Windows Agent: Not Supported

[ebonweaver]

The Windows agent 2.0.4 seems to be somewhat unreliable so far in testing. Having deployed it on 3 servers to monitor CPU and Memory, on one server the CPU query constantly stops with the message that it's not supported. Well, it IS supported quite obviously because it works some times on that same machine, and is working fine on two other machines (It's a linked template on all of them, so it's very much the same item). Is this a known stability problem that the agent is simply not reliable like this? Is there some other reason this particular item is unstable on one system? Item key is system.cpu.util[,system,avg1] for reference.

[ebonweaver]

I now have two servers, of entirely different models, who's agent claims that CPU is not supported. Other hardware of about the same type is fine. All systems in question use the same template. This is seeming like a bug, and it's disheartening that there seem to be no idea how to work around it.

Had no one run into this problem?

[tchjts1]

I don't use that counter for my Windows CPU utilization.

Instead, I use PerfCounters, which is basically the same information you would get if you went into the Windows native PerfMon function.

For instance, see my screenshot below for Processor Idle Time.

Then besides that one, I also use:
perf_counter[\Processor(_Total)\% Processor Time]
perf_counter[\Processor(_Total)\% User Time]

[ebonweaver]

Thanks for the reply, some good info there, and I can try that key to see if it works, though that still leaves the question of why the other does not in some cases.

Some questions about your use if you don't mind.

1. How did you come by the syntax of Processor(_Total) when that's not in the documentation or the registry values for the perf counters? Or is that on a version of windows other than server 2008?

2. I assume "Processor Time" in the key is "Total Time" on the graph? What is your understanding of these values? I ask because to me "total" would include "user" and therefore always be higher, but based on your graph that's not the case, and is puzzling to me. All I really care about is the total load per what you would see in the task manager for CPU utilization, but I want to be sure I have an accurate measure and not just a portion.

Thanks!

[tchjts1]

1. How did you come by the syntax of Processor(_Total) when that's not in the documentation or the registry values for the perf counters? Or is that on a version of windows other than server 2008?

I believe that one is fairly standard across any of the recent versions of Windows. The best way to get the list of PerfCounters is to pop a DOS box and run: typeperf -qx > perfcounters.txt
That will give you then entire list of counters and syntax to use. It will be large... maybe 30,000 - 50,000 lines. I like to use Notepad++ on those files.

2. I assume "Processor Time" in the key is "Total Time" on the graph? What is your understanding of these values? I ask because to me "total" would include "user" and therefore always be higher, but based on your graph that's not the case, and is puzzling to me. All I really care about is the total load per what you would see in the task manager for CPU utilization, but I want to be sure I have an accurate measure and not just a portion.

Thanks!

Don't confuse CPU load with CPU Utilization. Load is how many processes are currently engaging the CPU or waiting for the CPU to execute. Linear number.
CPU Utilization is a percentage of how busy the processor(s) are.

Yes, Processor Time is the key for Total Time. I would expect that it would include anything that uses CPU. I only include "User" because I like to know whether it is "User" or "System" if a problem crops up.

You would have to Google the counter to see maybe why there is a difference, as you have observed, but I do know that is is a very close representation of actual usage.

[ebonweaver]

Actually I think I might see why those values don't make sense to me. You are not specifying an interval in your key, which by document says it will default to 1 second average. However, your item is sampling only every 60 seconds, so this means you have 59 seconds of missing data each sample. Unless there is something else I'm missing or the doc is wrong, seems like your items need some tweaking to make them accurate.

I tried a few items, and found that the values that should work based on documentation of the counters in the registry, don't. Yours however does, so I'm even more curious where you got the (_Total) part as that is critical to having it work but doesn't appear anywhere. I also want to try to make sure that \Processor(_Total)\% Processor Time is in fact showing me the total load of all processes on the CPU, and it's not something squirrelly where it's only showing system and not user and that's why you have two different items you sample.

Thanks again!

[ebonweaver]

Ah! very handy command, thanks! You were more helpful than the MS kb articles on perf counters on that one

Yea looks like the actual syntax differs from the registry keys in some ways, most specifically due to multi proc multi core systems, hence the (total) now makes sense as a composite. This would all be some great additional info for the manual if anyone is watching

Thanks mate, I think this gets me operational. I'm going to run this key and the one I have been using on some systems to see how they compare, and call it a bug that the other value just fails some times.

[tchjts1]

This would all be some great additional info for the manual if anyone is watching

Thanks mate, I think this gets me operational. I'm going to run this key and the one I have been using on some systems to see how they compare, and call it a bug that the other value just fails some times.

Glad it works for you.

There are a few mentions of it in the Wiki...

https://www.zabbix.com/documentation/2.0/manual/config/items/perfcounters

https://www.zabbix.com/documentation/2.0/manual/config/items/itemtypes/zabbix_agent/win_keys

And quite a few mentions of it in the forums. For one:

https://www.zabbix.com/forum/showthread.php?t=12525

[ralopez58]

Hi ...

I got some problems with the windows agent too after upgrade version 1.8 to 2.0 the item host status show me not supported thanks in advance for the help

[tchjts1]

Are you talking about Zabbix agent active items not working?

If that is the problem, there is a new parameter for 2.x agents in the zabbix_agentd.conf file on the hosts...

##### Active checks related

### Option: ServerActive
#	List of comma delimited IP:port (or hostname:port) pairs of Zabbix servers for active checks.
#	If port is not specified, default port is used.
#	IPv6 addresses must be enclosed in square brackets if port for that host is specified.
#	If port is not specified, square brackets for IPv6 addresses are optional.
#	[B]If this parameter is not specified, active checks are disabled[/B].
#	Example: ServerActive=127.0.0.1:20051,zabbix.domain,[::1]:30051,::1,[12fc::1]
#
# Mandatory: no
# Default:
# ServerActive=

ServerActive=12.34.567.89   <-- Put IP or DNS name of Zabbix server here

If you make changes to this file, restart your agent.

[ralopez58]

thanks for replying

but i got problems with the item host status in the windows template I made the changes you mentioned and I'm still with the problem