Zabbix 7.0 is fast. Not only it is fast for the new performance it gained underneath, but it also feels fast even on a Raspberry Pi 4.
Me being me, the following truly is an idea brought to you by Bad Idea Panda. Please, please do not try this at home. Or actually, DO try this at home, or any lab, but under any circumstances, DO NOT try this in production. Clear? Clear. Good, let's continue.
Change every interval to one second
My home lab environment is very small for Zabbix. Only few thousand items in total, only about 10-20 new values per second. That's with Zabbix default settings.
But what would happen if I would change the default polling interval from one minute, 15 seconds, 30 seconds, whatever it is by default for different templates and items, to one second? Only one way to find out!
Here's Linux by Zabbix agent template after the change (did not make it compute /etc/passwd checksum every second, all the other items I did change)
... and here is the same for SNMP interfaces template, apart from interface type I changed all the items.
Does Zabbix server care?
No, of course it does not, as my environment is so tiny. Values processed per second did jump up quite a bit.
By default Zabbix server starts only one ICMP pinger poller, and that was too few -- yes, I did change even the SNMP generic template ICMP ping update interval to one second, which I know does not make any sense. That is exactly why I did it. After I upped the number of ICMP pollers to five, no problems.
For Zabbix internal processes, the spikes you see are from housekeeper.
Value cache seems to get some proper beating, but there's probably another reason for that -- we'll get into that just in a moment. Carry on reading.
Queues feel just fine -- the pending values are from my external test proxy, not Zabbix server itself. Why this happens, I have not checked, but this was going on before my little stunt.
... as you can see from queue details.
Let's change the Zabbix UI refresh interval to one second, too!
By default, the minimum update interval that Zabbix lets you to choose for its widgets is ten seconds. That's boringly slow. Luckily, that's easy to adjust to our ... or my liking. The refresh interval options are set in Zabbix installation directory include/classes/core/CWidget.php file, so let's inject the choice of one second there, too. This might also be the reason why the value cache hits did jump up so much earlier. I'm just guessing, no actual knowledge of that.
private static function getRefreshRates(): array {
return [
0 => _('No refresh'),
1 => _n('%1$s second', '%1$s seconds', 1),
10 => _n('%1$s second', '%1$s seconds', 10),
30 => _n('%1$s second', '%1$s seconds', 30),
60 => _n('%1$s minute', '%1$s minutes', 1),
120 => _n('%1$s minute', '%1$s minutes', 2),
600 => _n('%1$s minute', '%1$s minutes', 10),
900 => _n('%1$s minute', '%1$s minutes', 15)
];
}After that, Zabbix user interface now includes this:
Does user interface still work after that?
Yes, though Zabbix starts to have its own little rave party for my network gauge. It's still fast to use, but for example with these values gathered over SNMP, my router bits received widget goes a bit bonkers when I run internet speed test on fast.com.
Any side effects?
Well, apart from the fact that now my Zabbix is going to grow the database much, much more than before, not really. My Raspberry Pi 4 MariaDB instance seems to consume anything between 30-60% of the CPU all the time now, but I don't care. Raspberry Pi stays snappy, and everything else running on it, including Grafana and Home Assistant, still continues to run smoothly.
Required server performance is now a bit higher too compared to 10-20 new values per sec that was the status before I did the change.
Am I going to leave my Zabbix like this?
We will see. Part of me thinks that why not, I do have disk space for the database growth, and this seems to be working just fine. My sensible part thinks that I should revert the change. But not doing that right now.
