Ganglia installation running and accessible from the OpenNebula Frontend. Tested with ganglia version 3.1.7.

To enable ganglia monitoring these lines must be uncommented in oned.conf:

    IM_MAD = [
          name       = "im_ganglia",
          executable = "one_im_sh",
          arguments  = "ganglia" ]

You also need to add the hosts with im_ganglia as its im driver.

To enable ganglia in VMM drivers you mush add a parameter to the driver execution that will tell the driver to use a local script to gather the information. The parameter is -p poll_ganglia. For example for kvm you should change in oned.conf this:

    VM_MAD = [
        name       = "vmm_kvm",
        executable = "one_vmm_ssh",
        arguments  = "-t 15 -r 0 kvm",
        default    = "vmm_ssh/vmm_ssh_kvm.conf",
        type       = "kvm" ]

Into this:

    VM_MAD = [
        name       = "vmm_kvm",
        executable = "one_vmm_ssh",
        arguments  = "-t 15 -r 0 kvm -l poll=poll_ganglia",
        default    = "vmm_ssh/vmm_ssh_kvm.conf",
        type       = "kvm" ]

Virtual Machine information is not gathered by ganglia so we provide a script to get that information. The same probe that gets information for xen and kvm (/var/lib/one/remotes/vmm/xen/poll or /var/lib/one/remotes/vmm/kvm/poll) will be used to get this data and push it to ganglia. You have to copy this probe to each of the nodes or put it in a path visible by all the nodes. It will be executed by oneadmin user to take that into account when you decide where to put the file. We will assume that the probe resides in the home of oneadmin user (~).

This information needs to be pushed to a metric called OPENNEBULA_VMS_INFORMATION and we will be using gmetric command to push this information to ganglia:

    $ gmetric -n OPENNEBULA_VMS_INFORMATION -t string -v `$HOME/tmp/poll --kvm`

To make it refresh automatically you should add this command to cron subsystem. Every minute is nice but is up to you the frequency of its refresh.

Both im_ganglia and vmm's poll_local have at the start some values that can be changed:

    # host and port where to get monitoring information
    GANGLIA_HOST='localhost'
    GANGLIA_PORT=8649
 
    # If this variable is set the the information will be read from that file
    # otherwise it will get information from the ganglia endpoint
    # defined previously
    #GANGLIA_FILE='data.xml'

GANGLIA_HOST and GANGLIA_PORT should point to a suitable gmond that holds the information about all the hosts or to a central gmetad. If GANGLIA_FILE is defined instead of talking to gmond/gmetad directly the data will be read from there. This is useful for debugging or if you cannot access gmond/gmetad directly from the fronted.

Take also into account that the names of the hosts in ganglia and the names of the hosts in ONE should match. You can also use IP addresses instead of hosts.

Note: XEN hypervisor needs to collect special information from the hypervisor and needs an special configuration in the host, check 'Xen Information' chapter.

Ganglia information manager connects to ganglia and gets these monitoring values:

It will also get any metric from ganglia that start by OPENNEBULA_ and will add it to the monitoring data pushed to ONE as the name of the metric minus the prefix(OPENNEBULA_). For example, if we add to ganglia the metric OPENNEBULA_DISK_FREE that data will be pushed to ONE as DISK_FREE. You can also push metrics with the same name as the standard metrics in the previous table, that way you can change the way those metrics are computed. OPENNEBULA_TOTALCPU will be used instead the internal TOTAL_CPU computation it it is defined.

XEN host information can be pushed to ganglia using both the standard XEN probe and a helper script located in share/scripts/push_ganglia. Both scripts should be copied to each of the xen hosts and added to cron. The scripts needed:

• /usr/share/one/scripts/ganglia/push_ganglia
• /var/lib/one/remotes/im/xen.d/xen.rb

Place both scripts in the same directory and use them this way to push the info to ganglia:

<xterm> $ ./xen.rb | ./push_ganglia </xterm>

To automate the information refresh add those commands into oneadmin user cron.

This script should be executed adding a parameter that tells what hypervisor it needs to query. –xen for xen hypervisor or –kvm for kvm hypervisor. Executing it this way we will get something like this:

    $ ./poll --kvm
    .: 1: Can't open ./kvmrc
    LS0tIApvbmUtMDogCiAgOnN0YXRlOiBhCiAgOm5ldHR4OiAiOTQ
    wMjQiCiAgOnVzZWRjcHU6ICIwLjIiCiAgOm5hbWU6IG9uZS0wCi
    AgOnVzZWRtZW1vcnk6IDI2MjE0NAogIDpuZXRyeDogIjQ3MDM5M
    CIKb25lLTE6IAogIDpzdGF0ZTogYQogIDpuZXR0eDogIjkzNjMy
    IgogIDp1c2VkY3B1OiAiMC4yIgogIDpuYW1lOiBvbmUtMQogIDp
    1c2VkbWVtb3J5OiAyNjIxNDQKICA6bmV0cng6ICI0NjgwODQiCm
    9uZS0yOiAKICA6c3RhdGU6IGEKICA6bmV0dHg6ICI5MjY4MCIKI
    CA6dXNlZGNwdTogIjAuMiIKICA6bmFtZTogb25lLTIKICA6dXNl
    ZG1lbW9yeTogMjYyMTQ0CiAgOm5ldHJ4OiAiNDY0MTg2Igo=

You can safely disregard the message .: 1: Can't open ./kvmrc as it is written to STDERR. It tries to read kvmrc (or xenrc) from the same path the script resides to get configuration information on how to query the hypervisor but uses default configuration otherwise. You can put kvmrc or xenrc in the same directory as poll script to get rid of that message and change the way you want to query the hypervisor. These rc files are the ones located at the vmm remotes directory.

The big string written is the information about the VMs running in that host encoded in base64. In this example decoding it says:

    --- 
    one-0: 
      :state: a
      :nettx: "94024"
      :usedcpu: "0.2"
      :name: one-0
      :usedmemory: 263.04
      :netrx: "470390"
    one-1: 
      :state: a
      :nettx: "93632"
      :usedcpu: "0.2"
      :name: one-1
      :usedmemory: 263.04
      :netrx: "468084"
    one-2: 
      :state: a
      :nettx: "92680"
      :usedcpu: "0.2"
      :name: one-2
      :usedmemory: 263.04
      :netrx: "464186"

It is encoded as this string will be a value pushed to ganglia so it does not interfere with xml. The format is yaml and can be easily parsed with the multiple libraries available for various languages if you need to do so.

The command that pushes this information to ganglia is:

    $ gmetric -n OPENNEBULA_VMS_INFORMATION -t string -v `$HOME/tmp/poll --kvm`

You can check if this information is in ganglia after executing that to be sure that there is no problem.