[one-users] Some questions about overcommitment in OpenNebula 4.2 using KVM

[Michał Madziar]

On 2013-10-16 21:00, Gerry O'Brien wrote:
> On 16/10/2013 17:19, Michał Madziar wrote:
>> On 2013-10-16 14:42, Gerry O'Brien wrote:
>>> Hi,
>>>
>>>      Do you know if the CPU parameter has any effect on the amount of
>>> real cpu used, e.g. if I set the CPU parameter to 0.1 will this limit
>>> the actual cpu usage?
>> libvirt uses cgroups, so yes, it does.
> cgroups do not seem to be enabled with kvm. Am I correct in assuming
> this?
Check if /sys/fs/cgroup/cpu/libvirt/qemu/ exists and it's contents, if
cgroups are used cpu quotas can be checked with:
find /sys/fs/cgroup/cpu/libvirt/qemu/ -name cpu.shares -exec echo '{}'
\; -exec cat '{}' \;

>>
>>>      Also, can the monitor driver be  configured to report multiples of
>>> memory and allow memory overcommitment in the same way?
>> Just assign whatever you want to $total_memory in kvm.rb, but keep in
>> mind that if ballooned/deduplicated pages will start to differ either
>> swap will be used (which will probably kill the host) or oomkiller will
>> kick in.
>>
>> Example:
>> $total_cpu = 2000
>> $free_cpu = $total_cpu - $used_cpu
>> $total_memory = $total_memory.to_i+$free_memory.to_i
>> $free_memory = $total_memory - $used_memory.to_i
>> print_info("HYPERVISOR","kvm")
>> #...
>>
>> Dashboard charts will look strange but it seems to work.
>
>
> Hi,
>
> This didn't work for me. I'm running 4.2. The scheduler has a capacity
> test (see below). Any idea where the code for "get_requirements" lives?
I didn't touch the scheduler, just patched the reporting script.
On my system it exists in ~oneadmin/remotes/im/kvm.d/kvm.rb and is
copied by ON (idr when exactly - restart?)  to /var/tmp/one/im/kvm.d/ on
every host in cluster.
I just added above lines near the end of script - before printing,
copied it manually and about minute later changes were visible in chart.