We have a very busy server, 500 NVPS, and busy housekeeper, etc.

Randomly, but very often, our host time check which uses fuzzytime(120) triggers as a Problem, but the affected host is fine, time is exact.

This happens on direct server->agent data collection and via proxies.

We have added logs and instrumentation in the code, and think this is due to when the fuzzytime() is processed - we think the DB is getting busy and backed up, so the dbcache / sync system is delayed, maybe by minutes - then when the db code which also sets the triggers / evaluates functions, gets around to the check, the now() time is much later than when the data was collected, so the fuzzytime() function returns an error/difference, since now - lastvalue is very large.

Another possibility that we first thought it the data is from a proxy and very delayed over the Internet - the proxy collects data at 05:00 but the Zabbix server doesn't receive it until 05:05, and thus raises a trigger. But we are seeing this problem even on hosts which don't have proxies.

So we suspect a time delay between collection and trigger evaluation. We always assumed a trigger is evaluated as soon as data arrives/polled, but looking at the code this is clearly not the case - it seems to wait until the db sync code runs to actually evaluate functions and set triggers, so any delay in that process is deadly.

So, can you confirm when/how lastclock is set for direct and proxied data, and how fuzzytime() compares a time when there can be a delay in proxy data and/or in the db sync code on a busy system ?