uint where itemid=123 and clock<=1460364250 and clock>1459759450;
select clock,ns,value from history_uint where itemid=123 and clock=1460364197 order by ns desc limit 1;