I understand that having Multiple PROBLEM events generation enabled, inevitably causes a performance impact.

Anyhow, while reproducing the scenario described in ZBXNEXT-2430, by sending millions of log values to Zabbix, it turned out that Zabbix was not able to clear the text cache anymore. Even after recovering to normal operation (17:15) the value cache continues to fill (very) slowly.

~16:15 log items were activated
~17:15 all log values have been send
~21:20 disabling Multiple PROBLEM events generation

The observed behavior was (beside what already mentioned):

• generation of one PROBLEM event per second per trigger
• one of the (four) history syncers being busy for several minutes
• while being busy for minutes lots of the following sql statements were issued:

SELECT conditionid,
       conditiontype,
       operator,
       value
FROM   conditions
WHERE  actionid = <actionid>
ORDER  BY condition 

SELECT DISTINCT hg.groupid
FROM   hosts_groups hg,
       hosts h,
       items i,
       functions f,
       triggers t
WHERE  hg.host 

SELECT DISTINCT i.hostid
FROM   items i,
       functions f,
       triggers t
WHERE  i.itemid = f.itemid AND
       f.triggerid

SELECT templateid
FROM   triggers
WHERE  triggerid = <triggerid>

Just in case these statements are the cause for the performance difference, I wonder whether moving the related information into the configuration cache might speed up things...