We've run into and odd problem with packet loss to CentOS 6.3 guest VMs with the Zabbix agent running when they are hosted on a Windows Server 2012 Hyper-V host.
Most of our VMs are currently on Server 2008R2 Hyper-V hosts. We have about 40 CentOS 6.3 guests on those Hyper-V hosts. We've been monitoring those CentOS guests with Zabbix for some time now with good success. The agent version is v2.0.1 (revision 28455).
We've found that if we create a similar CentOS guest VM on a 2012 Hyper-V host with the 3.4 integration components (the most current version I could find when I last looked a few days ago) that we see between 2% and 40% packet loss to the CentOS guest if the Zabbix agent is started.
Also, stopping the agent alone will not correct the problem, it just reduces the problem down to about 2 to 5% packet loss. The problem goes away if the Zabbix agent is set not to start at boot and the VM is restarted. Essentially, as long as the Zabbix agent isn't started the packet loss problem doesn't show up, and if it is started, the only way to get the packet loss problem to stop is to restart the VM without having the agent start at boot.
I've tried compiling the 2.0.5 source and installing that agent. I've also tried running the 2.0.4 agent binaries that can be downloaded from Zabbix download section as well as a 1.8 agent from an EPEL repo. All of these agent version have exhibited the same behavior.
I don't see any port conflicts when I run a netstat from the command line. The agent listens on port 10050, and nothing else is. I also don't see any errors or indicators of a problem in the Zabbix or other logging on the CentOS guest, though I may be missing something.
Would anyone have an idea why the Zabbix agent would seemingly be causing packet loss when it's run on a guest VM hosted by a Win2012 Hyper-V server and not when it it's hosted on a Win2k8R2 host?
Most of our VMs are currently on Server 2008R2 Hyper-V hosts. We have about 40 CentOS 6.3 guests on those Hyper-V hosts. We've been monitoring those CentOS guests with Zabbix for some time now with good success. The agent version is v2.0.1 (revision 28455).
We've found that if we create a similar CentOS guest VM on a 2012 Hyper-V host with the 3.4 integration components (the most current version I could find when I last looked a few days ago) that we see between 2% and 40% packet loss to the CentOS guest if the Zabbix agent is started.
Also, stopping the agent alone will not correct the problem, it just reduces the problem down to about 2 to 5% packet loss. The problem goes away if the Zabbix agent is set not to start at boot and the VM is restarted. Essentially, as long as the Zabbix agent isn't started the packet loss problem doesn't show up, and if it is started, the only way to get the packet loss problem to stop is to restart the VM without having the agent start at boot.
I've tried compiling the 2.0.5 source and installing that agent. I've also tried running the 2.0.4 agent binaries that can be downloaded from Zabbix download section as well as a 1.8 agent from an EPEL repo. All of these agent version have exhibited the same behavior.
I don't see any port conflicts when I run a netstat from the command line. The agent listens on port 10050, and nothing else is. I also don't see any errors or indicators of a problem in the Zabbix or other logging on the CentOS guest, though I may be missing something.
Would anyone have an idea why the Zabbix agent would seemingly be causing packet loss when it's run on a guest VM hosted by a Win2012 Hyper-V server and not when it it's hosted on a Win2k8R2 host?