Incorrect Flapping and Wildly Incorrect Values from FreeBSD Agent?

[sirtech]

I am getting incorrect values coming from Zabbix agent v2.0.2 on FreeBSD.

FreeBSD 9.0-RELEASE
32bit
512MB RAM

The agent is often responding 20.48GB of RAM:

# zabbix_get -s localhost -k "vm.memory.size[total]"
21994364928

Incidentally, when it isn't responding 20.48GB, it is responding 768MB. I can't get it to see 512MB. The swap is 512MB so it isn't adding that in. Restarting the agent had no effect on correcting this value.

I am also getting incorrect flapping on the CPU idle parameter; it keeps flapping between 99% and 1% even when I am watching 'top' tell me that the idle percentage is 0% - 1% (i.e. the CPU is 99% busy). This seems to be another instance where the value is being reported incorrectly (it is flapping between a correct value and an incorrect value).

There are probably other values that are coming though incorrectly but I chose these two because the CPU idle value is preventing a trigger from firing, and the total memory because it is an easier value to reproduce.

[sirtech]

Update

An update on this; it does seem to be two separate problems, though there might be some linkage somewhere.

I upgraded to Zabbix agent 2.0.3 on this FreeBSD machine and have the same problem. I have tested on a FreeBSD 8.0-RELEASE machine with agent 1.8.1 and vm.memory.size[total] works fine. On reading the manual, I tried using the key vm.memory.size (no mode specified) on the 2.0.3 agent to see what it did -- it returns the correct amount of physical memory as it should.

Per the manual for vm.memory.size[mode]: "mode - one of total (default), active, anon, buffers, cached, exec, file, free, inactive, pinned, shared, wired, used, pused, available, pavailable". So vm.memory.size[total] is mapping to something other than vm.memory.size for me. I don't have another FreeBSD machine to test agent 2.x on so can't tell if it is the FBSD version or the agent version or both...

I still don't know why the CPU idle reading from Zabbix 2.0.x is flapping on this box; again the problem is not exhibited on a FBSD 8.0-RELEASE with agent 1.8.1 (they both peg out during backup times).

[tchjts1]

Looks like the memory issue is a fix that they have done in 2.0.4RC1

[ZBX-5552] fixed processing of vm.memory.size[total] under FreeBSD

https://support.zabbix.com/browse/ZBX-5552

[sirtech]

Looks like the memory issue is a fix that they have done in 2.0.4RC1

[ZBX-5552] fixed processing of vm.memory.size[total] under FreeBSD

https://support.zabbix.com/browse/ZBX-5552

Yes that looks about right. At least it's been fixed upstream then.

After more digging on the flapping / missing alert for CPU idle, it turns out the issue was the that the alert was getting lost in the email alert tracking system we use (we have a config issue that is causing Zabbix to generate too many alerts at present). I still don't know why Zabbix shows the CPU idle flapping in the 'latest data' values (maybe the host is really doing something weird), but the result at the end of the day is that the CPU use (not idle) graph shows the right thing, and the trigger for low CPU idle does work.